Você está na página 1de 223

Thermal and chemical behavior

of glass forming batches

Thermal and chemical behavior


of glass forming batches

PROEFSCHRIFT
ter verkrijging van de graad van doctor
aan de Technische Universiteit Eindhoven,
op gezag van de Rector Magnicus, prof.dr. R.A. van Santen,
voor een commissie aangewezen door het College voor Promoties
in het openbaar te verdedigen op woensdag 25 juni 2003 om 16.00 uur

door
Oscar Silvester Verheijen
geboren te Arnhem

Dit proefschrift is goedgekeurd door de promotoren:


prof.dr.ir. R.G.C. Beerkens
en
prof.dr.rer.nat. R. Conradt

CIP-DATA

LIBRARY TECHNISCHE UNIVERSITEIT EINDHOVEN

Verheijen, Oscar S.
Thermal and chemical behavior of glass forming batches / by Oscar S. Verheijen. Eindhoven : Technische Universiteit Eindhoven, 2003.
Proefschrift. - ISBN 90-386-2555-3
NUR 913
Trefwoorden: glasfabricage ; glasovens / glassmelten / warmteoverdracht /
fysisch-chemische simulatie en modellering / reactiekinetiek
Subject headings: glass technology ; glass furnaces / glass melting / heat
transfer / physicochemical simulation and modeling / reaction kinetics
Printed by: Universiteitsdrukkerij Technische Universiteit Eindhoven.

Voor Roelie

Summary
The quality of the glass melt produced in a glass furnace is to a large extent determined by
the temperature-time trajectory of the freshly molten glass melt in the melting tank of the glass
furnace. The residence time and the temperature of the glass melt need to be sufcient for
complete dissolution of raw materials, the release of dissolved gases from the glass melt and
the homogenization of the glass melt. To understand, control and optimize glass melt quality
and to improve glass furnace design, since the early 1980s, mathematical simulation models,
describing melting and heat transfer in glass furnaces, have been used. These mathematical
simulation models show that thickness, length, reactivity and thermal properties of glass forming batches have a large impact on the temperature distribution and glass melt ow patterns in
a glass furnace. Therefore, detailed models describing the thermal and chemical behavior of
glass forming batches are required in mathematical simulation models.
The melting of glass forming batches is a complex process involving different reaction types
such as dehydration reactions, crystalline inversions, solid-state reactions between different raw
material grains, decomposition reactions, melt forming reactions and dissolution processes. The
heating of the three-phase (reacting) glass forming batch is determined by both heat transfer
from the combustion space above and the glass melt underneath the glass batch and by heat
transport in the interior of the glass batch. Because of the complexity of the melting and heating process of glass batches and due to the lack of sensors and accurate analyzing techniques
to measure and monitor the progress of glass batch melting, no quantitative description of the
simultaneous heating and reactive melting of glass forming batches is known so far.
The objective of this study is to obtain a quantitative description of the heating process of
a glass forming batches and the conversion rate of the glass forming batch into the glass melt.
The heat transfer in the interior of a glass forming batch can be described by:
(m c p,m Tm )
Hchem

= ( g c p,g vg,z Tg ) (eff Tm ) + r,eff + (1 )


q
.
t
t

(1)

The left-hand-side term in the energy equation describes the local accumulation of heat. The
rst, second and third right-hand-side terms in the energy equation indicate convective, conductive and radiative heat transfer, respectively. The fourth right-hand-side term in the energy
equation indicates the energy consumption by chemical reactions. A detailed description of the
energy equation of the glass forming batch requires values for the temperature dependent glass
forming batch properties. In this study, the temperature dependent chemical energy demand
and the heat conductivity of glass forming batches are determined. To describe the conversion
rate of a glass forming batch into a glass melt, the dissolution rate of sand grains, which is considered as the most signicant criterion for the conversion of glass forming batches, has been
studied.
vii

viii
The glass forming batch properties have been determined by the following steps:
1. The identication of the main batch reactions in a typical oat and TV-panel glass batch
during heating by phase analysis on quenched samples of heat-treated glass batch mixtures.
2. The development of both experimental and mathematical techniques to determine quantitatively the kinetics of batch reactions and the heat penetration in typical industrial glass
batches.
3. The measurement and modelling of the temperature dependent chemical energy demand,
the effective heat conductivity and reaction kinetic parameters.
For a oat glass batch, containing silica sand, soda ash and dolomite, it was demonstrated
that the thermal calcination of dolomite and limestone and the reactive calcination of soda ash
occur (almost) independently from each other. The reactive calcination of soda ash with sand
determines the onset for batch-to-melt conversion of the oat glass batch, which starts at about
1070 K. The formation of this primary formed melt phase enhances the dissolution of sand,
(at least part of) MgO and intermediately formed binary and ternary silicates. In contrast to
the oat glass batch, the rate of the calcination reactions in TV-panel glass batches are dependent on other batch reactions such as the dissolution of intermediate formed crystalline silicates.
The chemical energy demand of a glass forming batch is mainly dependent on the energy
required for calcination reactions in the glass batch. Combination of measured kinetics of calcination reactions in a oat glass batch with the enthalpies for these calcination reactions, results
in an expression for the chemical energy required for complete calcination of the oat glass
batch as function of time, temperature and partial CO2 -pressure in the oat glass batch. To
determine the chemical energy demand of a oat glass batch, the kinetics of the thermal decomposition of dolomite and limestone and the kinetics of the reactive calcination of soda ash have
been measured.
The heat transport in the interior of a glass batch is determined by a combination of three
modes of heat transfer. To study the contribution of these three modes, viz. convection, conduction and radiation, an experimental set-up was developed to measure the temperatures in a
glass batch as function of time and position. A numerical-experimental technique was used to
derive heat conductivity data from the measured time- and position dependent temperatures.
With the numerical-experimental technique, the temperature dependent heat conductivity of
solid particle mixtures of individual glass batch components and mixtures of glass batch components were estimated. It is shown that the heat ow through a glass batch in the solid-state
regime can be regarded as a combination of a serial and parallel connection of thermal resistances.
The prediction of the effective heat conductivity of mixtures of glass batch components
based on the glass batch composition and the porosity of the glass batch, has become possible by regarding the glass batch as only a parallel connection of thermal resistances using
apparent values for the thermal heat conductivity of the glass batch components instead of the
intrinsic values for the thermal heat conductivity. Using this approach, the difference in estimated and predicted effective heat conductivity of a three-component mixture was less than

ix

0.01 W m1 K1 for a value of the heat conductivity between 0.15 en 0.40 W m 1 K1 .


The effect of the presence of cullet in a glass batch on the net effective heat conductivity
of the glass batch in the temperature range up to 1200 K has been studied with the numericalexperimental technique. The calculation of the contribution of radiative heat transport on the
net effective heat conductivity of a glass batch shows that the expected increase in effective heat
conductivity for a cullet containing glass batch is only observed (in case of a cullet particle size
between 4 and 8 mm) in case the cullet fraction exceeds 50 %.
The heating of glass forming batch oating on top of a glass melt is not only determined
by the heat penetration through the glass forming batch, but also by the heat transfer from the
combustion space above the glass forming batch to the top layer of the glass forming batch. It
is shown that the heat transfer towards the glass batch is mainly determined by radiative heat
transfer and less by free and forced convective ow of hot gasses over the glass forming batch.
In general, the degree of dissolution of sand grains during heating of glass forming batches
is considered as the most signicant criterion for glass batch conversion. The mathematical
formulation of this mainly diffusion governed process requires a description of phenomena at
micro scale, such as the wetting of single sand grains by low viscous melt phases, in combination with complex time-dependent boundary conditions. For the use of glass tank simulation
models as tool in the glass industry, it is desired to have a simple expression describing the timeand temperature dependent dissolution of sand grains in glass batches instead of using complex
dissolution models on micro scale. A prerequisite of the use of such a simple expression is that
this expression should reect the practical observed dissolution behavior of sand grains as function of e.g. particle size, heating rate and cullet fraction in the glass batch. It is a widespread
practice to use simple approximate theoretical models to describe complex processes similar to
the dissolution of sand grains.
It is shown that the best available approximate model describing the kinetics of a three
dimensional diffusion governed process, such as the dissolution of sand grains embedded in a
liquid reactive phase during heating of glass forming batches, is the Ginstling-Brounstein model
(GB-model). The applicability of the GB-model for describing the dissolution of sand grains
during the heating of glass batches was studied by comparing the results of the GB-model with
results of a more detailed rst principle numerical model. Under certain conditions, it appeared
to be possible to predict the dissolution of sand grains as function of time and temperature with
a modied GB-model. This is the case for large sand grains dissolving in a thick glass melt
layer surrounding the sand grain. For very thin layers, the prediction of the modied GB-model
appears to be rather inaccurate.
The use of the GB-model for the description of the dissolution of sand grains during heating
of glass batches has experimentally been validated by analysis of the conversion rate of sand
grains in oat glass batches. The degree of sand grain conversion as function of time and temperature has been analyzed by quantitative phase analysis using X-ray diffraction. For the oat
glass batch, the GB-model parameters have been determined as function of sand grain size,
heating rate and cullet fraction.

Contents
1

Introduction
1.1 Description of the industrial glass melting process . . . . . . . . . . .
1.2 The chemical behavior of glass forming batches . . . . . . . . . . . .
1.3 Modelling the thermal and chemical behavior of glass forming batches
1.4 Objectives and outline . . . . . . . . . . . . . . . . . . . . . . . . .
1.5 Nomenclature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.6 Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

1
1
4
6
8
10
11

Energy demand of glass forming batches


2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.2 Energy demand during heating of glass forming batches . . . . . . . . . .
2.3 Description of the kinetics of batch reactions . . . . . . . . . . . . . . . .
2.3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.3.2 Kinetics of homogeneous reactions . . . . . . . . . . . . . . . .
2.3.3 Kinetics of heterogeneous reactions . . . . . . . . . . . . . . . .
2.3.4 Heterogeneous reaction kinetics with participation of melt phases
2.3.5 Characteristics of calcination reactions . . . . . . . . . . . . . .
2.4 Experimental techniques . . . . . . . . . . . . . . . . . . . . . . . . . .
2.4.1 Thermogravimetric analysis . . . . . . . . . . . . . . . . . . . .
2.4.2 Differential thermal analysis . . . . . . . . . . . . . . . . . . . .
2.4.3 Phase analysis on partly reacted glass batches . . . . . . . . . . .
2.5 Calcination of a oat glass batch . . . . . . . . . . . . . . . . . . . . . .
2.5.1 Calcination of dolomite . . . . . . . . . . . . . . . . . . . . . .
2.5.2 Calcination of limestone . . . . . . . . . . . . . . . . . . . . . .
2.5.3 Calcination of soda ash . . . . . . . . . . . . . . . . . . . . . . .
2.5.4 Calcination of a mixture of silica sand, soda ash and limestone . .
2.5.5 Reaction mechanism of a oat glass batch . . . . . . . . . . . . .
2.5.6 Chemical energy demand of a oat glass batch . . . . . . . . . .
2.6 Concluding remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.7 Nomenclature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.8 Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

13
13
15
21
21
21
23
26
29
31
31
34
34
35
36
42
49
62
64
67
71
72
74

Dissolution of sand grains during heating of glass forming batches


3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.2 Mathematical and experimental descriptions of the sand grain dissolution process
3.3 Evaluation of the application of the Ginstling-Brounstein model . . . . . . . .
xi

.
.
.
.
.
.

77
77
80
92

xii

Contents

3.3.1
3.3.2

3.4
3.5

3.6
3.7
3.8

Derivation of the Ginstling-Brounstein model . . . . . . . . . . . . . . 92


Modelling of the dissolution of a single sand grain in a sodium silicate
melt . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
3.3.3 Applicability of the Ginstling-Brounstein model . . . . . . . . . . . . 103
Quantitative phase analysis with X-ray diffraction . . . . . . . . . . . . . . . . 109
Experimental determination of the apparent GB-model parameters . . . . . . . 116
3.5.1 Apparent GB-model parameters as function of the sand grain particle size116
3.5.2 Apparent GB-model parameters as function of the cullet fraction . . . . 120
Concluding remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122
Nomenclature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124
Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126

Heat conductivity of glass forming batches


4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.2 Phonon and photon conductivity . . . . . . . . . . . . . . . . . . . . . . . . .
4.2.1 Intrinsic phonon conduction of solid species . . . . . . . . . . . . . . .
4.2.2 Intrinsic photon conduction of solid species . . . . . . . . . . . . . . .
4.2.3 Phonon conduction of mixtures of solid species . . . . . . . . . . . . .
4.2.4 Photon conduction in mixtures of solid species . . . . . . . . . . . . .
4.3 Estimation of the heat conductivity from heat penetration experiments . . . . .
4.3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.3.2 Parameter estimation without a-priori knowledge of = f (T ) . . . . .
4.3.3 Parameter estimation with a-priori knowledge of = f (T ) . . . . . . .
4.4 Experimental set-up for measuring the heat penetration in solid particle mixtures
4.5 Experimental determination of heat conductivity in a particle bed . . . . . . . .
4.5.1 Heat conductivity of a silica sand batch . . . . . . . . . . . . . . . . .
4.5.2 Heat conductivity for multi-component mixtures . . . . . . . . . . . .
4.5.3 Heat conductivity of particle beds containing cullet . . . . . . . . . . .
4.6 Concluding remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.7 Nomenclature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.8 Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

129
129
132
132
133
135
137
139
139
141
143
156
160
161
164
166
171
172
174

Complete simulation model for the heating of glass forming batches


5.1 Energy conservation equation and temperature dependent glass batch properties
5.2 Heat transfer towards a batch blanket . . . . . . . . . . . . . . . . . . . . . . .
5.2.1 Forced convective heat transfer . . . . . . . . . . . . . . . . . . . . . .
5.2.2 Free convective heat transfer coefcient . . . . . . . . . . . . . . . . .
5.2.3 Radiative heat transfer coefcient . . . . . . . . . . . . . . . . . . . .
5.3 Nomenclature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
5.4 Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

177
177
180
180
183
184
186
188

A Calcination of a TV-panel glass batch


189
A.1 Description of the calcination mechanism of a TV-panel glass forming batch . . 189
A.2 Calcination of SrCO3 - and BaCO3 -grains . . . . . . . . . . . . . . . . . . . . 192
A.3 Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 196

Contents

xiii

B Estimation of model parameters by a least squares approach


197
B.1 Description of the least squares approach for parameter estimation . . . . . . . 197
B.2 Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 199
C Derivation of 2nd order derivative of position dependent temperature

201

Samenvatting

203

Dankwoord

207

Curriculum Vitae

209

xiv

Contents

Chapter 1
Introduction
1.1

Description of the industrial glass melting process

Nowadays, glass is mainly produced in continuously operating glass melting furnaces. In general, the glass melt ows through three segments of the furnace, i.e. the melting tank, the
working end and the feeder section. Figure 1.1 shows a schematic view of the melting tank and
the superstructure of an industrial glass furnace with a so-called throat.
Spring zone
Superstructure
Combustion room

Doghouse

Burner

Front wall

Back wall

Batch blanket
Glass melt

Throat

First circulation loop


Short cut flow
Second circulation loop

Figure 1.1: Scheme of the melting tank and the superstructure of an industrial glass furnace.

In the doghouse of the melting tank, a glass forming batch 1 is charged on top of the hot glass
melt and forms the so-called batch blanket or batch rolls (see gure 1.2) oating on top of the
glass melt. The batch blanket is mainly heated by radiative heat transport from the ames and
superstructure in the combustion chamber above the batch blanket and by conductive and ra1 The

glass forming batch is the mixture of raw material components, which forms a glass melt during heating.
Further in this thesis, a glass forming batch is denoted as glass batch.

Chapter 1. Introduction

diative2 heat transport from the hot glass melt underneath the batch blanket. During heating of
the glass batch, up to temperatures of even 1850 K, a complex process of chemical reactions
takes place transforming the solid raw material batch components into a liquid melt phase. The
freshly formed glass melt, still containing undissolved raw materials and gas bubbles, enters the
bulk glass melt underneath the batch blanket and follows the ow pattern(s) through the melting
tank towards the throat.

Figure 1.2: Batch rolls oating on top of the glass melt in an end-port red container glass furnace

The ow pattern of the glass melt in the melting tank is the result of both forced and free
convection driven ow. Forced convective ow is imposed on the glass melt by the pull rate 3
of the furnace. Due to an unequal fuel distribution over the burners above the glass melt and
the cooling effect of the relative cold batch blanket in the doghouse, temperature gradients are
present at the glass melt surface. These temperature gradients at the surface of the glass melt
lead to temperature gradients deeper in the glass melt by conductive, convective and radiative
heat transport. Free convective ow of the glass melt is caused by density gradients in the glass
melt as a consequence of the temperature gradients in the glass melt.
The combination of furnace geometry, pull rate and fuel distribution over the burners, may
in the ideal case cause two glass melt circulation loops in the melting tank. In the rst and often
largest circulation loop, the glass melt ows from the spring zone 4 in the direction of the batch
blanket and returns via the back wall (at the doghouse side) towards the throat. In the second
2 The

contribution of radiative heat transport to the overall heat transport from the hot glass melt to the batch
blanket is dependent on the emissivity of the glass that is produced.
3 Here, the pull rate of the furnace is dened as the amount of glass melt per period of time withdrawn from the
furnace for producing glass products.
4 The position at the surface of the glass melt at which the glass melt ow splits up in a forward (towards the
throat) and backward (towards the charging end) ow. In general, the spring zone is positioned close to the hot
spot of the glass melt, at which the position with the highest glass melt temperature is located.

1.1. Description of the industrial glass melting process

loop, the glass melt ows from the spring zone to the front wall and then downwards towards
the throat. A part of the glass melt enters the throat and a part follows the second circulation
loop towards the spring zone. In case this return ow of glass melt reaches the bottom of the
melting tank, a shortcut ow of glass melt from the batch blanket directly into the throat is
prevented. Now, the glass melt is forced to ow over the glass melt surface at which the highest
glass melt temperatures exist.
Next to the melting of the raw materials, also homogenization of the freshly formed nonhomogeneous melt phases is required, and ning and rening of the glass melt should take place
in the melting tank. During the ning process, bubbles which initially mainly contain CO 2 and
air, and dissolved gases, are removed from the glass melt. At elevated temperatures, deliberately added ning agents such as sodium sulfate and antimony oxide decompose, producing
SO2 and/or O2 . As an example, the thermal decomposition of chemically dissolved SO 2 in
4
oxidized glass melts is given by
1
T
SO2 (m) SO2 (g) + O2 (g) + O2 (m).
4
2

(1.1)

By the formation of the ning gases, the existing gas bubbles expand resulting in an enhanced rise of the bubbles towards the surface of the glass melt. At the surface of the glass
melt, the bubbles release their gases into the combustion space or may form a foam layer oating on top of the glass melt. During the take up of the ning gases in the gas bubbles, the initial
batch gases, which are enclosed in the bubbles, are diluted by the ning gases. The decrease in
the partial pressure of these gases in the bubbles is followed by diffusion of the dissolved gases
from the glass melt into the gas bubbles. In this way, dissolved gases are stripped from the glass
melt.
During the subsequent rening process, residual small gas bubbles are resorbed in the glass
melt due the shift of reaction 1.1 towards the left-side-side as a consequence of the temperature
decrease in later stages of the process. Via the throat of the melting tank, the glass melt enters
the working end and subsequently the feeder section(s), in which the glass melt is slowly cooled
down to the temperature required for the forming process of the nal glass products.
In order to understand, control and optimize the glass melting process, since the early
1980s, mathematical simulation models, based on computational uid dynamics, have been
used. These models describe the physical and chemical processes in industrial glass melting
furnaces [1]. The basic equations of these simulation models are the conservation laws for mass,
energy and momentum. With these equations glass melt temperatures and velocity vectors in
industrial glass furnaces are calculated. For describing typical features of the glass melting
process, additional models are added to the conservation laws such as a separate batch blanket
model. The batch blanket model describes the thermal, chemical and rheological behavior of
glass batches oating on top of a glass melt.
In industrial glass furnaces, the batch blanket has a direct impact on both the energy transport in the glass furnace and on the ow patterns of the glass melt in the melting tank. A batch
blanket can be regarded as a layer with a low heat conductivity and high heat reecting properties at the upper surface. Due to these properties, the batch blanket acts as an insulation layer
blocking direct heat input from the combustion room into the glass melt. Consequently, the
direct energy input from the combustion room in the glass melt itself is mainly limited to the
area of the glass melt, which is not covered with the batch blanket. The energy input per amount
of produced glass for given combustion conditions, is dependent on the batch blanket size and

Chapter 1. Introduction

shape.
The batch blanket has also an impact on the ow pattern of the glass melt in the melting
tank. The intensity of the rst circulation loop in the melting tank is determined by the temperature difference between the hot spot of the glass melt and the temperatures under the batch
blanket. An increase of this temperature difference results in a more intense circulation loop,
i.e. higher glass melt velocities, intensifying the bottom ow of glass melt from the back wall
towards the throat. In case the intensity of the second circulation loop is not sufcient, a short
cut ow of glass melt over the bottom from the rst circulation loop directly into the throat is
possible.
Despite the large impact of the batch blanket on the temperatures and glass melt ow in glass
furnaces, the current batch blanket models still represent a too much simplied description of
the physical and chemical processes occurring during heating glass batches. Improvement of the
glass tank simulation models requires a thorough investigation of the melting and heat transfer
processes in the glass batch layers, which is the main subject of the current study.

1.2

The chemical behavior of glass forming batches

The composition of a glass batch is dependent on the glass type to be produced. A glass
batch producing oat glass5 typically contains silica sand (SiO2 ), soda ash (Na2 CO3 ), dolomite
(MgCO3 CaCO3 ) and/or limestone (CaCO3 ), sodium sulfate (Na2 SO4 ) as ning agent and a
reducing agent such as carbon or the steel slag calumite. As mentioned by Hrma [2], the melting of glass batches is a complex process involving different reaction types such as dehydration
reactions, crystalline inversions, solid-state reactions between the different raw material grains,
decomposition reactions, melt forming reactions and dissolution processes. Below, a short description of the characteristics of these different reaction types is given.
Dehydration reactions: In the glass batch, water may be present as either molecular water and chemically bonded water (e.g. Na2 CO3 H2 O or water bonded in clays). Molecular water is added to the glass batch to lower the dusting of ne batch particles during
transport towards the glass furnace and to prevent extensive carry-over of loose batch
particles in the glass furnace. Molecular water, with a fraction of about 1.5-4 wt.%
in the glass batch, is released at temperatures below 373 K. The temperature at which
chemically bonded water is released, is dependent on the bonding strength of the water
molecule(s) to the batch constituents and may last up to temperatures of 880 K. The impact of dehydration reactions on the melting process of the glass batches is mainly the
consumption of extra energy during water evaporation.
Crystalline inversions: During melting of glass batches, structural modications of some
crystalline batch constituents occur. An example of a crystalline inversion in sand containing glass batches is the transformation of the -modication of quartz into the modication of quartz at 846 K. Crystalline inversions have no signicant impact on the
melting process of the glass batch.
Decomposition reactions: The main decomposition reactions during melting are the calcination reactions. The typical weight loss for a oat batch, without the presence of
5 Float

glass is at glass, e.g. window glass, that is produced via the so-called oat process.

1.2. The chemical behavior of glass forming batches

cullet6 , due to these calcination reactions equals about 15-20 wt.%. Alkaline (Na 2 CO3 ,
K2 CO3 ) and earth-alkaline carbonates (MgCO3 , CaCO3 , SrCO3 and BaCO3 ) either decompose due to temperature increase, i.e. thermal calcination above a certain temperature
level, or via a reaction with other batch constituents, i.e. reactive calcination. As an example, the thermal calcination of limestone is given by
T

CaCO3 (s) CaO(s) + CO2 (g).

(1.2)

The calcination temperature7 for the earth-alkaline carbonates, which is described in


chapter 2, is dependent on the bonding strength of CO 2 with the earth-alkaline oxide
and equals 687 K, 1173 K, 1513 K and 1646 K for MgCO 3 , CaCO3 , SrCO3 and BaCO3 ,
respectively.
An example of reactive calcination of a carbonate is the calcination of soda ash via a solidstate reaction with silica sand resulting in the formation of a crystalline sodium silicate
phase. The formation of sodium metasilicate (Na2 O SiO2 ) due to reactive calcination of
soda ash with silica sand is given by
Na2 CO3 (s) + SiO2 (s) Na2 O SiO2 (s) + CO2 (g).

(1.3)

Although the formation of Na2 O SiO2 is already thermodynamically favorable at temperatures below 620 K (see chapter 2), the kinetics of these solid-state reactions are low
even at temperatures of 1000 K. These low reaction kinetics are caused by slow solidstate diffusion of reacting species through the formed crystalline binary silicates towards
the interface of the reacting grains. Enhanced reactive calcination is observed at the presence of a melt phase [3] by which the transport of reactants and reactions products is
acceleration due to the higher diffusion coefcient of the formed oxides in the liquid state
compared with the solid-state.
Melt forming reactions: Melt phases are either formed by the melting of pure substances
(e.g. the melting of soda ash at 1129 K) or by eutectic melting of two solid species. An
example of eutectic melting is the melting of a mixture of Na 2 O SiO2 and Na2 O 2SiO2 at
1110 K. The formation of these so-called primary melt phases is regarded as the measurable onset for both reactive calcination of carbonates and the dissolution of solid oxides
in the glass batch such as e.g. silica (SiO2 ), zircon (ZrO2 SiO2 ) and alumina (Al2 O3 ).
Dissolution reactions: According to Conradt et al. [4], the dissolution rate of the oxides
such as silica is dependent on primary melt phase properties such as viscosity, surface
tension and composition. Next to these properties, the particle size of the oxide grains
has a large impact on the dissolution rate of the oxides in the glass melt [5]. Because the
main fraction of commercial glass batches is occupied by silica, the dissolution process
of silica is regarded as the most signicant criterion for the degree to which the melting
of a glass batch based on silica sand has advanced [6].
Concerning the melting of glass batches, two reaction paths are distinguished for soda-limesilicate glass batches [2, 4], i.e. the carbonate route and the silicate route (see also chapter 2).
6 Recycled

7 The

glass which acts as alternative raw material for glass production


temperature at which the thermodynamic equilibrium CO2 -pressure of the carbonate equals 1 bar.

Chapter 1. Introduction

The carbonate route is characterized by reactive dissolution of silica sand with a binary melt
phase of soda ash and limestone at temperatures below 1170 K. Soda ash and limestone may
form a double carbonate by a solid-state reactions according to
Na2 CO3 (s) + CaCO3 (s) Na2 Ca(CO3 )2 (s).

(1.4)

At 1090 K, the double carbonate forms a low viscous melt phase, with which silica grains may
form a sodium calcium silica melt:
Na2 Ca(CO3 )2 (l) + SiO2 (s) NCS(m)8 .

(1.5)

For the silicate route, the (eutectic) melting of crystalline sodium silicates is regarded as the
onset for batch melting. The eutectic melting of soda disilicate with silica at 1072 K is given by
Na2 O 2SiO2 (s) + SiO2 (s) NS(m)

(1.6)

The carbonate route is assumed to be predominant at rapid heating rates in the order of 200 K
min1 , whereas the silicate route is predominant at heating rates in the order of 10 K min 1 . The
presence of different reaction mechanisms as function of heating rate was clearly demonstrated
by Riedel [7], who studied the reactivity of mixtures of glass batch components with hot-stage
microscopy. The mechanism of the conversion of a typical oat and TV-panel batch as function
of temperature is discussed in more detail in chapter 2.

1.3

Modelling the thermal and chemical behavior of glass


forming batches

A glass batch is a mixture of a gas phase and solid particles. During heating of the glass batch,
the two-phase mixture transforms into a three-phase mixture also containing the melt phases,
which are formed by (reactive) melting of the glass batch. An accurate prediction of the energy
balance of a reacting glass batch requires knowledge of intrinsic properties of the solid phase,
the melt phase and the gas phase, such as e.g. heat capacity and heat conductivity. Also the
heat exchange between the different phases in the glass batch has to be known. Ungan and
Viskanta [9] assumed that the solid phase and the melt phase are, locally, in thermal equilibrium, which simplies the reacting glass batch to a two-phase mixture.
Wakao and Kaguei [8] presented different heat transfer models for gas-solid mixtures under
transient conditions. These heat transfer models, which could also be applied to glass batches
composed of a condensed phase and a gas phase, consist of the fundamental energy equations
for both phases in the two-phase mixture. The form of these idealized energy equations is dependent on the arrangement of the two phases in the two-phase mixture, the intrinsic properties
of both phases and the ow conditions of the gas phase through the porous solid phase. However, all energy equations used in these heat transfer models are approximations of the energy
equation taking into account all modes of heat transfer (conductive, convective and radiative
heat transfer) in the individual phases and the heat exchange between the two phases. As (experimental) discrimination between the temperature of both the gas phase and the condensed
phase in glass batches is very difcult or even impossible at the moment, in general a mean
8 NCS(l)

denotes a glass melt containing the oxides Na2 O (N), CaO (C) and SiO2 (S).

1.3. Modelling the thermal and chemical behavior of glass forming batches

temperature of the glass batch is obtained when measuring the temperature in the glass batch.
The overall energy balance of a glass batch during heating, with respect to a xed frame of
reference, can be described by

cp T
t

mn

= (1 p ) c cp,c Tc + p g cp,g Tg
vc
vg

(eff Tmn ) + r,eff + (1 p )


q

Hchem
.
t

(1.7)

in which

(1.8)
cp T mn =
(1 p ) c cp,c Tc + p g cp,g Tg .
t
t
The subscripts c and g denote the condensed phase and the gas phase, p is the porosity of the

glass batch, is the intrinsic density, c p is the heat capacity, T is temperature, t is time, is the
v
three-dimensional velocity, eff represents the effective heat conductivity of the glass batch with

a mean temperature Tmn , r,eff represents the effective three-dimensional radiative heat ux in
q
the glass batch and Hchem is the energy per unit volume of the glass batch required for batch
reactions. Assuming that the horizontal velocity of the condensed phase and the gas phase are
similar [9], the Lagrangian description of the energy balance of a glass batch is given by

cp T
t

Hchem
,
t
(1.9)
in which vg,z is the vertical velocity of the gas phase relative to the condensed phase. The heat
transport in the interior of the glass batch is determined by a combination of three modes of
heat transfer:
mn

q
= p g cp,g vg,z Tg (eff Tmn ) + r,eff + (1 p )

convective heat transport by the ascending gas phase (characterized by the rst righthand-side term in equation 1.9),
conductive heat transport by mutual contact between solid particles and between solid
particles and liquid phases (characterized by the second right-hand-side term in equation
1.9),
radiative heat transport through the transparent phases (characterized by the third righthand-side term in equation 1.9).
The mean and effective properties of the glass batch are dependent on the time- and temperature
dependent composition of the glass batch. Therewith, these glass batch properties are also timeand temperature dependent.
The melting of the initial solid glass batch producing a particle free glass melt proceeds via
a series of serial and parallel batch reactions. To describe the complete chemical conversion
process of a glass batch, for each reactant and (intermediate) reaction product, a single mass
conservation equation has to solved. In case accurate analyzing techniques and sensors would
be present to measure the kinetics of all individual batch reactions occurring simultaneously
during heating of a glass batch, still the complete description of the kinetics of all batch reactions
remains a very complex and time-consuming task. Therefore, rst the kinetics of the most
important batch reactions have to be determined. As mentioned in section 1.2, the dissolution
process of sand grains is regarded as the most signicant criterion for the degree to which the
melting of a glass batch based on silica sand has advanced. In chapter 3, the dissolution of sand
grains during glass batch melting is studied.

Chapter 1. Introduction

1.4

Objectives and outline

The objective of this study is to obtain a quantitative description of the heating process of a glass
batches and the conversion rate of the glass batch into the glass melt. To meet the objectives of
this study, the following activities are distinguished:
1. The identication of the main batch reactions in a typical oat and TV-panel glass batch
during heating.
2. The development of both experimental and mathematical techniques to determine quantitatively the kinetics of batch reactions and the heat penetration in typical industrial glass
batches.
3. The measurement and modelling of the temperature dependent chemical energy demand
Hchem , the effective heat conductivity eff and reaction kinetic parameters using the in this
study developed techniques.
A better quantitative understanding of these glass batch properties may lead to the improvement
of the performance of glass tank simulation models. The implementation of the glass batch
properties, which are determined during this study, in glass tank simulation models will result
in improved predictions of both the energy balance of the glass batch and the dimensions (size
and shape) of the batch blanket. These two aspects have a large impact on both the glass quality,
by inuencing the intensity of the rst circulation loop in the melting tank (see section 1.1), and
the energy consumption of the glass tank.
In this study, the experimental determination of the glass batch properties mentioned above,
are conned to a glass batch with a typical composition for the production of oat glass. Beside,
the conversion mechanism of a glass batch with a typical composition for the production of TVpanel glass is studied qualitatively. The melting experiments are performed with industrial
raw materials and at conditions (i.e. temperatures and heating rates) that are typical for the
industrial glass melting process. However, not the full range of industrial conditions is covered
due to experimental limitations.
One of the glass batch parameters that is required for solving the energy equation of the
glass batch is the source term Hchem representing the energy demand for chemical reactions
occurring during glass batch heating. In chapter 2, the chemical energy demand of the glass
batch producing oat glass is studied. According to Madivate et al. [10], the main energy
consuming batch reactions are the decomposition reactions during which batch gases such as
CO2 is released. Therefore, the major part of the activities described in chapter 2 focus on the
experimental determination of the calcination rate of the raw material carbonates present in the
oat glass batch. These calcination reactions are studied by both:
thermogravimetric analysis of (mixtures of) glass batch components, and by
identication of (intermediate formed) crystalline species in cooled samples of (mixtures
of) glass batch components, which have been imposed to a dened temperature program.
Also in this chapter, a qualitative description of the conversion of a TV-panel batch is discussed.
Because the dissolution process of silica sand is regarded as the most signicant criterion
for the degree to which the melting of a glass batch based on silica sand has advanced, the
mechanism and the rate of the sand grain dissolution process in a glass batch is studied (chapter

1.4. Objectives and outline

3). To measure the degree of sand grain dissolution in a glass batch as function of time and
temperature, a quantitative analyzing technique is required. This technique should be able to
determine the residual amount of crystalline sand in the partly molten glass batches. In this
study, the method of quantitative phase analysis with X-ray diffraction is applied as analyzing technique for measuring the crystalline silica content in partly molten glass batches. Next,
an approximate analytical model describing the sand grain dissolution during heating of glass
batches is proposed. The applicability of this approximate model is studied by comparing results of this approximate model with the results obtained with a more detailed numerical model.
Finally, the parameters required for this approximate model are determined for the oat glass
batch as function of the particle size of the sand grains and the cullet fraction in the oat glass
batch.
In chapter 4, the experimental determination of the effective heat conductivity of glass
batches is studied. To measure the heat penetration in a glass batch, an experimental set-up
was developed, with which at different positions in the glass batch the temperature as function
of time is monitored. Because the effective heat conductivity of a glass batch is temperature
dependent, the value for the effective heat conductivity of the glass batch cannot be derived
explicitly from the heat penetration experiments. To calculate the effective heat conductivity
from the measured heat penetration in a glass batch, a numerical-experimental technique is applied. With this technique, for different glass batch mixtures, the effective heat conductivity is
determined.
Finally, in chapter 5.1 the results of this study are summarized. To complete the description
of the heating process of glass batches, information of the heat transfer towards the boundaries
of glass batches is required. In section 5.2, a description is given of the heat transfer process to
the top layer of the batch blanket and the major mode of heat transfer is identied.

10

1.5

Chapter 1. Introduction

Nomenclature

Latin symbols
cp
g
Hchem
l
m
q
s
t
T
v

heat capacity
gas phase
chemical energy demand
liquid phase
dissolved in the melt phase
heat ux
solid phase
time
temperature
velocity

[J kg1 K1 ]
[J m3 ]
[W m2 ]
[s]
[K]
[m s1 ]

Greek symbols
p

porosity
heat conductivity
stoichiometric reaction coefcient
density

Subscripts
c
eff
mn
r
z

condensed phase
effective
mean
radiative
vertical direction

[-]
[W m1 K1 ]
[kg m3 ]

1.6. Bibliography

1.6

11

Bibliography

[1] R. Viskanta. Review of three-dimensional mathematical modeling of glass melting. J.


Non-Cryst. Sol., 177:347, 1994.
[2] P. Hrma. Complexities of batch melting. In: Proc. of the 1st International Conference on
Advances in the Fusion of Glass, pages 10.110.18, Alfred University, Alfred, New York,
June 14-17, 1988.
[3] C. Kr ger. Gemengereaktionen und Glasschmelze. Glastech. Ber., 25(10):307324, 1952.
o
[4] R. Conradt, P. Suwannathada, and P. Pimkhaokham. Local temperature distribution and
primary melt formation in a melting batch heap. Glastech. Ber. Glass Sci. Technol.,
67(5):103113, 1994.
[5] R.G.C. Beerkens, H.P.H. Muijsenberg, and Heijden van der T. Modelling of sand grain
dissolution in industrial glass melting tanks. Glastech. Ber. Glass Sci. Technol., 67(7):179
188, 1994.
[6] L. Bodalbhai and P. Hrma. The dissolution of silica grains in isothermally heated batches
of sodium carbonate and silica sand. Glass Technology, 27(2):7278, 1986.
[7] L. Riedel. Die Benetzung von Kalk und Quarz durch schmelzende Soda - Eine ph nomea
nologische Studie. Glastechn. Ber., 35(1):5356, 1962.
[8] N. Wakao and S. Kaguei. Topics in chemical engineering: Vol. 1 - Heat and mass transfer
in packed beds. Gordon and Breach, Science Publishers, Inc., New York-London-Paris,
1st edition, 1982.
[9] A. Ungan and R Viskanta. Melting behavior of continuously charged loose batch blankets
in glass melting furnaces. Glastech. Ber., 59(10):279291, 1986.
[10] C. Madivate, F. M ller, and W. Wilsmann. Thermochemistry of the glass melting process
u
- energy requirement in melting soda-lime-silica glasses from cullet-containing batches.
Glastech. Ber. Glass Sci. Technol., 69(6):167178, 1996.

12

Bibliography

Chapter 2
Energy demand of glass forming batches
2.1

Introduction

As mentioned in chapter 1, a glass forming batch can be regarded as a mixture of a condensed


phase, which encloses the solid batch particles and the formed melt phases, and a gas phase.
The Lagrangian description of the energy equation of the two-phase glass batch is given by

cp T
t

Hchem
,
t
(2.1)
in which cp T mn represents the mean value of the enthalpy of the glass batch, g and cp,g
are the density and heat capacity of the gas phase, Tmn is the mean temperature of the glass
batch, t is the time, p is the porosity of the glass batch, vg,z is the vertical velocity of the gas
phase relative to the condensed phase1 , eff is the effective heat conductivity of the glass batch,

q r,eff is the effective radiative heat ux through the glass batch, and Hchem is the temperature
dependent energy per unit volume of the glass batch required for batch reactions. For a complete description of the heating of glass batches, values for the parameters p , g , cp,g , vg,z , eff ,
, and H

q r,eff
chem are required.
In this chapter, the time- and temperature dependent glass batch property Hchem is determined for a glass batch producing oat glass. Madivate et al. [2] dened the net chemical
energy demand of a glass batch as the enthalpy difference of the glass batch at 298 K and the
formed melt phase with the released gases after cooling from the batch melting temperature
down to 298 K. For a cullet-free glass batch, the net chemical energy demand for commercial
container glass batches was estimated to be approximately 15 % of the total net energy required
for batch melting. The residual 85 % is used for heating of the unreacted glass batch, the formed
melt phase and the released gases. Next, Madivate et al. [2] distinguished two types of batch
reactions:
mn

q
= p g cp,g vg,z Tg (eff Tmn ) + r,eff + (1 p )

1. Gas-releasing decomposition reactions, during which H 2 O, CO2 , NOx , SOx and/or O2 is


released and during which solid oxides are formed, such as the calcination of limestone
given by CaCO3 (s) CaO(s) + CO2 (g).
2. Mixing and fusion reactions of the solid oxides forming a glass melt.
1 It

is assumed that the horizontal velocity of the gas phase is similar to the horizontal velocity of the condensed
phase [1].

13

14

Chapter 2. Energy demand of glass forming batches

The rst type of batch reactions are endothermic reactions and require approximately 40 % of
the total energy demand of a glass batch, whereas the latter type of batch reactions are exothermic reactions, which compensate about 55 % of the energy required for the endothermic batch
reactions. In practice, the batch reactions may already start at about 300 K and may range up to
about 1700 K for the dissolution of oxides. The chemical energy required for batch reactions is
temperature dependent. A detailed description of the chemical energy demand of a glass batch
containing silica sand, soda ash, limestone and dolomite is given in section 2.2.
The major part of the gas-releasing batch reactions are the calcination reactions during
which CO2 is released. Therefore, the temperature dependent net chemical energy demand
during melting of glass batches is to a large extent determined by the rate of the calcination
reactions. The net rate of consumption of chemical energy per unit volume of the glass batch
for all calcination reactions is given by
nc
Hr,i w0 c i
Hchem
i
,
=
t
Mi
t
i=1

(2.2)

in which nc is the total number of carbonates in the glass batch, Hr,i is the enthalpy required
for the calcination of carbonate i expressed in J mole1 , w0 is the initial weight fraction of
i
carbonate i in the condensed phase, c is the density of the condensed phase, Mi is the molar
mass of carbonate i, i is the degree of conversion2 of carbonate i, and t is the time. To determine
the time- and temperature dependent energy demand for the calcination reactions, the rate of
the calcination reactions expressed by i has to be determined. In section 2.3, the general
t
approach for describing the reaction rate of homogeneous and heterogeneous reactions with
and without the presence of a non-ideal liquid phases such as glass melts is discussed. Based on
the equations presented in this section, the rate of the calcination reactions which occur during
heating of a typical oat glass batch is studied in section 2.5.
The identication of the calcination reactions in the oat glass batch and the determination
of the calcination rate of these decomposition reactions are studied by
analysis of the (intermediate formed) crystalline phases which are present in cooled heattreated oat glass batch samples, and
thermogravimetric analysis.

The backgrounds and procedure for these two experimental techniques is discussed in section
2.4.
The calcination behavior of dolomite, limestone and soda ash is discussed in sections 2.5.1,
2.5.2 and 2.5.3, respectively. The calcination behavior of the complete oat glass batch containing these batch components is discussed in section 2.5.4. In section 2.5.5, the conversion
mechanism of the oat glass batch, based on literature data and on the results of the calcination
experiments described in this chapter, is described. The temperature dependent energy demand
of the oat glass batch, calculated from the measured rate of the calcination reactions and the
calcination enthalpies of these decomposition reactions, is presented in section 2.5.6. In section
2.6 concluding remarks are presented concerning the reaction mechanism, the calcination rate
and the chemical energy demand of the oat glass batch are given. Also a few comments are
given to the calcination behavior of a TV-panel glass batch.
2 The

degree of conversion of a carbonate is dened as the ratio of the weight loss with respect to the total
weight loss when the carbonate has completely been dissociated.

PSfrag replacements

H(Tnal )
H(Tr,end )

2.2. Energy demand during heating of glass forming batches

E
H3

r3
D

H2

r2,1
r2,2 r2,3

H(Tr,onset )

B
H1

H(T0 )

15

r2

r1

A
Tr,onset

T0

Tr,end

Tnal

Figure 2.1: Schematic representation of the energy demand of a glass batch as function of temperature. A
indicates the enthalpy of the glass batch at the initial glass batch temperature T0 . E indicates
the enthalpy of the formed glass melt and gas phase after heating and melting the glass batch
up to Tnal . B, C and D are intermediate enthalpy states of the reacting glass batch. Tr,onset
and Tr,end are the onset and end temperature for the batch reactions. The routes along which
the energy demand of a glass batch may run are indicated by r 1 , r2 , r2,1 or r2,2 or r2,3 and r3 .
It is remarked that the energy required for evaporation of water, which is in general present
in glass batches, is not indicated in this gure.

2.2

Energy demand during heating of glass forming batches

In this section, a general description is given of the energy demand of a glass batch during
heating. Next, the temperature dependent energy demand of a glass batch producing oat glass
is calculated based on literature data. It is shown that for an accurate description of the temperature dependent energy demand of a glass batch, the onset temperature and the kinetics of
the (mainly) energy consuming batch reactions is required.
The energy demand of a glass batch is determined by the energy required for heating the raw
materials, the formed melt phases and the released gases, and the energy required for the endothermic and exothermic batch reactions. The melting of a glass batch can be described by
H

(1 + ) batch(T0 ) 1 melt(Tnal ) + gas(Tnal ),

(2.3)

16

Chapter 2. Energy demand of glass forming batches

in which T0 is the initial temperature of the glass batch, Tnal is the temperature to which the
glass batch is heated3 , H is the energy demand of heating and melting the glass batch, and
is the mass of gas that is released during heating of a glass batch with mass (1 + ).
With respect to the energy demand during the melting process of glass batches, three temperature ranges (see gure 2.1) are distinguished:
1. The temperature range between the initial temperature of the glass batch T0 and the onset
temperature for the batch reactions Tr,onset . The reaction occurring in this temperature
range is given by
H

1
(1 + ) batch(T0 ) (1 + ) batch(Tr,onset ).

(2.4)

During heating of the glass batch in this temperature range, the enthalpy of the glass batch
runs via route r1 from A towards B in gure 2.1. The energy demand of this reaction,
expressed in J per (1 + ) kg of batch, is given by
H1 = H(Tr,onset ) H(T0 ) =

Tr,onset nb
T0

mi cp,i dT,

(2.5)

i=1

in which nb is the number of batch components, mi is the mass of batch component i


per (1 + ) kg of batch, and cp,i is the temperature dependent heat capacity of batch
component i.
2. The temperature range between the onset for batch reactions Tr,onset and Tr,end , during
which the endo- and exothermic batch reactions occur, as described by
H

2
(1 + ) batch(Tr,onset ) 1 melt(Tr,end ) + gas(Tr,end ).

(2.6)

In case no batch reactions occur in this temperature regime, the enthalpy change of the
glass batch runs via route r2 towards C. However, because of the occurrence of (mainly)
energy consuming batch reactions, the enthalpy change runs via the arbitrarily drawn
routes r2,1 , r2,2 or r2,3 from B towards D. The route that is followed is dependent on
the temperature and time dependent chemical energy consumption during batch melting.
The enthalpy change in this temperature range, which is dependent on the kinetics of the
energy consuming and producing batch reactions, expressed in J per (1 + ) kg of batch
is described by
H2 = Hr,end Hr,onset = Hr +

Tr,end nb

mi cp,i dT,

Tr,onset i=1

(2.7)

in which Hr is the chemical energy demand for the batch reactions, nb is the number of
batch components, mi is the temperature and time dependent mass of batch component
i per (1 + ) kg of batch and cp,i is the temperature dependent heat capacity of batch
component i, which is either a solid specie, a gaseous specie or a melt phase. The mixture
of all species i at temperature T and time t represents the composition of the reacting
glass batch at temperature T and time t.
3 Here

identical.

it is assumed that the nal heating temperature of the generated melt phase and the released gases is

2.2. Energy demand during heating of glass forming batches

17

3. The temperature range between Tr,end and the nal heating temperature Tnal , during
which the formed melt phases and the released gases are heated as described by
H

3
1 melt(Tr,end ) + gas(Tr,end ) 1 melt(Tnal ) + gas(Tnal ).

(2.8)

This equation describes the enthalpy change via route r3 from D towards E. The total
energy demand of this reaction is given by
H3 =

Tnal
Tr,end

mmelt c p,melt dT +

Tnal ngc

mg cp,g dT,

Tr,end g=1

(2.9)

in which ngc is the number of gaseous components, mmelt is the mass of melt phase produced during melting of (1 + ) kg of batch, c p,melt is the temperature dependent heat
capacity of the formed melt phase, mg is the temperature dependent mass of gaseous
specie g per (1 + ) kg of batch, and cp,g is the temperature dependent heat capacity of
gaseous component g. The rst right-hand-side term in equation 2.9 represents the difference in sensible heat of the melt phase at T = Tnal and T = Tr,end , whereas the second
right-hand-side term in equation 2.9 represents the difference in sensible heat of the gas
phase at T = Tnal and T = Tr,end .
According to equation 2.7, for the determination of the chemical energy demand of the glass
batch Hr , the value for H2 and information of the temperature dependent composition of the
reacting glass batch are required. According to gure 2.1, the value for H2 can be estimated
from
H2 = (H(Tnal ) H(T0 )) (H1 + H3 ) = Ht (H1 + H3 ) .
(2.10)
The value for H3 is deduced from

H3 = Hmelt + Hgas ,

(2.11)

in which is the mass of the gas phase released per unit mass of glass melt and Hmelt and
Hgas are the differences in sensible heat between Tr,end and Tnal of the formed glass melt and
the released gases, respectively. Combining equations 2.7, 2.10 and 2.11 results in an expression
for the chemical energy demand of the glass batch:
Hr = Ht H1 + Hmelt + Hgas

Tr,end nb

mi cp,i dT.

Tr,onset i=1

(2.12)

For a given onset and end temperature for batch reactions, the values for H1 and Hgas can
be calculated from thermodynamic tables [3]. The value for Hmelt can either be derived from
an empirical formula describing the temperature dependent heat capacity of melt phases [4] or
by thermodynamic models [5]. Another possibility is to determine Hmelt from the measured
sensible heat of the glass melt as function of temperature [2]. The total energy demand for
heating and melting of the glass batch Ht requires knowledge of the formation enthalpies
of the glass batch at T0 ,
the formed melt phase at Tnal , and

18

Chapter 2. Energy demand of glass forming batches

H(Tnal )
H(Tr,end )

E
H3

r3
D

H2

r2,2
Hr (Tonset )

H(Tr,onset )

H(T0 )
r2

Hr (Tend )

r2,1
B

H1

r2,4

r2,3

r1
Reaction zone

A
T0

Tr,onset

Tr,end

Tnal

Figure 2.2: Schematic representation of the energy demand of a glass batch as function of temperature.

the released gases at Tnal .


The enthalpy of both the glass batch and the released gases can be derived from thermodynamic
tables [3]. The formation enthalpy of the glass melt can be estimated from models describing
the thermodynamic behavior of glass melts [58] or it can be measured directly. Now, the only
remaining unknown parameter required for calculation of the chemical energy demand of a
glass batch by equation 2.12 is the sensible heat of the glass batch components between Tonset
and Tend .
In gure 2.2, the temperature dependent heat capacity of the unreacted glass batch in the
reaction zone is characterized by the slope of line r2,3 , whereas the heat capacity of the formed
glass melt and the released gases in the reaction zone is characterized by the slope of line r 2,2 . In
case these slopes are identical, the chemical energy demand of the glass batch is independent on
the reaction path followed in the reaction zone. This means that Hr (Tonset ) equals Hr (Tend ).
However, since these heat capacities are in general not identical, the chemical energy demand
of a glass batch depends on the temperature range during which the batch reactions occur. The
dependency of the chemical energy demand Hr on the temperature range in which the batch
reactions take place is shown by the following example. For the ease of the calculation, it
is assumed that the glass batch reacts at a distinct reaction temperature Tr and not during a
temperature range. This simplies equation 2.12 because the last right-hand-side term becomes
zero.
Madivate et al. [2] measured the total energy demand of a glass batch, with the composition
listed in table 2.1. For this glass batch, the value for the mass of gas that is released during
heating of a glass batch with mass (1 + ) equals 0.212. The total energy demand of this glass

2.2. Energy demand during heating of glass forming batches

19

Table 2.1: Glass batch composition for which the chemical energy demand is calculated.
Batch component

Concentration [wt. %]

SiO2
Na2 CO3
CaCO3
MgCO3 CaCO3

58.21
18.76
5.24
15.89

batch with respect to 298 K is given by


Ht = 514.9 + 1.818 T,

(2.13)

in which the total energy demand is expressed in kJ per kg of glass and the temperature is
expressed in K. The sensible heat of the glass melt is given by
Hmelt = 614.0 + 1.373 T.

(2.14)

Both H1 and Hgas are determined from thermodynamic tables [3]. Figure 2.3 shows the
chemical energy demand of the glass batch expressed in J per kg of glass as function of the
batch reaction temperature. The chemical energy demand at 298 K, which equals 471 kJ kg 1 ,
melt
is almost equal to the values presented by Conradt and Pimkhaokham [7] and Madivate et al. [2].
From this gure it is clearly seen that the chemical energy demand of the glass batch decreases with increasing reaction temperature. The reason for this is the larger heat capacity of
the unreacted glass batch components compared to the heat capacity of the mixture of the melt
phase and the released gases. This means that the slope of line r2,3 in gure 2.2 is steeper than
the slope of line r2,2 .
Figure 2.4 shows the total energy demand during heating and melting of the glass batch for
both a reaction temperature of 1080 K (solid line) and a reaction temperature of 1400 K (dotted
line). Line 1 indicates the total energy demand of the glass batch during heating from the initial
temperature up to 1080 K. Line 2 describes the instantaneous chemical energy consumption at
1080 K, which is followed by line 3, which describes total energy demand during heating of
the formed melt phase and the released gases. In case the reaction temperature equals 1400 K
instead of 1080 K, the energy consumption during heating of the glass batch runs via line 1
and followed by line 4. At 1400 K, the chemical energy is consumed instantaneously and the
enthalpy change runs along line 5.
Resuming, an accurate prediction of the net chemical energy demand of a glass batch as
function of time and temperature requires knowledge of the onset temperature and the kinetics
of the energy consuming and producing batch reactions. As mentioned in the previous section,
the main energy consuming batch reactions are the calcination reactions. In section 2.5, the
kinetics of the calcination reactions occurring during heating of a oat glass batch is studied.
The determination of the calcination rate of these batch reactions is based on kinetic equations
which are discussed in the next section.

20

Chapter 2. Energy demand of glass forming batches

Chemical energy demand [J kg1 ]


melt

5x10

4x10

3x10

2x10

1x10

0
400

PSfrag replacements

600

800

1000

1200

1400

Reaction temperature [K]

Figure 2.3: Chemical energy demand of the glass batch as function of the reaction temperature for the
composition given in table 2.1 and with respect to 298 K.

PSfrag replacements

Total energy demand [J kg1 ]


melt

2.5x10

2.0x10

1.5x10

Hr (1400 K)
3
Hr (1080 K)

1.0x10

5.0x10

0.0
200

400

600

800

1000

1200

1400

1600

Temperature [K]

Figure 2.4: Total energy demand of the glass batch as function of temperature for both a batch reaction
temperature of 1080 K (solid line) and 1400 K (dotted line).

2.3. Description of the kinetics of batch reactions

2.3

21

Description of the kinetics of batch reactions

In this section, the kinetics of both homogeneous and heterogeneous chemical reactions, with
and without the presence of a non-ideal liquid phase such as a glass melt is discussed. Based
on the equations presented in this section, the kinetics of calcination reactions occurring in a
oat glass batch are studied in section 2.5.

2.3.1 Introduction
During heating and melting of glass batches, different types of reactions occur such as solidstate reactions, gas-solid reactions, gas-liquid reactions and solid-liquid reactions. In general,
the overall rate of each of these heterogeneous batch reactions is dependent on the rate of three
separate process steps, viz.:
the transport of reactants to the reaction interface,
the reaction rate at the reaction interface, and
the transport of reaction products away from the reaction interface.
The overall rate of chemical reactions is governed by the process step with the lowest rate. In
general, three reaction types are distinguished, i.e. reactions that are governed by mass transfer
processes, reactions that are governed by reaction kinetics and reactions that are governed by
thermodynamic driving forces. To describe mass transfer governed processes knowledge is required about the diffusive and convective transport of reactants and reaction products towards
and from the reaction interface. In general, for a mass transfer governed process, it is assumed
that the reaction rate is fast with respect to the rate of mass transfer. This means that for mass
transfer governed processes it is assumed that at the reaction interface thermodynamic equilibrium exists between the reactants and the reaction products.
Reaction kinetic governed processes require knowledge of reaction kinetic parameters. For
heterogeneous reactions, such as the main glass batches, these reaction kinetic parameters are
described in section 2.3.3. The description of kinetics of these heterogeneous reactions is based
on the description of reaction governed homogeneous reactions such as gas phase reactions.
Although homogeneous batch reactions are not expected to contribute to a large extend to batch
melting, a description of homogenous reaction kinetics is given in section 2.3.2, because some
aspects of the description of homogeneous reaction kinetics are applied for the description of
heterogeneous reaction kinetics. Section 2.3.4 describes the effect of the presence of non-ideal
liquid phases, such as glass melts, on the kinetic description of the rate of a chemical reaction.

2.3.2 Kinetics of homogeneous reactions


The chemical conversion of reactants A and B into the reaction products C and D is given by
A A + B B

kf
kb

C C + D D,

(2.15)

in which k f and kb are the rate constants of the forward and the backward reaction, and A ,
B , C and D are the stoichiometric reaction coefcients. Assuming that reaction 2.15 is a

22

Chapter 2. Energy demand of glass forming batches

reversible and elementary reaction4 , the overall reaction rate of reaction 2.15 is given by

r = r f rb = k f aA aB kb aCC aD ,
D
A B

(2.16)

in which aA , aB , aC and aD are the activities of the species A, B, C and D. The rates of the
forward and the backward reaction are indicated by r f and rb , respectively. For pure condensed
phases, such as solids and pure liquids, the activity is dened as unity. The activity of a component i in a gas phase mixture, which is known as the fugacity f of component i, is given
by
f i = i pi ,
(2.17)
in which i is the activity (or fugacity) coefcient of component i and p i is the partial pressure
of component i in the gas phase. For ideal gases, the fugacity or activity coefcient for each
specie equals unity and the fugacity of component i equals the partial pressure of component i.
Similar to a gas phase, the activity of a component i in a liquid solution is given by
a i = i xi ,

(2.18)

in which i is the activity coefcient of component i and xi is the mole fraction of component
i in the liquid solution. For ideal liquid solutions, the activity coefcient of each specie equals
unity and the activity of component i equals the mole fraction of component i. However, for
non-ideal solutions such as glass melts, the activity coefcient deviates signicantly from unity.
For example, the activity coefcient for Na2 O in a binary melt phase with a mole fraction of
Na2 O equal to 0.25 and a mole fraction of SiO2 equal to 0.75 equals 2.4 1011 at 1173 K [5].
Usually, a chemical reaction is the result of a number of successive elementary reaction
steps. In case reaction 2.15 is the overall reaction of the two successive elementary reactions,
i.e.

and
I I

kf
kb

kf

I I,

(2.19)

C C + D D,

(2.20)

A A + B B

kb

in which I indicates an intermediate reaction product, two reaction rates, r 2.19 and r2.20 , are
obtained, viz.
r2.19 = k f ,2.19 aA aB kb,2.19 aI ,
(2.21)
I
A B
and

r2.20 = k f ,2.20 aI kb,2.20 aCC aD .


I
D

(2.22)

The overall reaction rate of the reaction steps 2.19 and 2.20 is now governed by the rate of the
rate governing reaction step, which is either r2.19 or r2.20 . Therefore, the determination of the
kinetics of chemical reactions starts with the identication of the rate governing reaction step.
At thermodynamic equilibrium, the reaction rate of a chemical reaction equals zero, because
the rate of the forward reaction equals the rate of the backward reaction. The ratio of the reaction
4A

chemical reaction which proceeds exactly as expressed by the stoichiometric reaction equation [9]. However, usually a chemical reaction is the result of a number of successive elementary reactions as will be described by
equations 2.19 and 2.20. Than the chemical conversion proceeds via (a number of) intermediate reaction products.

2.3. Description of the kinetics of batch reactions

23

rate constant of the backward and the forward reaction equals the reaction equilibrium constant
Keq . The reaction equilibrium constant for reaction 2.15 is dened by

k f ,2.15 aCC aD
rb,2.15
= 1 Keq,2.15 =
= A DB .
r f ,2.15
kb,2.15
aA a
B

(2.23)

Combining equations 2.16 and 2.23 results in a general expression for the reaction rate of a
reversible elementary chemical reaction as function of the rate constant of the forward reaction
k f , the reaction equilibrium constant Keq , the activities of nr participating reactants and the
activities of nrp participating reaction products:
r = kf

nr

i=1

ai
i

1
Keq

nrp

aj j

(2.24)

j=1

The reaction equilibrium constant Keq is obtained from the standard chemical potential 0 of
the reactants and the reaction products according to
1
ln Keq =
RT

nrp

nr

j 0j i 0
i

j=1

i=1

G0
r
,
RT

(2.25)

in which R is the gas constant, T is the absolute temperature and G 0 is the standard Gibbs free
r
energy of reaction. The values for the standard standard chemical potential for a large number
of species are tabulated in thermodynamic tables [3].
For gas phase reactions, the rate constants k f and kb are temperature dependent according
to the so-called Arrhenius equation, which is given by
Ea

k = A e R T ,

(2.26)

in which A is the pre-exponential factor and Ea is the reaction activation energy. The exponential
temperature dependency of the gas phase rate constants is based on the exponential energy
distribution over the gas molecules in a system at a specic temperature, which is known as the
Boltzmann-distribution.
Resuming, the rate of a reversible homogeneous chemical reaction is described by equation
2.24, in which the rate constant is given by an Arrhenius equation. In the next section, the
kinetics of heterogeneous reactions, which is the major reaction type occurring during melting
of glass batches, is discussed.

2.3.3 Kinetics of heterogeneous reactions


The description of heterogeneous reactions originates from the kinetic description of solid-state
reactions, which are either diffusion or reaction governed processes. Both for diffusion and reaction kinetics governed solid-state reactions, the overall rate of the reaction is dependent on the
geometry of the reacting species. Dependent on the size and the shape of the reacting species,
four different processes are generally distinguished, viz. a one-, two- or three-dimensional diffusion governed process or a reaction governed process. Therefore, for the description of the

24

Chapter 2. Energy demand of glass forming batches

kinetics of solid-state reactions, different geometrical models have been developed. In general,
the rate of a heterogeneous reaction is described by
i
= k f (i ) ,
t

(2.27)

in which i is the conversion of specie i 5 , k is the reaction rate constant, and f (i ) is the
so-called reaction mechanism function for specie i. Table 3.1 lists the main generally applied
reaction types and their function f () derived from the reaction mechanism.

Table 2.2: Code, reaction type and reaction mechanism function (see for an overview of these reaction
mechanism functions [10]).
Code

Reaction type
th

f ()
(1 )nro

Fn

nro order reaction

D1

One-dimensional diffusion

1
2

D2

Two-dimensional diffusion

1
ln(1)

D3

Three-dimensional diffusion (Janders type)

1.5(1)2/3
1(1)1/3

D4

Three-dimensional diffusion (Ginstling-Brounstein type)

1.5
(1)1/3 1

R2

Two-dimensional phase boundary reaction

2(1 ) 1/2

R3

Three-dimensional phase boundary reaction

3(1 ) 2/3

Similar to the homogeneous gas phase reaction kinetics, the reaction rate constant k is assumed to be temperature dependent according to the Arrhenius equation. In contrast to gas
phase reactions, the energy distribution amongst the, in this case, immobilized constituents of
(crystalline) species is not represented by the Boltzmann-distribution. For this reason, the applicability of the Arrhenius equation, describing the exponential temperature dependency of the
rate constants for reactions at which solids participate, has been questioned for a long time by
Galwey and Brown [11]. In solids, energy transfer is determined by either phonon or photon
transfer. Recently, Galwey and Brown [11] reported that the energy distributions of these two
energy transfer modes also show an exponential temperature dependency, which allows the use
of the Arrhenius equation for describing the exponential temperature dependency of the rate
constants for solid-state reactions.
When applying equation 2.27, the reaction activation energy E a and the pre-exponential factor A for a chemical reaction can be derived by plotting the left-hand-side term of equation 2.28
5 The

conversion i of a specie i is dened as the ratio of the actual amount of specie i with respect to the initial
amount of specie i.

2.3. Description of the kinetics of batch reactions

25

versus the reciprocal absolute temperature.


ln

Ea

ln f () = ln A
t
RT

(2.28)

The intercept and the slope of this plot provide the values for ln A and E a /R, respectively. In
general, the selection of the reaction mechanism function f () for a specic reaction is based
on the correlation coefcient describing the accuracy of t of the reaction mechanism function
with the measured conversion data. The reaction mechanism function f () providing the largest
value for the correlation coefcient is identied as the reaction mechanism function describing
the conversion mechanism for the reaction that is investigated. According to Opfermann [12],
the selection of the reaction mechanism function based on the correlation coefcient does not
always seem to be statistically well-founded. The difference in correlation coefcients for the
different reaction mechanism functions is often insignicant. It is also mentioned that the reaction mechanism based on this selection approach is likely to provide no accurate information
about the real reaction mechanism. For this reason, the reaction kinetic parameters are not expected to have a physical meaning. Therefore, the reaction activation energy determined by this
approach will be denoted throughout this chapter as an apparent reaction activation energy.
Equation 2.27 can be regarded as a kinetic equation describing the reaction rate of a heterogeneous reaction which occurs far from thermodynamic equilibrium. In order to predict the
kinetic behavior of a heterogeneous reaction close to thermodynamic equilibrium, the reaction
equilibrium has to be taken into account in equation 2.27. Pokol [13] discussed different approaches to take thermodynamic equilibrium into account. Similar to homogeneous reactions,
the reaction rate of a heterogeneous reaction can be described by an equation similar to equation 2.24. Equation 2.29 describes the rate of a heterogeneous reaction both far and close to
thermodynamic equilibrium in which Keq is the reaction equilibrium constant and Ka , given
by equation 2.30, describes the ratio of the actual activities of the reactants and the reaction
products.
nr
Ka
r = k f f () ai 1
(2.29)
i
Keq
i=1
nrp

Ka =

j=1 a j

r
i=1 ai i

(2.30)

The forward reaction of a reversible reaction is favored in case Ka < Keq , whereas the backward
reaction is favored in case Ka > Keq . For example, for the reversible calcination of limestone,
which is given by CaCO3 (s) CaO(s) + CO2 (g), equation 2.29 simplies to equation 2.31,
because the activities of solid species equal unity by denition. The partial CO 2 -pressure in
the atmosphere surrounding the limestone particle is given by p CO2 ,a , whereas the temperature dependent equilibrium partial CO2 -pressure during limestone decomposition is given by
pCO2 ,eq .
pCO2 ,a
r = k f f () 1
(2.31)
pCO2 ,eq
In case the actual partial CO2 -pressure in the atmosphere surrounding the limestone particles
exceeds the thermodynamic equilibrium partial CO2 -pressure, the carbonization reaction, i.e.
CaO(s) + CO2 (g) CaCO3 (s), is favored, whereas in case pCO2 ,a is lower than pCO2 ,eq , the
calcination reaction, CaCO3 (s) CaO(s) + CO2 (g), is favored.

26

Chapter 2. Energy demand of glass forming batches

Resuming, the kinetics of reversible heterogeneous reactions is described by equation 2.29.


Up to now, the kinetics of chemical reactions at which reactants participate with a xed stoichiometry are discussed. However, melt phases formed during heating of glass batches have no
xed composition. In the next section, the effect of melt phases on the kinetics of heterogeneous
reactions is discussed.

2.3.4 Heterogeneous reaction kinetics with participation of melt phases


The reactive decomposition of soda ash at the presence of a binary sodium silicate melt is given
by

(1 + y)

y Na2 CO3 (s) + xNa2 O (1-x)SiO2 (l)


x+y
1x
Na2 O
SiO2 (l) + y CO2 (g)
1+y
1+y

(2.32)

Similar to equation 2.29, the kinetics of the soda ash calcination can be described by
r = k f f () amelt1 (l)
with:
Keq = e
and

1+y

Ka =

G0
r
R T

Ka
,
Keq

(2.33)

(2.34)

amelt2 pCO2 (g)


amelt1

(2.35)

in which amelt1 is the activity of the reacting melt phase and amelt2 is the activity of the
formed melt phase. The calculation of the activity of the reacting and formed melt phases is not
trivial. In order to explain the effect of melt phases on the thermodynamics of batch reactions,
the approach for describing the thermodynamics of molten oxide systems, such as glass melt
systems, presented by Shakhmatkin et al. [8] is discussed.
In general, the energy state of a solid, liquid or gas phase mixture is given by the Gibbs free
energy of the mixture according to
nms

nms

i=1

i=1

G = xi 0 + R T
i

xi ln xi + GE ,

(2.36)

in which G is the molar Gibbs free energy of the mixture, n ms is the number of species in the
mixture, xi is the mole fraction of specie i in the mixture, 0 is the standard chemical potential of
i
specie i, R is the universal gas constant, T is the absolute temperature and G E is the excess Gibbs
free energy. For ideal mixtures, such as gas phase mixtures, the excess Gibbs free energy equals
zero and the Gibbs free energy of the mixture can be calculated from the tabulated values of the
standard chemical potential of the mixture components and their mole fraction in the mixture as
given by equation 2.36. However, glass melts do not behave as an ideal mixture of the individual
oxides. The Gibbs free energy of a glass melt is given by
nms

nms

nms

i=1

i=1

i=1

G = xi 0 + R T
i

xi ln xi + R T

xi ln i,

(2.37)

2.3. Description of the kinetics of batch reactions

27

in which i is the activity coefcient of oxide i in the glass melt. For species in an ideal mixture,
the activity coefcient equals unity. The activity coefcient of Na 2 O in the non-ideal binary
melt phase with a mole fraction of Na2 O equal to 0.25 and a mole fraction of SiO2 equal to 0.75
equals 2.4 1011 at 1473 K [5].
To describe the thermodynamics of glass melts, an expression for the activity coefcient of
all species in the glass melt as function of glass melt composition and temperature is required,
i.e. ln i = f (x1 , .., xi , .., xn , T ). The main problem with non-ideal solutions is that setting-up a
model describing the interaction between solution components and the determination of these
interaction parameters is a complex and laborious task. In general, thermodynamic models
describing the deviation from ideal behavior of a solution formulate expressions for the activity coefcients of each specie i in the non-ideal solution as function of the composition of
the non-ideal solution and the temperature, i.e. ln i = f (x1 , .., xi , .., xn , T ). For a glass melt,
these expressions are based on assumed structural models of the glass melt, such as the quasichemical model [5]. The parameters of these models describing the interaction between the
oxides in the glass melt are derived from phase diagrams of (subsystems of) the complete glass
melt and from measured thermodynamic properties of the glass melt such as for example partial
vapor pressures of the glass melt oxides.
Shakhmatkin et al. [8] developed a thermodynamic model, with which the activity coefcients of glass melt oxides can be predicted based on thermodynamic properties of pure
components, which are tabulated in several references such as Knacke et al. [3]. According to
Shakhmatkin et al. [8], a glass melt can be considered as an ideal solution of so-called stoichiometric compounds in the vitreous state. These stoichiometric compounds have a stoichiometry,
which is similar to the crystalline compounds existing in the phase diagrams of the subsystems of the glass melt. For example, for a binary melt phase containing the oxides SiO 2 and
Na2 O, the main important stoichiometric compounds of the melt phase are SiO 2 , Na2 O 2SiO2 ,
Na2 O SiO2 , 2Na2 O SiO2 and Na2 O (see gure 2.5). The Gibbs free energy of a glass melt is
now described by
nms

nms

i=1

sc,i=1

G = xsc,i 0 + R T
sc,i

xsc,i ln xsc,i ,

(2.38)

in which sc denotes the stoichiometric compound in the vitreous state 6 . Minimization of the
Gibbs free energy with respect to the mole fractions of the stoichiometric compounds and taking
into account the element balances of the oxides in the glass melt, results in the thermodynamic
most stable composition of the glass melt expressed in the mole fractions of the stoichiometric
compounds.
Conradt proposed a similar model [15], in which the composition of a glass melt can be approximated by an ideal mixture of the so-called nearest stoichiometric compounds 7 , which are
in thermodynamic equilibrium with the pure oxides of the glass melt. For example, for a sodium
silicate melt phase with a molar SiO2 fraction of 0.75, the glass melt mainly contains the nearest stoichiometric compounds SiO2 (in which xSiO2 =1) and Na2 O 2SiO2 (in which xSiO2 =0.67).
Now, the glass melt in thermodynamic equilibrium can be described by an equilibrium between
6 In

general, Shakhmatkin et al. use the crystalline state of the stoichiometric compounds as reference state
instead of the vitreous state.
7 Conradt uses the vitreous state of the stoichiometric compounds as reference state. Below the liquidus temperature of a glass melt, Conradt estimates the Gibbs free energy of the vitreous state.

28

Chapter 2. Energy demand of glass forming batches

Figure 2.5: Binary phase diagram of SiO2 and Na2 O [14]. The temperature is expressed in C.

the stoichiometric compounds according to


Na2 O (l) + 2 SiO (l) Na2 O 2SiO (l),
2
2

(2.39)

Na2 CO3 (s) + 2SiO (l) Na2 O 2SiO (l) + CO2 (g).
2
2

(2.40)

in which Na2 O (l), SiO (l) and Na2 O 2SiO (l) represent the stoichiometric compounds in the
2
2
vitreous state. The reactive calcination of solid soda ash with a binary melt phase in which the
mole fraction of SiO2 equals 0.75 can now be described by
Using equation 2.29, the calcination rate of soda ash is given by
2
r = k f f () xSiO

in which
Ka =

Ka
,
Keq

xNa2 O 2SiO pCO2


2
2
xSiO
2

(2.41)

(2.42)

2.3. Description of the kinetics of batch reactions

29

and xSiO and xNa2 O 2SiO are the mole fractions of the stoichiometric compounds SiO 2 and
2
2
Na2 O 2SiO2 in the reacting glass melt. Resuming, using the approach for modelling the thermochemistry of glass melts proposed by Conradt and Shakhmatkin et al., the kinetics of thermodynamic driving force governed batch reactions at which melt phases participate can be
described.
Next to the presence of oxides in a glass melt, also dissolved gases can be present. A gaseous
specie can either be physically or chemically dissolved in the glass melt [16, 17]. The physical
dissolution of gases in a glass melt concerns the settling of gas atoms or molecules in holes in
the network of the glass melt matrix. On the other hand, gases can dissolve in the glass melt via
a reaction with glass melt components. For example, CO2 may react with free oxygen present
in the glass melt according to
CO2 (g) + O2 (m) CO2 (m).
3

(2.43)

The thermodynamic model presented by Shakhmatkin et al. and Conradt, incorporate the chemically dissolved CO2 by allowing the stoichiometric compound Na2 CO3 to be present in the
glass melt. The description of reactive dissolution of Na 2 O in the binary glass melt according
to equation 2.32, requires both the thermodynamic equilibrium data of equation 2.40 and of
Na2 CO3 (s) Na2 CO (m),
3
and

(2.44)

Na2 CO (m) Na2 O (m) + CO2 (g),


3

(2.45)

in which Na2 CO (m) is the stoichiometric compound Na2 CO3 in the melt phase, which repre3
sents the chemical dissolved CO2 . Now, the calcination rate of soda ash is not only given by
equation 2.41, which describes the reactive calcination of soda ash with the melt phase, but also
by
Ka

r = k f f () xNa2 CO3 1
,
(2.46)
Keq
which describes the thermal calcination of chemically dissolved Na 2 CO3 in the glass melt. The

chemically solubility of Na2 CO3 is given by xNa2 CO3 and Ka is given by


Ka =

xNa2 O (m) pCO2 (g)


.
xNa2 CO (m)
3

(2.47)

2.3.5 Characteristics of calcination reactions


Calcination reactions, such as the decomposition of dolomite (MgCO 3 CaCO3 ) and limestone
(CaCO3 ) during melting of oat and container glass batches and the decomposition of SrCO 3
and BaCO3 during melting of TV-panel batches, are the major decomposition reactions occurring during glass batch melting. According to Paulik and Paulik [18] and Young [19], the
onset for calcination reactions is the formation of nucle from which CO2 can be released. The

calcination reaction leaves a porous layer, containing the initially formed decarbonated oxide.
Through this porous layer, the CO2 resulting from the calcination reaction can easily escape.
However, structural ordering of the formed oxides may result in a second reaction layer (see
gure 2.6) in which recrystallized oxides are present with a lower porosity. This low porosity
layer hinders the CO2 transport from the reaction interface through the oxide layer and this may

30

Chapter 2. Energy demand of glass forming batches

retard the overall calcination rate.


The rate of calcination reactions is dependent on three processes (see gure 2.6 for the
calcination of MCO3 , in which M represents an alkaline-earth atom such as Ca or Mg), viz.:
the heat transport q, which is required for the endothermic calcination reaction, from the
outer radius r0 , through the reaction layer towards the reaction interface r i ,
the calcination at the reaction interface ri , and

ri

MCO3 (s)

Surrounding

ro

Crystallized MO

PSfrag replacements

Non crystallized MO

MCO3 grain

the (diffusive) CO2 gas ux JCO2 from the reaction interface ri through the porous reaction
layer into the surrounding atmosphere.

MO(s)+CO2 (g)

MCO3 grain
q

porous layer

JCO2

Figure 2.6: Schematic representation of the different processes occurring during calcination of a MCO 3
particle, at which M represents an alkaline-earth atom such as Ca or Mg. The initial radius
of the MCO3 particle is indicated by r0 , whereas the radius of the unreacted MCO3 particle
is indicated by ri . The heat ux through the oxide layer containing MO is indicated by q,
whereas the CO2 ux from the reaction interface at r = ri through the oxide layer is indicated
by JCO2 . In general, the oxide layer is composed of two layers. The rst layer contains the
initial formed MO, whereas the second layer contains the recrystallized MO.

Because calcination reactions are accompanied with weight loss, a suitable technique for the
determination of the kinetics of the calcination reactions is thermogravimetric analysis (TGA),
which is discussed in section 2.4.1. The reaction kinetic parameters A i and Ea,i describing the
calcination of carbonate i can be derived from transient TGA with a constant heating rate
using equation 2.28, which results in:
ln

i
T

+ ln ln f (i ) = ln Ai

Ea,i
RT

(2.48)

The intercept and the slope of this plot provide the values for ln A i and Ea,i /R, respectively.
To determine the rate of a calcination reaction, the experimental conditions during TGA have

2.4. Experimental techniques

31

to be selected such that additional mass and heat transfer problems are avoided as much as
possible. Accurate and reliable determination of the kinetic parameters requires a low heating
rate, a small sample volume and a vacuum technique for a quick removal of the released gases.
A good alternative for vacuum is the use of a carrier gas which is purged over the sample with a
high ow rate for a fast removal of the released gases from the sample. However, accumulation
of gases in the small pores in the reaction zone will not be prevented.

2.4

Experimental techniques

This section describes the experimental techniques and procedures, which are used for the identication of the calcination reactions in a oat glass batch and the determination of the calcination rate of oat glass batch. Sections 2.4.1 and 2.4.2 describe the backgrounds of thermogravimetric and differential thermal analysis, whereas in section 2.4.3 the identication procedure
for analysis of (intermediate formed) crystalline phases in quenched heat-treated oat glass
batch samples is discussed.

2.4.1 Thermogravimetric analysis


During thermogravimetric analysis8 , the weight change of a single component or a mixture of
components during heating is recorded. As an example, gure 2.7 shows the measured normalized mass of a single CaCO3 particle during ramp heating with 10 K min1 and applying a 100
ml min1 N2 -ow in a TGA apparatus9 . The onset temperature for calcination reactions derived
from TGA is dened as the intersection of the tangent of the weight loss curve with the zero
percent weight loss line. The tangent of the weight loss curve is determined at the temperature
at which the weight loss rate is maximal.
In gure 2.7, the onset temperature for CaCO3 calcination equals about 953 K, although weight
loss is observed already at 823 K. Complete calcination of CaCO 3 is achieved at 1073 K.
Although TGA is the most common technique for studying decomposition reactions, the
applicability and results of TGA-experiments are inuenced by the following encountered phenomena:
The self-cooling effect:
Because decomposition reactions are in general strong endothermic reactions, cooling of
the reactants and reaction products during decomposition is possible. Due to cooling,
the decomposition kinetics decrease, which results in a wrong estimation of the kinetic
decomposition parameters determined with equation 2.48 [20].
The Topley-Smith effect:
The heating of a sample in a TGA-apparatus is caused by convective and radiative heat
transport from a hot purge gas ow. At high temperatures, the sample in the TGAapparatus is mainly heated by radiative heat transfer. Topley and Smith discovered anomalous variation of the dehydration rate with increasing partial water vapor pressure. The
8 Further

in this thesis, thermogravimetric analysis will be denoted as TGA.


TGA-experiments in this study are performed with the SDT 2960 apparatus from TA Instruments combined with the Universal Analysis software package. An empty Platinum crucible was used as reference, whereas
the amount of sample to be investigated varied between 0.5 mg up to 50 mg.
9 All

32

Chapter 2. Energy demand of glass forming batches

Normalized mass of a limestone particle [-]

1.0

0.8

0.6

0.4

0.2

0.0

PSfrag replacements

800

900

1000

1100

1200

Temperature [K]

Figure 2.7: Measured weight loss during heating of CaCO3 when ramp heated with 10 K min1 in a 100
ml min1 N2 -ow.

dehydration rate decreases with increasing partial water vapor pressure on one hand. On
the other hand, the dehydration rate increases due to enhanced heating of the sample by
improved radiative heat transfer with increasing partial water vapor pressure. The presence of the Topley-Smith effect also results in less accurate estimation of the reaction
mechanism and kinetic decomposition parameters. The Topley-Smith effect has not only
been observed with dehydration reactions, but also during calcination reactions, like the
decomposition of CaCO3 [20].
The kinetic compensation effect:
The shape of a TG-curve is not only dependent on the kinetics of the decomposition reaction, but also on the experimental conditions at which the TGA-experiment is performed.
Gallagher and Johnson [21] measured a dependency of the apparent reaction activation
energy for the calcination of CaCO3 on both sample weight and heating rate. An increase
in sample weight and heating rate resulted in a decrease of the apparent reaction activation energy. Also a decrease of the apparent reaction activation energy was observed in
case of a decrease in partial CO2 -pressure in the atmosphere surrounding the CaCO3 particle. The experimental conditions inuence the shape of the TG-curve and therewith the
apparent reaction activation energy deduced from it. In general, the steeper a TG-curve,
the higher the value for the calculated apparent reaction activation energy. Because of
the large variety in experimental conditions at which the calcination of CaCO 3 is studied,
also a large variation in apparent reaction activation energies are presented in literature

2.4. Experimental techniques

33

for the calcination of CaCO3 . However, it appeared that plotting the natural logarithm
of the pre-exponential factor versus the apparent reaction activation energy resulted in a
linear relation. This linear relation is known as the kinetic compensation effect, which is
described by
ln A = ckc1 + ckc2 Ea ,
(2.49)
in which ckc1 and ckc2 are constants.
An increase in the apparent reaction activation energy is compensated by an increase in
the value for the pre-exponential factor A. Combining equation 2.49 with the Arrhenius
equation describing the temperature dependency of the reaction rate constant, results in
1
Ea ,
(2.50)
RT
in which k is the reaction rate constant. From the slope of the plot of the natural logarithm
of the pre-exponential factor versus the apparent reaction activation energy, a value for T
is obtained. This temperature T reects a reasonable value for the average temperature of
the temperature range from which the kinetic compensation plot is made. The intercept
of the plot reects the value for the reaction rate at the average temperature.
The dependency of the apparent reaction activation energy on experimental conditions
such as sample size, heating rate and partial CO2 -pressure indicates that the reaction of
interest is governed by either heat transfer or mass transfer limitations. The reaction kinetic parameters appear to be strongly dependent on the experimental conditions at which
the thermal analysis are performed. Plotting the natural logarithm of the pre-exponential
factor versus the apparent reaction activation energy provides information about the reaction mechanism as function of experimental conditions. In case the reaction kinetic
parameters for different experimental conditions satisfy the kinetic compensation equation given by equation 2.50, it can be assumed that the reaction mechanism is independent
on the experimental conditions. The kinetic compensation effect reects the effect of experimental conditions on the reaction kinetic parameters and can be used as a check for
evaluation the reaction mechanism as function of experimental conditions. The real reaction kinetic parameters are the values for the apparent reaction activation energy and
the pre-exponential factor at experimental conditions at which both mass and heat transfer limitations are excluded as much as possible. These conditions are small amounts of
sample, slow heating rates and high purge gas ows.
ln A = ln k +

In section 2.5.2, the reaction activation energy for limestone decomposition is determined as
function of heating rate assuming a rst order reaction mechanism. It will be shown that there
is no signicant dependency of the reaction activation energy on the heating rate. Therefore,
the presence of the kinetic compensation effect during the study can be neglected. The heating
rate and the sample size are chosen such that the heat transport towards the reaction interface is
not the rate limiting step.
To evaluate if the ow rate of the carrier gas has an impact on the calcination rate of limestone, the onset temperature for calcination as function of ow rate has been evaluated. Increasing the ow rate from the standard value of 100 ml min1 did not show a signicant change in
onset temperature for calcination. Therefore it can be assumed that no gases are accumulated
in the crucible containing the sample. However, this study does not provide information about
accumulation of gases in the interior of the particle because the diffusive transport of CO 2 is
also dependent on the porosity of the reaction layer.

34

Chapter 2. Energy demand of glass forming batches

2.4.2 Differential thermal analysis


Next to TGA-experiments, differential thermal analysis (DTA) are performed. From DTAexperiments the occurrence of endo- and exothermic reactions are observed. The temperature at
which these DTA-peaks occur provide information about the mechanism of the reaction which is
investigated. The TGA- and DTA-experiments are performed at the same time in the SDT2960
apparatus from TA Instruments.

2.4.3 Phase analysis on partly reacted glass batches


Next to the use of thermal analysis, the mechanism of calcination reactions is supported by
phase analysis on quenched samples of partly reacted mixtures of glass batch components. Below, both a description is given of the melting experiments for the preparation of the samples for
phase analysis and the experimental conditions at which the phase analysis with X-ray diffraction were performed.
For the batch melting experiments, about 5 grams of a mixture of the glass batch components
is prepared by hand mixing of the batch components10 . The mixed batch sample is brought in a
conic Pt-crucible with a bottom diameter of 30 mm, a top diameter of 55 mm and a height of 55
mm. Charging the 5 grams of batch in the Pt-crucible results in a thickness of the batch sample
of about 5 mm. The Pt-crucible, containing the batch sample, is positioned at room temperature
in an electrically heated laboratory furnace (Nabertherm HT16/17). The furnace temperature
is controlled by a thermocouple, which is located about 120 mm above the Pt-crucible. The
furnace temperature is programmed with a dened heating rate up to a nal heating temperature. The temperature of the batch sample itself is monitored by an S-type (Pt/95%Pt-5%Rh)
thermocouple, from which the joint is completely immersed in the batch sample during heating.
Throughout this chapter, the indicated temperature of the batch sample is the temperature measured by this S-type thermocouple. The set-up for the batch melting experiments is depicted in
gure 2.8.
However, despite the small amount of batch used during the current study, temperature differences have been measured throughout the batch sample. Above 1100 K, the temperature
difference in radial direction between the center of the batch sample and the crucible is less
than 14 K. This temperature difference will be denoted as the uncertainty in the batch sample
temperature. After ramp heating of the batch sample up to a specied temperature, the Ptcrucible containing the heated batch sample is withdrawn from the furnace and directly cooled
by dipping the Pt-crucible in a water bath at room temperature. Direct contact of the heated
batch sample with the cooling water is prevented. The cooling rate during the quenching period
is estimated to be approximately 20 K s1 . Because of this high cooling rate, it is assumed that
only the pure molten substances, such as soda ash or non-silicate salts are capable of recrystallizing. Directly after quenching, the residual weight of the partly molten glass forming batch
sample is measured. The value of the weight loss during heating, which is mainly caused by
the release of CO2 during heating, is required for quantitative analysis of the total amount of
crystalline quartz in the thermal treated glass batch sample.
10 Repeated

melting experiments and analysis of the fraction crystalline quartz for glass batches with identical
compositions showed that the differences in measured fraction crystalline quartz were less than 1 % absolute.
This indicates that the preparation of the glass forming batch samples by hand mixing provides a reproducible
homogeneity.

2.5. Calcination of a oat glass batch

TIC

35

Insulation bricks

TI

Closed control
thermocouple
S-type
thermocouple
260 mm

Heating elements
55 mm
Pt-crucible

55 mm

Batch
sample

5 mm
30 mm

280 mm

Figure 2.8: Front view of the experimental set-up for batch melting experiments.

After quenching, the batch sample is crushed with a tungsten carbide mortar resulting in
a residual particle size of the crushed batch sample below 10 m. From this crushed sample,
a tablet is pressed, which is used for qualitative phase analysis with X-ray diffraction. Identication of the phases present in a mixture of crystalline species is achieved by comparison
of the X-ray diffraction patterns from the unknown sample with an internationally recognized
database containing a large amount of reference patterns. The diffraction patterns are collected
using a Philips XPert X-ray diffractometer using Cu-K radiation. In general, the diffraction
data are collected from 2 equals 50 up to 800 in steps of 0.020 using a counting time of 1.0
s step1 .

2.5

Calcination of a oat glass batch

In this section, the calcination behavior of both single oat batch components and mixtures
of a complete oat glass batch is studied. The raw material components of the oat glass
batch are silica sand (SiO2 ), soda ash (Na2 CO3 ) and dolomite (MgCO3 CaCO3 ). Section
2.5.1 describes the calcination behavior of the MgCO3 -part of dolomite. In section 2.5.2, the
calcination behavior of limestone is discussed. The decomposition of soda ash is described in
section 2.5.3, whereas the calcination behavior of the complete oat glass batch is discussed
in section 2.5.4. The mechanism of melting the oat glass batch is discussed in section 2.5.5.
Finally, the temperature dependent chemical energy demand during heating of the oat glass
batch is described in section 2.5.6.

36

Chapter 2. Energy demand of glass forming batches

2.5.1 Calcination of dolomite


According to Ozao et al. [22], mineral dolomite contains both MgCO 3 CaCO3 and a small
amount of free MgCO3 present as magnesite11 . The thermal decomposition of mineral dolomite
is regarded as a three-step decomposition process:
1. The thermal decomposition of MgCO3 present as magnesite:
MgCO3 (s) MgO(s) + CO2 (g)

(2.51)

2. The thermal decomposition of the MgCO3 -part of the dolomite:


MgCO3 CaCO3 (s) MgO(s) + CO2 (g) + CaCO3 (s)

(2.52)

3. The thermal decomposition of the CaCO3 -part of the dolomite:


CaCO3 (s) CaO(s) + CO2 (g)

(2.53)

The general form of the Gibbs free energy of reaction is given by


Gr =

nrp

nr

j=1

i=1

i 0 j 0j + R T ln Ka = G0 + R T ln Ka,
i
r

(2.54)

in which m is the number of reactants, n is the number of reaction products, i is the stoichiometric reaction coefcient of specie i, 0 and 0 are the standard molar Gibbs free energy of
i
j
species i and j, Ka is the ratio of the actual activities of the reactants and reaction products
(see equation 2.30) and G0 is the standard Gibbs free energy of reaction. At thermodynamic
r
equilibrium, Ka is given by Keq and the Gibbs free energy of reaction equals zero. Equation
2.54 simplies to:
Gr = 0 = G0 + R T ln Keq .
(2.55)
r
Because the activities of solid species equal unity by denition, the reaction equilibrium constant Keq for the three calcination reactions described above, equals the partial equilibrium
CO2 -pressure. Now, the equilibrium partial CO2 -pressure surrounding the carbonates at thermodynamic equilibrium is given by
G0
r

pCO2 ,eq = e R T .

(2.56)

In case the equilibrium partial CO2 -pressure surrounding the (partly decomposed) dolomite,
calculated with equation 2.56, exceeds the partial CO2 -pressure in the atmosphere above the
dolomite particle, dolomite starts to decompose. Figure 2.9 shows both the standard Gibbs free
energy of reaction G0 for these three decomposition reactions and isobars for CO 2 . The calr
cination reactions occur in the case that the partial CO2 -pressure surrounding the carbonate,
characterized by the isobars, is positioned above the line describing the standard Gibbs free
energy of reaction. The decomposition temperatures for MgCO 3 , MgCO3 CaCO3 and CaCO3 ,
in a 1 bar CO2 -atmosphere equal 576 K, 687 K and 1170 K, respectively. Above this decomposition temperature the carbonates decompose completely.
11 Mineralogic

name of free MgCO3

1.0x10

37

104

Standard Gibbs free energy of reaction [J mol1 ]

g replacements

2.5. Calcination of a oat glass batch

8.0x10

6.0x10

Ca

CO

4.0x10

Ca
O(s

)+C

O2

2.0x10

103

3 (s

Carbonization is favored
3 C

Mg
C

3 (s

3 (s

100
100.7

Mg

Mg
O

O(
s)

(s)
+C

+C
O

-6.0x10

101
100.3

aC

-4.0x10

Calcination is favored

Mg
CO

0.0
-2.0x10

102

(g)

2 (g

101

)+C

aC
O

600

700

800

2 (g

102

3 (s

900

1000

1100

1200

1300

Temperature [K]

Figure 2.9: Standard Gibbs free energy of decomposition reactions of MgCO 3 , MgCO3 CaCO3 and
CaCO3 . The dashed lines are isobars for the partial CO2 -pressure in bar surrounding the
carbonates.

Thermal calcination of the MgCO3 -part of dolomite


The thermal behavior of dolomite was studied by thermogravimetric analysis. Figure 2.10
shows the degree of dolomite calcination12 as function of temperature in both a N2 - and a 1
bar CO2 -atmosphere.
It is remarked that during these TGA of dolomite no calcination peak of free MgCO 3 is
observed. However, in general free magnesite is present in dolomite as is shown in gure
2.11. In the CO2 -atmosphere, a two-step decomposition process is observed. Point A in gure 2.10 marks the measured onset temperature for CaCO3 calcination, which is almost similar
to the thermodynamic onset temperature for calcination of individual limestone. In the N 2 atmosphere, complete calcination is already observed at 1030 K, which indicates that the calcination of CaCO3 is dependent on the partial CO2 -pressure. The small temperature range in
which the calcination of limestone is completed in a CO2 -atmosphere indicates high calcination
kinetics, which will be discussed in section 2.5.2.
12 At this position the degree of dolomite calcination is dened as the ratio of the current weight loss with respect

to the weight loss after complete calcination of dolomite in MgO and CaO.

38

Chapter 2. Energy demand of glass forming batches

1.0

MgCO3 CaCO3 (s)

0.8

1
MgCO3 CaCO3 (s)

0.6

MgO(s)+CO2 (g)+CaCO3 (s)

0.4

0.2

CaCO3 (s) CaO(s)+CO2 (g)

PSfrag replacements

Degree of dolomite calcination [-]

MgO(s)+CaO(s)+2CO2 (g)

0.0
800

900

1000

1100

1200

1300

Temperature [K]

Figure 2.10: Measured degree of dolomite calcination (MgCO 3 CaCO3 (s) MgO(s) + CaO(s) +
2 CO2 (g)) as function of temperature in both a N2 -atmosphere (1) and a 1 bar CO2 atmosphere (2). The applied heating rate equals 10 K min 1 .

Calcination of free MgCO3

PSfrag replacements

Degree of MgCO3 calcination [-]

0.9

Calcination of MgCO3
part of dolomite

0.8

0.025

0.020

0.7
0.6

0.015

0.5
0.4

0.010

0.3
0.2

0.005

Dolomite calcination rate [K1 ]

1.0

0.1
0.0
850

900

950

1000

1050

0.000
1100

Temperature [K]

Figure 2.11: Measured degree of calcination of MgCO3 in dolomite and calcination rate as function of
temperature in a CO2 -atmosphere.

2.5. Calcination of a oat glass batch

39

Although the thermodynamic calcination temperature for the MgCO 3 -part of the dolomite
in a 1 bar CO2 -atmosphere equals 687 K, the onset temperature for calcination of dolomite is
observed to be at about 800 K. The fact that this measured calcination onset temperature is far
above the thermodynamic calcination temperature and the observation of a broad temperature
range during which calcination of dolomite occurs, is indicative for a reaction kinetic or mass
transfer governed process. Although accumulation of CO 2 near the dolomite grains is not expected13 , accumulation of CO2 in the pores of the partly calcinated MgCO3 -part of dolomite
may be present. The diffusive transport of CO2 through the pores may govern the calcination
of the MgCO3 -part of dolomite.
During heating of dolomite usually a third calcination peak is observed. Figure 2.11 shows
the calcination of MgCO3 present in dolomite in a CO2 -atmosphere (The calcination of CaCO3
in dolomite is not observed in this gure, because the onset for CaCO 3 calcination in a 1 bar
CO2 atmosphere equals 1173 K). The rst broad weight loss peak is assumed to be attributed to
the calcination of the free MgCO3 in the dolomite (see equation 2.51), whereas the second calcination peak is attributed to the calcination of the MgCO 3 -part of dolomite (see equation 2.52).
For the determination of the reaction kinetic parameters for the calcination of the MgCO 3 part of dolomite, a rst-order reaction mechanism F1 is assumed. The calcination rate of the
MgCO3 -part of dolomite is given by
Ea

r = A eR T

1 MgCO3

Ka
,
Keq

(2.57)

in which MgCO3 is the degree of calcination of the MgCO3 -part of the dolomite. Because
the calcination of MgCO3 occurs far from thermodynamic equilibrium, i.e. Ka << Keq , the
calcination of the MgCO3 -part of the dolomite is given by
Ea

r =A eR T

1 MgCO3 .

(2.58)

The apparent reaction kinetic parameters A and Ea are determined from TGA of dolomite
heated in a 1 bar CO2 -atmosphere, because a pure CO2 -atmosphere facilitates the discrimination between the calcination of the MgCO3 -part of dolomite and the residual CaCO3 . For
non-isothermal calcination of dolomite, the values for the apparent activation energy and the
pre-exponential factor are derived by plotting [ln r + ln ln (1 )] versus the reciprocal absolute temperature according to
ln r + ln ln (1 ) = ln A

Ea
,
RT

(2.59)

in which is the applied constant heating rate. Figure 2.12 shows the plot of
[ln r + ln ln (1 )] versus the reciprocal of the absolute temperature for heating rates of 10
K min1 and 20 K min1 . According to equation 2.59, the intercept of the plot equals the
natural logarithm of the pre-exponential factor A and the apparent activation energy can be calculated from the slope of the plot.
From gure 2.12, it can be seen that the apparent reaction activation energy is different for
both heating rates. In case that the effect of the heating rate on the values for the reaction kinetic
13 In

section 2.4.1, it was shown that the N2 gas ow during TGA did not effect the calcination kinetics of
limestone. It was concluded that this indicates that no CO2 , which is produced during calcination, accumulates
during the TGA.

40

Chapter 2. Energy demand of glass forming batches

ln r + ln ln (1 ) [ln s1 ]

-4.0

=10 K min1 :
ln A=51.6, Ea =513 kJ mol1 , r2 =0.9943

-4.4
-4.8
-5.2
-5.6

PSfrag replacements

-6.0
-6.4
0.00096

=20 K min1 :
ln A=56.8, Ea =560 kJ mol1 , r2 =0.9836

0.00097

0.00098

0.00099

0.00100

0.00101

Reciprocal temperature [K1 ]

Figure 2.12: [ln r + ln ln (1 )] versus the reciprocal absolute temperature describing the calcination
of the MgCO3 -part of dolomite. Line 1 and line 2 indicate [ln r + ln ln (1 )] in case
of a heating rate of 10 K min1 and 20 K min1 , respectively.

parameters is caused by the kinetic-compensating effect, which was discussed in section 2.4.1,
it is expected that the apparent reaction activation energy increases with decreasing heating rate.
However, gure 2.12 shows the opposite relation. It can be concluded that the non-isothermal
calcination curves do not provide sufcient information for unambiguous determination of the
kinetic parameters.
Olszak-Humienik and Mozejko [23] selected the one-dimensional diffusion type D1 as reaction mechanism for the calcination of the MgCO3 -part in dolomite. The value for the apparent
reaction activation energy determined by these authors with the D1 model (see table 3.1) equals
219 kJ mol1 , which is already much smaller than the apparent reaction activation energy determined from gure 2.12 (590 kJ mol1 ). However, describing the MgCO3 calcination measured
in this study with the D1 model, instead of the F1 model, results in a value of 254 kJ mol 1 ,
which is much closer to the value presented by Olszak-Humienik and Mozejko. This example
indicates that the values of derived reaction kinetic parameters is strongly dependent on the selected reaction type.
For technological purposes, the description of the MgCO 3 calcination with a rst order reaction type is sufcient in case this reaction type describes the calcination reaction accurately.
Figure 2.13 shows the measured and the simulated calcination of the MgCO 3 -part of dolomite
for a heating rate of 10 K min1 and the two sets of kinetic parameters (which were determined
at two different heating rates). It can be seen that the maximum difference in degree of calcination is less than 0.05 for the set of kinetic parameters determined from the measured calcination
during ramp heating with 10 K min1 . Applying the set of kinetic parameters determined from
the measured calcination during ramp heating with 20 K min 1 results in a maximum difference
in degree of calcination of 0.10.

2.5. Calcination of a oat glass batch

41

PSfrag replacements

Degree of dolomite calcination [-]

1.0

0.8

0.6

0.4

0.2

0.0
980

1000

1020

1040

1060

1080

Temperature [K]

Figure 2.13: Measured (1) versus simulated (2, 3) calcination of MgCO 3 for a heating rate of 10 K
min1 . Line 2 describes the calcination with ln A =55.7 and Ea =513 kJ mol1 , whereas line
3 the calcination describes with ln A =60.9 and Ea =560 kJ mol1 .

1220

Calcination onset temperature [K]

1200
1180
1160
1140
1120
1100
1080
1060
1040

PSfrag replacements

0.0

0.2

0.4

0.6

0.8

1.0

Partial CO2 -pressure [bar]

Figure 2.14: Measured () and calculated ( ) calcination onset temperatures of limestone as function of
the partial CO2 -pressure in the carrier gas during TGA.

42

Chapter 2. Energy demand of glass forming batches

Conradt [24] studied the effect of dolomite, limestone and soda ash on the formation of an
interconnected melt phase during heating of glass batches containing int cullet. The formation
of an interconnected melt phase in melting glass batches is regarded as the onset of primary
melt phase formation. In the primary melt phase, the mobility of reactants and reaction products, which are formed during solid-state reactions, increases. This increased mobility enhances
the, in general, mass transfer limited solid-state batch reactions.
Conradt observed that these glass batch components affected the formation of the interconnected melt phase layer. This seems to indicate that (int) cullet is not inert to these glass
batch components. However, thermogravimetric analysis of mixtures of oat glass cullet and
dolomite, which have been performed in this study, did not show a change in the onset temperature for MgCO3 calcination. This indicates that oat cullet has no or only a minor effect on
the calcination of dolomite.
Further throughout this thesis, the calcination of MgCO 3 in a oat glass batch is regarded
as a single thermal decomposition reaction.

2.5.2 Calcination of limestone


In contrast to MgCO3 , the calcination of limestone is dependent on the partial CO 2 -pressure.
Figure 2.14 shows both the measured14 and the calculated onset temperature for calcination of
individual limestone grains as function of the partial CO 2 -pressure.
According to Gallagher et al. [25], most studies concerning the determination of the mechanism and the values for the kinetic parameters of the thermal decomposition of limestone
were, in general, accompanied with mass and/or heat transfer problems within the experimental
equipment for thermogravimetric analysis. Dependencies of the kinetics parameter values with
sample size, heating rate and furnace geometry were observed. Criado et al. [26] found that, at
experimental conditions at which mass and/or heat transfer problems were limited, the thermal
decomposition of limestone follows 1st order kinetics. They also observed an inuence of the
particle size of the limestone grains on the thermal decomposition of limestone. The values
for the apparent reaction activation energy increased with particle size and ranged from 142 kJ
mol1 (25 m< dlimestone <50m) to 163 kJ mol1 (100 m< dlimestone <160 m).
Kinetic parameters describing the thermal calcination of limestone
The reaction kinetic parameters for the calcination of limestone are determined assuming a rst
order reaction type. The limestone grains were ramp heated with a heating rate of 20 K min 1
in a 100 ml min1 N2 -ow. Figure 2.15 shows [ln r + ln ln (1 )] versus the reciprocal
absolute temperature describing the thermal calcination of limestone.
Table 2.3 shows the calculated apparent reaction activation energy and the pre-exponential factor for limestone calcination in a N2 -atmosphere at three different heating rates. It is observed
that the heating rate does not affect the derived apparent reaction activation energy, which indicates the absence of the kinetic-compensation effect.
Figure 2.16 shows both the measured and the simulated thermal calcination of limestone for
the three different heating rates. The maximum difference in temperature for the calculated and
simulated calcination level equals 14 K in the conversion range from 0 up to 0.9.
14 The

onset temperature for calcination is determined by the approach described in section 2.4.1. However, the
onset for weight loss commences already at lower temperatures, which explains the difference in measured and
calculated calcination onset temperatures

2.5. Calcination of a oat glass batch

43

ln r + ln ln (1 ) [ln s1 ]

-4

PSfrag replacements

ln A=15.9, Ea =190 kJ mol1


r2 =0.9902

-5

-6

-7

-8

-9
0.00090

0.00095

0.00100

0.00105

0.00110

Reciprocal absolute temperature [K1 ]

Figure 2.15: [ln r + ln ln (1 )] versus the reciprocal absolute temperature characterizing the calcination kinetics of limestone during ramp heating with 20 K min 1 in a 100 ml min1
N2 -ow.

The kinetic equation describing the thermal calcination of solid limestone during which solid
CaO is formed is described by
r = k f (1 CaCO3 )

pCO2 ,a
,
pCO2 ,eq

(2.60)

in which pCO2 ,a is the partial CO2 -pressure in the atmosphere surrounding the limestone grains.
At thermodynamic equilibrium, the standard Gibbs free energy of the thermal calcination reaction is given by
0
G0 = Hr0 T Sr = R T ln Keq = R T ln pCO2 ,eq ,
r

(2.61)

0
in which Hr0 is the standard reaction enthalpy and Sr is the standard reaction entropy. The
onset temperature for thermal calcination of limestone as function of the actual partial CO 2 pressure surrounding the limestone grains pCO2 ,a is given by
0

Tonset =

H r

Sr R ln (pCO2 ,a )

(2.62)

The average value for both the reaction enthalpy and entropy in the temperature range between
0
975 K and 1175 K, which are determined from thermodynamic tables [3], are H r = 175.3

44

Chapter 2. Energy demand of glass forming batches

Table 2.3: Apparent reaction activation energy and pre-exponential factor for limestone calcination.
Heating rate [K min1 ]

Ea [kJ mol1 ]

ln A [ln s1 ]

5
10
20
20
20

188.1
190.8
187.7
187.2
190.6

16.58
16.75
16.50
16.23
17.05

kJ mol 1 and Sr = 150.7 J mol 1 K1 , respectively. The thermal calcination kinetics of


limestone at a given partial CO2 -pressure is now given by
T Tonset : r = 0

(2.63)

T > Tonset : r = k f (1 CaCO3 )

pCO2 ,a
.
pCO2 ,eq

(2.64)

Reactive calcination of limestone


The general equation for limestone calcination, assuming a rst-order reaction mechanism, is
given by
aCaO pCO2 ,a
Ka
r = k f (1 CaCO3 ) aCaCO3 1
.
(2.65)
; with Ka =
Keq
aCaCO3
As mentioned in section 2.3.3, the calcination reaction is favored over the carbonisation reaction in case Ka < Keq . As can be seen from equation 2.65, at a constant partial CO 2 -pressure
and a constant CaO activity, the ratio of Ka relative to Keq increases with a decreasing limestone
activity. This indicates that in case of a limestone activity less than unity, it is expected that
the calcination of limestone retards. On the other hand, in case the activity of the formed oxide
CaO is less than unity, and the limestone activity remains constant, the value of the ratio of
Ka relative to Keq decreases, which results in an increase of the calcination rate of limestone.
Below, it is studied if either of these two cases is likely to be present in industrial glass batches
during heating.
Limestone calcination in the presence of soda ash
In a mixture containing both limestone and soda ash, the double carbonate Na 2 Ca(CO3 )2 may
be formed via a solid-state reaction [27]:
Na2 CO3 (s) + CaCO3 (s) Na2 Ca(CO3 )2

(2.66)

The double carbonate melts at 1090 K [14], whereas a liquid phase may already be formed
at 1058 K by eutectic melting of the double carbonate with an excess of soda ash. The onset
temperature for limestone calcination in a 1 bar CO2 -atmosphere as function of the limestone
activity in the liquid solution of limestone and soda ash is given by
0

Tonset =
0

H r

Sr + R ln (aCaCO3 )

(2.67)

,
0

in which in which H r is the standard reaction enthalpy and Sr is the standard reaction entropy
of the calcination of limestone.

2.5. Calcination of a oat glass batch

45

PSfrag replacements

in 1
Km
=2
0

in 1
Km

0.6

=1
0

Km

in 1

0.8

=5

Degree of thermal limestone calcination [-]

1.0

0.4

0.2

0.0
850

900

950

1000

1050

1100

Temperature [K]

Figure 2.16: Measured (solid lines) and simulated (dotted lines) thermal calcination of limestone as function of temperature for three different heating rates in a N2 - atmosphere.

Figure 2.17 shows that the calculated onset temperature for the limestone calcination, which
is 1175 K for the calcination of a pure solid limestone grain, equals 1423 K for a limestone
activity of 0.05 in the liquid solution. This indicates that the presence of soda ash in a glass
batch may retard the calcination of limestone.
Sheckler and Dinger [28] studied the reaction mechanism of glass batches containing silica
sand, soda ash and limestone qualitatively. Both TGA and DTA were performed on the separate
batch components and on the complete glass batch as function of particle sizes of the batch
components. With X-ray diffraction, the formation of intermediate crystalline phases and the
disappearance of the initial crystalline batch components was followed. The presence of the
crystalline double carbonate was observed during heating a mixture of soda ash and limestone
with particle sizes below 53 m. The calcination behavior of limestone in this binary mixture
was measured with TGA and compared with the calcination behavior of limestone in a mixture
of limestone with silica. In the mixture of soda ash and limestone, the calcination reaction
lasted up to 1153 K, whereas the calcination of limestone in the mixture with silica was already
completed at about 1063 K. The slower limestone calcination in the mixture of limestone with
soda ash can be explained by a limestone activity less than unity in the liquid double carbonate
solution.
Although the retarding effect of the double carbonate on the calcination of limestone is

46

Chapter 2. Energy demand of glass forming batches

Calcination onset temperature [K]

1450

1400

1350

1300

1250

1200

PSfrag replacements

0.0

0.1

0.2

0.3

0.4

0.5

CaCO3 activity [-]

Figure 2.17: Calculated calcination onset temperature of CaCO 3 in a liquid mixture of Na2 CO3 and
CaCO3 in case pCO2 equals 1 bar as function of the CaCO3 activity.

observed, this effect may not be present in practice, because the double carbonate is only formed
at special conditions. Kautz and Stromburg [27] identied the crystalline phases present in a
partly reacted glass batch containing silica, soda ash, limestone and dolomite which was melted
in a CO2 -atmosphere. It was observed that the double carbonate formation depends on three
conditions:
1. The particle size of both limestone and dolomite. For a mixture of soda ash with limestone
it was observed that the crystalline double carbonate was identied in case the limestone
particle size was below 100 m. Also during experiments with coarse dolomite instead
of limestone no double carbonate was identied.
2. The type of soda ash (i.e. heavy or light soda). The formation of the double carbonate
was observed to occur at about 30 K lower temperatures in case heavy soda was applied
instead of light soda. This was explained by the larger reactive surface area of heavy soda
compared to light soda due to the smaller crystals in heavy soda.
3. The water content of the glass batch. The presence of the double carbonate was observed
to be more pronounced in case limestone was humidied.
Kautz and Stromburg [27] explained these observations by describing the formation of the double carbonate as a surface reaction of soda ash with ne distributed limestone. Therefore, the
formation of the double carbonate is dependent on the contact area and the contact time between
soda ash and limestone or dolomite.
Up to now, the effect of the partial CO2 -pressure on the formation of the double carbonate

2.5. Calcination of a oat glass batch

47

has not been discussed. The formation of the double carbonate is dependent on the presence
of both soda ash and limestone. Because the presence of limestone is dependent on the partial
CO2 -pressure, it is expected that the formation of the double carbonate also depends on the
partial CO2 -pressure. The effect of the CO2 -pressure on the formation of the double carbonate
has been evaluated by three experiments, viz.:
combined TGA and DTA of soda ash heated in a 1 bar CO2 -atmosphere,
combined TGA and DTA of a mixture of soda ash and limestone heated in a N 2 -atmosphere,
and
combined TGA and DTA of a mixture of soda ash and limestone heated in a 1 bar CO 2 atmosphere.
The DTA-signals of these three experiments are shown in gure 2.18. Curve 1 shows the DTAsignal of soda ash heated in a CO2 -atmosphere. At 1129 K, an endothermic peak is observed,
which characterizes the melting of soda ash. Curve 2 shows the DTA-signal for the mixture of
soda ash and limestone heated in a CO2 -atmosphere. Already at 1058 K, an endothermic peak
is observed, which coincides with the eutectic melting temperature of soda ash and limestone.
However, during heating of a mixture of soda ash and limestone in a N 2 -atmosphere, the endothermic peak of soda ash melting at 1129 K is observed instead of the endothermic peak at
1058 K, which indicates that the double carbonate is not formed in the N 2 -atmosphere. This can
be explained by the observation (see section 2.5.2) that in a N 2 -atmosphere limestone already
starts to decompose below 923 K. The early release of CO2 prevents the formation of the double
carbonate. Therefore, the formation of the double carbonate is more pronounced in a CO 2 - than
in a N2 -atmosphere.
During reheating of the mixture of soda ash and limestone, which was heated in a 1 bar CO 2 atmosphere, two endothermic peaks were observed at 659 K and 703 K, respectively. These
endothermic peaks correspond to two crystalline inversions of the double carbonate, which was
observed by Wilburn et al. [29] and Taylor and Rowan [30]. During the TGA of the mixture
of limestone with soda ash heated in a 1 bar CO2 -atmosphere up to 1273 K, no calcination
was observed. This supports the observation that the double carbonate was formed, which may
retard the calcination temperature of limestone up to temperatures above the thermodynamic
calcination temperature in a CO2 -atmosphere. However, for the formation of the double carbonate, also soda ash is required. As will be shown in the next section, the calcination of soda
ash is already almost completed at about 1133 K. Therefore, it is expected that the retarded
calcination of limestone due to the formation of the double carbonate is not expected in glass
forming batches.
Reactive limestone calcination in the presence of oat glass cullet
As mentioned at the beginning of this section, the calcination reaction of limestone is favored
above the carbonization reaction of CaO in case the activity of the formed CaO is less than unity.
The CaO activity in a glass melt containing 75 wt.% SiO2 , 15 wt.% Na2 O and 10 wt.% CaO
equals 5.7 106 at 1473 K. Therefore, it is expected that the onset temperature for calcination
of solid limestone with simultaneous dissolution of CaO in a present glass melt shifts to lower
temperatures. The effect of the presence of a glass melt on the calcination of limestone is studied by TGA of a mixture of limestone with oat glass cullet. Figure 2.19 shows the measured

48

Chapter 2. Energy demand of glass forming batches

1
T =80 K

3
DTA signal

PSfrag replacements

1000

1050

1100

1150

1200

Temperature [K]

Figure 2.18: DTA signal during heating of Na2 CO3 in a 1 bar CO2 -atmosphere (1), a mixture of Na2 CO3
and CaCO3 in a 1 bar CO2 -atmosphere (2) and a mixture of Na2 CO3 and CaCO3 in a N2 atmosphere (3).

calcination rate of limestone in the mixture of limestone and crushed oat cullet ramp heated
with 10 K min1 in a 100 ml min1 CO2 -ow.
It can be seen that the calcination of limestone already starts at about 1020 K 15 , which is
about 160 K lower than the thermal calcination of limestone in a 1 bar CO 2 -atmosphere. At
about 1053 K, the calcination of limestone stops. At this temperature, the presence of a crystalline sodium calcium silicate, 2Na2 O CaO 3SiO2 , was identied with X-ray diffraction. This
ternary crystalline silicate is formed in case the CaO-content in the sodium silicate melt phase
exceeds the maximum CaO-solubility in the sodium silicate melt phase. The formation of the
crystalline silicate may prevent direct contact between the limestone grain and a melt phase with
a low CaO-content, by which the reactive calcination of limestone slows down or even stops.
Now, the reactive calcination of limestone during further heating of the glass batch depends
on
the supply of a melt phase with a low CaO-content in which the crystalline sodium calcium silicate is able to dissolve, and
the temperature dependent CaO-solubility in the sodium silicate melt phase.
15 Similar

to a mixture of dolomite and int cullet, Conradt [24] observed that limestone affects the formation of
an interconnected melt phase during heating int cullet. This may indicate that int cullet is not inert to limestone
as is shown in gure 2.19.

2.5. Calcination of a oat glass batch

49

PSfrag replacements

Calcination rate of limestone [K1 ]

0.014

Thermal calcination
of limestone

0.012
0.010
0.008
0.006

Reactive calcination
of limestone

0.004
0.002
0.000
1000

1050

1100

1150

1200

1250

Temperature [K]

Figure 2.19: Calcination of CaCO3 at the presence of oat glass cullet in a CO2 -atmosphere.

At 1173 K, the regular thermal calcination of limestone in a CO 2 -atmosphere is observed. This


indicates that limestone is mainly decomposed via thermal dissociation. At this point, it is
concluded that the presence of a melt phase does not inuence the calcination behavior of
limestone signicantly.

2.5.3 Calcination of soda ash


In this section, the calcination behavior of individual soda ash grains and soda ash grains
in mixtures with other glass batch components is studied in both a N 2 - and a 1 bar CO2 atmosphere. From thermogravimetric analysis, the kinetics of reactive soda ash calcination
is determined.
Dissociative calcination of individual soda ash grains
Figure 2.20 shows both the degree of soda ash calcination16 and the DTA-signal measured during heating of soda ash grains in both a 100 ml min1 N2 -ow and a 100 ml min1 CO2 -ow.
It is observed that up to the melting temperature of soda ash, which equals 1128 K, hardly
any calcination of soda ash is observed in both atmospheres. Above the melting temperature
of soda ash, a slow weight loss is observed in the N2 -atmosphere. The partial CO2 -pressure,
which is in equilibrium with the soda ash, is calculated using thermodynamic tables [3] and
equals 1.8 107 bar at 1200 K. This indicates that thermal calcination of soda ash is not ex16 The

degree of soda ash calcination is dened as the ratio of the actual weight loss of soda ash with respect to
the weight loss of soda ash after complete calcination.

50

Chapter 2. Energy demand of glass forming batches

0.10

0.10

0.08

0.05

0.07

0.00

0.06

-0.05

0.05
-0.10

0.04
0.03

-0.15

0.02

DTA-signal [a.u.]

PSfrag replacements

Degree of soda ash calcination [-]

0.09

-0.20

0.01
-0.25

0.00

-0.01
950

1000

1050

1100

1150

1200

-0.30
1250

Temperature [K]

Figure 2.20: Measured degree of soda ash calcination and DTA-signals during heating of soda ash in a
100 ml min1 N2 -ow and in a 100 ml min1 CO2 -ow. Line 1 indicates the degree of
soda ash calcination during heating in the N2 -atmosphere, line 2 indicates the degree of
soda ash calcination in the CO2 -atmosphere, line 3 indicates the DTA-signal of soda ash
heated in the N2 -atmosphere and line 4 indicates the DTA-signal of soda ash heated in the
CO2 -atmosphere. Point A marks a small endothermic peak located at 963 K.

pected during melting of glass batches. The thermal calcination of soda ash is not observed
in the CO2 -atmosphere, which is likely because the CO2 -pressure in the atmosphere above
the soda ash grains in the TGA/DTA-apparatus exceeds the thermodynamic equilibrium partial
CO2 -pressure.
In the N2 -atmosphere, a small endothermic peak is observed at 963 K, which is not observed during heating of soda ash in the CO2 -atmosphere. However, additional experiments
with mixtures of soda ash and silica sand heated in a CO2 -atmosphere did show the presence
of this small endothermic peak, whereas during heating of these mixtures in a N 2 -atmosphere
this endothermic peak was not observed. During the quantitative determination of the reactive
calcination of soda ash, which is described later in this section, it seems that this endothermic
peak is related to accelerated calcination of soda ash. A more intense endothermic peak at 963
K resulted in a shift of the onset temperature for reactive calcination of soda ash with silica sand
towards 973 K. The presence of the endothermic peak at 963 K seems to be more pronounced
in a CO2 -atmosphere than in a N2 -atmosphere. However, because the focus in this section is
on the quantitative determination of the kinetics of soda ash calcination, without identication
of the detailed calcination mechanism, the cause of this endothermic peak is not investigated in
more detail.

2.5. Calcination of a oat glass batch

51

Reactive calcination of soda ash grains


TGA-experiments of soda ash combined with dolomite, limestone or silica sand showed that
soda ash calcination only occurs via reactive dissociation with silica sand. During the reactive
calcination of soda ash with silica, several crystalline sodium silicate phases can be formed below 1072 K as can be seen from the binary phase diagram of Na2 O and SiO2 (see gure 2.5).
The predominant sodium silicates in the binary Na2 O-SiO2 phase diagram are sodium disilicate
(2Na2 O SiO2 ), sodium metasilicate (Na2 O SiO2 ) and sodium orthosilicate (Na2 O 2SiO2 ).
Figure 2.21 shows the standard Gibbs free energy of the calcination reactions of soda ash by
silica sand resulting in the formation of these sodium silicates. It can be seen from this gure
that the calcination temperatures17 are 556 K, 632 K and 1011 K for the reactions 3.18, 3.19
and 3.20, respectively.
Na2 CO3 (s) + 2 SiO2 (s)
Na2 CO3 (s) + SiO2 (s)
Na2 CO3 (s) + 0.5 SiO2 (s)

Na2 O.2SiO2 (s) + CO2 (g)

(2.68)

Na2 O.SiO2 (s) + CO2 (g)

(2.69)

0.5 2Na2 O.SiO2 (s) + CO2 (g)

(2.70)

Figure 2.22 shows the degree of soda ash calcination and the DTA-signals in case of heating
binary mixtures of soda ash and silica sand in both a N2 - and a CO2 -atmosphere. For the
composition of these binary mixtures and for the experimental conditions at which the combined TGA/DTA-experiments are performed, is referred to table 2.4. Lines 1 and 2 indicate
the degree of soda ash calcination in the N2 - and the CO2 -atmosphere, respectively. Lines 3
and 4 indicate the DTA-signals during heating of the binary mixtures in the N 2 - and the CO2 atmosphere, respectively. It is observed that the onset for reactive soda ash calcination in the
CO2 -atmosphere occurs at lower temperatures than in the N2 -atmosphere. It is also observed
that during heating of the binary mixture in the CO 2 -atmosphere, the endothermic peak at 963
K is present, which is absent in the N2 -atmosphere.
Based on the measured degree of soda ash calcination, it is concluded that reactive soda ash
calcination hardly occurs below 973 K, which is signicantly higher than the thermodynamic
onset temperatures for reactive calcination of soda ash with silica sand according to gure 2.21.
The onset temperature for calcination of soda ash in a CO2 - and a N2 -atmosphere appeared
to be approximately 973 K and 1073 K, respectively. The latter nding was also observed by
Kr ger [31], who studied isothermally the reaction kinetics of binary mixtures of silica and soda
o
ash by measuring the release of CO2 during the heating of mixtures of silica sand and soda ash

with particle sizes ranging from 150 m to 250 m. During his experiments, Kr oger made the
following observations:
1. The reactive calcination of soda ash with silica sand steeply increased between 1098 K
and 1127 K for SiO2 :Na2 CO3 -ratios in the binary mixtures of 1:1 and 2:1.
2. The calcination rate of soda ash increased with increasing fraction of silica sand in the
initial mixture of soda ash and silica sand.
3. The calcination rate of soda ash decreased with increasing partial CO 2 -pressure.
17 Here, the calcination temperature is dened as the temperature at which the thermodynamic equilibrium partial

CO2 -pressure equals 1 bar.

52

Chapter 2. Energy demand of glass forming batches

1.0x10

8.0x10

6.0x10

4.0x10

2.0x10

frag replacements

Standard Gibbs free energy of reaction [J mol1 ]

104

103

Na

2 CO
3 (s)+

0.5 S

102

iO2 (

s)
0

.5 [2
Na

2O

SiO

2 ](s)

Na

2 CO
3 (s)

0.0

+ Si

O2 (

Na

-2.0x10

-4.0x10

+2

s)

-6.0x10

2 CO
3 (s)

Na

2O

SiO

2 (s)+

+CO

101

2 (g)

100.3
100
100.7

500

600

SiO

2 (s)

a2 O

700

2Si

O2 (s

CO

2 (g)

)+C

800

O2 (g
)

900

101
102

1000

Temperature [K]

Figure 2.21: Standard Gibbs free energy of the calcination reactions of soda ash with silica sand. The
dashed lines are isobars, expressed in bar, for the partial CO 2 -pressure above the carbonates.

4. During isothermal calcination of soda ash at temperatures above 1138 K, a part of the
soda ash was not decomposed even after long isothermal periods. The amount of the soda
ash that was not decomposed appeared to be dependent on temperature.
In the temperature range between 1098 K and 1127 K, primary melt phase formation is possible due to eutectic melting of sodium metasilicate with sodium disilicate at 1108 K, melting
of sodium disilicate at 1129 K and melting of soda ash at 1128 K. The reason for the absence,
or very slow, reactive calcination of soda ash at temperatures below 973 K are likely the low
solid-state diffusion coefcients of the reactants and reaction products. The rate of solid-state
reactions is namely governed by the slow diffusion of reactants and reaction products from the
reaction interface. Because the liquid-state diffusion coefcients of the SiO 2 and Na2 O are several orders of magnitude larger than their solid-state diffusion coefcients, it is expected that
the calcination of soda ash is enhanced at the temperature at which the primary melt phase is
formed.
The second, third and fourth observation of the above mentioned observations that were
made by Kr ger, can be explained as follows. According to equation 2.41, the reactive calcio
nation rate of soda ash is proportional to the square of the mole fraction of the stoichiometric
compound SiO2 in the melt phase, which increases with increasing mole fraction of silica sand

2.5. Calcination of a oat glass batch

53

1.0

0.8

0.35

3
4

0.30

0.6
0.25
0.20

0.4

DTA-signal [a.u.]

PSfrag replacements

Degree of soda ash calcination [-]

0.40

0.15
0.2

0.10

1
0.0
900

950

1000

1050

1100

1150

1200

0.05
1250

Temperature [K]

Figure 2.22: Degree of soda ash calcination and DTA-signals in case of heating soda ash with 10 K
min 1 in a 100 ml min1 N2 -ow and in a 100 ml min1 CO2 -ow. Line 1 indicates the
degree of conversion of soda ash heated in the N2 -atmosphere, line 2 indicates the degree
of conversion of soda ash in the CO2 -atmosphere, line 3 indicates the DTA-signal of soda
ash heated in the N2 -atmosphere and line 4 indicates the DTA-signal of soda ash heated in
the CO2 -atmosphere.

in the initial mixture of soda ash and silica sand. Therefore, the calcination rate of soda ash is
expected to be increased with increasing silica sand content in the binary mixture.
Assuming that the chemically dissolved CO2 in the glass melt is characterized by the stoichiometric compound Na2 CO (m) (see section 2.3.4), the mole fraction of this stoichiometric
3
compound is given by
aNa2 O(m) pCO2 (g)
xNa2 CO3 (m) =
,
(2.71)
K
in which K is the thermodynamic reaction equilibrium constant. It is seen that the mole fraction
of the stoichiometric compound Na2 CO (m), and therewith the amount of CO2 dissolved in
3
the glass melt, increases with increasing partial CO2 -pressure above the sodium silicate melt,
which was observed by Kr ger. The temperature dependent amount of soda ash, which was not
o
decomposed during an isothermal reaction of a mixture of soda ash and silica sand is given by
the temperature dependent chemical solubility of CO2 in the binary glass melt.
As mentioned above, the onset temperature for soda ash calcination in a 1 bar CO 2 -atmosphere
appeared to be lower than the onset temperature for calcination in a N 2 -atmosphere. This lowered calcination temperature can not be explained by the thermodynamic description of the
calcination of soda ash in the mixture of soda ash and silica sand. At this moment, the cause of
the lowered calcination temperature of soda ash is not claried.
During thermogravimetrical analysis of mixtures of soda ash and silica sand, above 1123

54

Chapter 2. Energy demand of glass forming batches

Concentration of Na2 O SiO2 [a.u]

0.25

0.20

0.15

0.10

0.05

PSfrag replacements

0.00
1120

1140

1160

1180

1200

1220

1240

1260

Temperature [K]

Figure 2.23: Measured concentration of crystalline Na2 O SiO2 in a partly molten mixture of SiO2 and
Na2 CO3 .

K a decrease of the calcination rate of soda ash was observed. This temperature is close to the
eutectic melting temperature of Na2 O 2SiO2 with Na2 O SiO2 (i.e. 1108 K). According to the
binary phase diagram of Na2 O and SiO2 , beyond this temperature crystalline sodium metasilicate (Na2 O SiO2 (s)) can be formed by a reaction of soda ash with the eutectic melt phase.
The presence of crystalline sodium metasilicate in quenched samples of partly molten mixtures
of soda ash and silica sand was identied during phase analysis with X-ray diffraction. Figure
2.23 shows the measured semi-quantitative concentration of crystalline Na 2 O SiO2 , in arbitrary
units, in a partly molten mixture of SiO2 and Na2 CO3 .
Similar to the system of limestone reacting with oat cullet, the presence of the crystalline
specie surrounding the soda ash grains may act as a barrier preventing contact between soda ash
and silica sand. This hampered contact may result in a decrease of the calcination rate of soda
ash, which was observed from the TGA-experiments above 1123 K.
Resuming, the calcination of soda ash by a reaction with silica sand, mainly occurs above
the onset temperature for primary melt phase formation. In the following, the kinetic parameters describing the reactive calcination of soda ash by reaction with silica sand are determined
from TGA-experiments of binary mixtures of soda ash and silica sand as function of
the ratio of the amount of soda ash with respect to the amount of silica sand,
the particle size of the soda ash and silica sand grains, and
the atmosphere composition in which the binary mixtures are heated.

2.5. Calcination of a oat glass batch

55

Determination of the kinetic parameters of reactive calcination of soda ash


As mentioned above, the reactive calcination of soda ash with silica sand mainly occurs in
case a sodium silicate melt phase is present. According to the binary phase diagram of SiO 2
and Na2 O (see gure 2.5), the temperature at which the rst sodium silicate melt is formed
equals 1053 K, at which silica sand eutectic melts with sodium disilicate. The rst formed melt
phase is regarded as a mixture of the stoichiometric compounds SiO 2 and sodium disilicate
(Na2 O 2SiO2 ). Now, the reactive calcination of soda ash in contact with the rst formed melt
phase, neglecting the chemical dissolution of CO2 in the melt phase, is described by
Na2 O.2SiO2 (m) + CO2 (g),

Na2 CO3 (s) + 2 SiO2 (m)


or by

Na2 CO3 (s) + Na2 O.2SiO2 (m)

2 Na2 O.SiO2 (m) + CO2 (g)

(2.72)
(2.73)

in which the superscript denotes the stoichiometric compound. The general equation describing the kinetics of heterogeneous reactions is given by (see section 2.3.3)
r = k f f ()

nr

ai
i
i=1

Ka
.
Keq

(2.74)

Because the calcination temperature for both reactions 2.72 and 2.73 are far below 1053 K,
the calcination of soda ash occurs far from thermodynamic equilibrium, which indicates that
Ka < Keq . The reactive calcination of solid soda ash is now described by
r = k f f () ar ,
r

(2.75)

in which ar is the activity of the reactant with which soda ash reacts, which is either silica
or sodium disilicate. Estimation of the reaction kinetic parameters for the reactive calcination of soda ash requires knowledge of the temperature and composition dependent activity of
the stoichiometric compounds SiO2 and Na2 O 2SiO2 during the reactive calcination reaction.
However, from the measured calcination rate of soda ash with TGA, the apparent soda ash
calcination rate, as dened by equation 2.76 is measured.
r

= r = r = k f f () ,
t
ar

(2.76)

The thermogravimetrical analysis does not provide information on the activities of the stoichiometric compounds during the calcination reaction.
The reaction mechanism function, which provides the best t with the measured data of
the reactive calcination of soda ash is determined by plotting [ln r + ln ln f ()] versus the
reciprocal absolute temperature for different reaction mechanism functions. Figure 2.24 shows
this plot in the temperature range from 1068 K up to 1139 K for the reaction mechanism functions F0, F1, D3, D4, R2 and R3 for the binary mixture nr. 4 listed in table 2.4.
It can be seen that the pattern for all reaction mechanism functions is similar. Between
1083 K and 1123 K, an almost linear relation between [ln r + ln ln f ()] and the reciprocal
absolute temperature is observed. However, above 1123 K the linear relation is absent. This
indicates that above 1123 K the calcination of soda ash is governed by an other process than
in the temperature range between 1083 K and 1123 K. As mentioned before, the formation of
crystalline sodium metasilicate is likely the cause of this change in calcination rate of soda ash.

56

Chapter 2. Energy demand of glass forming batches

2
2

PSfrag replacements

[ln r + ln ln f ()] [ln min1 ]

-1

3
4

-2
-3

5
6

-4
-5
-6
-7
-8
-9

-10
-11
0.88

1123 K

1083 K

0.90

0.92

Reciprocal absolute temperature [ 103 K1 ]

0.94

Figure 2.24: Plot of [ln r + ln ln f ()] versus the reciprocal absolute temperature for reactive calcination of soda ash with silica sand in the temperature range from 1068 K up to 1139 K as
function of the reaction mechanism functions for the binary mixture nr. 4 from table 2.4.
Line 1 describes the F0-mechanism function, line 2 describes the F1-mechanism function,
line 3 describes the D3-mechanism function, line 4 describes the D4-mechanism function,
line 5 describes the R2-mechanism function and line 6 describes the R3-mechanism function.

Table 2.4: Experimental conditions for TGA of mixtures of silica sand and soda ash.
Nr.
1
2
3
4
5
6
7

Ratio silica:
soda ash

dSiO2
[m]

dNa2 CO3
[m]

Atmosphere

Heating rate
[K min1 ]

1:0.225
1:0.285
1:0.347
1:0.295
1:0.281
1:0.214
1:0.170

160 - 250
160 - 250
160 - 250
63 - 125
160 - 250
160 - 250
160 - 250

250 - 500
250 - 500
250 - 500
250 - 500
106 - 125
106 - 125
250 - 500

N2
N2
N2
N2
N2
CO2
CO2

20
20
20
20
20
20
20

2.5. Calcination of a oat glass batch

57

Figure 2.25 shows a detailed part of gure 2.24 in the temperature range from 1083 K
up to 1123 K. For this temperature range, the correlation coefcient of the linear relation of
[ln r + ln ln f ()] versus the reciprocal absolute temperature is determined for the six reaction mechanism functions. The correlation coefcients for these six reaction mechanism
functions are listed in table 2.5.

[ln r + ln ln f ()] [ln min 1 ]

-1
-2

-3

-4

PSfrag replacements

-5
-6
-7
0.895

0.900

0.905

0.910

0.915

Reciprocal absolute temperature

[ 103

0.920

0.925

K1 ]

Figure 2.25: Plot of [ln r + ln ln f ()] versus the reciprocal absolute temperature for reactive calcination of soda ash with silica sand in the temperature range from 1083 K up to 1123 K as
function of the reaction mechanism functions for the binary mixture nr. 4 from table 2.4.
Line 1 describes the F0-mechanism function, line 2 describes the F1-mechanism function,
line 3 describes the D3-mechanism function, line 4 describes the D4-mechanism function,
line 5 describes the R2-mechanism function and line 6 describes the R3-mechanism function.

It can be seen that the reaction mechanism functions D3 and D4 show the highest value for
the correlation coefcient. The reaction mechanism function D3, which is known as the Jander
equation, was used by Kr ger [31] to describe the reactive calcination of soda ash by a reaction
o
with silica. However, Kr ger mentioned that the experimental data for reactive calcination of
o
soda ash could only be described by the Jander equation in a limited conversion range. This
indicates that it is likely that reactive calcination of soda ash does not proceed via only one
reaction.
As mentioned in section 2.3.3, the approach for selecting the most accurate reaction mechanism function based on the highest value for the correlation coefcient is arbitrarily and does not
necessary provide information about the reaction mechanism. This is supported by gure 2.26,
which describes the simulated calcination of soda ash as function of the reaction mechanism
functions with the kinetic reaction parameters as presented in table 2.5.

58

Chapter 2. Energy demand of glass forming batches

Table 2.5: Correlation coefcients for six reaction mechanism functions describing the reactive calcination of soda ash by silica sand in the temperature range from 1083 K up to 1123 K.
Reaction mechanism function

Correlation coefcient

Ea
kJ mol1

ln A
ln s1

F0
F1
D3
D4
R2
R3

0.9939
0.9846
0.9972
0.9988
0.9936
0.9911

657.4
939.6
1699.6
1605.5
798.5
845.5

66.57
97.76
177.72
167.32
81.48
86.27

Degree of soda ash calcination [-]

1.0

PSfrag replacements

0.8

0.6

0.4

0.2

0.0
1080

1090

1100

1110

1120

Temperature [K]

Figure 2.26: Simulated reactive calcination of soda ash with silica sand when ramp heated with 20 K
min1 in a N2 -atmosphere and as function of the reaction mechanism function. Because
simulated curves are very close to each other, the individual lines are not labelled with their
reaction mechanism function.

It is noticed that the difference in the simulated versus the measured degree of soda ash
calcination is less than 0.02. This indicates that the choice of the reaction mechanism function
for describing the reactive calcination of soda ash is arbitrarily. For the description of the
reactive calcination of soda ash with silica sand, the F1 reaction mechanism function is applied.
The reaction kinetic parameters have been determined for 7 binary mixtures. The composition
of these mixtures and the experimental conditions at which the TGA-experiments are performed
are listed in table 2.4. The values for the apparent reaction activation energy and the natural
logarithm of the pre-exponential factor are listed in table 2.6 for the F1 reaction mechanism
function.

2.5. Calcination of a oat glass batch

59

Table 2.6: Apparent reaction activation energy, pre-exponential factor, correlation coefcient in the temperature range from 1083 K up to 1123 K for seven calcination experiments of soda ash
assuming a F1 reaction mechanism function.
Nr.
1
2
3
4
5
6
7

Ea
[kJ mol1 ]

ln A
[ln s1 ]

r2

953.0
1028.6
919.0
939.6
780.0
499.4
512.7

99.1
107.6
95.5
97.8
80.3
49.9
51.5

0.9808
0.9968
0.9857
0.9846
0.9915
0.9828
0.9762

120

PSfrag replacements

Natural logarithm of A [ln min1 ]

A
100

80

60

40

20
200

400

600

800

1000

Apparent reaction activation energy [kJ mol1 ]

Figure 2.27: Kinetic compensation effect: plot of the natural logarithm of the pre-exponential factor
versus the reaction activation energy for the reactive calcination of soda ash (A indicates
the experiment 1, 2, 3 and 4; B indicates experiment 5; C indicates the experiments 6 and
7) and the calcination of limestone indicated by D. According to 2.50, the slope of the
line equals (R T )1 and the intercept equals ln A. The values for both slope and intercept
are 1.09 104 mole J1 and 0.58, respectively. The unit of A is min1 . The correlation
coefcient of this line equals 0.9995.

60

Chapter 2. Energy demand of glass forming batches

Figure 2.27 plots the relation between the apparent reaction activation energy versus the
natural logarithm of the pre-exponential factor as is described by equation 2.49, which describes
the kinetic compensation effect.
The reaction kinetic parameters derived from the TGA-experiments performed in a N 2 atmosphere are indicated by the dots in the square indicated by A, representing the values for
the binary mixtures 1, 2, 3 and 4, and by the dot indicated by B, representing TGA-experiment
5. From this gure it seems that the particle size of the silica grains does not have a signicant
impact on the calcination of soda ash. However, the particle size of the soda ash seems to affect
the reactive calcination of soda ash by silica sand.
Figure 2.28 shows the simulated reactive calcination of soda ash with silica sand when
ramp heated with 20 K min1 in both a N2 - and a 1 bar CO2 -atmosphere and as function of the
reaction kinetic parameters presented in table 2.6. Although it seemed that the particle size of
soda ash has an impact on the reaction kinetic parameters, the simulated reactive calcination of
soda ash for experiment 5 is almost similar to the simulated reactive calcination of soda ash for
the experiments 1, 2, 3 and 4.

PSfrag replacements

Degree of calcination of soda ash [-]

1.0

0.8

0.6

0.4

0.2

B
A

0.0
1040

1060

1080

1100

1120

1140

Temperature [K]
D

Figure 2.28: Simulated reactive calcination of soda ash with silica sand when ramp heated with 20 K
min1 in both a N2 - and a 1 bar CO2 -atmosphere and as function of the reaction kinetic
parameters presented in table 2.6. A indicates the simulated reactive calcination of soda ash
for the experiments 1, 2, 3 and 4. B indicates the simulated reactive calcination of soda
ash for experiment 5. C indicates the simulated reactive calcination of soda ash for the
experiments 6 and 7.

2.5. Calcination of a oat glass batch

61

According to gure 2.27, the presence of CO2 in the atmosphere surrounding the binary
mixture shows the lowest value for the apparent reaction activation energy. The different reaction kinetic parameters for reactive calcination of soda ash in a CO 2 -atmosphere show a signicant difference with the calcination of soda ash in the N 2 -atmosphere. From the slope of
the kinetic compensation curve presented in gure 2.27, a temperature value can be calculated
which, according to Gallagher and Johnson [21], represents the average temperature over which
the reaction kinetic parameters are derived. For the reactive calcination of soda ash this value
equals 1103 K. However, in case the reaction kinetic parameters for the calcination of limestone are added to gure 2.27, the slope of the kinetic compensation curve does not change,
whereas the limestone calcination in a N2 -atmosphere occurs, dependent on the heating rate,
in the temperature range from 873 K up to 1073 K (see gure 2.16). This indicates that the
kinetic compensation curve is quite insensitive for changes in reaction temperatures of 100-200
K. Therefore, the explanation of different reaction activation energies presented in literature for
the same chemical reaction by the kinetic-compensation effect is questioned.
Because it is expected that the partial CO2 -pressure is high during melting of glass batches
containing carbonated raw materials, the calcination of soda ash is described by
r =

500000
(1 ) e(54 R T ) ,

(2.77)

in which the pre-exponential factor and the apparent reaction activation energy describe the
reactive calcination of soda ash with silica sand in a CO2 -atmosphere. This equation does
not include the calcination of soda ash at temperatures above 1123 K at which the soda ash
calcination is retarded. However, because the main part of the soda ash reacts below 1123 K,
equation 2.77 is used as a rst approximation of the reactive calcination of soda ash by silica
sand in a CO2 containing atmosphere.

62

Chapter 2. Energy demand of glass forming batches

2.5.4 Calcination of a mixture of silica sand, soda ash and limestone


As described in the previous sections, the calcination of dolomite, limestone and soda are described as separate reactions. In this section, the measured calcination 18 of a mixture of silica
sand, soda ash and limestone is compared with the simulated calcination of this mixture.
Table 2.7 lists the batch composition and the experimental conditions during heating of four
glass batches.
Table 2.7: Experimental conditions for TGA of mixtures of silica sand, soda ash and limestone.
Nr.
1
2
3
4

Ratio silica:
soda ash: limestone

dSiO2
[m]

dNa2 CO3
[m]

dCaCO3

Atmosphere

Heating rate
[K min1 ]

1:0.254:0.452
1:0.294:0.197
1:0.256:0.224
1:0.295:0.264

250 - 500
250 - 500
250 - 500
250 - 500

106 - 125
106 - 125
106 - 125
106 - 125

coarse
coarse
coarse
coarse

N2
N2
N2
CO2

20
20
20
20

The values for the apparent reaction activation energy and the natural logarithm of the preexponential factor for the calcination of limestone and for the calcination of soda ash in both a
N2 - and a 1 bar CO2 -atmosphere are listed in table 2.8. These values are only valid for describing the calcination of limestone and soda ash by a rst-order reaction mechanism.
Table 2.8: Reaction activation energy and natural logarithm of the pre-exponential factor for calcination
of limestone and calcination of soda ash in both a N2 - and CO2 -atmosphere.
Specie

Atmosphere

Ea
kJ mol1

ln A
ln s1

limestone
soda ash
soda ash

N2 and CO2
N2
CO2

188.9
953.0
500.0

16.6
98.9
49.9

Figure 2.29 shows the measured (dotted line) and the calculated (solid line) degree of calcination of the ternary mixture nr. 1 heated in a N2 -atmosphere. It can be seen that the maximum
temperature difference between both curves equals 20 K.
Figure 2.30 shows the simulated versus the measured degree of calcination for the experiments 1, 2 and 3. It can be seen that the maximum difference between simulated and measured
degree of calcination, which is observed for experiment 3, is less than 0.20. This is caused by a
maximum temperature difference between both curves of 25 K. Figure 2.31 shows the measured
(dotted line) and the calculated (solid line) degree of calcination of the ternary mixture nr. 4
heated in a 1 bar CO2 -atmosphere.
Resuming, with the reaction kinetic parameters describing the calcination rate of the individual carbonates, the calcination of a oat glass batch can be predicted with an uncertainty in
the degree of calcination of the complete oat glass batch less than 0.20.
18 Here

the degree of calcination is dened as the ratio of the actual weight loss with respect to the weight loss
after complete calcination of limestone and soda ash

2.5. Calcination of a oat glass batch

63

1.0
0.9

Degree of calcination [-]

0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0.0
900

PSfrag replacements

950

1000

1050

1100

1150

Temperature [K]

Figure 2.29: Measured (dotted line) and simulated (solid line) degree of calcination as function of temperature for glass batch mixture 1 from table 2.7.

1.0

PSfrag replacements

Calculated degree of calcination [-]

0.9
0.8
0.7

0.6

0.5
0.4
0.3
0.2
0.1
0.0
0.0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1.0

Measured degree of calcination [-]

Figure 2.30: Simulated versus the measured degree of calcination of the glass batch mixture 1, 2 and 3
from table 2.7.

64

Chapter 2. Energy demand of glass forming batches

1.0
0.9

Degree of calcination [-]

0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0.0

PSfrag replacements

900

950

1000

1050

1100

1150

1200

1250

Temperature [K]

Figure 2.31: Measured (solid line) and simulated (dotted line) degree of calcination as function of temperature for glass batch mixture 5 from table 2.7.

2.5.5 Reaction mechanism of a oat glass batch


According to Conradt et al. [32], the melting of commercial glass batches, with as major components silica sand, soda ash, limestone and/or dolomite, proceeds via either the carbonate or
the silicate route (see also Riedel [33]). During the carbonate route, the formation of the primary melt phase is governed by the melting of the double carbonate of soda ash and limestone,
Na2 Ca(CO3 )2 , or by eutectic melting of a mixture of soda ash and limestone (see section 2.5.2).
During the silicate route, the formation of the primary melt phase is governed by eutectic melting of sodium disilicate with silica.
This section describes, based on experimental studies performed by Kautz and Stromburg
[27], Sheckler and Dinger [28] and Savard and Speyer [34], the conditions for the occurrence
of these two reaction routes. Additionally, the reaction mechanism of the oat glass batch, for
which the calcination kinetics was studied in the previous sections, is determined by phase analysis on quenched oat glass batch samples, which were heated up to temperatures in the range
of 973 K to 1673 K.
Study on the reaction mechanism of soda-lime-silica glass by Kautz and Stromburg [27]
Kautz and Stromburg [27] identied the crystalline phases in quenched glass batch samples after heating the subsystems and the complete mixture of a glass batch composed of sand, soda
ash and limestone or dolomite. The mixture of the glass batch components was ramp heated
with 10 K min1 in a CO2 -atmosphere. Both during heating of a mixture of soda ash and
limestone and during heating of the complete glass batch, the presence of the double carbonate

2.5. Calcination of a oat glass batch

65

Na2 Ca(CO3 )2 was observed in the temperature range from 773 K up to 1113 K.
Similar as was mentioned in section 2.5.2, the double carbonate was only formed in case
of small particle sizes (d< 100 m) of the soda ash grains and the limestone grains and in case
the glass batch was humidied. The observed presence of the double carbonate in a pure CO 2 atmosphere is supported by the own observation presented in section 2.5.2, which showed the
effect of the partial CO2 -pressure on the double carbonate formation.
During heating a of mixture of silica sand, soda ash and limestone, crystalline CaO was
identied at temperatures above 1173 K, which equals the thermodynamic onset temperature
for limestone calcination in a 1 bar CO2 -atmosphere. The presence of crystalline CaO at 1173
K indicates that limestone is not (completely) decomposed by reactive calcination with the primary formed sodium silicate melt phase, which was also observed in section 2.5.2.
Phase analysis on quenched samples after both heating the complete glass batch and a mixture of soda ash and silica, showed the presence of crystalline sodium metasilicate in the temperature range from 1123 K up to 1223 K. The presence of sodium metasilicate was also observed
by own measurements presented in gure 2.23.
At about 1073 K, the presence of the ternary crystalline silicate 2Na 2 O CaO 3SiO2 was
identied. This ternary silicate remained in the partly molten glass batch up to 1373 K, whereas
the maximum concentration of the ternary silicate was observed at 1163 K. Using ne limestone and soda ash particles (d< 100 m) instead of coarser particles (d> 100 m), showed that
the ternary silicate was present, but sodium metasilicate could not be identied. Crystalline
MgO was observed to remain present in the partly molten batch up to temperatures of 1373 K,
whereas crystalline cristobalite was observed starting at about 1273 K.
Study on the reaction mechanism of soda-lime-silica glass Sheckler and Dinger [28]
Sheckler and Dinger [28] identied in the temperature range from 1048 K up to 1138 K
the crystalline phases present in quenched samples after heating of a mixture of silica, soda
ash and limestone. The particle sizes of both the soda ash and the limestone were less than 53
m. Three different particle size distributions for the silica have been tested, viz. d<53 m,
125 m<d<180 m and 425 m<d<500 m. In all mixtures the presence of the double carbonate was observed, which was expected due to the small particle sizes of both soda ash and
limestone. However, in case of ne silica particles, only a very slight amount of the double
carbonate was identied.
Starting from 1078 K, different ternary silicates were observed. The Na 2 O 2CaO 3SiO2
phase was identied at 1093 K and was more pronounced with smaller particle size of the silica sand grains. At slight higher temperatures, the presence of crystalline 2Na 2 O CaO 3SiO2
was observed for the two coarser silica particle distributions. Reactive dissolution of limestone
resulted in the formation of crystalline ternary silicates with relative high silica content such
as 2Na2 O CaO 3SiO2 and Na2 O CaO 3SiO2 . Reactive decomposition of limestone even resulted in the formation of wollastonite.
Study on the reaction mechanism of soda-lime-silica glass Savard and Speyer [34]
Savard and Speyer [34] identied the crystalline phases of quenched glass batch samples after
ramp heating with 10 K min1 in an ambient atmosphere. The glass batch mixture contained
sand, soda ash, calcite, dolomite and feldspar. The chemical behavior of two different particle
size distributions were evaluated, viz. 125 m<d<250 m and d<45 m. In case of the coarser
batch, mainly sodium metasilicate was identied and also a slight amount of sodium disilicate.

66

Chapter 2. Energy demand of glass forming batches

With decreasing particle size, the presence of sodium disilicate became more pronounced and
sodium metasilicate disappeared. Next to these crystalline silicates no other binary or ternary
silicates were identied.
Conditions for occurrence of the carbonate and the silicate route
Based on the observations presented by Kautz and Stromburg, Sheckler and Dinger and Savard
and Speyer, the conditions required for following either the carbonate route or the silicate route
during melting of glass batches containing as major components silica sand, soda ash and limestone and/or dolomite are:
The carbonate route:
The carbonate route requires the formation of the double carbonate Na 2 Ca(CO3 )2 , which
is favored in case of a small particle size of limestone and/or dolomite, improved wetting
of the glass batch mixture and a high partial CO2 -pressure. At the presence of the double
carbonate, a primary melt phase composed of Na2 CO3 and CaCO3 , can be formed from
1053 K. From this mixture, limestone decomposes via reactive calcination of this primary formed melt phase with silica grains, which results in the formation of a crystalline
ternary silicate, e.g. Na2 O 2CaO 3SiO2 . In contrast to the silicate route, the formation
of crystalline sodium metasilicate is not observed or only as a minor component. With
decreasing particle size of the silica particles, the onset for melt formation shift from the
carbonate melt towards the sodium silicate melt.
The silicate route:
In case large limestone grains are present in the glass batch mixture, it is expected to
no double carbonate is formed. The primary melt phase formed during heating of these
glass batch mixtures is likely to be caused by eutectic melting of sodium disilicate with
silica starting at about 1053 K. The sodium disilicate is formed by reactive calcination of
soda ash with silica sand grains. Upon further heating, crystalline sodium metasilicate is
formed above the eutectic melting temperature of sodium disilicate and sodium metasilicate. Because of the large particle size of the limestone, reactive calcination of limestone
is only moderate, as can be seen from gure 2.19 for a mixture of limestone and oat cullet. The ternary crystalline silicate will only be present as minor component. The main
part of the limestone decomposes above the thermodynamic calcination temperature of
limestone, which is dependent on the partial CO2 -pressure.
Identied crystalline phases during heating of a oat glass batch
Figure 2.32 shows the identied crystalline phases as function of temperature. The particle sizes
of both soda ash and dolomite ranged between 63 m and 150 m, while the particle size of the
silica sand was between 212 m and 300 m. Phase analysis of the quenched batch samples
did not show the presence of the double carbonate. The absence of the double carbonate during
melting the oat glass batch indicates that either the dolomite grains are too large or that the
release of CO2 caused by the low temperature thermal calcination of dolomite prevents direct
contact of the CaCO3 grains with the soda ash grains. Because the glass batch is heated in a
N2 -atmosphere, it is expected that both MgO and CaO are present at temperatures of 973 K.
Because the melting of this glass batch at the melting conditions indicate that the melting of
the glass batch proceeds via the silicate route, it is expected that sodium disilicate is formed.

2.5. Calcination of a oat glass batch

67

However, no sodium disilicate is identied. The main intermediate crystalline phase that is
observed is sodium metasilicate which is present at about 1123 K. Starting at about 1173 K, the
ternary silicate Na2 O 2CaO 3SiO2 is observed. The formation of the ternary silicate at higher
temperatures indicate the reaction of (decomposed) limestone with the primary formed sodium
silicate melt. MgO remains in the melt phase till about 1423 K, which was also observed Kautz
and Stromburg. Cristobalite is formed starting at about 1373 K.

Na2 O 2CaO 3SiO2


Na2 O SiO2
MgO
CaO
CaCO3

PSfrag replacements

MgCO3 CaCO3
Na2 CO3
SiO2 (cr)
SiO2 (q)
1000

1100

1200

1300

1400

1500

1600

1700

Temperature [K]

Figure 2.32: Identied crystalline species during heating of a oat glass batch.

2.5.6 Chemical energy demand of a oat glass batch


In the previous sections, the calcination rate of dolomite, limestone and soda ash is determined.
Combining the equation describing the rate of these calcination reactions with the energies required for these calcination reactions provide expressions for the temperature dependent energy
demand of the calcination reaction for carbonate i, Hi , expressed in kJ K1 mole1 . The total
i
T
temperature dependent energy demand for the calcination of a glass batch is given by
H
T

= xi
i=1

Hr,i
,
T

(2.78)

in which xi is the amount of moles of carbonate i per kg of glass batch. However, next to
the calcination reactions, the total temperature dependent chemical energy demand of a glass
forming batch is also dependent on the endo- and exothermic reactions for the dissolution of
the oxides CaO, MgO and SiO2 is the primary formed melt phases. In this section, the kinetics
of these reactions are not evaluated, because:

68

Chapter 2. Energy demand of glass forming batches

the main part of the energy required for chemical reactions is necessary for the calcination
reactions,
quantitative analysis of the amount of undissolved oxides in the primary melt phases is
not trivial and requires advanced analyzing techniques, and
the dissolution energy for the oxides is both temperature and composition dependent,
which requires a sophisticated thermodynamic model.
In the following, the temperature dependent energy demand for the calcination reactions for the
glass batch with the composition given in table 2.1 is calculated. The temperature dependent
calcination energy of the glass forming batch is compared with the total temperature dependent
chemical energy demand as was presented in gure 2.3.
Chemical energy demand for the thermal calcination of the MgCO3 -part of dolomite
From gure 2.13 it can be seen that calcination of the MgCO3 -part of dolomite occurs in the
temperature range between 873 K and 1073 K. The average calcination enthalpy in this temperature range, which is derived from thermodynamic tables, equals 128.8 kJ mol 1 . The energy
required for calcination of the MgCO3 -part of dolomite, expressed in kJ K1 mole1 3 , is
MgCO
given by
513000
H
128.8
(2.79)
=
1 MgCO3 e(55.7 R T ) .
T MgCO3

Chemical energy demand for the thermal calcination of the CaCO3


From gure 2.16 it can be seen that the thermal calcination of the CaCO 3 occurs in the temperature range between 873 K and 1173 K. The average calcination enthalpy, which is derived from
thermodynamic tables, in this temperature range equals 171.8 kJ mol 1 . The energy required
for CaCO3 calcination, expressed in kJ K1 mole1 3 , is given by
CaCO
H
T

=
CaCO3

171.8

190800
(1 CaCO3 ) e(20.8 R T ) .

(2.80)

Chemical energy demand for the reactive calcination of Na2 CO3


In contrast to the thermal calcination of both MgCO3 and CaCO3 , the explicit reaction equation
describing the reactive calcination of soda ash can not be given. The reason for this is that soda
ash mainly reacts with the primary formed melt phase, which composition is dependent on both
time and temperature. In order to describe the energy required for reactive calcination of soda
ash, it is assumed that the reactive calcination of soda ash can be described by
Na2 CO3 (s) + 2 SiO2 (s) Na2 O 2SiO2 (l) + CO2 (g).

(2.81)

For the determination of the calcination energy of the calcination reaction, this reaction is described by two serial reactions, viz. the formation of crystalline sodium disilicate
Na2 CO3 (s) + 2 SiO2 (s) Na2 O 2SiO2 (s) + CO2 (g),

(2.82)

and the melting of the crystalline sodium disilicate according to


Na2 O 2SiO2 (s) Na2 O 2SiO2 (l).

(2.83)

2.5. Calcination of a oat glass batch

69

The reaction enthalpy of reaction 2.82 in the temperature range from 973 K up to 1123 K equals
82.8 kJ per mol Na2 CO3 , whereas the melting of the crystalline sodium disilicate requires 35.6
kJ per mol Na2 CO3 . The energy required for reactive calcination of soda ash is now given by
H
T

=
Na2 CO3

118.3

500000
(1 Na2 CO3 ) e(54.0 R T ) ,

(2.84)

assuming that the reactive calcination of soda ash occurs in a CO 2 -atmosphere.


Total energy demand for the calcination of the oat glass batch
Figure 2.33 shows the simulated temperature dependent total energy demand as function of temperature for the oat batch listed in table 2.1. In contrast to gure 2.4, the total energy demand
is expressed in J kg1 instead of J kg1 . The energy demand of the oat batch is simulated
batch
melt
for a heating rate 20 K min1 in both 0 bar CO2 -atmosphere and an 1 bar CO2 -atmosphere.
During the simulation it is assumed that the dissolution of the oxides CaO, MgO and SiO 2 will
occur after the (reactive) calcination of the oat batch is nished.

PSfrag replacements

Total energy demand [J kg1 ]


batch

2.0x10

1.5x10

1.0x10

C
B

pCO2 =0 bar

pCO2 =1 bar
5.0x10

Onset temperature for


calcination reactions
0.0
400

600

800

1000

1200

1400

Temperature [K]

Figure 2.33: Simulated total energy demand as function of temperature during ramp heating with 20
K min1 of the oat batch with the composition as listed in table 2.1 in case p CO2 =0 bar
and pCO2 =1 bar. A indicates the end temperature for the calcination of the oat batch, B
indicates the enthalpy of the completely molten oat batch according to Madivate et al. [2]
and C indicates the enthalpy of the oat batch in case the dissolution of the oxides CaO,
MgO and SiO2 in the sodium silicate melt phase is not taken into account.

70

Chapter 2. Energy demand of glass forming batches

Up to a temperature of 973 K, the energy demand of the oat batch is determined by the
heat capacity of the raw material components of the oat batch. The onset temperature for
calcination of the oat batch equals 973 K at which the calcination of the MgCO 3 -part of the
dolomite starts. The curved lines show the change in energy demand of the oat batch in the
temperature range from 973 K up to 1183 K due to heating of the glass batch components
and the endothermic calcination reactions. During the simulations, at 1183 K the glass batch
contains a binary sodium silicate melt phase, crystalline MgO, CaO and SiO 2 and the CO2
which is released during the calcination reactions. The total energy demand of a completely
molten glass melt at 1443 K is xed (point B) and is derived from [2].
In case the solid oxides would not dissolve in the sodium silicate melt, the energy demand
of the oat batch would follow line 1 up to point C. The difference in energy demand of point C
and point D is the net energy which is released during the dissolution of the oxides and mixing
the glass melt, which equals about 0.1 106 J kg1 .
batch
It can be seen from gure 2.33 that the atmosphere composition does not have a large impact
on the temperature dependent energy demand of the oat glass batch. Also the effect of the
heating rate of the oat batch on the temperature dependent energy demand of the oat batch
(see gure 2.34) is only small.

PSfrag replacements

Total energy demand [J kg1 ]


batch

2.0x10

1.5x10

C
B

5 K min1
1.0x10

5.0x10

20 K min1
Onset temperature for
calcination reactions

0.0
400

600

800

1000

1200

1400

Temperature [K]

Figure 2.34: Simulated total energy demand as function of temperature during ramp heating of the oat
batch with the composition as listed in table 2.1 in case pCO2 =1 bar for both a heating rate of
5 K min1 and 20 K min1 . A indicates the end temperature for the calcination of the oat
batch, B indicates the enthalpy of the completely molten oat batch according to Madivate
et al. [2] and C indicates the enthalpy of the oat batch in case the dissolution of the oxides
CaO, MgO and SiO2 in the sodium silicate melt phase is not taken into account.

2.6. Concluding remarks

2.6

71

Concluding remarks

As discussed in section 2.1, the chemical energy demand of glass batches is to a large extent
dependent on the energy required for decomposition reactions, such as calcination reactions.
For the estimation of the time and temperature dependent chemical energy demand of a glass
batch, the kinetics of the different decomposition reactions are required. For the oat glass
batch, which was discussed in section 2.4, the calcination of dolomite, limestone and soda ash
occurred (almost) independently from each other. This allows the determination of the kinetics
of the individual calcination reactions to describe the calcination behavior of the complete oat
glass batch. Combination of the measured kinetics of the individual calcination reactions with
the calcination enthalpies, which were derived from thermodynamic tables [3], resulted in an
expression for the chemical energy required for complete calcination of the oat glass batch as
function of time, temperature and partial CO2 -pressure.
Kr mer [35] and Kawachi et al. [36] measured the release of batch and ning gases as funca
tion of temperature for a TV-panel glass batch. It was observed that the release of CO 2 occurs
in a broad temperature range from 700 K up to 1500 K with a maximum CO 2 release between
1100 K and 1200 K. To determine the chemical energy required for complete calcination of
a TV-panel glass batch as function of time, temperature and partial CO 2 -pressure, similar to
the oat glass batch, the kinetics of the different calcination reactions occurring during heating
of the TV-panel glass batch need to be determined. In appendix A, the calcination behavior
of a TV-panel glass batch, with as major constituents silica sand, soda ash, potash (K 2 CO3 ),
potassium nitrate (KNO3 ), nepheline (Na2 O Al2 O3 SiO2 ), strontium carbonate (SrCO3 ), barium carbonate (BaCO3 ), and zircon silicate (ZrO2 SiO2 ), is investigated.
In contrast to the oat glass batch, at which only soda ash decomposed via reactive calcination, during heating of a TV-panel batch, it appeared that soda ash, potash, strontium carbonate
and barium carbonate decompose via reactive calcination. These reactive calcination reactions
does not allow the easy determination of the kinetics of the individual calcination reactions in
the TV-panel glass batch. Therefore, in contrast to the oat glass batch, an expression for the
chemical energy required for complete calcination of the TV-panel glass batch as function of
time, temperature and partial CO2 -pressure cannot easily be determined.
Because the detailed analysis of the calcination rate of the complete TV-panel batch is timeconsuming, the chemical energy demand of the TV-panel batch is not easy to determine. Before
the kinetics of the individual calcination reactions in the TV-panel glass batch is studied in detail, it should be known how important a detailed description of the chemical energy demand is
for an accurate prediction of the heating process of a glass batch.
For the determination of the impact of the chemical energy demand on glass batch heating,
information of other glass batch properties which affect the heating of a glass batch is required.
In chapter 4, the heat conductivity of the glass batch will be discussed.

72

2.7

Chapter 2. Energy demand of glass forming batches

Nomenclature

Latin symbols
a
A
cp
ckc1
ckc2
d
Ea
f
f ()
G
GE
H
Hchem
Hr
g, i, j
k
Ka
Keq
l
m
M
nro
nb
nc
ngc
nms
nr
nrp
p
q
r
r
R
s
t
T
T0
Tnal
Tr
Tr,onset

activity
pre-exponential factor
heat capacity
kinetic compensation constant (equation 2.49)
kinetic compensation constant (equation 2.49)
diameter
(apparent) reaction activation energy
fugacity
reaction mechanism function
Gibbs free energy
excess Gibbs free energy
enthalpy
chemical energy demand
reaction enthalpy
indicators
reaction rate constant
ratio of the activities of reaction products and
reactants given by equation 2.30
reaction equilibrium constant
liquid phase
mass
molar weight
reaction order
number of glass batch components
number of carbonates in a glass batch
number of gas phase components
number of mixture species
number of reactants
number of reaction products
pressure
heat ux
reaction rate
apparent reaction rate
universal gas constant
solid phase
time
temperature
initial temperature of a glass batch
nal heating temperature of a glass batch and
the released gases
glass batch reaction temperature
onset temperature for batch reactions

[-]
[s1 ]
[J kg1 K1 ]
[m]
[J mol1 ]
[Pa]
[-]
[J mol3 ]
[J mol3 ]
[J mol3 ]
[J m3 ]
[J mol1 ]
[s1 ]
[-]
[-]
[kg]
[kg mol1 ]

[Pa]
[W m2 ]
[s1 ]
[s1 ]
[J K1 mole1 ]
[s]
[K]
[K]
[K]
[K]
[K]

2.7. Nomenclature

Tr,end
v
w
x
y

end temperature for batch reactions


velocity
weight fraction
mole fraction
mole fraction

73

[K]
[m s1 ]
[-]
[-]
[-]

Greek symbols

Gr
G0
r
H
Ht
H 0
S0
p

mass of gas formed that is released


per mass (1 + ) of glass batch
heating rate
Gibbs free energy of reaction
standard Gibbs free energy of reaction
enthalpy change
total energy demand of a glass batch
standard reaction enthalpy
standard reaction entropy
porosity
activity coefcient
heat (phonon) conductivity
chemical potential
density
stoichiometric reaction coefcient
degree of conversion

Sub- and superscripts


a
b
c
eff
eq
f
g
mn
r
sc
z
0

actual
backward
condensed phase
effective
equilibrium
forward
gas phase
mean
radiative
stoichiometric compound
vertical direction
initial

[-]
[K s1 ]
[J mol1 ]
[J mol1 ]
[J kg1 ]
[J kg1 ]
[J mol1 ]
[J mol1 ]
[-]
[-]
[W m1 K1 ]
[J mol1 ]
[kg m3 ]
[-]

74

2.8

Bibliography

Bibliography

[1] A. Ungan and R. Viskanta. Melting behavior of continuously charged loose batch blankets
in glass melting furnaces. Glastech. Ber., 59(10):279291, 1986.
[2] C. Madivate, F. M ller, and W. Wilsmann. Thermochemistry of the glass melting process
u
- energy requirement in melting soda-lime-silica glasses from cullet-containing batches.
Glastech. Ber. Glass Sci. Technol., 69(6):167178, 1996.
[3] O. Knacke, O. Kubaschewski, and K. Hesselmann. Thermochemical Properties of Inorganic Substances. Springer-Verlag Berlin, Heidelberg, Germany, 2 nd edition, 1991.
[4] D.E. Sharp and L.B. Ginther. Effect of composition and temperature on the specic heat
of glass. J. Am. Ceram. Soc., 34(9):260271, 1951.
[5] Manual FactSage 5.1. Centre for Research in Computational Thermochemistry, Ecole
Polytechnique (Universit de Montr al), Montreal, Quebec, Canada.
e
e
[6] C. Kr ger. Theoretischer W rmebedarf der Glasschmelzprozesse.
o
a
26(7):202214, 1953.

Glastech. Ber.,

[7] R. Conradt and P. Pimkhaokham. An easy-to-apply method to estimate the heat demand
for melting technical silicate glasses. Glastech. Ber., 63K:134143, 1990.
[8] B.A. Shakhmatkin, N.M. Vedishcheva, and C.A. Wright. Thermodynamic properties: A
reliable instrument for predicting glass properties. In Proc. Int. Congr. Glass, volume 1,
pages 5260, Edinburgh, Scotland, 1-6 July 2001.
[9] R.A. Van Santen and J.W. Niemantsverdriet. Chemical kinetics and catalysis. Lecture
notes, Eindhoven University of Technology, Eindhoven, The Netherlands, June 1992.
[10] H. Salmang and H. Scholze. Die physikalischen und chemischen Grundlagen der
Keramik. Springer-Verlag, Berlin/Heidelberg, Germany, 5 edition, 1968.
[11] A.K. Galwey and M.E. Brown. Application of the arrhenius equation to solid state kinetics: can this be justied? Thermochim. Acta, 386:9198, 2002.
[12] J. Opfermann. Kinetic analysis using multivariate non-linear regression I. Basic concepts.
J. Thermal Anal. Cal., 60:641658, 2000.
[13] G. Pokol. The thermodynamic driving force in the kinetic evaluation of thermoanalytical
curves. J. Thermal Anal. Cal., 60:879886, 2000.
[14] E.M. Levin, C.R. Robbins, and H.F. McMurdie. Phase Diagrams for Ceramics. The
American Ceramic Society, 1964.
[15] R. Conradt. A simplied procedure to estimate thermodynamic activities in multicomponent oxide melts. Molten Salt Forum, 5-6:155162, 1998.
[16] M.L. Pearce. Solubility of carbon dioxide and variation of oxygen ion activity in sodasilica melts. J. Am. Ceram. Soc., 47(7):342347, 1964.

Bibliography

75

[17] P.R. Laimb ck. Foaming of glass melts. PhD thesis, Eindhoven University of Technology,
o
1998.
[18] F. Paulik and J. Paulik. Investigations under quasi-isothermal and quasi-isobaric conditions by means of the derivatograph. J. Thermal Anal., 5:253270, 1973.
[19] D.A. Young. The international encyclopedia of physical chemistry and chemical physics,
Topic 21: Solid and surface kinetics, Volume 1: Decomposition of solids. Pergamon Press
LTD ., London, UK, 1st edition, 1966. Editor: P.C. Tomkins.
[20] B.V. Lvov. Mechanism and kinetics of thermal decomposition of carbonates. Thermochim. Acta, 386:116, 2002.
[21] P.K. Gallagher and D.W. Johnson. Kinetics of the thermal decomposition of CaCO 3 in
CO2 and some observations on the kinetic compensation effect. Thermochimica Acta,
14:255261, 1976.
[22] R. Ozao, M. Ochiai, A. Yamazaki, and R. Otsuka. Thermal analysis of ground dolomites.
Thermochimica Acta, 183:183198, 1991.
[23] M. Olszak-Humienik and J. Mozejko. Kinetics of thermal dolomite of dolomite. J. Thermal Anal. Cal., 56:829833, 1999.
[24] R. Conradt. Melting behavior of batches containing ground cullets. In Fundamentals of
Glass Science and Technology, pages 290297, V xj , Sweden, June 9-12, 1997.
a o
[25] P.K. Gallagher and D.W. Johnson. The effects of sample size and heating rate on the
kinetics of the thermal decomposition of CaCO3 . Thermochimica Acta, 6:6783, 1973.
[26] J.M. Criado and A. Ortega. A study of the inuence of particle size on the thermal decomposition of CaCO3 by means of constant rate thermal analysis. Thermochimica Acta,
195:163167, 1992.
[27] K. Kautz and Stromburg G. Untersuchungen der Vorg nge beim Einschmelzen von
a
Glasgemengen im Gradientofen. Glastech. Ber., 42(7):309317, 1969.
[28] C.A. Sheckler and D.R. Dinger. Effect of particle size distribution on the melting of sodalime-silica glass. J. Am. Ceram. Soc., 73(1):2430, 1990.
[29] F.W. Wilburn, S.A. Metcalfe, and R.S. Warburton. Differential thermal analysis, differential thermogravimetric analysis, and high temperature microscopy of reactions between
the major components of a sheet glass batch. Glass Technology, 6(4):107114, 1965.
[30] T.D. Taylor and K.C. Rowan. Melting reactions of soda-lime-silicate glasses containing
sodium sulphate. Comm. Am. Ceram. Soc., pages C227C228, 1983.
[31] C. Kr ger. Gemengereaktionen und Glasschmelze. Glastech. Ber., 25(10):307324, 1952.
o
[32] R. Conradt, P. Suwannathada, and P. Pimkhaokham. Local temperature distribution and
primary melt formation in a melting batch heap. Glastech. Ber. Glass Sci. Technol.,
67(5):103113, 1994.

76

Bibliography

[33] L. Riedel. Die Benetzung von Kalk und Quarz durch schmelzende Soda - Eine ph nomea
nologische Studie. Glastechn. Ber., 35(1):5356, 1962.
[34] M.E. Savard and R.F. Speyer. Effect of particle size on the fusion of soda-lime-silicate
glass containing NaCl. J. Am. Ceram. Soc., 76(3):671677, 1993.
[35] F. Kr mer.
a
Gasprolmessungen zur Bestimmung
Glasschmelzproze. Glastechn. Ber., 53(7):177188, 1980.

der

Gasabgabe

beim

[36] S. Kawachi, M. Kato, and Y. Kawase. Evaluation of reaction rate of rening agents.
Glastech. Ber. Glass Sci. Technol., 72(6):182187, 1999.

Chapter 3
Dissolution of sand grains during heating
of glass forming batches
3.1

Introduction

As mentioned by Hrma [1], the melting of glass batches is a complex process involving different reaction types such as dehydration reactions, crystalline inversions, solid-state reactions
between the different raw material grains, decomposition reactions, melt forming reactions and
dissolution processes. The dissolution of sand grains in the generated melts is regarded as the
most signicant criterion for the degree to which the melting of a glass batch based on silica
sand has advanced [2]. In general, during the glass melting process, two dissolution stages of
sand grains are distinguished, viz.:
1. the dissolution of sand grains in the batch blanket, and
2. the dissolution of sand grains in the bulk of the glass melt.
The rst dissolution stage concerns the dissolution of sand grains in the primary formed melt
phases during batch heating1 . Although the major part of a sand grain dissolves in the batch
blanket, a minor part of the sand grain may dissolve in the glass melt underneath the batch
blanket. The return ow of glass melt from the spring zone in the melting tank underneath
the bottom of the batch blanket (see gure 1.1 in chapter 1), may drag undissolved sand grains
from the batch blanket into the bulk of the glass melt. The major part of the experimental and
modelling studies of sand grain dissolution presented in literature, focusses on the description
of the dissolution of the sand grains in a large glass melt volume as function of time and temperature. These studies from literature represent the dissolution of sand grains in the bulk of the
glass melt in industrial glass melting furnaces. Primarily, this dissolution stage is investigated
because undissolved sand grains in the bulk of the glass melt are potential causes for product
reject in case the time-temperature trajectory of sand grains in the melting tank are insufcient
for complete or late dissolution. Because the objective of this thesis is to describe the behavior
of glass batches, this section focusses on the more complex description of the dissolution of
sand grains in the batch blanket.
1 The

primary formed melt phases are the rst melt phases formed during heating of a glass batch. The onset
temperature for primary melt phase formation, which is dependent on the composition of the glass batch, equals
approximately 1073 K for soda-lime-silica glass types (see section 2.5.3).

77

78

Chapter 3. Dissolution of sand grains during heating of glass forming batches

In section 3.2, a literature review of mathematical models is given describing the sand grain
dissolution during heating of glass batches. From the literature review, it is concluded that the
sand grain dissolution process is a fairly complex process of different simultaneously occurring processes, which depend on a large variety of glass batch properties such as for example
homogeneity of the glass batch, particle size (distribution) of the raw material components,
viscosity and surface tension of the glass melts formed during glass batch heating. Because it
is (almost) impossible to describe such a complex process, due to e.g. the lack of sensors to
monitor the processes or to measure different properties that determine the rate of this process,
it is a widespread practice to use simple approximate theoretical models [3] for the description
of such a complex process. An approximate theoretical model is a simplication of a detailed
rst-principles model, which can provide a detailed description of the process that is studied.
To derive an approximate theoretical model, simplications in both the governing equations and
the boundary conditions of the detailed rst-principles model are applied. The main advantage
of using an approximate theoretical model above the detailed rst-principles model for describing the dissolution of sand grains during glass batch heating is that the approximate theoretical
model provides a far more simple expression for the dissolution rate of a sand grain. A disadvantage of the approximate theoretical model is that, because of the assumptions that are made
to obtain the simple expression, the prediction of the dissolution rate of the sand grain is likely
to be less accurate than the predictions obtained with the numerical model.
The general form of an approximate theoretical model (see chapter 2) is given by
r = k f (),

(3.1)

in which r describes the rate of the process that is studied and k is the rate constant of the
process, which is in general given by an Arrhenius type equation given by
Ea

k = A eR T ,

(3.2)

in which A is a pre-exponential factor, Ea is an apparent reaction activation energy, R is the


universal gas constant and T is the temperature in K. The parameter f () in equation 3.1 is the
so-called reaction mechanism function. Table 3.1 lists the main generally applied reaction types
and their function f () derived from the reaction mechanism.
The reaction mechanism functions f () in table 3.1 characterize different types of processes
such as for example reaction kinetic limited reactions and mass transfer governed processes.
The most suitable approximate model describing the dissolution of sand grains during heating
of glass batches depends on the rate governing step of the dissolution process. In section 3.2, a
description is given of the dissolution mechanism of sand grain in glass batches during heating.
It appears that it is most likely that the dissolution of spherical sand grains 2 can be regarded as a
three dimensional diffusion governed process in the phase surrounding the sand grain. For this
type of process, two approximate models are listed in table 3.1, viz. the Jander model and the
Ginstling-Brounstein model3 . According to Frade and Cable [3], the GB-model is favored over
the Jander model. The reason for this is that in the derivation of the Jander model, equations
which are based on different shapes of the dissolving particle (planar and spherical) are combined to predict the degree of conversion of the spherical particle (see also Carter [5]). In this
2 Although

sand has different shapes, it is assumed for modelling of the sand grains dissolution behavior that
sand grains have a spherical shape.
3 Further in this chapter, the Ginstling-Brounstein model will be denoted as the GB-model.

3.1. Introduction

79

Table 3.1: Code, reaction type and reaction mechanism function (see for an overview of these reaction
mechanism functions [4]).
Code

Reaction type
th

f ()
(1 )nro

Fn

nro order reaction

D1

One-dimensional diffusion

1
2

D2

Two-dimensional diffusion

1
ln(1)

D3

Three-dimensional diffusion (Janders type)

1.5(1)2/3
1(1)1/3

D4

Three-dimensional diffusion (Ginstling-Brounstein type)

1.5
(1)1/3 1

R2

Two-dimensional phase boundary reaction

2(1 ) 1/2

R3

Three-dimensional phase boundary reaction

3(1 ) 2/3

chapter, the applicability of the approximate GB-model for the description of the dissolution of
sand grains during glass batch heating is studied.
The applicability of the GB-model for describing the dissolution of sand grains during glass
batch heating is studied by comparison of simulation results obtained with the GB-model with
the simulation results obtained with a more detailed numerical model describing the same process. In section 3.3, it is shown that the GB-model is not capable of predicting accurately the
degree of conversion of a sand grain dissolving in a surrounding medium in case the true values for the physical and chemical properties used in the GB-model are applied. However, it
is also observed that the shapes of the calculated conversion rate of the sand grain as function
of time with the GB-model and the more detailed numerical model are similar. This indicates
that a modied GB-model can be used for the prediction of the sand grain dissolution process,
although the values for the GB model parameters do not have a chemical and physical meaning
anymore.
To measure the degree of sand grain conversion as function of time and temperature, a quantitative analyzing technique is required. This technique should be able to determine the residual
amount of crystalline silica in the partly molten glass batches. In section 3.4, the method of
quantitative phase analysis with X-ray diffraction, which is used as analyzing technique for
measuring the residual crystalline silica content in partly molten glass batches, is described.
In section 3.5, the experimental determination of the GB-model parameters as function of
the initial particle size of the sand grain, and the cullet fraction in a oat glass batch is discussed. The oat glass batch, with which the experiments are performed contain silica sand,
soda ash and dolomite4 . Although minor batch components such as sodium sulphate, calumite 5
and water may have an impact on the sand grain dissolution process, the effect of these minor
components has not been studied in detail during this study.
4 Because
5A

of condentiality of the composition of the oat glass batch, no detailed batch recipe is provided.
steel slag often used as raw material during soda-lime-silica glass production.

80

Chapter 3. Dissolution of sand grains during heating of glass forming batches

3.2

Mathematical and experimental descriptions of the sand


grain dissolution process

In this section, a literature review is given of the mathematical description of sand grain dissolution during heating of glass batches. First, the mathematical description presented by
M hlbauer and Neme [6] and Beerkens et al. [7] is described, followed by the description
u
c
given by Hrma and co-authors [2, 810]. Next, a description is given of the dissolution mechanism of sand grains during the initial stage of glass batch heating. This description is based on
literature data and on own experimental results.

Dissolution models presented by Muhlbauer and Neme and Beerkens et al.


c

During heating of a glass batch, Muhlbauer and Neme [6] and Beerkens et al. [7] distinguished
c
three different dissolution stages in the reacting glass batch (see gure 3.1), viz.:
1. A reactive stage, during which the sand grains are consumed by a reaction with other
batch components such as the reactive calcination of soda ash, which is given by
Na2 CO3 (s,l) + SiO2 (s,l) Na2 O SiO2 (s,l) + CO2 (g).

(3.3)

2. A transient stage, during which the dissolution of the sand grains is determined by the
diffusion of SiO2 in the formed melt phase surrounding the sand grains. For this diffusion
governed process, the shrinkage of the sand grains is determined by the SiO 2 concentration gradient and the inter SiO2 diffusion coefcient in the melt phase at the surface of
the sand grain according to
s

CSiO2
rs
= DSiO2
t
r

(3.4)

r=rs

in which s is the constant density of the sand grain, rs is the radius of the sand grain,
t is the time, DSiO2 is the temperature dependent inter diffusion coefcient of SiO 2 in
the glass melt and CSiO2 is the time, temperature and position dependent concentration of
SiO2 expressed in kg m3 in the melt phase. In the transient stage, the SiO2 concentration
gradient in the melt phase at the surface of the sand grain depends on time.
3. A quasi-stationary stage, during which the dissolution of the sand grains is also determined by the SiO2 diffusion rate through the formed melt phase. However, in the quasistationary stage, the SiO2 concentration gradient in the melt phase at the surface of the
sand grain is (almost) independent on time.
According to Beerkens et al. [7], the shrinkage of a sand grain during heating of a glass batch
(see gure 3.2 for a schematic representation of the dissolution of SiO 2 in a glass melt surrounding the sand grain) is described by
s

rs
= h m,i wi m,b wb ,
t

(3.5)

in which h is the time- and temperature dependent effective mass transfer coefcient, m,i and
m,b are the temperature dependent density of the glass melt at the interface of the sand grain

in which 1/hr is the resistance for reactive sand grain dissolution and 1/h d is the resistance for
SiO2 diffusion through the formed melt phase.

For the diffusion governed stage, both Muhlbauer and Neme [6] and, Beerkens et al. [7]
c
distinguished between a mass transfer coefcient for the transient stage and a mass transfer co
efcient for the quasi-stationary stage. Muhlbauer and Neme [6] dened a characteristic time
c
1
1
1
= + ,
h hr hd

(3.6)

and in the bulk of the glass melt, respectively. The temperature dependent weight fraction of
SiO2 in the glass melt at the interface of the sand grain and in the bulk of the glass melt are
given by wi and wb , respectively.
The reciprocal of the effective mass transfer coefcient h characterizes the total resistance
of mass transfer of SiO2 from the surface of the sand grain at r = rs towards the bulk of the glass
melt. Similar to M hlbauer and Neme [6], Beerkens et al. [7] described the total resistance of
u
c
mass transfer, in case that reaction kinetic limitations as well as SiO 2 diffusion determine the
overall mass transfer of SiO2 into the bulk of the glass melt, as the sum of the resistance against
reactive sand dissolution at the interface of the sand grain and the resistance against diffusive
transport of SiO2 through the melt phase as is given by
Figure 3.1: Schematic representation of different sand grain dissolution stages during heating of glass
batches containing sand grains and soda ash grains. The sand grain is represented by A,
whereas the soda ash grain is represented by B. C represents the reaction product formed
after reactive dissolution of SiO2 according to equation 3.3. M represents a melt phase. The
curved lines in the melt phase in the lower gure represent the SiO 2 concentration proles in
the melt phase as function of time. CSiO2 ,rs is the SiO2 concentration at the surface of the sand
grain rs , which is constant for a constant ratio of Na2 O and CaO in the melt for a constant
temperature.

Diffusion
of SiO2

CSiO2 ,rs

Transient and
quasi-stationary
stage:

Reaction interface

PSfrag replacements
Reactive stage:

3.2. Mathematical and experimental descriptions of the sand grain dissolution process 81

Chapter 3. Dissolution of sand grains during heating of glass forming batches

m w

82

Sand grain

Glass melt

s
Interface sand grain with melt phase

m,i wi

PSfrag replacements

Bulk of the glass melt

m,b wb

rs

Figure 3.2: Schematic representation of the dissolution of SiO 2 in a glass melt surrounding a sand grain.

H for settling the quasi-stationary diffusion layer surrounding the silica grain and described the

reciprocal overall mass transfer coefcient by

1
1
1
= +
h hr hd

1 eH

(3.7)

Beerkens et al. [7] described the mass transfer coefcient for the diffusion process by
hd = DSiO2

1
+
rs

1
DSiO2 t

(3.8)

The rst right-hand-side term between brackets characterizes the quasi-stationary diffusion
stage, whereas the transient stage is described by the second right-hand-side term between
brackets. Equation 3.8 is based on the analytical description of the diffusive transport of SiO 2
from the surface of a spherical sand grain into a semi-innite glass melt volume described by
CSiO2
DSiO
CSiO2
r2
= 22
t
r r
r

(3.9)

3.2. Mathematical and experimental descriptions of the sand grain dissolution process 83

in which DSiO2 is the inter diffusion coefcient of SiO2 in the melt phase, which is taken independently from the melt phase composition. The derivation of this equation is given in section
3.3.1.
According to Beerkens et al. [7], the dissolution of sand grains in a glass melt can be regarded as a mass transfer process of SiO2 by diffusion enhanced by convective ow of the glass
melt. Free convective ow in the glass melt surrounding the sand grains is present in case
the glass melt density varies with the radial position in the melt phase. In general, glass melt
density is dependent on both glass melt composition and temperature. Therefore, a radial dependent glass melt density can be caused by a radial temperature and/or composition gradient
in the glass melt. The contribution of the free convective ow of the glass melt surrounding the
sand grain on the mass transfer coefcient is described by the Sherwood number Sh according
to
1
Sh 1
+
hd = DSiO2
.
(3.10)
2
rs
DSiO2 t
The Sherwood number characterizes the effect of both ow conditions and properties of the
glass melt on the mass transfer of SiO2 away from the sand grain interface. In case of mass
transfer without convection, the Sherwood number equals the value 2 and equation 3.8 is retained. In case of free and forced convective ow of the glass melt surrounding the sand grain,
the Sherwood number exceeds 2. The Sherwood number describing the contribution of free and
forced convective ow on the mass transfer process from a spherical particle is given by
Sh = 2 + 0.89 Re Sc + (Gr Sc3/4 )

1/3

(3.11)

in which Re, Sc and Gr are the Reynolds, Schmidt and Grasshof number, respectively. For a
detailed description and derivation of these dimensionless numbers is referred to Bird et al. [11].
Next to the contribution of convective ow of the glass melt on the mass transfer process of
SiO2 , Beerkens et al. [7] also take into account the effect of a so-called moving boundary on
the dissolution of sand grains6 . In case a moving boundary is taken into account, the Sherwood
number is corrected according to Ready and Cooper [12], resulting in
Sh =

Sh
(1

CSiO2 ,rs
CSiO2 ,s )

(3.12)

in which CSiO2 ,rs is the SiO2 concentration in the glass melt at the interface of the sand grain
and CSiO2 ,s is the SiO2 concentration in the sand grain. Combining equations 3.10, 3.11 and
3.12 results in equation 3.13 describing the mass transfer coefcient for SiO 2 transport in the
diffusive governed stage taking into account both SiO 2 diffusion, free and forced convective
ow of the glass melt and the effect of a so-called moving boundary:
2 + 0.89 Re Sc + (Gr Sc3/4 )
hd = DSiO2

CSiO ,i
(1 CSiO 2,s )
2

1/3

1
+
rs

1
DSiO2 t

(3.13)

Combination of equations 3.5, 3.6 and 3.13 results in an expression describing the shrinkage of
a sand grain dependent on the free and forced convective ow conditions in the surrounding melt
6 The

phenomenon of the moving boundary is explained in section 3.3.2

84

Chapter 3. Dissolution of sand grains during heating of glass forming batches

phase, the melt phase properties and the resistance towards reactive dissolution of sand grains
characterized by hr . Beerkens et al. [7] did not provide an explicit expression for the mass
transfer coefcient resistance hr for reactive sand grain dissolution. Based on experimental data
of residual crystalline silica as function of time and temperature in a reacting batch, Beerkens et
al. [7] proposed the empirically derived equation for the shrinkage of a sand grain in the glass
batch:
3
K2 T
rs 1 rs,0
=
K1 e K3 ,
(3.14)
2
t
3 rs
in which rs,0 is the initial radius of the sand grain and K1 , K2 and K3 are constants which depend
on the composition of the glass batch.
M hlbauer and Neme [6] determined the values for hr , hd and H from experimental data
u
c
for a oat glass batch. However, the procedure for the determination of these values is not
given. It is also not known whether these values are dependent on glass batch properties such
as particle size of the raw materials.
Dissolution model presented by Hrma
Hrma [9] described the dissolution process/reaction of sand grains in mixtures of sand and soda
ash during which ve different stages are distinguished:
1. An initial stage, which is controlled by a surface reaction of sand with soda ash during
which a sodium silicate melt phase is formed, which surrounds the sand grain.
2. A transient stage, which is controlled by both the surface reaction and non steady-state
diffusion of reactants and reaction products. During this stage, the thickness of the melt
phase around the sand grain increases with time. The transport of reactants and reaction
products in the melt phase is determined by diffusion and convective ow of the glass
melt surrounding the sand grain. The convection of the glass melt can either be caused
by free (buoyancy) convection due to density gradients in the glass melt or by forced
convection due to the effect of ascending gas bubbles on the melt.
3. A stationary stage, during diffusion and convection is the rate governing step. The SiO 2
concentration at the sand grain interface is in thermodynamic equilibrium with the SiO 2
in the sand grain.
4. A disappearance stage, during which the sand grain dissolution is enhanced by the reducing sand grain size.
5. A homogenization stage, during which local variations in SiO 2 concentration in the formed
melt phase are smoothed out.
For the initial stage, Hrma [8] described the silica dissolution during heating of a mixture of
silica sand and soda ash based on observations and identication of intermediate formed crystalline sodium silicates presented in literature. During the calcination of soda ash, Hrma assumes that molten soda ash reacts with silica sand grains according to
Na2 CO3 (l) + SiO2 (s) Na2 O SiO2 (s,l) + CO2 (g).

(3.15)

3.2. Mathematical and experimental descriptions of the sand grain dissolution process 85

The shrinkage of the silica grain during the calcination of molten soda ash can be derived from
the CO2 release given by
rs
Vs
= sa J
,
(3.16)
t
V CO2
in which sa is the silica grain surface fraction which is wetted by the molten soda ash, is
the stoichiometric reaction coefcient of reaction 3.15, J is the volume ux of evolving CO 2
per unit of grain surface which is wetted by the molten soda ash, V s is the molar volume of
solid silica and V CO2 is the molar volume of the released CO2 at the reaction temperature.
According to experimental studies (e.g. [1315]), the predominant stoichiometric crystalline
sodium silicate, which is formed in the temperature range between 973 K and 1173 K, is sodium
metasilicate (Na2 O SiO2 ). Now, describing the shrinkage of the sand grain requires the rate of
the calcination reaction of soda ash at the silica surface forming sodium metasilicate.
Finally, knowledge of the degree of wetting of the silica grains by molten sodium carbonate
and the sodium silicate melt phase would complete the information necessary for describing the
shrinkage of silica grains.
The wetting of sand grains by liquid phases is dependent on several parameters such as
the particle size of the soda ash particles, gas evolution, surface tension of the liquid phases
and humidity and homogeneity of the glass batch. The evolution of gases during heating of
glass batches may, on one hand, retard the dissolution of sand grains by preventing contact
between the sand grain and the melt phase. On the other hand, gas evolution may also enhance
the dissolution of sand grains by improved micro mixing of the primary formed melt phases
surrounding the sand grains. This results in an enhanced transport of SiO 2 away from the sand
grain surface. The wetting of the sand grains will also be improved in case that the surface
tension of the liquid phase formed during heating of the glass forming batch is lowered.
Humidication of the glass batch prior to heating may also enhance the dissolution rate of
sand grains by for example improving the contact of the sand grains with the soda ash layers.
During humidication of a glass batch, part of the soda ash will dissolve in the water. During
heating, the soda ash will recrystallize and thereby forming a ner distributed layer throughout
the glass batch. Local variations in the composition of the glass batch are likely because of:
segregation of batch components due to differences in the particle size(s) distribution of
the different glass batch components, and
the ascension of undissolved batch grains in the formed melt phases due to density differences or due to the ascension of undissolved batch grains with attached bubbles.
Because of the processes mentioned above, it is not likely that an unambiguous prediction of
the wetting behavior and the local homogeneity of the glass batch is possible.
According to Hrma, after complete calcination of the soda ash, which is given by equation
3.15, the remaining species in the melting glass batch are the primary formed melt phase, the
residual silica sand grains and the intermediate crystalline sodium silicates. The dissolution of
silica grains is now regarded as a diffusion governed process for which the shrinkage of the
silica grains is described by
wSiO2
rs
= s DSiO2
,
(3.17)
t

in which s is the silica grain surface fraction which is wetted by the liquid sodium silicate
melt, wSiO2 is the difference in SiO2 weight fraction at the surface of the silica grain and the

86

Chapter 3. Dissolution of sand grains during heating of glass forming batches

the outer boundary of the sodium silicate (melt) phase, and is the Nernst boundary layer thickness between the silica grain and the surface of the temporary (liquid) sodium metasilicate layer.

This equation is similar to the equation presented by Muhlbauer and Neme and Beerkens et al.
c
for the diffusion governed stage. Again, the wetting of the silica grains is an important parameter in the equation describing the shrinkage of the silica grains, which can hardly be predicted
beforehand.
Discussion of the conversion mechanism of silica sand during heating of glass batches
According to the mathematical models presented above, the conversion of sand grains during
heating of glass batches is determined by a combination of a reaction kinetic governed process
and a SiO2 diffusion governed process. In general, a chemical reaction can also be governed by
thermodynamic driving forces. For the sand grain conversion in a glass batch, it is easily proved
that thermodynamics of the reactions do not govern this process.
Based on thermodynamics (see section 2.5.3), in a 1 bar CO 2 -atmosphere, the stoichiometric sodium silicates Na2 O SiO2 and Na2 O 2SiO2 are likely to be formed as solid phases at
temperatures above 632 K by a reaction of soda ash with sand according to
Na2 CO3 (s) + 2 SiO2 (s)

Na2 O.2SiO2 (s) + CO2 (g),

(3.18)

Na2 O.SiO2 (s) + CO2 (g).

(3.19)

and
Na2 CO3 (s) + SiO2 (s)

At 1011 K, also the formation of sodium disilicate is thermodynamically favorable:


2 Na2 CO3 (s) + SiO2 (s)

2Na2 O.SiO2 (s) + 2 CO2 (g).

(3.20)

From thermogravimetrical analysis of mixtures of sand and soda ash, which are described in
section 2.4, it was observed that the onset temperature for soda ash calcination 7 , during heating
of a mixture of sand and soda ash in a N2 -atmosphere, is determined on approximately 1073
K. This indicates that the calcination of soda ash occurs far from thermodynamic equilibrium.
Therefore, it is concluded that this reaction is not determined by thermodynamics, but that either reaction kinetics and/or mass transfer limitations govern these conversions.
Kr ger and co-authors performed an extensive experimental study on the (solid-state) kineto
ics of soda ash calcination in both binary mixtures with sand grains and in ternary and quater
nary systems at different temperatures [1626]. Kroger observed that at temperatures at which
only solid state reaction products are thermodynamically stable, the rate of the solid-state reactions provided by reactions 3.18, 3.19 and 3.20 are slow. An enhanced calcination rate of soda
ash was observed after the formation of a liquid phase. Thermogravimetric analysis performed
during the current study (see section 2.5.3), showed a measured calcination onset temperature
for soda ash of 1073 K, which is almost equal to the eutectic melting temperature of sodium

disilicate (Na2 O 2SiO2 ) and SiO2 at 1072 K (see also gure 3.3). This agrees with Krogers
observation that the presence of a melt phase enhances the calcination of soda ash.
Assuming that the eutectic melting temperature of Na2 O 2SiO2 and SiO2 determines the
onset for soda ash calcination, it is expected that crystalline sodium disilicate (Na 2 O 2SiO2 ) is
present as intermediate crystalline specie during heating of glass batches containing silica sand
7 The

onset temperature for soda ash calcination, which is determined by thermogravimetrical analysis, is characterized by the observed onset temperature for weight loss of the soda ash grain during heating.

3.2. Mathematical and experimental descriptions of the sand grain dissolution process 87

1800
1700

L1

PSfrag replacements

Temperature [K]

1600
1500
1400
1300

L3

1200

L2

1100

w2

1000
0.5

0.6

w1
0.7

0.8

0.9

1.0

SiO2 mole fraction [-]

Figure 3.3: Part of the binary phase diagram of Na2 O and SiO2 for a mole fraction of SiO2 in the binary
system larger than 0.5. This phase diagram is redesigned after [27]. L 1 , L2 , and L3 are the
liquidus lines of SiO2 , Na2 O 2SiO2 and Na2 O SiO2 , respectively. w1 and w2 indicate the
composition of the melt phases formed after eutectic melting of SiO 2 with Na2 O 2SiO2 and
Na2 O 2SiO2 with Na2 O SiO2 , respectively. The values for w1 and w2 are 0.735 and 0.62,
respectively.

and soda ash. However, in the current study, sodium disilicate has neither been identied in
quenched samples of partly molten mixtures of soda ash and sand grains nor in partly molten
oat glass and TV-panel glass batches. Similar to this observation, during most melting experiments with glass batches containing sand and soda ash presented in literature (e.g. [1315]),
also no sodium disilicate was identied. Only during a few experimental studies (e.g. [28, 29]),
crystalline sodium disilicate or both crystalline sodium disilicate and sodium metasilicate were
identied at temperatures around 1073 K. From the results presented by Savard and Speyer [28],
it seems that the preference of either of these two sodium silicates is dependent on the particle
size of the raw materials. The reason that no sodium disilicate is found during the current study,
is probably that the conditions at which the melting experiments are performed (e.g. the particle
size of the raw material components) did not allow a sufcient amount of sodium disilicate to
be formed in order to be detected with X-ray diffraction. Phase analysis on heated samples of
mixtures of sand and soda ash with systematic variations of parameters such as particle size of
the reactants and heating rate, combined with microscopic observations of the heated samples
would reveal information on the detailed mechanism of interaction of sand and soda ash. However, this detailed information of the microscopic reaction mechanism of silica sand and soda
ash is not required to determine the rate of this reaction.

88

Chapter 3. Dissolution of sand grains during heating of glass forming batches

w=1
wSiO2 ,L1

wSiO2 ,L2

PSfrag replacements

wNa2 O,L2

wSiO2 (r)

wNa2 O (r)

wNa2 O,L1

w=0
NC

C
Figure 3.4: Simplied schematic representation of the different phases present during heating of a mixture containing silica sand and soda ash. NC, C, S and M indicate Na 2 CO3 , the intermediate
crystalline sodium disilicate, SiO2 and a melt phase, respectively. wi,L1 and wi,L2 indicate the
weight fraction of component i given by the liquidus lines L 1 and L2 , respectively (see gure
3.3).

At temperatures above the measured onset temperature for soda ash calcination (1073 K),
the primary formed melt phase separates the sand grain from a thin layer of intermediate formed
crystalline sodium disilicate. Assuming that at the surface of the silica sand grain and the
crystalline sodium disilicate phase thermodynamic equilibrium exist with the melt phase, the
reacting mixture of silica sand and soda ash can schematically be represented by gure 3.4.
The overall rate of the reaction of soda ash with sand grains is now determined by the following
three process steps:
1. The dissolution of silica at the interface of the silica sand grain with the sodium silicate
melt phase.
2. The diffusion of SiO2 from the surface of the sand grain through the melt phase towards
the intermediate crystalline sodium disilicate and the diffusion of Na 2 O in opposite direction.
3. The dissolution of the crystalline sodium silicate in the melt phase, which is assumed
to be instantaneously followed by the reactive calcination of soda ash at the interface of
crystalline sodium silicate.

3.2. Mathematical and experimental descriptions of the sand grain dissolution process 89

The conversion rate of the sand grain can now be determined by the measured calcination
rate of soda ash according to
0
ENa2 CO3
MSiO2 mNa2 CO3 wSiO2 ,m
SiO2
=
ANa2 CO3 e R T (1 Na2 CO3 ) ,
t
MNa2 CO3 m0
1 wSiO2 ,m
SiO2

(3.21)

in which SiO2 is the degree of conversion of the sand grain, MNa2 CO3 and MSiO2 are the molar
weights of soda ash and silica sand, respectively. The initial composition of the binary component glass batch is given by m0 2 and m0 2 CO3 . The mean SiO2 weight fraction in the melt
SiO
Na
phase formed above 1073 K, which is dependent on the SiO 2 concentration prole is the melt
phase, is given by wSiO2 ,m . The reaction kinetics parameters describing the reactive calcination
of soda ash are ANa2 CO3 and ENa2 CO3 , respectively (see section 2.5.3). The degree of calcination
of soda ash is given by Na2 CO3 . The conversion of silica after complete calcination of soda ash
is given by
m0 2CO3 MSiO2
wSiO2 ,m
Na
.
(3.22)
SiO2 =
0
mSiO MNa2CO3 1 wSiO2 ,m
2

The mean SiO2 weight fraction in the melt phase is at most equal to the maximum SiO 2
solubility given by the liquidus line of silica. The minimum value for the mean SiO 2 weight
fraction in the melt phase is equal to the SiO2 solubility given by the liquidus compositions of
the crystalline sodium silicates. The degree of conversion of silica sand in a oat glass batch, in
the temperature range during which still soda ash is present, is evaluated by quantitative phase
analysis of the amount of sand grains8 in partly molten oat glass batch samples. Figure 3.5
shows both the measured and the calculated conversion of the sand grains in the partly molten
oat glass batch as function of temperature.
According to the phase diagram of SiO2 and Na2 O, in the temperature range from 1072 K
up to 1110 K, the maximum and minimum SiO2 solubility in the melt phase is given by the
liquidus composition of silica (wL1 ) and sodium disilicate (wL2 ), respectively. Above 1110 K,
the minimum SiO2 solubility is given by the liquidus line of sodium metasilicate (w L3 ). From
gure 3.5, it is seen that the measured sand grain conversion at temperatures above 1110 K is
lower than is expected in case the sand grain conversion would have followed line C during
heating. This indicates that, assuming that recrystallization of SiO 2 from the sodium silicate
melt phase is not likely to occur, at temperatures below 1110 K, a melt phase is present with a
lower SiO2 weight fraction as is expected based on the SiO2 and Na2 O phase diagram.
Because the mean weight fraction of SiO2 in the melt phase is not equal to the SiO2 weight
fractions given by one of the liquidus lines, it is concluded that sand grain dissolution during
melting of glass batches is at least partly determined by the diffusion of silica through the pri
mary formed melt phase, which is supported by Rottenbacher and Engelke [30].
Based on the experimental evaluation of the mechanism of the sand grain dissolution process
in glass batches and based on literature, the dissolution process of a sand grain in a surrounding
glass melt is described as a diffusion governed process of SiO 2 from the surface of the sand
grain through the melt phase surrounding the sand grain. Assuming that the sand grain has a
spherical shape and that the sand grain is completely wetted by the melt phase, gure 3.6 shows
a schematic representation of the sand grain dissolution process in a primary formed melt phase.
8 For

the description of the quantitative analysis of the amount of sand grains in partly molten glass batches is
referred to section 3.4

90

Chapter 3. Dissolution of sand grains during heating of glass forming batches

PSfrag replacements

Degree of sand grain conversion [-]

0.5

A
0.4

0.3

0.2

C
B

0.1

0.0
1080

1100

1120

1140

1160

1180

Temperature [K]

Figure 3.5: Measured (blacks dots) and simulated (solid lines) conversion of the sand grains during heating a oat glass batch as function of temperature. A indicates the simulated sand grain conversion in case the average SiO2 weight fraction in the melt phase equals wL1 (T ) , B indicates
the simulated sand grain conversion in case the average SiO 2 weight fraction in the melt
phase equals wL3 (T ) and C indicates the simulated sand grain conversion in case the SiO 2
weight fraction in the melt phase equals wL2 (T ) .

The situations represented by A, B and C describe the dissolution of the sand grain at the presence of still undecomposed soda ash. Throughout the dissolution process of the sand grain, it is
assumed that at the surface of the sand grain, the SiO2 concentration is given by the temperature
dependent maximum solubility of SiO2 in the sodium silicate melt phase CSiO2 ,L1 (T) . As long as
the soda ash is not completely decomposed, the SiO2 concentration at the surface of the soda
ash grain is given by the temperature dependent solubility of SiO 2 CSiO2 ,L2 (T) . At the surface
of the sand grain, SiO2 dissolves in the sodium silicate melt phase and the sand grain size r s
decreases. At the surface of the soda ash grain, soda ash reacts with the sodium silicate melt
phase during which CO2 is released. Because of mass loss during the release of CO2 , the radius
of the total system rt decreases. After complete calcination of soda ash, the radius of the total
system remains constant. Because a melting glass batch can be regarded as a three dimensional
arrangement of sand grains dissolving in a surrounding glass melt as is indicated in gure 3.6,
the SiO2 concentration at the outer radius of the system can be approximated by a zero gradient
boundary condition. The concentration proles in the sodium silicate melt phase attens with
increasing time.
Concluding remarks
Because of the complex dissolution mechanism of sand grains in glass batches, Frade and
Cable [3] mentioned that a model with the correct mass balances together and the appropriate time-dependent boundary conditions, describing the time- and temperature-dependent sand

3.2. Mathematical and experimental descriptions of the sand grain dissolution process 91

CSiO2 ,s

CSiO2 ,s
CSiO2 ,L1

CSiO2 ,L1

CSiO2 ,L2

PSfrag replacements

II

III

rs ro

CSiO2 ,L2

II

I
rt

III
ro rt

rs

CSiO2 ,s

CSiO2 ,s
CSiO2 ,L1

CSiO2 ,L1

II

I
0

rs

CSiO2 ,L2

ro
C

II

I
0

rs

ro
D

Figure 3.6: Schematic representation of the sand grain dissolution process in a surrounding melt phase.
The x-axis represents the radial distance from the center of the sand grain and the y-axis
represents the SiO2 concentration, The sand grain, the melt phase and the soda ash grain
are indicated by I , II and III , respectively. The SiO2 concentration in the sand grain, at the
surface of the sand grain and at the surface of the soda ash grain is given by CSiO2 ,s , CSiO2 ,L1 ,
and CSiO2 ,L2 , respectively. Here, it is assumed that the soda ash grain is covered with a thin
layer of crystalline sodium disilicate. The sand grain size, the outer radius of the melt phase
and the outer radius of the total system are given by rs , ro and rt , respectively.

grain dissolution process is fairly complex. Because of the large number of parameters and processes, which inuence the sand grain dissolution process, a generally applicable microscopic
model describing the sand grain dissolution is at the moment not feasible. For the description
of such complex processes and phenomena, it is a widespread practice to use simple approximate theoretical models [3]. Table 3.1 listed the main generally applied approximate theoretical
models. The approximate models which agree the best with the dissolution mechanism of sand
grains described above, are the Jander and the Ginstling-Brounstein model, because these models are based on three dimensional diffusion governed mass transfer processes similar to the
dissolution of sand grains. According to Frade and Cable [3], the GB-model is favored over the
Jander model. The reason for this is that in the derivation of the different Jander model equations, which are based on different shapes of the dissolving particle (planar and spherical), are
combined to predict the degree of conversion of the spherical particle. Therefore, the Ginstling-

92

Chapter 3. Dissolution of sand grains during heating of glass forming batches

Brounstein is used as approximate model for the description of the dissolution of sand grains
during glass batch.
However, the Ginstling-Brounstein model contains several approximations, which are likely
to be invalid for the overall description of the dissolution process of sand grains during heating
of glass batches. In the next section, it is studied whether and to which extent, the GinstlingBrounstein is able to describe the sand grain dissolution process. The applicability of this
model is studied by comparison of simulation results of the GB-model with simulation results
obtained with a more detailed model. In the next section, rst the derivation and assumptions
of the Ginstling-Brounstein model are described. Section 3.3.2 describes the governing equations of the numerical sand grain dissolution model. In section 3.3.3, the applicability of the
GB-model is studied by comparison of the results of the GB-model with results from a more
detailed numerical model and with the results of experiments.

3.3

Evaluation of the application of the Ginstling-Brounstein


model

In this section, the applicability of the Ginstling-Brounstein model for the description of the
dissolution of sand grains during the melting of glass batches is studied. To evaluate and
interpret the applicability of the Ginstling-Brounstein model, the dissolution of a sand grain in
a surrounding nite melt phase volume is simulated with two mathematical models, viz.:
the Ginstling-Brounstein model, with which an approximate description of the sand grain
dissolution process can be simulated, and
a numerical model, describing the sand grain dissolution process with more complex
(real) melt properties and more realistic boundary conditions.
Section 3.3.1 describes the derivation of the Ginstling-Brounstein model. Section 3.3.2 describes the governing equations of the numerical sand grain dissolution model. The applicability of the Ginstling-Brounstein model for the description of the sand grain dissolution during
heating of glass batches is evaluated in section 3.3.3 by comparison of results of the numerical
model and the results of the Ginstling-Brounstein model and by the interpretation of the found
differences.

3.3.1 Derivation of the Ginstling-Brounstein model


The Ginstling-Brounstein model was developed for describing the kinetics of diffusion governed processes such as solid-state reactions [3, 13, 31]. The GB-model describes the shrinkage
of a spherical particle by dissolution in a innite surrounding phase. At the interface of the
spherical particle, a reaction layer is formed separating the spherical particle and the surrounding phase. The growth of the reaction layer is assumed to be governed by diffusion of reactants
and reaction products through the formed diffusion layer. Figure 3.7 gives a schematic representation of this diffusion governed process. The radius of the spherical particle is denoted
by rs,0 , whereas the outer radius of the reaction zone is given by rs . With time, the spherical
particle shrinks accompanied by an increase of the thickness of the reaction layer.
With the GB-model, a quasi steady-state solution for the diffusion governed process in a phase

3.3. Evaluation of the application of the Ginstling-Brounstein model

93

Diffusion layer
Innite phase

PSfrag replacements
Ci,rs,0

Spherical particle
composed of specie i

Ci,0

t
rs

t =0
t =t
CSiO2 ,rs
CSiO2 ,rt

rs

rs,0

rs,0

Figure 3.7: Schematic representation of the diffusion governed dissolution of a spherical sand grain in a
surrounding glass melt.

surrounding a spherical particle is calculated. In this section, the GB-model will be applied for
describing the dissolution rate of a sand grain in a nite shell of melt phase surrounding the sand
grain. With the description of this diffusion governed process, the dissolution of sand grains during the melting of glass batches will be simulated. The main advantage of the GB-model above
the numerical model presented in the next section, is that the GB-model will provide a far more
simple expression for the dissolution rate of a sand grain. A disadvantage of the GB-model may
be that, because of the assumptions that are made to obtain the simple expression, the prediction
of the dissolution rate of the sand grain is likely to be less accurate than the predictions obtained
with the numerical model.
The mathematical starting point of the GB-model, when applied to the dissolution of a sand
grain in a glass melt surrounding the sand grain, is the equation describing the transient diffusion
of SiO2 through the surrounding glass melt given by
CSiO2
2CSiO2 2DSiO2 CSiO2
+
= DSiO2
,
t
r2
r
r

(3.23)

in which CSiO2 is the weight concentration of SiO2 in the glass melt, t is the time, DSiO2 is the
inter diffusion coefcient of SiO2 and r is the radial position in the glass melt. The derivation
of the GB-model based on equation 3.23, is based on the following assumptions:

94

Chapter 3. Dissolution of sand grains during heating of glass forming batches

For the determination of the SiO2 concentration at the boundary of the


sand grain, which governs the dissolution rate of the sand grain, it is
assumed that the radius of the sand grain is constant with time.
The outer radius of the melt phase, rs,0 , is constant.
The SiO2 concentrations at the inner and outer radius of the reaction zone
CSiO2 ,rs and CSiO2 ,rs,0 remain constant.
The initial concentration of SiO2 in the melt phase, CSiO2 ,0 , is constant
throughout the melt phase.
The inter SiO2 diffusion coefcient in the melt phase is independent on
the local composition of the melt phase.
The density of the sand grain is similar to the density of the melt phase.

The analytical solution, describing the radial dependent weight concentration of SiO 2 in the
glass melt layer as function of time and taking into account the assumptions listed above, is
given by Carslaw and J ger [32]:
a
rs CSiO2 ,rs (rs,0 CSiO2 ,rb rs CSiO2 ,rs ) (r rs )
+
+
r
r (rs,0 rs )
2 rs,0 (1)n CSiO2 ,rs,0 rs CSiO2 ,rs
n(r rs )
+
sin

r n=1
n
rs,0 rs

CSiO2 =

n(r rs )
2 rs,0 (1)nCSiO2 ,0 + rs CSiO2 ,0
sin

r n=1
n
rs,0 rs

en
en

2 2

2 2

(3.24)

The symbol is the so-called Fourier number:


=

DSiO2 t
.
(rs,0 rs )2

(3.25)

The term (r SiO2 )2 is the reciprocal process time needed for a noticeable change in SiO 2 cons,0 rs
centration at the outer radius of the spherical glass melt shell r s,0 . If the time t is less than the
D
process time (r SiO2 )2 , the glass melt surrounding the sand grain can be regarded as an innite
s,0 rs
glass melt shell.
The equation describing the shrinkage of a sand grain in a nite melt phase surrounding the
sand grain, which is derived in the next section, is given by
C

SiO2
DSiO2
r
rs
r=rs
=
t
CSiO2 ,s CSiO2 ,rs

(3.26)

in which CSiO2 ,s is the SiO2 concentration in the sand grain. The shrinkage of the sand grain is
dependent on the radial concentration gradient of SiO2 at the surface of the sand grain. Although
equation 3.24 is only valid for non-shrinking cores, the GB-model uses this equation for the
determination of the radial concentration gradient of SiO 2 at the surface of the shrinking sand

3.3. Evaluation of the application of the Ginstling-Brounstein model

95

grain, which is given by


CSiO2
r

=
r=rs

2
rs (rs,0 rs )

n=1

rs,0 (CSiO2 ,rs,0 CSiO2 ,rs )


+
rs (rs,0 rs )

(3.27)

rs,0 (1)n (CSiO2 ,rs,0 CSiO2 ,0 ) + rs (CSiO2 ,0 CSiO2 ,rs ) en

2 2

In case the SiO2 concentration at the outer boundary of the glass melt equals the initial concentration of SiO2 throughout the glass melt, i.e. CSiO2 ,rs,0 = CSiO2 ,0 , equation 3.27 simplies
to
CSiO2
r

r=rs

(CSiO2 ,rs,0 CSiO2 ,rs )


=
rs

2rs
(rs,0 rs )

rs,0

en + (rs,0 rs)
2 2

(3.28)

n=1

The time dependent radial concentration gradient of SiO2 at the surface of the sand grain is
determined by two terms characterizing different diffusion stages. The rst term between the
square brackets in equation 3.28 characterizes the transient stage, which was mentioned by
both M hlbauer and Neme [6] and Beerkens et al. [7]. The latter term represents the quasiu
c
stationary stage during which the SiO2 concentration gradient in the melt phase at the surface of
the sand grain is (almost) independent on time. This indicates that for sufcient large values of
t, the time itself has no signicant effect on the concentration prole of SiO 2 at the sand grain
boundary. During this stage, the radial concentration gradient of SiO 2 in the glass melt at the
surface of the sand grain is mainly determined by the size of the residual sand grain.
2 2
r
2rs
Figure 3.8 shows the term (rs,0 rs ) en + (rs,0s,0 s ) versus the reciprocal value of
n=1
r
the square root of the time for the dissolution of a sand grain, with a grain size of 50 m, in
the surrounding glass melt layer with a thickness of 950 m. The inter diffusion coefcient is
set on 1 1012 m2 s1 . The time period during which the quasi-stationary stage predominates
the dissolution process is indicated by A. The transition period between the quasi-stationary
stage and the transient stage is indicated by B, whereas the time period during which the transient stage predominates the dissolution process is indicated by C. In the transient stage of the
dissolution process (C), the radial concentration gradient in the melt phase at the surface of the
sand grain is proportional to the reciprocal of the square root of the time. This is similar to the
transient term presented by Beerkens et al. [7]. In case the quasi-stationary stage predominates
the dissolution process (A), the radial concentration gradient in the glass melt at the surface of
the silica sand grain is independent of time.
To evaluate which dissolution stage predominates during heating of glass batches, the end
time of the transient stage as function of particle size and melt phase thickness is determined.
Dening the end time of the transient stage by the time at which the transient term in equation
3.28 is less than 1 % of the quasi-stationary term, the Fourier number at the end of the transient
stage, te , satises

2 2
1 rs,0
(3.29)
en te = 200 rs .
n=1
The time required for reaching the end of the transient stage is calculated from equation 3.25
and is given by
te (rs,0 rs )2
.
(3.30)
te =
DSiO2

96

Chapter 3. Dissolution of sand grains during heating of glass forming batches

+ (rs,0s,0 s ) [-]
r

1.35

PSfrag replacements

1.25

A B C

1.20
1.15

2rs
(rs,0 rs )

en
n=1

2 2

1.30

1.10
1.05
0.000

0.002

0.004

0.006

t0.5

Figure 3.8:

2 2

0.008

0.010

0.012

[s0.5 ]

en + (rs,0s,0 s ) versus the reciprocal value of the square root of the time.
n=1
r
A indicates the quasi-stationary stage, B indicates the transition period and C indicates the
transient range for sand grain dissolution in a surrounding melt phase in case the radius of
the sand grain equals 50 m, the thickness of the surrounding melt phase equals 950 m and
the inter diffusion coefcient equals 1 1012 m2 s1 .
2rs
(rs,0 rs )

Figure 3.9 shows the time required for reaching the end of the transient stage as function of the
thickness of the glass melt rs,0 rs surrounding the sand grain for both a radius of 50 m and
100 m. The value for te is calculated from equation 3.25, with a constant diffusion coefcient
of 1 1012 m2 s1 . It is clearly seen that the the duration time of the transient stage increases
with increasing thickness of the melt phase surrounding the sand grain. In case of thin glass
melt layers surrounding the sand grain, the dissolution process of sand grains mainly occurs in
the quasi-stationary stage. For a thickness of the surrounding melt phase of 25 m and 50 m
surrounding the sand grain with a radius of 50 m, the duration time of the transient stage equals
5 min and 20 min, respectively. This indicates that the duration time of the transient stage is
linear proportional to the square of the thickness of the melt phase.
Resuming, in case convection of the melt phase surrounding the sand grains is absent, according to the GB-model the dissolution of sand grains in a large volume of glass melt mainly
occurs in the transient stage. The dissolution of sand grains in the primary formed thin layers
of melt phase during melting of a glass batch, mainly occurs in the quasi-stationary stage. In
2
(r rs )
general, for the quasi-stationary stage, t > s,0SiO , equation 3.28 simplies to
D
2

CSiO2
r

=
r=rs

(CSiO2 ,rs,0 CSiO2 ,rs )


rs,0
.
rs
(rs,0 rs )

(3.31)

In case of a thick glass melt phase surrounding the silica sand grain and a very long residence

3.3. Evaluation of the application of the Ginstling-Brounstein model

97

PSfrag replacements

Duration time of the transient stage te [min.]

1200

1000

Dissolution process in
the quasi-stationary stage

800

600

400

Dissolution process in
the transient stage

200

0
0

100

200

300

400

500

Thickness of melt phase rs,0 rs [m]

Figure 3.9: The time required for reaching the end of the transient stage t e as function of the thickness
of the melt phase rs,0 rs surrounding the sand grain for both a radius of 50 m (solid line)
and a radius of 100 m (dotted line). The value for te is calculated from equation 3.25, with
a constant diffusion coefcient of 1 1012 m2 s1 .

time of the silica grain in the surrounding glass melt, equation 3.31 further simplies to
CSiO2
r

=
r=rs

(CSiO2 ,rs,0 CSiO2 ,rs )


.
rs

(3.32)

Then the concentration gradient in the melt phase at the surface of the sand grain is only dependent on the actual size of the silica sand grain, which was also described by Beerkens et al. (see
equation 3.8).
The GB-model describes the conversion rate of the silica sand grain during dissolution in a
nite glass melt surrounding the silica sand grain in the quasi-stationary stage. Combination of
equations 3.26 and 3.31 results in

in which is given by

1 rs,o
rs
= DSiO2
,
t
rs rs,o rs
=

Integration of equation 3.33 results in


2
2
rs rs,0

CSiO2 ,rs CSiO2 ,rs,0


.
CSiO2 ,s CSiO2 ,rs

3
3
rs,0
rs
+
= D t.
3 rs,0 3 rs,0

(3.33)

(3.34)

(3.35)

98

Chapter 3. Dissolution of sand grains during heating of glass forming batches

Expressing the degree of silica conversion by


= 1

rs
rs,0

(3.36)

and combination of equations 3.35 and 3.36 results in the GB-model equation:
2 DSiO2
2
1 (1 )2/3 =
t
2
3
rs,0

(3.37)

Resuming, the GB-model is an approximate model describing the conversion of diffusion governed solid-state reactions in a nite layer surrounding the spherical particle. In the GB-model,
the effect of the grains size on the SiO2 concentration proles is neglected. Then, the conversion of sand grains depends on the initial sand grain size, the inter SiO 2 diffusion coefcient
and the SiO2 solubility characterized by . To evaluate whether and when this model is capable of describing the dissolution of sand grains during heating of glass batches, the results of
simulations with the GB-model will be compared with the results of simulations with a more
realistic numerical model described in the next section.

3.3.2 Modelling of the dissolution of a single sand grain in a sodium silicate melt
For the mathematical description of the diffusion governed dissolution of sand grains during
heating of glass batches, the dissolution of a single spherical sand grain, which is completely
wetted by a nite shell of a sodium silicate melt, is considered. In this section, the governing
equations describing this dissolution process and the numerical solution method for these equations by the commercial nite element code SEPRAN is described.
Figure 3.10 shows a schematic representation of a sand grain surrounded by a sodium silicate
melt9 . The dissolution rate of the sand grain in the NS-melt, characterized by the time dependent weight change of the sand grain is, for a diffusion governed dissolution rate, determined
by the transient SiO2 -concentration gradient at the interface of the sand grain with the NS-melt
according to
mSiO2
CSiO2
= As DSiO2
,
(3.38)
t
r
r=rs
in which mSiO2 is the mass of the spherical sand grain, t is the time, As is the surface area of the
sand grain, which is assumed to be completely wetted with the NS-melt, D SiO2 is the time- and
temperature dependent inter diffusion coefcient of SiO 2 in the NS-melt expressed in m2 s1 ,
CSiO2 is the time- and position dependent SiO2 -concentration in the NS-melt and r is the radial
distance from the center of the sand grain.
In gure 3.10, the SiO2 concentration in the sand grain is denoted by CSiO2 ,s and equals the
density of the sand grain. At the interface of the sand grain and the binary glass melt at position
r = rs , it is assumed that thermodynamic equilibrium exists between the sand grain and the
NS-melt. This means that the SiO2 concentration at r = rs is given by
CSiO2 ,rs = m,rs wL1 .
9 Further

in this section, the sodium silicate melt is denoted as NS-melt.

(3.39)

CSiO2

3.3. Evaluation of the application of the Ginstling-Brounstein model

99

rs,0

rs

CSiO2 ,s
CSiO2 ,r=rs

PSfrag replacements
Sodium
silicate
melt

Sand grain

CSiO2
r

=0

rt

rs

Figure 3.10: Schematic representation of the diffusion governed process of silica dissolution in a sodium
silicate melt phase.

in which m,rs is the density of the NS-melt at r = rs and wL1 is the SiO2 weight fraction given
by the liquidus line of SiO2 in the NS-melt. In this model, it is also assumed that the glass
batch can be regarded as a three dimensional arrangement of spherical sand grains surrounded
by a melt phase. Then, the radial gradient of the SiO2 concentration at the outer radius of the
NS-melt equals zero:
CSiO2
r

= 0.

(3.40)

r=rt

The concentration gradient of SiO2 at the surface of the sand grain can be determined by solving
the continuity equation for SiO2 in the NS-melt given by
CSiO2

+ ( SiO2 ) = 0,
n
t

(3.41)

in which SiO2 is the mass ux of SiO2 through the NS-melt expressed in kg m2 s1 . The
n
mass ux vector of SiO2 is given by

n SiO2 = CSiO2 SiO2 ,


v

(3.42)

100

Chapter 3. Dissolution of sand grains during heating of glass forming batches

in which SiO2 is the average velocity vector of SiO2 in the NS-melt. The velocity of SiO2
v
through the NS-melt is composed of a convective and a diffusive contribution according to

j SiO2

v SiO2 =
+ ,
v
CSiO2

(3.43)

in which j SiO2 is the SiO2 mass ux relative to the mass-average velocity of the NS-melt,
v
and describes the diffusion of SiO2 through the NS-melt. Convection of the NS-melt, which is

indicated by the mass-average velocity , is caused by melt expansion. Melt expansion results
v
from three phenomena, viz.:
a difference in specic volume of SiO2 in the silica grain and in the melt phase,
a difference in specic volume of dissolved SiO2 and dissolved Na2 O, which results in a
radial density gradient in the melt phase, and
an inhomogeneous temperature distribution during non-isothermal dissolution.
Although the glass melt density is likely to be dependent on the glass melt composition (see e.g.
Shartsis et al. [33]), in the numerical model the glass melt density is set constant. A constant
value for the glass melt density in the numerical model is required to be able to validate the
GB-model in which the glass melt density is set constant as well .
In the following, the dissolution behavior of a sand grain in a NS-melt phase at constant
temperature is evaluated by numerical simulation of the diffusion of SiO 2 through the NSmelt following an instantaneous reaction of dissolution of silica at the silica interface. Because
only the isothermal dissolution of silica is investigated by numerical simulations, the effect of
melt phase expansion by temperature change on the convection of the NS-melt is excluded.
According to Bird [11], the diffusive mass ux of SiO 2 through the binary melt phase is given
by
CSiO2

,
(3.44)
j SiO2 = m DSiO2

in which m is the density of the melt phase. Because the density of the binary melt phase is
dened to be independent on the composition of the binary melt phase, equation 3.44 simplies
to

j SiO2 = DSiO2 CSiO2 .


(3.45)
Combination of equations 3.41, 3.42, 3.43 and 3.45 results in
CSiO2

= ( CSiO2 ) + ( DSiO2 CSiO2 ) .


v
t

(3.46)

For a spherical silica grain and in the case that no concentration gradients of SiO 2 in the - and
-direction are present, the transient concentration prole of SiO 2 in the binary melt phase is
given by
CSiO2
CSiO2
1
1
r2 DSiO2
,
(3.47)
= 2
CSiO2 r2 vr + 2
t
r r
r r
r

3.3. Evaluation of the application of the Ginstling-Brounstein model

101

in which vr is the radial convective ow of the binary melt phase. The convective radial velocity
of the NS-melt is derived from a mass balance of the system of both the sand grain and the NSmelt surrounding the sand grain. This mass balance is given by
(sVs + mVm,r )
= 0,
t

(3.48)

in which s is the constant density of the silica grain, Vs is the volume of the silica grain, m
is the constant density of the melt phase and Vm,r is the volume of the melt phase at position r.
Because both the density of the solid silica grain and the binary melt phase are assumed to be
constant during isothermal dissolution, equation 3.49 describes the expansion rate of the melt
phase as a function of the shrinkage of the silica particle during isothermal dissolution of silica
in the binary melt phase.
rs 2 rs
m s
(3.49)
vr =
m
rt
t
The shrinkage rate of the sand grain is calculated by setting up a mass balance over the sand
grain as is shown in gure 3.11. The dissolution of the sand grain is given by equation 3.38,
which describes the change in mass of the sand grain as function of the radial concentration
gradient of SiO2 in the NS-melt at the interface of the sand grain and the NS-melt, the inter
diffusion coefcient of SiO2 in the NS-melt and the surface area of the sand grain. For the
current model, it is assumed that, for a small time-step, a fresh thin NS-melt layer is formed
with a constant composition (Area B in gure 3.11). This composition equals the liquidus
composition of SiO2 in the sodium silicate melt phase CSiO2 ,rs . This implies that the amount of
SiO2 , which diffuses from the surface of the silica grain into the binary melt phase is represented
by Area A in gure 3.11.
The mass balance of SiO2 is now given by
mSiO2
t

+
grain

mSiO2
t

nfm

= As DSiO2

CSiO2
r

(3.50)

,
r=rs

in which the subscript nfm indicates the new formed melt phase. Rewriting equation 3.50 results
in
2
4 rs CSiO2 ,s

rs
t

2
+ 4 rs CSiO2 ,wL1

rs
t

2
= 4 rs DSiO2

CSiO2
r

(3.51)

r=rs

From equation 3.51, an expression for the shrinkage of the silica grain is derived:
C

SiO2
DSiO2
r
rs
r=rs
=
t
CSiO2 ,s CSiO2 ,wL1

(3.52)

The simulation of the dissolution process of the silica grain in the sodium silicate melt phase,
by solving equations 3.47, 3.49 and 3.52, requires values for the model parameters D SiO2 , s ,
m , CSiO2 ,s and CSiO2 ,wL1 . Similar to the glass melt density, the inter diffusion coefcient is dependent on the composition of the glass melt [14,34]. Because in the GB-model a constant inter
SiO2 diffusion coefcient is used, in the numerical model the inter SiO 2 diffusion coefcient
is set constant. The temperature dependent SiO2 solubility in the NS-melt is derived from the

Chapter 3. Dissolution of sand grains during heating of glass forming batches

CSiO2

102

CSiO2 ,s
A: Amount of silica which
dissolves in the melt
at r > rs (t).

PSfrag replacements
CSiO2 ,rs

B: Amount of silica which


forms a melt with
CSiO2 =CSiO2 ,rs .

Silica grain

CSiO2 (r)
CSiO2
r

rs (t = t + t)

r=rs

ro

rs (t = t)

Figure 3.11: Schematic representation of dissolution of SiO2 in a sodium silicate melt phase. The concentration of SiO2 in the silica grain is CSiO2 ,s . The concentration of SiO2 at the interface of
the silica grain and binary melt phase is CSiO2 ,s .

commercial thermodynamic software package FACT [35] and is tted with a 2 nd order polynomial. The concentration of SiO2 in the sand grain equals the density of the sand grain and
is constant. Assuming that the density of the silica grain equals the density of the melt phase,
s = m = , equation 3.47 further simplies to
CSiO2
CSiO2
1
r2 DSiO2
= 2
t
r r
r

(3.53)

Resuming, the isothermal dissolution of silica in the sodium silicate melt phase is simulated by
solving equation 3.53. The shrinkage of the silica grain is given by equation 3.26. The boundary
conditions are given by equations 3.54 and 3.55, whereas the initial condition given by equation
3.56.

3.3. Evaluation of the application of the Ginstling-Brounstein model

CSiO2 (rs (t),t) = wSiO2 ,L1


CSiO2
r

(3.54)

(rt ,t) = 0

CSiO2 (r, 0) =

103

(3.55)

CSiO2 ,L2 , r = rs,0


CSiO2 ,0 , rs,0 < r rt

(3.56)

The equations listed above are implemented in the nite element code SEPRAN. Equation 3.53
is solved for this domain using a grid consisting of 200 nodes. The time-step for solving this
equation equals 103 s.

3.3.3 Applicability of the Ginstling-Brounstein model


In the previous section, the GB-model describing the dissolution of a sand grain in a nite glass
melt layer surrounding the sand grain, under the consideration of several assumptions, was
derived. Because the assumptions of the GB-model are not generally valid for the dissolution
of sand grains during the melting of glass batches, the applicability of the GB-model for predicting the sand grain dissolution rate is questioned.
In this section, the applicability of the GB-model is evaluated by comparison of the results
of the numerical simulation of the sand grain dissolution process with the results derived from
the GB-model.
Table 3.2 lists the values for the model parameters of the numerical and the GB-model, viz.
the initial radius of the sand grain rs,o , the SiO2 concentration in the sand grain CSiO2 ,s , the SiO2
concentration in the melt phase at the interface with the sand grain CSiO2 ,rs , the initial SiO2 concentration throughout the melt phase CSiO2 ,0 , the inter diffusion coefcient of SiO2 in the melt
phase DSiO2 and the parameter , which are used during simulation studies.
Table 3.2: Values for the model parameters of the numerical and the GB-model.
Parameter

Unit

rs,0

[m]
[kg m3 ]
[kg m3 ]
[kg m3 ]
[m2 s1 ]
[-]

CSiO2 ,s
CSiO2 ,rs
CSiO2 ,0
D

Value
35
2200
1668
1440
1 1012
0.43

As an example of a result obtained from a numerical simulation of the sand grain dissolution
process, gure 3.12 shows both the radius and the degree of conversion of a sand grain with an
initial radius of 35 m surrounded by a melt phase with an initial thickness of 25 m as function
of time. It is observed that the sand grain completely dissolves in the surrounding melt phase
within 1625 s.

Chapter 3. Dissolution of sand grains during heating of glass forming batches

35

1.0

Radius of the sand grain rs [m]

30

PSfrag replacements

0.8
25
0.6

20
15

0.4

10
0.2
5
0
0

250

500

750

1000

1250

1500

1750

Degree of conversion of sand grain [-]

104

0.0
2000

Time [s]

Figure 3.12: Calculated radius and degree of conversion of a sand grain with an initial radius of 35
m surrounded by a melt phase with an initial thickness of 25 m (For the other model
parameters is referred to table 3.2) as function of time.

According to the GB-model, the plot of 1 2 (1 )2/3 versus time would give a constant
3

2
slope equal to 2 D rs,0 (see equation 3.37). To study the applicability of the GB-model for
describing the sand grain dissolution process during heating of glass batches, the results ob2
tained from numerical simulations studies are reported by the plot of 1 3 (1 )2/3 versus time. In gure 3.13, the results of numerical simulations, using the model parameter values
listed in table 3.2, for ve different values of the initial thickness of the melt phase surrounding
the sand grain (10 m (case 1), 20 m (case 2), 25 m (case 3), 465 m (case 4) and 965 m
(case 5)) are shown. Also in gure 3.13, the result of the simulation using the GB-model is
shown.
At t = 0, the value for 1 2 (1 )2/3 equals zero. Complete dissolution of the sand
3

2
grain is represented by = 1 and results in a value for 1 3 (1 )2/3 equal to 1/3. It
is observed that for case 1, no complete dissolution of the sand grain in the surrounding melt
phase is reached, i.e. = 1/3. For this case, the initial amount of melt phase surrounding the 35
m sand grain is insufcient for complete dissolution of the sand grain. Complete dissolution
of the sand grains is observed for the cases 2, 3, 4 and 5, which indicates that for these cases the
initial amount of melt phase surrounding the sand grain is sufcient for complete dissolution.
It is also observed that an increase of the thickness of the melt phase surrounding the sand
grain from 20 m to 465 m results in an increase in the dissolution rate of the sand grains.
However, a larger value for the initial melt phase thickness than 465 m is not accompanied
with a further increase in dissolution rate. The reason for this is explained as follows:

3.3. Evaluation of the application of the Ginstling-Brounstein model

1 2 (1 )2/3 in case = 1
3

0.35

GB-model

1 2 (1 )2/3 [-]
3

0.30

PSfrag replacements

105

0.25

Case 3

Case 4, 5

Case 2

0.20
0.15
0.10
0.05

Case 1
A B

0.00
0

200

400

600

800

1000 1200 1400 1600 1800 2000

Time [s]

2
Figure 3.13: Plot of 1 3 (1 )2/3 versus time representing the calculated conversion of a sand
grain with an initial radius of 35 m during dissolution in a surrounding glass melt according
to the Ginstling-Brounstein model and the numerical model. The initial thickness of the
melt phase surrounding the sand grain equals 10 m (case 1), 20 m (case 2), 25 m (case
3), 465 m (case 4) and 965 m (case 5). A and B indicate the end time for the transient
stage for case 1 and case 2.

Figure 3.14 shows a schematic representation of the SiO2 concentration proles in the glass
melt surrounding a sand grain with an initial radius of 45 m as function of time. These concentration proles are calculated with the numerical model. At t = 0, the surface of the sand
grain is given by rs (t = 0), which equals 45 m. From the surface of the sand grain, at which
the SiO2 concentration is constant and given by CSiO2 ,rs =1668 kg m3 , SiO2 diffuses through
the melt phase towards the outer boundary of the melt phase at r = rt , which equals 53 m
in this example case. The initial SiO2 concentration in the glass melt CSiO2 ,0 equals 1100 kg
m3 . With time, the SiO2 concentration proles in the glass melt follow the proles indicated
in gure 3.14. The concentration prole in the glass melt in case the time equals the Fourier
DSiO
number (r r 2)2 , which is also referred as the end time of the transient stage t e . When the time t
t
s
is smaller than the Fourier number, no change in the SiO2 concentration at the outer boundary
of the glass melt is observed, whereas at t > te , SiO2 accumulates at the boundary of the melt
phase. As long as t < te , the initial thickness of the melt phase does not affect the dissolution
process. For both an initial glass melt thickness of 465 m and 965 m, the dissolution rate calculated with the numerical model is identical, which indicates that for both cases the dissolution
time t is less than te and that the surrounding melt phase can be regarded as a semi-innite melt
phase.

106

Chapter 3. Dissolution of sand grains during heating of glass forming batches

1700

1600

CSiO2 ,rs
CSiO2 [kg m3 ]

1500

1400

PSfrag replacements

1300

rs,0

rt

1200

CSiO2 ,0
1100
3.4

3.6

3.8

te
4

4.2

4.4

4.6

4.8

Radial position [m]

5.2

5.4
5

x 10

Figure 3.14: Schematic representation of the SiO2 concentration proles in the glass melt as function
of time. The parameter te represents the end time of the transient dissolution stage. In
case t < te , the SiO2 concentration at r = rt is constant, whereas in case t > te , the SiO2
concentration at r = rt increases.

In section 3.3.1, the equations with which the duration time of the transient stage t e can be
calculated are given. The duration time for the transient stage for the different values for the
melt phase thickness are 3 min. (=20 m), 5 min. (=25 m), 964 min. (=465 m) and 3064
min. (=965 m), respectively. The high values for the duration time of the transient stage
longer than the total dissolution time agree with the assumption that t < t e for the cases 4 and
5. For the cases 2 and 3, the duration time are plotted in gure 3.13 and marked by A and B,
respectively. At the end of the transient stage, the SiO2 concentration proles start to deviate
from the curves for cases 4 and 5.
In gure 3.13, it is clearly seen that the results of the simulations obtained from the numerical model deviate signicantly from the simulation result obtained from the GB-model. The
reason for these large deviations is caused by the difference in SiO 2 concentration gradient at
the surface of the sand grain as function of time for the two models. As can be seen from gure
3.14, the concentration proles in the glass melt surrounding the sand grain attens with time.
The attening of the concentration proles is caused by two effects:
1. The stretching of the concentration prole at the surface of the sand grain due to the
shrinkage of the sand grain.
2. The accumulation of SiO2 at the outer boundary of the glass melt.
In contrast to the numerical model, in the GB-model it is assumed that both boundaries of the
melt phase, rs and rt , remain constant during the dissolution process. It is also assumed that the

3.3. Evaluation of the application of the Ginstling-Brounstein model

107

SiO2 concentration at these boundaries are constant. Therefore, the SiO 2 concentration proles
at the boundary of the sand grain will be steeper in the GB-model compared to the numerical
model. Combining equations 3.31 and 3.36 results in an expression for the SiO 2 gradient at the
surface of the sand grain for the GB-model:
CSiO2
r

=
rs

CSiO2 ,rs,0 CSiO2 ,rs

rs,0 (1 )1/3

(3.57)

1 (1 )1/3

Figure 3.15 shows the concentration gradient at the surface of a sand grain with an initial grain
size of 35 m surrounded by a melt phase calculated with both the numerical model and the
GB-model during the rst second of the dissolution process. The values for CSiO2 ,s , CSiO2 ,rs ,
CSiO2 ,0 , D and are listed in table 3.2. It is observed that the SiO 2 concentration gradient calculated with the GB-model is steeper than the SiO2 concentration gradient calculated with the
numerical model. This explain the deviation of the degree of conversion of the sand grain calculated with the numerical model relative to the result of the GB-model (see gure 3.13).

0.0

rs

-2.0x10

-2.5x10

-3.0x10

-3.5x10

-4.0x10

CSiO2
r

-1.5x10

PSfrag replacements

-1.0x10

[kg m4 ]

-5.0x10

-4.5x10

-5.0x10

GB-model
Numerical model

0.0

0.2

0.4

0.6

0.8

1.0

Time [s]

Figure 3.15: Concentration gradient at the surface of a sand grain with an initial grains size of 35 m
surrounded by a melt phase calculated with both the numerical model and the GB-model
during the rst second of the dissolution process. The values for CSiO2 ,s , CSiO2 ,rs , CSiO2 ,0 , D
and are listed in table 3.2.

From gure 3.13, it is observed that during the dissolution of sand grains in large glass melt
2
volumes, which is for example characterized by case 4 and case 5, 1 3 (1 )2/3 is
(almost) linear proportional to time, which is similar to the GB-model. Also for case 3, an
almost linear relation is observed. The proportionality constant for the GB-model, relating

108

Chapter 3. Dissolution of sand grains during heating of glass forming batches

0.0035
0.0030

knm = kGB

0.0025

2
rs,0

knm [-]

0.0020

0.0015

0.0010

PSfrag replacements
0.0005
0.0000
0.000

0.002

0.004

kGB =

0.006
2D
rs,0

0.008

0.010

[-]

Figure 3.16: knm versus kGB as function of , D and rs,0 calculated for case 3 (see gure 3.13).

2
1 2 (1 )2/3 versus time and dened as kGB , equals 2 D rs0 , whereas the propor3
tionality constant for the numerical model, dened as knm , is given by the slope of the lines
shown in gure 3.13. In case the difference of kGB with respect to knm is independent on the
values for , D and rs,0 , a modied GB-model can be dened, with which the sand grain dissolution process, simulated with the numerical model can be approximated. In this modied
GB-model, the proportionality constant kGB is multiplied by the constant knm /kGB . However,
for sand grains dissolving in a very thin glass melt layer, the ratio of k nm /kGB is non-linear,
which can for example be seen from the curved line for case 2 in gure 3.13. In case the
amount of glass melt surrounding the sand grain is insufcient for complete dissolution of the
sand grain (see case 1 in gure 3.13), the GB-model is not able to predict the dissolution rate of
the sand grain.
To evaluate whether the modied GB-model is able to predict the sand grain dissolution
as function of the GB model parameters , D and rs,0 , the effect of these parameters on the
value for knm as function of kGB is studied for case 3. Figure 3.16 shows the results of a systematic variation of these model parameters. The used values for these model parameters are
D=1 1012 m2 s1 , 2 1012 m2 s1 and 5 1012 m2 s1 , =0.43, 2.83 and 5.60, and rs,0 =10
m, 15 m, 25 m and 35 m.
It is observed that the absolute difference between kGB and knm decreases with increasing
initial sand grain radius, decreasing diffusion coefcient, decreasing value for , i.e. with decreasing SiO2 solubility in the glass melt. It is also observed that knm /kGB is independent of

3.4. Quantitative phase analysis with X-ray diffraction

109

2
D and rs,0 . Only for very high (and non-realistic values for the glass melting process 10 ) values
for , no linear relation between knm and kGB is observed. Excluding high values for , the
linear relation between knm versus kGB indicates that the sand grain dissolution calculated with
the modied GB-model can be used to estimate the sand grain dissolution process as calculated
with the numerical model. However, the ratio of knm /kGB will be a function of the ratio of the
thickness of the melt phase surrounding the sand grain, /r s,o , which becomes non-linear for
low ratios of /rs,0 . In practice, the value for /rs,0 will depend on a large series of parameters
such as the bulk composition of the glass batch, the homogeneity of the glass batch, the wetting
of the sand grains by the primary formed melt phase, the presence of gas bubbles preventing
contact between glass melt and sand grain. Because of this complexity, the applicability of the
modied GB-model for the description of sand grain dissolution in industrial glass batches is
questioned.

In the section 3.5, it is studied whether the modied GB-model can still be used for the description of the dissolution of sand grains in a multi-component glass batch. The modied GB
model proportionality constant kGB is determined from non-isothermal dissolution experiments
of sand grains in a oat glass batch. The non-isothermal dissolution of sand grains as function
of temperature is described by

kGB
=
f () .
(3.58)
T

The reaction mechanism function f () for the GB-model is derived by differentiation of equation 3.37, which results in
1.5
.
(3.59)
f () =
(1 )1/3 1

Assuming that the temperature dependency of the GB model proportionality constant k GB is


given by an Arrhenius type equation, the apparent Arrhenius parameters A GB and EGB are derived from

EGB
ln
+ ln ln f () = ln AGB
.
(3.60)
T
RT
To measure the degree of sand grain fusion in glass batches as function of time and temperature,
a quantitative analyzing technique is required. This technique should be able to determine the
residual amount of crystalline silica in partly molten glass batch samples. In the next section,
the method of quantitative phase analysis with X-ray diffraction is described.

3.4

Quantitative phase analysis with X-ray diffraction

To determine the fusion rate of sand grains in glass batches during heating, the content of residual crystalline silica in glass batch samples, which have been imposed to a dened time/
temperature-program, is measured. For the preparation of the heated glass batch samples, about
5 grams of a mixture of glass batch components is heated in an electrical laboratory furnace.
After ramp heating of the batch sample up to a specied temperature, the heated glass batch
10 A

value for equal to 5.60 indicates a SiO2 weight fraction in the glass melt of 0.95 in case the initial SiO2
weight fraction in the melt equals 0.65.

110

Chapter 3. Dissolution of sand grains during heating of glass forming batches

sample is withdrawn from the furnace and rapidly cooled 11 . After cooling, the glass batch
sample is crushed with a tungsten carbide mortar resulting in a powder mixture with a residual
particle size below 10 m. This small particle size is required for accurate quantitative phase
analysis with X-ray diffraction. To the crushed batch sample, a known amount of an internal
standard, which is discussed later in this section, is added. From the mixture of the crushed
glass batch sample and the internal standard, a tablet is pressed, which is used for quantitative
phase analysis by X-ray diffraction.
Figure 3.17 shows the diffraction pattern of silica sand in the quartz structure for 5 0 <2<800 .
This diffraction pattern is collected using a Philips XPert X-ray diffractometer using Cu-K
radiation. The data are collected from 50 up to 800 in steps of 0.020 using a counting time of
1.0 s step1 . These diffractometer settings are used for all qualitative and quantitative phase
analysis during the current study.
Identication of the phases present in a mixture of crystalline species is achieved by comparison of the X-ray diffraction patterns from the unknown sample with an internationally recognized database containing a large amount of reference patterns.
Next to qualitative phase analysis with X-ray diffraction, also quantication of the amount of
crystalline species is possible with X-ray diffraction. As mentioned by Madsen et al. [36], there
are several methods based on X-ray diffraction, which can be applied for quantitative phase
analysis. The method used for the quantitative determination of the amount of crystalline SiO 2
in this study, is the internal standard Reference Intensity Ratio (RIR) method. As mentioned
above, each crystalline specie can be characterized by its typical diffraction pattern. Figure
3.17 depicts the diffraction pattern of silica sand, which shows the intensities of the different
diffraction lines as a function of the diffraction angle 2.
The intensity of the diffraction line i of quartz particles in a multi-component mixture is described by
Ke Ki,SiO2 (q)
s
Ii,SiO2 (q) =
xSiO2 (q) ,
(3.61)
s
SiO2 (q)
in which Ke is a constant for the particular X-ray diffraction system, Ki,SiO2 (q) is a constant
for diffraction line i for quartz, s is the mass absorption coefcient of X-rays in the multicomponent mixture, s is the density of the multi-component mixture, SiO2 (q) is the density
of quartz and xSiO2 (q) is the weight fraction of quartz in the mixture. For the determination of
the required weight fraction of quartz (xSiO2 (q) ) in the multi-component mixture, the so-called
mass attenuation coefcient of the mixture sample, (s /s ), has to be known. The wavelength
dependent mass attenuation coefcient is given by

i=1

xi ,

(3.62)

in which n equals the total number of different species in the multi-component mixture. Thus,
determination of the weight fraction of quartz in the multi-component mixture requires information of the composition of the mixture, which is not known a-priori. To overcome this problem,
several approaches have been presented in literature, from which the so-called Internal Standard
Method is applied during the current study. This method is based on the elimination of the mass
attenuation coefcient of the multi-component mixture in equation 3.61 by relating the intensity
11 For

a detailed description of the melting procedure of the glass batch samples is referred to section 2.4.3.

3.4. Quantitative phase analysis with X-ray diffraction

111

30000

Peak intensity [counts s1 ]

25000

20000

15000

10000

5000

PSfrag replacements

10

20

30

40

50

60

70

80

Diffraction angle 2 [o ]

Figure 3.17: XRD-pattern for silica sand for 50 <2<800 . The different lines indicate the different
diffraction lines for quartz.

of the diffraction lines of quartz to the intensity of the diffraction lines of an internal standard 12 ,
which has been added to the multi-component mixture in a known amount. In this study, pure
silicon (Si) powder is used as internal standard material.
The ratio of the intensity of diffraction line i of quartz is now related to the intensity of the
diffraction line j of the internal standard Si according to
Ii,SiO2
Ki,SiO2 Si xSiO2
xSiO2
xSiO2
=
=K
= RIRSiO2 (q),Si
,
I j,Si
K j,Si SiO2 xSi
xSi
xSi

(3.63)

in which the constant K is the Reference Intensity Ratio (RIRSiO2 (q),Si ) of SiO2 (q) with respect
to Si. A mixture containing quartz to which a known amount of the internal standard is added
permits quantitative analysis of the amount of quartz in this multi-component mixture by measuring the peak intensities of SiO2 (q) and Si and applying equation 3.63. The RIR of SiO 2 (q)
with respect to Si, RIRSiO2 (q),Si , is derived from mixtures of SiO2 and Si with known compositions.
The accuracy of the determined Reference Intensity Ratio is dependent on the accuracy of
the measured intensities of the diffraction lines and on the accuracy of the weight fractions of
SiO2 (q) and Si in the binary standard mixtures. According to Jenkins and Snyder [37], there are
four types of factors that affect the measured intensity of a diffraction line:
1. Crystal structure-sensitive factors.
2. Instrument-sensitive factors.
12 A

compound with a well known X-ray diffraction pattern.

112

Chapter 3. Dissolution of sand grains during heating of glass forming batches

3. Sample-sensitive factors.
4. Measurement-sensitive factors.
The crystal structure-sensitive factors are specic for each diffraction line i for each specie s
and are described by the diffraction line constant Ki,s as given in equation 3.61. In case no crystal structural changes occurs in quartz during heating of the glass batch samples, the relative
intensities of the diffraction lines of quartz remain unaltered and structure-sensitive factors do
not contribute to errors in the calculated weight fraction of quartz.
The main instrument-sensitive factors are the intensity of the incident beam and the wavelength of the X-ray beams. These factors affect the absolute diffraction line intensities, which
means that the intensities of the diffraction lines of both SiO 2 (q) and Si are modied to the
same extent. Because for quantitative phase analysis of quartz, the diffraction line intensities
of SiO2 (q) are always measured with respect to the intensities of the diffraction lines of Si, the
contribution of instrument-sensitive factors to an inaccurate measurement of the diffraction line
intensities can also be neglected.
Sample-sensitive factors are able to inuence both absolute and relative diffraction line intensities. The main sample-sensitive factors are preferred orientation, extinction and microabsorption. Preferred orientation is the phenomenon that a crystal exhibits a preferred orientation of crystal planes within the sample during preparation of the tablet for X-ray diffraction
analysis. There are sample preparation methods and instrumental techniques to reduce preferred
orientation. During the current study, preferred orientation is opposed by spinning the sample
tablet during the diffraction measurement.
Extinction is the phenomenon that the intensity of the most intense diffraction lines of a
crystal decreases by interference of diffraction X-rays with reected X-rays from upper lattice
planes. Extinction mainly occurs in perfect crystals. In glass batches containing industrial raw
materials, extinction is not expected to have a signicant effect on the X-ray diffraction lines of
thermal treated glass batch samples, because the glass batch components are not perfect crystals.
Micro-absorption of X-rays in a sample occurs
in case a large difference in absorption coefcient of the different phases in the sample
are present, and
in case a large difference in crystal size of the different phases in the sample are present.
Micro-absorption can be reduced by grinding the sample as ne as possible, but should at least
be down to a grain size below 10 m.
The main important measurement-sensitive factors are the denition of the onset of the
diffraction line (i.e. A in gure 3.18 for diffraction line 2 for silica sand), the end of the diffraction line (i.e. B in gure 3.18 for diffraction line 2 for silica sand) and the baseline of the
diffraction line, which gives the area to be integrated (i.e. C in gure 3.18 for diffraction line 2
for silica sand).
The error in the weight fraction quartz calculated from X-ray diffraction patterns by using
equation 3.63, is dependent on both systematic and random errors in the different parameters
listed in equation 3.63. The presence of random errors is prevented as much as possible by
using the same instrument settings, sample preparation method and peak integration procedure
for all diffraction measurements. The accuracy of the weight fraction of quartz in the glass

3.4. Quantitative phase analysis with X-ray diffraction

113

2
4

PSfrag replacements

Peak intensity [counts s1 ]

10

1
3

10

10

C
1

10

A
20

30

40

[o ]

Diffraction angle 2

Figure 3.18: XRD-pattern for silica sand for 150 <2<450 . A is the onset of diffraction line 2, B is the
end of diffraction line 2, C is the baseline of diffraction line 2.

batch sample is mainly dependent on the random errors induced by sample- and measurementsensitive factors. According to Coleman and Steele [38], in case a parameter Y is dependent on
n independent parameters X, the uncertainty or error in the parameter Y due to random errors in
the n parameters X is given by
Y =

i=1

Y
Xi

X 2 ,
i

(3.64)

in which Y is the uncertainty or error in the parameter Y and Xi is the uncertainty in the
parameter Xi . The uncertainty in the parameter Xi is given by the value of the standard deviation
in the parameter Xi .
To calculate with equation 3.63 the weight fraction of quartz in heated glass batches to
which a known amount of internal standard is added, the RIR SiO2 (q),Si has to be known. In
order to reduce the uncertainty in RIRSiO2 (q),Si , it is recommended to use multiple diffraction
lines of both quartz and the internal standard Si. For each diffraction line of quartz and the
internal standard Si, the relative peak intensity has been determined from the pure individual
components. The relative peak intensity of a diffraction line i is the peak intensity of this
diffraction line with respect to the most intense diffraction line for that substance. For quartz the
most intense diffraction peak is positioned at 2 = 26.6 o , whereas the most intense diffraction
peak for internal standard Si is positioned at 2 = 28.4 o . The relative diffraction peak intensities
for quartz and Si are listed in table 3.3 and table 3.4, respectively. The standard deviation of the
relative peak intensity of a diffraction line, I rel
or I rel , is determined from the measured
i,SiO2 (q)

j,Si

114

Chapter 3. Dissolution of sand grains during heating of glass forming batches

Table 3.3: Relative peak intensities for quartz.


Diffraction line
1
2
3
4
5

Iirel

2
20.87
26.62
36.51
39.43
40.26

I rel [%]
i

0.150
1.000
0.097
0.085
0.050

0.1
0.2
0.2
0.1

Table 3.4: Relative peak intensities for the internal standard Si


Diffraction line

I rel
j

I rel [%]

1
2
3

28.44
47.30
56.12

1.00
0.98
0.65

0.8
0.7

peak intensities of the diffraction line for ten different samples of either silica sand or the internal
standard Si.
The Reference Intensity Ratio for quartz with respect to the internal standard Si can now be
calculated from a quartz tablet to which a known amount of the internal standard Si is added:
1 n
RIRSiO2 (q),Si =

n i=1

Ii,SiO2 (q)

1
m

rel
Ii,SiO2 (q)

j=1

I j,Si
I rel
j,Si

XSi
XSiO2 (q)

(3.65)

in which n and m are the numbers of diffraction lines of quartz and the internal standard Si,
rel
which are used for the determination of the RIRSiO2 (q),Si . Ii,SiO2 (q) and I rel are the relative peak
j,Si
intensities of line i of quartz and line j of the internal standard Si with respect to the most
intense diffraction line and xSiO2 (q) and xSi are the weight fractions of quartz and the internal
standard in the mixtures of quartz and Si with known composition.
The RIRSiO2 (q),Si has been determined by plotting the quotient of the average weighted intensities of the quartz and Si diffraction lines versus the ratio of the weight fractions of quartz
and Si for three binary mixtures of quartz and Si according to
RIRSiO2 (q),Si =
in which I SiO2 (q) equals

1 n
n i=1

Ii,SiO2 (q)
rel
Ii,SiO (q)
2

I SiO2 (q)

XSi

I Si

XSiO2 (q)

and I Si equals

(3.66)

1 m
m j=1

I j,Si
I rel
j,Si

. The results are

plotted in gure 3.19. The slope of the line, which equals RIR SiO2 (q),Si , is 0.89, with a regression
coefcient of 0.9988. In a similar way as presented in equation 3.64, the standard deviation in
the RIRSiO2 (q),Si is determined, which resulted in a value of 0.04. So, the Reference Intensity
Ratio RIRSiO2 (q),Si equals 0.890.04.
From equation 3.63, the weight fraction of quartz in the batch sample diluted with the internal standard can be calculated. Equation 3.67 provides the weight fraction of quartz in the
batch sample before dilution with the internal standard and in case multiple diffraction lines are

3.4. Quantitative phase analysis with X-ray diffraction

115

9
8

I SiO2 (q) /I Si [-]

7
6
5
4
3
2

PSfrag replacements

1
1

10

xSiO2 (q) /xSi [-]

Figure 3.19: The ratio of the average diffraction line intensities versus the ratio of the weight fractions
for quartz and Si in three binary mixtures.

used for both quartz and the internal standard.


xSiO2 (q) =

I SiO2 (q)

RIRSiO2 (q),Si

I Si

xSi
1 xSi

(3.67)

For three non-heated mixtures of silica sand, soda ash and dolomite, the weight fraction of
quartz in these mixtures is determined with X-ray diffraction and compared with the known
weight fraction of quartz in the ternary mixtures. Figure 3.20 shows the calculated weight
fraction of quartz, the 95 % condence limits of these values and the known weight fraction of
quartz. The calculated weight fractions of quartz fall clearly within the condence limits which
has been calculated with equation 3.64.
The standard deviation in the calculated quartz content equals 5 % relative. This value will
be taken as the uncertainty in the calculated SiO2 content with X-ray diffraction. The degree of
fusion of the silica sand in the heated glass batch samples can now be calculated with
SiO2 (q) = 1

xSiO2 (q) m0
b
,
0
xSiO (q) mb

(3.68)

0
in which xSiO2 (q) is the initial weight fraction of quartz in the glass batch sample, m 0 is the initial
b
mass of the glass batch sample and mb is the actual mass of the glass batch sample.
As was shown in section 2.5.5, starting at about 1300 K, SiO 2 is not only present in the
quartz modication, but also in the cristobalite modication. To determine the total content
of crystalline SiO2 in heat-treated glass batch samples, also the Reference Intensity Ratio of
SiO2 (cr) with respect to Si has been determined.

116

Chapter 3. Dissolution of sand grains during heating of glass forming batches

Known weight fraction of quartz [-]

80

PSfrag replacements

75

70

65

60

55

50
45

50

55

60

65

70

75

Calculated weight fraction of quartz [-]

Figure 3.20: Calculated SiO2 weight fraction versus real SiO2 weight fraction in a glass batch sample
consisting of silica sand, soda ash and dolomite.

3.5

Experimental determination of the apparent GB-model


parameters

In this section, the apparent Ginstling-Brounstein model parameters A GB and EGB are determined for the dissolution of silica in a glass batch producing oat glass. The GB model parameters are easily determined from measured conversion of silica during heating of the oat
glass batch. The values for the GB model parameters for the oat glass batch are determined
as function of the initial particle size of the sand grains in the oat batch and the cullet fraction
in the oat batch.

3.5.1 Apparent GB-model parameters as function of the sand grain particle size
Figure 3.21 shows the measured degree of conversion of silica sand in the oat glass batch as
function of temperature during ramp heating with 25 K min 1 in an ambient atmosphere. Line
1 and line 2 indicate the measured degree of silica conversion for particle sizes of the silica
grains between 212 m and 300 m and between between 63 m and 150 m, respectively. The
onset for silica conversion in the oat glass batch is observed at approximately 1023 K, which
is almost similar to the onset temperature for reactive soda ash calcination (see section 2.5.3).
Complete conversion of the silica sand grains is observed at 1723 K.

3.5. Experimental determination of the apparent GB-model parameters

117

PSfrag replacements

Degree of silica conversion [-]

1.0

0.8

0.6

0.4

0.2

0.0
1000

1100

1200

1300

1400

1500

1600

1700

1800

Temperature [K]

Figure 3.21: Measured degree of silica conversion during ramp heating of a oat batch with 25 K min 1
in an ambient atmosphere. Line 1 and line 2 present the measured silica conversion as
function of temperature for particle sizes of the silica grains between 212 m and 300 m
and between between 63 m and 150 m, respectively.

+ ln ln f () [ln min1 ]

-3

1673 K

-4

-5

1173 K

-6

-7

ln

PSfrag replacements

1073 K

-8
0.55

0.60

0.65

0.70

0.75

0.80

0.85

Reciprocal absolute temperature

0.90

[ 103

0.95

K1 ]

Figure 3.22: ln T + ln ln f () characterizing the experimentally derived conversion of silica


grains in the oat glass batch as function of temperature during ramp heating with 25 K
min1 in an ambient atmosphere and in case the particle size of the silica grains varies
between 212 m and 300 m.

118

Chapter 3. Dissolution of sand grains during heating of glass forming batches

It is observed that a decrease of the particle size of the sand grains results in a slightly higher
conversion rate of the sand grains in the oat batch. The reason for this is the larger surface area
per unit volume of the sand grains, in case of smaller sand grains, which promotes the rate of
the diffusion controlled dissolution of the sand grains in the primary formed melt phases during
heating of the oat batch.
From the combination of equations 3.58 and 3.59, the non-isothermal conversion of silica
sand during heating of glass batches can be derived
EGB

1
1.5
= AGB e R T
,
T

(1 )1/3 1

(3.69)

in which is the degree of silica conversion, T is the temperature in K, is the applied constant
heating rate and AGB and EGB are the apparent GB model parameters. Figure 3.22 shows the

1.5
, versus the reciprocal absoplot of ln T + ln ln f () , with f () given by
1/3
(1)

lute temperature in the temperature regime between 1073 K and 1723 K. It is observed that

above 1173 K, an (almost) linear relation between ln T + ln ln f () and the reciprocal


absolute temperature is present, which allows the calculation of the apparent GB model parameters.
-3.0

-3.5

+ ln ln f () [ln min1 ]

ln AGB = 0.324; EGB /R = 6256; r2 = 0.9861

-4.0

-4.5

PSfrag replacements

ln

-5.0

-5.5

-6.0
0.55

ln AGB = 0.583; EGB /R = 7608; r2 = 0.9749

0.60

0.65

0.70

0.75

Reciprocal absolute temperature

0.80

[ 103

0.85

K1 ]

Figure 3.23: ln T + ln ln f () characterizing the experimentally derived conversion of silica


grains versus the reciprocal absolute temperature for two different silica grain particle sizes.

The and the indicate ln T + ln ln f () in case 212 m < dSiO2 < 300 m and

63 m < dSiO2 < 150 m, respectively. The heating rate equals 25 K min 1 .

The apparent GB model parameters AGB and EGB , are determined with equation 3.60 for the

3.5. Experimental determination of the apparent GB-model parameters

119

two oat glass batches with different particle size of the silica grains. Figure 3.23 shows the

plot of ln T + ln ln f () versus the reciprocal absolute temperature starting from 1173


K. The intercept and the slope of the lines in gure 3.23 provide the values for ln A GB and
EGB /R, respectively. The experimental derived values for the apparent GB model parameters
are 3.0 102 s1 and 63.3 kJ mol1 in case 212 m < dSiO2 < 300 m and 2.3 102 s1 and
52.0 kJ mol1 in case 63 m < dSiO2 < 150 m.
In case the modied GB-model is valid for the description of the sand grain dissolution pro2
cess in the oat glass batch, knm would be proportional to D rs,0 . Because the composition
of both glass batches is identical, both the inter SiO2 diffusion coefcient D and the parameter
are assumed constant. Now, it is expected that knm is proportional to the square of the reciprocal
initial sand grain radius rs,0 :
kGB,ds
ln
kGB,dl

rs,0,dl
= ln
rs,0,ds

(3.70)

in which kGB,ds is the proportionality constant for the oat glass batch with the small particles,
kGB,dl is the proportionality constant for the oat glass batch with the large particles, r s,0,dl is the
average initial radius of the large sand grains and rs,0,ds is the average initial radius of the small
sand grains. However, because of the broad particle size distribution, no unambiguous value
for the initial radius of the sand grains in both oat glass batches can be given. The measured
k

rs,0,d

is between 0.7 and 3.2. Based on


difference in ln kGB,ds equals 0.7. The value for ln rs,0,dl
GB,dl
s
these results, the applicability of the GB-model for the description of the sand grain dissolution
process in glass batches is not proved unambiguously. To test whether the modied GB-model
is able to predict the effect of particle size on the dissolution rate of sand grains, it is recommended to perform melting experiments with a more narrow particle size distribution despite
that fact that industrial glass batches may contain a broad particle size distribution for the sand
grains.
The accuracy of the prediction of the degree of silica conversion in the oat batch for different heating rates is evaluated by comparison of the measured and the predicted (calculated
with equation 3.69) silica conversion at 1323 K as function of the heating rate of the oat batch
containing sand grains with 212 m < dSiO2 < 300 m and in case the values for AGB and EGB
are 3.0 102 s1 s1 and 63.3 kJ mol1 , respectively. The measured degree of silica conversion at 1223 K for the different heating rates is dened as the starting point for the simulation
of the silica conversion at 1323 K. The measured values for the silica conversion at 1223 K, the
measured values for the silica conversion at 1323 K and the predicted values, from the modied
GB-model and the measured degree of conversion at 1323 K, for the silica conversion at 1323
K are listed in table 3.5.
The predicted and the measured degree of silica conversion at 1323 K in case of a heating
rate of 25 K min1 are identical, because the apparent GB model parameters describing the
dissolution rate of the sand grains in the oat batch are derived from the measured conversion
of silica in the oat glass batch when ramp heated with 25 K min 1 . It is observed that at low
heating rates the GB-model overestimates the silica conversion, whereas at high heating rates,
the silica conversion is underestimated. Similar to the prediction of the effect of the particle size
of the sand grains described above, the effect of heating rate on the dissolution of sand grains is
not accurately described by the modied GB-model.

120

Chapter 3. Dissolution of sand grains during heating of glass forming batches

Table 3.5: The measured silica conversion at 1223 K and 1323 K and the predicted silica conversion at
1323 K as function of the heating rate of the oat glass batch.
Heating rate
K min1

Measured
at 1223 K [-]

Measured
at 1323 K [-]

Calculated
at 1323 K [-]

5
10
25
35

0.41
0.34
0.33
0.32

0.63
0.56
0.48
0.46

0.78
0.62
0.48
0.44

3.5.2 Apparent GB-model parameters as function of the cullet fraction


Figure 3.24 shows the measured degree of conversion of silica sand during ramp heating of
the oat glass batch with 25 K min1 in an ambient atmosphere as function of the content of
crushed cullet in the oat glass batch.
It is observed that the onset for silica conversion in the oat glass batch increases with
increasing cullet content. The reason for the shift in onset temperature for silica conversion to
higher values is the lower additional solubility of SiO 2 in the oat glass cullet compared to the
primary formed melt phase. The liquidus temperature, by the crystallization of SiO 2 , of the
glass melted from the oat glass batch is calculated with FACT [35] and equals 1309 K. This
indicates that up to 1309 K, no silica will dissolve in the oat glass cullet.

PSfrag replacements

Degree of silica conversion [-]

1.0

0.8

0.6

0.4

2
3

0.2

4
0.0
1000

1100

1200

1300

1400

1500

1600

1700

1800

Temperature [K]

Figure 3.24: Measured degree of silica conversion during ramp heating a oat glass batch with 25 K
min1 in an ambient atmosphere. Lines 1, 2, 3 and 4 present the degree of silica conversion
in case the cullet content in the oat glass batch equals 0 wt.%, 25 wt.%, 50 wt.% and 75
wt.%, respectively. The silica grain particle size in the oat glass batch is between 212 m
and 300 m.

3.5. Experimental determination of the apparent GB-model parameters

121

Table 3.6: The apparent GB model parameters AGB and EGB for the oat batch as function of the oat
cullet content.
Percent cullet
in oat batch [%]

AGB
[s1 ]

EGB
[kJ mol1 ]

r2
[-]

0
25
50
75

3.0 102
8.5 100
3.2 101
1.3 105

63
120
144
245

0.9749
0.9995
0.9940
0.9929

Figure 3.25 shows the plot of ln T + ln ln f () versus the reciprocal absolute temperature for the four oat batches with different oat cullet content itself. The apparent GB model
parameters AGB and EGB for the oat batch as function of the oat cullet content are listed in
table 3.6.

-3.0

+ ln ln f () [ln min1 ]

-3.5
-4.0
-4.5
-5.0

-5.5

-6.0

ln

-6.5

PSfrag replacements

-7.0

-7.5
-8.0
0.55

0.60

0.65

0.70

0.75

Reciprocal absolute temperature

0.80

[ 103

0.85

K1 ]

Figure 3.25: ln T + ln ln f () characterizing the dissolution of silica grains versus the reciprocal
absolute temperature for two different cullet fractions. Line 1, line 2, line 3 and line 4

present ln T + ln in case the cullet content of the oat batch equals 0 wt.%, 25 wt.%,
50 wt.% and 75 wt.%, respectively. The silica grain particle size was between 212 m and
300 m. The applied heating rate was 25 K min1

Figure 3.26 shows the relation of the apparent GB model parameters E GB and ln AGB as
function of the cullet fraction. With these relations, the dissolution rate of sand grains, with
the same sand grain particle size and about similar heating rates, as function of the content of
crushed cullet can be simulated.

122

Chapter 3. Dissolution of sand grains during heating of glass forming batches

24

250

22

EGB =2.28 xc +57.5; r2 =0.9372

200

20

EGB [kJ mol1 ]

16
150

14
12
10

100

ln AGB [ln min1 ]

18

6
50

ln AGB =0.19 xc +0.48; r2 =0.9310

PSfrag replacements

2
0

0
0

10

20

30

40

50

60

70

80

Cullet fraction xc [-]

Figure 3.26: Apparent GB model parameters EGB and ln AGB describing the dissolution of sand grains as
function of the cullet fraction in a oat glass batch. The apparent GB model parameters are
determined for sand grains with a average particle size of 250 m and for a heating rate of
25 K min1 .

3.6

Concluding remarks

In this chapter, the applicability of the relative simple Ginstling-Brounstein model for the prediction of the sand grain dissolution process in soda rich glass batches is studied. The accuracy
of the prediction of the sand grain dissolution process with the GB-model is determined by
comparison of simulation results obtained from the GB-model with simulation results obtained
from a more detailed numerical model. It appeared that with the GB-model, no quanitative
prediction of the sand grain dissolution process can be performed in case the true chemical and
physical properties of the sand grain and the surrounding thin melt phases are used. However,
for sand grains surrounded by a large volume of glass melt, the sand grain dissolution process
simulated with the numerical model can be approximated by a modied GB-model. In the modied GB-model, the proportionality constant kGB is multiplied with a constant less than unity.
For a small amount of glass melt surrounding a sand grain, the ratio between the constants k GB
and knm is not dependent on kGB . For these systems, the GB-model is not suitable for describing
the sand grain dissolution process.
The main important parameter which determines the linearity between the proportionality
constants kGB and knm , is the ratio of the melt phase thickness surrounding the sand grain
and the initial sand grain radius rs,0 . In the numerical model, an idealized situation of a completely and uniform wetted sand grain is modelled. In practice, the ideal situation is disturbed
by e.g. incomplete wetting of the sand grains, an inhomogeneous glass batch, the presence of

3.6. Concluding remarks

123

gas bubbles separating the sand grain and the glass melt. Therefore, it was also questioned to
which extent the modied GB-model is capable of describing the sand grain dissolution process
in industrial glass melts. The applicability of the modied GB-model for describing the sand
grain dissolution process in industrial glass batches is studied by comparison of the measured
GB proportionality constant kGB from a oat glass batch with different sand grain sizes with
the expected proportionality constant. From the experiments, the applicability of the GB-model
could not be proved unambiguously. Although the conversion rate of sand grains in the oat
glass batch could be described well with the GB-model resulting in apparent values for GB
model parameters EGB and AGB , the prediction of the dissolution rate of sand grains as function
of particle size and as function of the heating rate, using the GB-model and the derived E GB - and
AGB -values, show relative large differences with the measured dissolution rates. The reason for
this is likely the complexity of the glass melting process and the broad particle size distribution
present in the oat glass batch.
By the experiments which are performed in this chapter, it is possible to predict, for a given
particle size and heating rate (for which EGB and AGB are determined experimentally), the temperature dependent conversion rate. Also the presence of (crushed) cullet in the oat glass batch
with a dened particle size of the sand grains can be simulated. To evaluate whether the modied GB-model is capable of describing the sand grain dissolution, additional experiments with
well-dened and narrow particle size distributions are required. It is recommended to start with
simple glass batches such as binary systems before extending the melting experiments to multicomponent batches. This approach will provide more detailed information of the dissolution
mechanism of sand grains in glass batches.
Another option to simulate the sand grain dissolution process in glass batches is to develop
a model based on a large series of experiments. This model does not need to be based on rstprinciples like the GB-model.

124

Chapter 3. Dissolution of sand grains during heating of glass forming batches

3.7

Nomenclature

Latin symbols
As
A
CSiO2
DSiO2
Ea
f ()
g
Gr
h
hd
hr

Ii, j

j
J
k
Ki, j
K1
K2
K3
Ke
l
m
M
n
nro
r
r
ro
rs
rt
R
Re
RIR
s
Sc
Sh

surface area
pre-exponential factor
SiO2 concentration
inter diffusion coefcient of SiO2
(apparent) activation energy
reaction mechanism function
gas phase
Grasshof number
mass transfer coefcient
mass transfer coefcient in the diffusive stage
mass transfer coefcient in the reactive stage
settling time
intensity of diffraction line i for crystalline
specie j
mass ux relative to mass-average velocity
gas volume ux
reaction rate constant
constant for diffraction line i for crystalline
specie j
constant in equation 3.14
constant in equation 3.14
constant in equation 3.14
X-ray diffraction constant
liquid phase
mass
molar weight
mass ux
reaction order
reaction rate
radius
radius of a sand grain and the surrounding melt
phase shell
sand grain radius
radius of a sand grain, the surrounding melt
phase shell and the surrounding soda ash shell
universal gas constant
Reynolds number
Reference Intensity Ratio
solid phase
Schmidt number
Sherwood number

[m2 ]
[s1 ]
[kg m3 ]
[m2 s1 ]
[J mol1 ]
[-]
[-]
[m s1 ]
[m s1 ]
[m s1 ]
[s]
[counts s1 ]
[kg m2 s1 ]
[m3 m2 ]
[s1 ]
[s1 ]
[K]
[K]
[-]
[kg]
[kg mol1 ]
[kg m2 s1 ]
[-]
[s1 ]
[m]
[m]
[m]
[m]
[J K1 mole1 ]
[-]
[-]
[-]
[-]

3.7. Nomenclature

Sh
t
te
T
v
vr
V
V
w
x

Sherwood number corrected for moving boundary


time
duration time of the transient stage
temperature
velocity
radial velocity
volume
molar volume
weight fraction
weight fraction

125

[-]
[s]
[s]
[K]
[m s1 ]
[m s1 ]
[m3 ]
[m3 mol1 ]
[-]
[-]

Greek symbols

heating rate
melt phase thickness
dimensionless concentration (see equation
3.34)
density
uncertainty
diffraction angle
Fourier number
fraction
stoichiometric reaction coefcient
mass absorption coefcient
degree of conversion

Sub- and superscripts


b
c
dl
ds
GB
i
L1
L2
L3
m
n
nm
0
r
s
sa
t

bulk
cullet
large diameter
small diameter
Ginstling-Brounstein
interface
liquidus line SiO2
liquidus line Na2 O 2SiO2
liquidus line Na2 O SiO2
melt
number
numerical model
initial
radial
sand grain
soda ash
total

[K s1 ]
[m]
[-]
[kg m3 ]
[-]
[o ]
[-]
[-]
[-]
[-]

126

3.8

Bibliography

Bibliography

[1] P. Hrma. Complexities of batch melting. In: Proc. of the 1st International Conference on
Advances in the Fusion of Glass, pages 10.110.18, Alfred University, Alfred, New York,
June 14-17, 1988.
[2] L. Bodalbhai and P. Hrma. The dissolution of silica grains in isothermally heated batches
of sodium carbonate and silica sand. Glass Technology, 27(2):7278, 1986.
[3] J.R. Frade and M. Cable. Reexamination of the basic theoretical model for the kinetics of
solid-state reactions. J. Am. Ceram. Soc., 75(7):19491957, 1992.
[4] H. Salmang and H. Scholze. Die physikalischen und chemischen Grundlagen der Keramik.
Springer-Verlag, Berlin/Heidelberg, Germany, 5 edition, 1968.
[5] R.E. Carter. Kinetic model for solid-state reactions. J. Chem. Phys., 34(6):20102015,
1961.
[6] M. M hlbauer and L. Nemec.
u

64(11):14711475, 1985.

Dissolution of glass sand.

Am. Ceram. Soc. Bull.,

[7] R.G.C. Beerkens, H.P.H. Muijsenberg, and Heijden van der T. Modelling of sand grain
dissolution in industrial glass melting tanks. Glastech. Ber. Glass Sci. Technol., 67(7):179
188, 1994.
[8] P. Hrma. Reaction between sodium carbonate and silica sand at 874 o C<T<1022o C.
J.Am.Ceram.Soc., 68(6):337341, 1985.
[9] P. Hrma. A kinetic equation for interaction between grain material and liquid with application to glass melting. Silik ty, 24(1):716, 1980.
a
[10] P. Hrma, J. Barton, and T.L. Tolt. Interaction between solid, liquid and gas during glass
batch melting. J. Non-Cryst. Solids, 84:370380, 1986.
[11] R.B. Bird, W.E. Stewart, and E.N. Lightfoot. Transport phenomena. John Wiley & Sons,
Inc., New York, 1960.
[12] D.W. Ready and A.R. Cooper. Molecular diffusion with a moving boundary and spherical
symmetry. Chem. Engng. Sci., 21:917922, 1966.
[13] R.V. Harrington, J.R. Hutchins, and J.D. Sherman. The kinetics and mechanisms of subliquidus alkali carbonate-silica reactions. Advances in Glass Technology, 8-14 July, 1962.
USA Washington DC.
[14] M. Cable and D Martlew. Effective binary diffusivities for the dissolution of silica in melts
of the sodium carbonate - silica system. Glass Technology, 26(5):212217, 1985.
[15] K. Kautz and Stromburg G. Untersuchungen der Vorg nge beim Einschmelzen von
a
Glasgemengen im Gradientofen. Glastech. Ber., 42(7):309317, 1969.

Bibliography

127

[16] C. Kr ger. Gemengereaktionen und Glasschmelze. Glastechn. Ber., 25(10):307324,


o
1952.
[17] C. Kr ger.
o
Die tern ren und quatern ren Systeme Alkalioxyd-CaOa
a
SiO2 , Gleichgewichte, Reaktionsgeschwindigkeiten und ihre Beziehung zum
Glasschmelzprozess, Teil I. Glastechn. Ber., 15(9):335346, 1937.
[18] C. Kr ger.
o
Die tern ren und quatern ren Systeme Alkalioxyd-CaOa
a
SiO2 , Gleichgewichte, Reaktionsgeschwindigkeiten und ihre Beziehung zum
Glasschmelzprozess, Teil I. Glastechn. Ber., 15(10):371379, 1937.
[19] C. Kr ger.
o
Die tern ren und quatern ren Systeme Alkalioxyd-CaOa
a
SiO2 , Gleichgewichte, Reaktionsgeschwindigkeiten und ihre Beziehung zum
Glasschmelzprozess, Teil I. Glastechn. Ber., 15(11):403416, 1937.
[20] C. Kr ger.
o
Die tern ren und quatern ren Systeme Alkalioxyd-CaOa
a
SiO2 , Gleichgewichte, Reaktionsgeschwindigkeiten und ihre Beziehung zum
Glasschmelzprozess, Teil II. Glastechn. Ber., 22(12):248261, 1949.
[21] C. Kr ger.
o
Die tern ren und quatern ren Systeme Alkalioxyd-CaOa
a
SiO2 , Gleichgewichte, Reaktionsgeschwindigkeiten und ihre Beziehung zum
Glasschmelzprozess, Teil II. Glastechn. Ber., 22(15):331338, 1949.

[22] C. Kr ger and G. Ziegler. Uber die Geschwindigkeiten der zur Glasschmelze fuhreno
den Reaktionen. II. Die Umsetzung von Natriumdisilikat mit Soda und von Quarz mit
Kalkstein. Glastechn. Ber., 26(11):346353, 1953.

[23] C. Kr ger and G. Ziegler. Uber die Geschwindigkeiten der zur Glasschmelze fuhrenden
o
Reaktionen. III. Reaktionsgeschwindigkeiten im quatern ren System Na2 O-CaO-SiO 2 a
CO2 . Glastechn. Ber., 27(6):199212, 1954.

[24] C. Kr ger and F. Marwan. Uber die Geschwindigkeiten der zur Glasschmelze fuhrenden
o
Reaktionen. IV. Die Druckabh ngigkeit der Umsetzgeschwindigkeiten im quatern ren
a
a
System Na2 O-CaO-SiO 2 - CO2 . Glastechn. Ber., 28(2):5157, 1955.

[25] C. Kr ger and F. Marwan. Uber die Geschwindigkeiten der zur Glasschmelze fuhrenden
o
Reaktionen. VI. Der Einu von Zus tzen auf die Reaktionsgeschwindigkeit eines Sodaa
Kalkstein-Quarz-Grundgemengen. Glastechn. Ber., 29(7):275289, 1956.

[26] C. Kr ger. Uber die Geschwindigkeiten, den Mechanismus und die Phasenneubilding bei
o

den unter Schmelzbilding ablaufenden Festkorperreaktionen. Glastechn. Ber., 30(2):42


52, 1957.
[27] E.M. Levin, C.R. Robbins, and H.F. McMurdie. Phase Diagrams for Ceramics. The
American Ceramic Society, 1964.
[28] M.E. Savard and R.F. Speyer. Effect of particle size on the fusion of soda-lime-silicate
glass containing NaCl. J. Am. Ceram. Soc., 76(3):671677, 1993.
[29] L. Stoch and S. Kraishan. Interface phenomena accompanying the early stages of glass
batch reactions - a model study. Glastech. Ber. Glass. Sci. Technol., 70(10):298305, 1997.

128

Bibliography

[30] R. R ttenbacher and H.E. Engelke. Reaktionsvorg nge zwischen SiO 2 und einer
o
a

Natriumsilikatschmelze, Teil 1. Auosung von SiO 2 in der Schmelze. Glastechn. Ber.,


49(11):257263, 1976.
[31] A. Shimizu and Y. Hao. Inuence of particle contact on the estimation of powder reaction
kinetics of binary mixtures. J. Am. Ceram. Soc., 80(3):557568, 1997.
[32] H.S. Carslaw and J.C. Jaeger. Conduction of heat in solids. Clarendon Press, Oxford,
U.K., 2nd edition, 1959.
[33] L. Shartsis, S. Spinner, and W. Capps. Density, expansivity, and viscosity of molten alkali
silicates. J. Am. Ceram. Soc., 35(6):155160, 1952.
[34] J. Hermans. Personal communication.
[35] Manual FactSage 5.1. Centre for Research in Computational Thermochemistry, Ecole
Polytechnique (Universit de Montr al), Montreal, Quebec, Canada.
e
e
[36] I.C. Madsen, N.V.Y. Scarlett, L.M.D. Cranswick, and T. Lwin. Outcomes of the international union of crystallography commision on powder diffraction round robin on quantitative phase analysis: samples 1a to 1h. J. Appl. Cryst., 34:409426, 2001.
[37] R. Jenkins and R.L. Snyder. Introduction to X-ray powder diffractometry. John Wiley &
Sons, Inc., 605 Third Avenue, New York, NY, USA, 1996.
[38] H.W. Coleman and W.G. Steele. Experimentation and uncertainty analysis for engineers.
John Wiley & Sons, Inc., 2nd edition, 1999.

Chapter 4
Heat conductivity of glass forming batches
4.1

Introduction

As mentioned in chapter 1, a glass forming batch can be regarded as a mixture of a condensed


phase, which encloses the solid batch particles and the formed melt phases, and a gas phase.
The Lagrangian description of the energy equation of the two-phase glass batch is given by

cp T
t

Hchem
,
t
(4.1)
in which cp T mn represents the mean value of the enthalpy of the glass batch, g and cp,g
are the density and heat capacity of the gas phase, Tmn is the mean temperature of the glass
batch, t is the time, p is the porosity of the glass batch, vg,z is the vertical velocity of the gas
phase relative to the condensed phase1 , eff is the effective heat conductivity of the glass batch,

q r,eff is the effective radiative heat ux through the glass batch, and Hchem is the temperature
dependent energy per unit volume of the glass batch required for batch reactions.
The heat transport in the interior of the glass batch is determined by a combination of three
modes of heat transfer:
mn

= p g cp,g vg,z Tg (eff Tmn ) + r,eff + (1 p )


q

convective heat transport by the ascending gas phase (characterized by the rst righthand-side term in equation 4.1),
conductive heat transport by mutual contact between solid particles and between solid
particles and liquid phases (characterized by the second right-hand-side term in equation
4.1),
radiative heat transport through the transparent phases (characterized by the third righthand-side term in equation 4.1).
Since (experimental) discrimination between these modes of heat transfer in glass batches is
complex, in general the heat penetration in a glass batch is described by an effective heat conductivity eff , which encloses the contributions of the three modes of heat transfer and the
energy required for the chemical batch reactions. In the past, several attempts have been made
1 It

is assumed that the horizontal velocity of the gas phase is similar to the horizontal velocity of the condensed
phase [1].

129

130

Chapter 4. Heat conductivity of glass forming batches

to measure and describe the thermal behavior of glass batches during heating. In general, the
thermal behavior of glass batches is studied by measuring the temperature at different positions
in the glass batch as function of time during heating of the glass batches [26] 2 . However,
neither of these studies provided values for the temperature dependent heat conductivity of the
glass batches. Jack and Jacquest [2], Faber et al. [5] and Conradt et al. [6] calculated a (temperature dependent) thermal diffusivity3 of glass batches derived from the measured temperatures
in the glass batches during heating. The calculated temperature dependent thermal diffusivity
for different batches showed that at temperatures above 1173 K, a sudden increase in the thermal diffusivity of the glass batches is observed. This increase in thermal diffusivity is attributed
to the formation of primary formed melt phases, which are transparent for radiative energy and
therefore enhance the heating of glass batches [6]. At the onset temperature for primary melt
phase formation, radiative heat transport (or photon conductivity) through the glass batch predominates over conductive heat transport (or phonon conductivity) in the glass batch. Conradt
et al. [6] also observed that the onset for enhanced thermal diffusion in a glass batch is not
only dependent on temperature, but also on the position in the glass batch. The position dependent onset temperature for enhanced thermal diffusion was explained by local differences in the
amount of entrapped batch gases in the melting glass batch, which decrease the mean free path
for radiative energy in the glass batch. This indicates that the thermal diffusivity (and therewith
the heat conductivity) is dependent on local conditions in the glass batch and therefore these
properties are time- and temperature.
A quantitative study to the temperature dependent heat conductivity of glass batches was
performed by Kr ger and Eligehausen [8]. The heat conductivity of glass batch components
o
and complete glass batches was measured with a so-called static hollow cylinder technique.
With this technique, two cylinders with different diameters are placed in each other. The space
between both cylinders is lled with glass batch (components). During heating of this system,
both the temperature difference over the glass batch shell and the heat ow at the outer boundary
of the glass batch shell are measured. The heat conductivity of the glass batch is calculated
by
Q
ln(d1 /d2 )
=
,
(4.2)
Tr2 Tr1
2 L
in which Q is the heat ow through the glass batch, Tr1 and Tr2 are the temperatures at the
boundaries of the glass batch, d1 and d2 are the diameters of both cylinders, and L is the length
of the cylinders over which heat is transferred. In this experimental set-up, the thickness of the
glass batch was approximately 17 mm. It was observed that in the temperature range at which
no melt phases are generated, which is approximately below 1073 K, the temperature dependent
heat conductivity can mostly be described by a rst order temperature dependency.
Because the presence and the amount of generated melt phases in a glass batch, which enhances the diffusion of heat through the glass batch, is dependent on the kinetics of melt phase
forming and dissolution reactions, the effective heat conductivity of glass batches is both timeand temperature dependent. Therefore, a drawback of the use of a static technique to determine
the effective heat conductivity of glass batches is that time dependent effects (caused by the
2 Next

to the determination of the heat conductivity of glass batches by measuring the temperature at different
positions in the glass batch as function of time, Krzoska et al. [7] developed an alternative method with which the
local thermal diffusivity of melting glass batches can be determined using a pulsing heat source.
3 The

thermal diffusivity is dened as

c p

4.1. Introduction

131

time dependent melting of glass batches) on the heat conductivity of glass batches are not taken
into account.
The purpose of this chapter is to develop a transient method, with which the temperature
dependent heat conductivity of glass batches can be determined quantitatively. This transient
method encloses both the development of an experimental set-up and a mathematical method
with which the temperature dependent heat conductivity of glass batches can be estimated from
the experiments performed in the experimental set-up. Similar to most experimental studies,
in this study it is chosen to study the heat conductivity of glass batches by measuring the temperatures at different positions in a glass batch as function of temperature. Figure 4.1 shows
an example of the measured temperatures at different positions in a silica sand batch which is
measured in this study. Throughout this section, the heating experiments of mixtures of glass
batch components is denoted as heat penetration experiments.

1600

1
1400

Temperature [K]

1200

1000
800

600
400

PSfrag replacements

1000

2000

3000

4000

5000

6000

7000

8000

9000

Time [s]

Figure 4.1: Measured temperatures at different positions from the hot boundary of a silica sand batch.

In section 4.2, a description is given of the phonon and photon conductivity of individual
species and of mixtures of species. In section 4.3, the most suitable mathematical method for
estimating the effective heat conductivity of glass batches from heat penetration experiments is
studied. The developed experimental set-up for performing these heat penetration experiments
is discussed in section 4.4. Section 4.5 describes the estimation of the temperature dependent
heat conductivity of solid particle mixtures up to a temperature of maximum 1400 K and the
modelling of the temperature dependent heat conductivity of these mixtures. Section 4.6 provides a summary of the results obtained in this section.

132

4.2

Chapter 4. Heat conductivity of glass forming batches

Phonon and photon conductivity

The effective heat conductivity of a mixture of solid particles surrounded by a static gas phase
is determined by both phonon and photon conduction through the gas-solid mixture. In sections
4.2.1 and 4.2.2, the intrinsic phonon and photon conductive properties of both crystalline and
amorphous solid species is discussed. Sections 4.2.3 and 4.2.4 describe the phonon and photon
conductive properties of a mixture of solid particles surrounded by a static gas phase.

4.2.1 Intrinsic phonon conduction of solid species


Phonon conduction in a solid specie is the transfer of vibrational energy of atoms via lattice
waves, i.e. phonons, through the solid specie. The phonon conductivity phonon , as described
by Kingery [9], is given by
phonon =

1
3

c() v() l() d,

(4.3)

in which c is the heat capacity of the solid, expressed in J kg1 m3 , v is the phonon velocity
in the solid4 , l is the mean free path of the phonons, and is the frequency of the phonons.
According to Kingery et al. [9], the heat capacity of a solid increases with temperature up to a
maximum limiting value expressed in J mole1 K1 equal to (3 N R), in which N is the number
of atoms per molecule and R is the universal gas constant. The mean free path of the phonons
is limited by the scattering of phonons at lattice imperfections. Because in general the number
of lattice imperfections increases with temperature, the phonon conductivity decreases with
temperature and becomes independent on the frequency . As an example, Slifka et al. [10]
measured the phonon conductivity of polycrystalline MgO over the temperature range from
400 K up to 1300 K. In this temperature range, the phonon conductivity decreases from 30 W
m1 K1 at 400 K down to 8 W m1 K1 at 1290 K. According to gure 4.2, the phonon
conductivity is linear dependent on the reciprocal value of the absolute temperature.
At high temperatures, the mean free path of the phonons in crystalline solids approaches
a value similar to the lattice spacing of the crystalline solid. Now, the mean free path of the
phonon is independent on the phonon frequency. For a constant mean free path of the phonons
and writing the phonon velocity as the ratio of the velocity of light in vacuum, v vacuum , and the
refractive index of the solid n, the temperature dependency of the phonon conductivity is given
by
vvacuum l
phonon =
c,
(4.4)
3n
in which c is the temperature dependent heat capacity of the solid. Because at high temperatures, the heat capacity of a solid is constant, also the phonon conductivity attains a constant
value. Figure 4.3 shows the measured phonon conductivity of crystalline silica, which is given
by Eligehausen and Kr ger [11], as function of the reciprocal absolute temperature. The lino
ear dependency of the phonon conductivity with the reciprocal absolute temperature in the low
temperature range clearly diminishes at higher temperatures.
Because in amorphous species, such as glass and glass melts, no short-range structural ordering of atoms is present, the mean free path for phonons in amorphous species is limited. The
4 The

phonon velocity in the solid is given by the ratio of the velocity of light and the refractive index of the
solid, which is only slightly dependent on the wavelength of the phonon.

4.2. Phonon and photon conductivity

Phonon conductivity [W m1 K1 ]

30

133

T=400 K

25

20

15

10

T=1290 K

PSfrag replacements

5
0.6

0.8

1.0

1.2

1.4

1.6

1.8

2.0

2.2

2.4

2.6

Reciprocal absolute temperature [ 103 K1 ]

Figure 4.2: Phonon conductivity of polycrystalline MgO as function of the reciprocal absolute temperature [10]. The temperature dependency of the phonon conductivity is given by phonon =
1.67 + 1.269 104 T 1 .

temperature dependency of the phonon conduction in amorphous species is mainly given by


the temperature dependency of the specic heat of the amorphous specie, which increases with
temperature. Mann et al. [12] measured the intrinsic phonon conductivity of oat glass cullet
up to about 780 K, which is given by
int,phonon,oat cullet = 0.97 + 6.24 104 T,

(4.5)

in which T is expressed in K.

4.2.2 Intrinsic photon conduction of solid species


In addition to vibrational energy transfer in solids, energy is also transferred by electromagnetic
radiation, which is called photon conduction. According to Kingery [9], the intrinsic photon
conductivity int,photon is given by
16
n2 T 3 lr ,
(4.6)
3
in which is the Stefan-Boltzmann constant, n is the refractive index, T is the temperature in
K, and lr is the mean free path of photons in the solid. Single defect free crystals are, to a
int,photon =

134

Chapter 4. Heat conductivity of glass forming batches

Phonon conductivity [W m1 K1 ]

T=273 K

T=1373 K

PSfrag replacements

3
0.5

1.0

1.5

2.0

2.5

3.0

3.5

4.0

Reciprocal absolute temperature [ 103 K1 ]

Figure 4.3: Intrinsic phonon conductivity of crystalline silica as function of the reciprocal absolute tem
perature given by Eligehausen and Kroger [11].

certain extent, transparent in the electromagnetic spectrum for wavelengths between 1 m and
4 m. Both at wavelengths shorter than 1 m and longer than 4 m, the crystals are opaque
for radiative energy. In the UV-range of the electromagnetic spectrum (at wavelengths larger
than 4 m), the transmission of radiative energy is lost by electron excitation, whereas in the
IR-range, strong absorption is present due to atomic vibrational phenomena. In the visible
region of the electromagnetic spectrum, the mean free path for a single alumina crystal in the
temperature range from 500 K up to 2000 K increases from 1 cm up to 10 cm [9]. However, in
case pores are present in polycrystalline alumina, the photon conduction is strongly limited due
to the scattering of photons at the surface of the pores. The transmission of photons through
polycrystalline alumina, containing 1 volume percent of pores with a diameter of 2 m, equals
about 1 % of the transmission of photons through a single alumina crystal [9]. Therefore, in
general polycrystalline solids can be regarded as opaque for photon conduction.
For translucent species, such as glass and glass melts, the photon conductivity photon is
given by
16 n2 T 3
,
(4.7)
int,photon,translucent =
3
a
in which a is the wavelength dependent absorption coefcient, which is the reciprocal of the
mean free path of the photons. The absorption coefcient of a glass (melt) is mainly dependent

4.2. Phonon and photon conductivity

135

on the absorption of radiation by electronic transitions of transition elements such as Fe, Cr.
Each valency state of these polyvalent ions, such as Fe2+ and Fe3+ , absorbs radiation of a typical
own wavelength (e.g. Fe2+ around 1050 nm and Fe3+ around 380 nm) and gives a temperature
dependent absorption spectrum. The valency state of the polyvalent ions in the glass (melt) is
coupled to the oxygen activity in the glass (melt). Therefore, the photon conductivity of a glass
(melt) is dependent on both the total amount of transition elements and the oxygen activity in
the glass (melt).

4.2.3 Phonon conduction of mixtures of solid species


In the solid-state regime5 , a glass batch is regarded as a mixture of a solid phase and a gas phase.
For the description of the heat conductivity of gas-solid mixtures, different geometrical models
are proposed in literature [13, 14]. These geometrical models predict the heat conductivity of
the gas-solid mixture, assuming a static gas phase, as a function of
the volume fraction of the gas phase,
the heat conductivity of the gas phase,
the heat conductivity of the solid phase(s), and
the structural arrangement of the two phases.

In general, the geometrical models describe the heat conductivity through a gas-solid mixture 6
by a heat ow through a three dimensional arrangement of thermal resistances. For a solid
particle mixture, each individual thermal resistance characterizes the resistance towards heat
transport through either a solid particle or through the gas phase between the solid particles.
The heat ow Q through each thermal resistance is described by the ratio of the temperature
difference over the resistance, T2 T1 , and value of the heat resistance RT according to
Q=

T2 T1
.
RT

(4.8)

The heat resistance RT is dened by

L
,
(4.9)
A
in which L is a characteristic length scale for heat transport through a solid particle or a static
gas phase volume, A is a characteristic cross-section of the thermal resistance perpendicular on
the heat ow direction, and is the heat conductivity of the solid particle or gas phase. The
two extreme arrangements of the thermal resistances are shown in gures 4.5(a) and 4.5(b).
Figure 4.5(a) shows the arrangement of the thermal resistances in case the heat ow in the solid
particle mixture is regarded as a parallel process of heat transfer through the solid and the gas
phase. Figure 4.5(b) shows the arrangement of the thermal resistances in case the heat ow in
the solid particle mixture is regarded as a serial process of heat transfer through the solid and
RT =

5 The

solid-state regime is the temperature regime during which no melt phases are present. A typical value for
the onset temperature for melt phase formation, which nishes the solid-state regime, for a soda-lime-silica glass
equals approximately 1073 K.
6 Further throughout this thesis, a gas-solid mixture is denoted as a solid particle mixture.

136 PSfrag replacements


Chapter 4. Heat conductivity of glass forming batches

T1

T2

A
Q

Resistance RT
L

Figure 4.4: Schematic representation of the heat ow through a thermal resistance in which Q is the heat
ow, T2 T1 is the temperature difference over the thermal resistances, R T is the thermal resistance, L is a characteristic length scale for heat transport through a solid particle or a static
gas phase volume, A is a characteristic cross-section of the thermal resistance perpendicular
on the heat ow direction

the gas
PSfrag replacementsphase. The mathematical formulation describing the heat transport through these two
PSfrag replacements
extreme arrangements is denoted as the Parallel Network Model (PNM) or the Serial Network
Solid phase
Solid phase
model (SNM).
Gas phase
Gas phase
RTg
T1

T2
Q

RTs
T1

(a) Parallel arrangement of thermal resistances

RTg

RTs
T2

(b) Serial arrangement of thermal resistances

Figure 4.5: Models of the arrangements of thermal resistances in a solid particle mixture, in which Q is
the heat ow, T2 T1 is the temperature difference over the thermal resistances, R Tg is the
thermal resistance of the gas phase, and RTs is the thermal resistance of the solid phase.

The effective heat conductivity of the packed bed derived from both models is given by
PNM = p g + (1 p ) s ,
and
SNM =

1
p
g

1 p
s

(4.10)
(4.11)

in which PNM is the heat conductivity according to the parallel network model, SNM is the
heat conductivity according to the serial network model, s is the heat conductivity of the solid
phase, g is the heat conductivity of the gas phase, and p is the volume fraction of the gas
phase in the solid particle mixture.
According to Kingery [9], a mathematical formulation of the heat conductivity of the solid
particle mixture based on a more realistic structure of a solid particle mixture, which is shown
in gures 4.6(a) and 4.6(b), is proposed. Figure 4.6(a) shows the structure of a solid particle
mixture in case the major phase is a continuous phase in which the minor phase is dispersed.
Figure 4.6(b) shows the structure of a solid particle mixture in case the minor phase is a continuous phase (gas) in which the major phase (solid particles) is dispersed. The mathematical
formulations describing the heat transport process through these two structures is denoted as the
Continuous Phase Network model (CPNM) and the Dispersed Phase Network model (DPNM).

4.2. Phonon and photon conductivity

(a) Major phase is continuous phase

137

(b) Major phase is dispersed phase

Figure 4.6: Schematic structural arrangements for solid particle mixtures.

Kingery applied the so-called Eucken-model for the description of the heat conductivity of
these continuous-dispersed models. The heat conductivity of a solid particle mixture containing
spherical particles is given by this model is
Eucken = c

1 + 2 p
1 p

1 c \ 2
d

1 c \
d

c
d

c
d

+1

+1

(4.12)

in which c is the heat conductivity of the continuous phase, and d is the heat conductivity of
the dispersed phase.
Resuming, the effective phonon conductivity of a mixture of solid particles surrounded by
a gas phase is dependent on the intrinsic properties of the solid and gas phase species and
on the three dimensional arrangement of the solid particle mixture. The identication of the
most suitable structural model for the description of the phonon conductivity of solid particle
mixtures composed of industrial grade raw material components of glass batches is considered
in section 4.5.1.

4.2.4 Photon conduction in mixtures of solid species


The photon conduction in mixtures of solid species is determined by the mean free path for
photons in the mixture, which is dependent on:
the absorption and transmission of photons in and through the solid particles and the
surrounding gas phase, and
the reection of photons at the surface of solid particles and pores in the mixture.
Most of the industrial raw material components of glass batches, such as silica sand and limestone, are opaque for radiative energy transfer and will therefore not transmit radiation. In case
cullet, which is (partly) transparent for radiation, is used as alternative raw material component

138

Chapter 4. Heat conductivity of glass forming batches

during glass batch melting, the mean free path for photons in the solid particle may increase
resulting in a higher value for the photon conductivity of the solid particle mixture.
According to Kingery [9], the reecting properties of solid particles is characterized by a
so-called scattering coefcient. The degree of photon scattering at the surface of particles is
dependent on
the ratio of the particle diameter to the wavelength of the incident photons,
the relative refractive index, which is dened as the ratio of the refractive index of the
particles and the refractive index of the medium surrounding the particles.
The degree of photon scattering increases with the relative refractive index. At a particle size
less than the wavelength of the incident photons, the degree of scattering increases up to a
maximum value, which is attained at a particle size equal to the wavelength of the incident
photons. At a particle size larger than the wavelength of the incident photons, the degree of
scattering becomes proportional to the reciprocal particle size.
According to Holman [15], the photon conductivity of a pore is given by
photon,pore = 4 d p n2 g T 3 ,
g

(4.13)

in which d p is the diameter of the pore, ng is the refractive index of the gas phase present in
the pore, and g is the emissivity of the radiating gas. The radiative conductivity of the pore
increases with pore size, emissivity and temperature. Therefore, especially small pores in a
glass batch act as barriers for radiative heat transfer.
Resuming, the photon conductivity of a glass batch is dependent on
the intrinsic photon conductivity of, especially, the glass cullet and the radiative gas in the
pores,
the dimensions of the glass cullet and the pores,
the surface roughness of the glass cullet, and
the volume fraction of the glass cullet and the pores per unit volume of the glass batch.
In general, the contribution of the intrinsic photon conductivity of the glass cullet to the heat
conductivity of a glass batches decreases with decreasing particle size, increased surface roughness of the glass cullet and with decreasing volume fraction of the glass cullet in the glass
batch. Faber et al. [5] remarked that the benecial effect of glass cullet on the heating rate of
glass batches is apparent for cullet fractions above 50 %.
The heat conductivity of a glass batch containing glass cullet can hardly be predicted from
the intrinsic photon conductivity of the glass cullet because of the complexity concerning different batch components, cullet sizes, gas pore sizes, etc, . The photon conductivity of a glass
batch containing glass cullet is mainly dependent on the presence of scattering sources for photon conduction, such as small bubbles and very ne glass powder and glass cullet with a rough
surface. The fusion of glass cullet, the disappearance of small bubbles in the glass batch and the
formation of new melt phases will determine the actual photon conductivity of the glass batch.
This indicates that knowledge of the reactive melting of the raw materials of glass batches will
not be sufcient to predict the photon conductivity of the glass batch.

4.3. Estimation of the heat conductivity from heat penetration experiments

4.3

139

Estimation of the heat conductivity from heat penetration experiments

In this section, different mathematical methods are studied to estimate the temperature dependent effective heat conductivity of a solid particle mixture from heat penetration experiments.
From the accuracy of the estimated effective heat conductivity, which is derived from a ctive
heat penetration experiment using a known value for the heat conductivity to reconstruct the
heat penetration in the solid particle mixture, the most accurate mathematical method is identied.
After a general introduction in section 4.3.1, section 4.3.2 describes the accuracy of estimating the effective heat conductivity from heat penetration experiments in case a-priori information of = f (T ) is missing. The inuence of a-priori information of = f (T ) on the accuracy
of the estimated heat conductivity is studied in section 4.3.3. In this section, three different
mathematical methods are described, viz. a differential approach, an integral approach, and a
numerical-experimental approach.

4.3.1 Introduction
Different mathematical methods can be used for the determination of the temperature dependent
heat conductivity of a solid particle mixture7 from heat penetration experiments. In this section,
the most accurate mathematical method is identied from the calculated accuracy of the heat
conductivity of a solid particle mixture derived from ctive heat penetration experiments with
a known value for the heat conductivity of the solid particle mixture. To identify the most
accurate mathematical method, the following steps are performed:
1. Simulation of the heat penetration in a solid particle mixture
2. Registration of the temperature as function of time at predened positions in the solid
particle mixture.
3. Estimation of the known temperature dependent heat conductivity based on the ctive
temperature measurements.
The ctive heat penetration experiment is obtained by numerical simulation of the heat penetration process in a solid particle mixture, for which a schematic representation is given in gure
4.7. The indicated locations in the solid particle mixture denote the positions of the innite
small ctive thermocouples.
The numerical simulations are performed with the nite element code SEPRAN [16] by
solving

T
T
=
cp

,
(4.14)
t
x
x
which describes the one-dimensional heat penetration in a non-reacting solid particle mixture.
At both boundaries of the solid particle mixture, the temperature is prescribed, i.e. at x=0 and
7 Throughout

this section is referred to a solid particle mixture instead of a glass batch. The reason for this is 1)
the mathematical methods can be applied to solid particle mixtures in general and not only to glass batches, and 2)
for the evaluation of the most accurate mathematical method, the heat penetration in a non-reacting solid particle
mixture is simulated.

140

Chapter 4. Heat conductivity of glass forming batches

x = dspm

Tb
Packed bed

PSfrag replacements

Center line
x4 =40 mm
x3 =30 mm
x2 =20 mm
x1 =10 mm
x=0

Tx4
Tx3
Tx2
Tx1
Tb

Figure 4.7: Schematic representation of the geometry of the solid particle mixture for which the heat
penetration is simulated. The indicated locations in the solid particle mixture denote the
positions of ctive thermocouples, from which the temperature as function of time are shown
in gure 4.8.

x=dspm , the temperature equals Tb . The initial solid particle mixture temperature is T0 . The
values for the solid particle mixture properties , c p and , the thickness of the solid particle
mixture dspm and the initial and boundary temperatures T0 and Tb , for the heat penetration simulations are listed in table 4.1.
Figure 4.8 shows the calculated temperatures as function of time at the ctive thermocouple
positions. For the evaluation of the different mathematical methods to estimate the heat conductivity from these heat penetration curves, it is assumed that the solid particle mixture properties
and c p , the initial and boundary temperatures, the temperatures at the indicated positions in
the solid particle mixture and the indicated positions self are known a-priori and are exact. The
temperatures at the predened positions in the solid particle mixture are stored each 5 s. From
these stored temperatures, the heat conductivity of the solid particle mixture is estimated. First,
in section 4.3.2, the temperature dependent heat conductivity is estimated using a differential
method and in case a-priori information on = f (T ) is missing. In section 4.3.3, the estimation
of the heat conductivity is performed in case beforehand is known that the heat conductivity of
the solid particle mixture is given by = ca + cb T . For this estimation, three mathematical
methods are used, i.e. a differential method, an integral method and a numerical-experimental
technique.
The deviation between the estimated and the true value of ca and cb are an indication for
the accuracy of the mathematical method, with which the estimation is performed. Next to the
estimation of ca and cb in case the other model parameters and c p , the temperatures and

4.3. Estimation of the heat conductivity from heat penetration experiments

141

Table 4.1: Physical properties of the solid particle mixture, thickness of the solid particle mixture and
boundary and initial temperatures used for simulating the heat penetration in the solid particle
mixture as function of time. These simulations are used for the evaluation of the accuracy of
different mathematical methods for the estimation of the temperature dependent heat conductivity.
Bulk density
Heat capacity c p
Heat conductivity
Thickness dspm
Initial temperature T0
Boundary temperature Tb

1500
1200
0.2+3.0 104 T
0.10
298
1673

[kg m3 ]
[J kg1 K1 ]
[W m1 K1 ]
[m]
[K]
[K]

the positioning of the ctive thermocouples are exact, also the effect of uncertainties 8 in these
model parameters on the estimated value for ca and cb is studied. The mathematical method
which shows the lowest uncertainty in the temperature dependent heat conductivity is chosen to
be the most accurate method and will be used further throughout this chapter.

4.3.2 Parameter estimation without a-priori knowledge of = f (T )


In case a-priori information of the temperature dependency of the heat conductivity of the solid
particle mixture is missing, a rst estimate of the temperature dependency of the heat conductivity can be derived by neglecting the second right-hand-side term in
cp

T
2 T
=
t
x2

T
.
x x

(4.15)

Figure 4.9 shows a schematic representation of the geometry of the solid particle mixture. In
case the spatial distribution of thermocouples is uniform, i.e. x 1 = x2 = x, the heat conductivity at position i and temperature T can be calculated from three spatial distributed temperatures according to
Ti

Ti+1 + Ti1 2Ti


,
(4.16)
/
=
cp
t
x2
in which Ti is the temperature at position i in the solid particle mixture, Ti1 and Ti+1 are the
temperatures at the positions at x and x distance from xi , and Ti is the temperature difference at position xi in the solid particle mixture after a time step of t [5]. The denominator in
equation 4.16 is derived by assuming that the spatial temperature distribution between the three
thermocouples is described by a 2nd order polynomial (see Appendix C).
Plotting the heat conductivity calculated with equation 4.16 versus temperature provides the
temperature dependent heat conductivity. Figure 4.10 shows both the true and the estimated
heat conductivity of the solid particle mixture as function of temperature in case x=10 mm
and for xi =20 mm, xi =30 mm and xi =40 mm.
8 According to Coleman and Steele [17], the uncertainty in a parameter value is dened as the standard deviation

of the parameter value.

142

Chapter 4. Heat conductivity of glass forming batches

1600

x=10 mm
1400

x=20 mm

Temperature [K]

1200

x=30 mm

1000
800

x=40 mm

600

PSfrag replacements

400

1000

2000

3000

4000

5000

6000

7000

8000

9000

Time [s]

Figure 4.8: Calculated temperature as function of time at different positions in a solid particle mixture
simulated with SEPRAN [16]. The solid particle mixture properties, the solid particle mixture thickness and the initial and boundary conditions are listed in table 4.1

It is observed that large deviations in the temperature dependent heat conductivity are obtained in case a-priori knowledge of the temperature dependency of the heat conductivity is
missing. Combining equations 4.15 and 4.16 show that the accuracy of the estimated heat conductivity at position i is dependent on:
the accuracy of approximating the second order derivative of the temperature to position
given by the denominator in equation 4.16,
the accuracy of approximating the heating rate by the rst order derivative of the temperature to position, and
the value for the neglected term from equation 4.15, which equals c b
= c a + c b T .
2

T
x

in case

Because T = limx0 Ti+1 +Ti1 2Ti , it is expected that the approximation of the second
x2
x2
order derivative of the temperature to position by the denominator in equation 4.16 improves
in case the spacing of the ctive thermocouples (x) decreases. Figure 4.11 shows the effect
of the spacing of the ctive thermocouples on the estimated value for the heat conductivity. As
expected, a closer spacing of the thermocouples results in a smaller difference between the true

4.3. Estimation of the heat conductivity from heat penetration experiments

PSfrag replacements

143

x = dspm
Packed bed

Ti+1
Ti
Ti1

xi+1
xi

x1

xi1

x2
x=0

Figure 4.9: Schematic representation of the geometry of the solid particle mixture through which the
heat penetration is simulated.

and estimated heat conductivity.


In case the approximation of the heating rate by a rst order derivative of the temperature
to time affects the estimation of the heat conductivity, a higher sampling frequency of the measured temperatures is expected to result in a better estimation of the heat conductivity. However,
additional simulations did not show a signicant change in the estimated heat conductivity with
increasing sampling frequency, which indicates that the approximation of the heating rate does
not have a signicant effect on the estimated heat conductivity in case the sampling frequency
is less than 0.2 s1 .
The effect of the temperature dependency of the solid particle mixture on the accuracy of
the estimated heat conductivity, which is given by the parameter c b in = ca + cb T , is
shown in gure 4.12. This gure shows the heat conductivity of the solid particle mixture estimated from a ctive heat penetration experiment for which the true heat conductivity is given
by = 0.2 + 1 103 T instead of = 0.2 + 3 104 T . It is observed that for this stronger
temperature dependent heat conductivity, non-realistic values for the heat conductivity are obtained. From gures 4.10, 4.11, and 4.12, it is concluded that in case a-priori knowledge of
the temperature dependency of the heat conductivity is missing, non-realistic values form the
true heat conductivity are obtained. The condence of the estimated heat conductivity by this
method depends on the solid particle mixture properties (especially the heat conductivity), the
temperature gradient at the position of thermocouple for which the heat conductivity is estimated and the spacing of the three thermocouples, which are required for the estimation of the
heat conductivity.

4.3.3 Parameter estimation with a-priori knowledge of = f (T )


This section describes the estimation of the temperature dependent heat conductivity from heat
penetration experiments in case beforehand is known that = c a + cb T . The parameter estimation is performed by a differential method, an integral method and a numerical- experimental
method.
The differential method

144

Chapter 4. Heat conductivity of glass forming batches

2.0
1.8

Heat conductivity [W m1 K1 ]

1.6

PSfrag replacements

1.4
1.2
1.0
0.8

0.6

0.4

True heat conductivity

0.2
0.0
400

600

800

1000

1200

1400

1600

Temperature [K]

Figure 4.10: True (dotted line) and estimated (solid lines) heat conductivity of a solid particle mixture as
function of temperature. A, B and C indicate the estimated heat conductivity for a ctive heat
penetration experiment in case a-priori information of = f (T ) is missing for x i =20 mm,
xi =30 mm and xi =40 mm, respectively. The spacing between the ctive thermocouples x is
10 mm. The properties of the solid particle mixture and the initial and boundary conditions,
with which the ctive heat penetration experiment is simulated, are listed in table 4.1.

In case the temperature dependency of the heat conductivity is known beforehand and is described by = ca + cb T , equation 4.14 can be written as
T
= c a
cp
t

2 T
x2

+ c b

T
x

+T

2 T
x2

(4.17)

Describing the spatial dependency of the temperature with a 2 nd order polynomial, the best
estimates for the parameters ca and cb can be obtained from a least squares approach. Equation
4.17 is now rewritten to
k1,i = ca k2,i + cb k3,i
(4.18)
in which i refers to the i-th time period, k1,i is the left-hand-side term of equation 4.17 at t = ti ,
k2,i is the rst right-hand-side term of equation 4.17 at t = t i , and k3,i is the second right-handside term of equation 4.17 at t = ti . The least squares approach requires a criterion qls , which
quanties the difference between the measured and the simulated temperatures. The criterion

4.3. Estimation of the heat conductivity from heat penetration experiments

145

2.0
1.8

Heat conductivity [W m1 K1 ]

1.6

PSfrag replacements

1.4
1.2
1.0
0.8

0.6
0.4

True heat conductivity

0.2
0.0
400

600

800

1000

1200

1400

1600

Temperature [K]

Figure 4.11: True (dotted line) and estimated (solid lines) heat conductivity of a solid particle mixture as
function of temperature. A and B indicate the estimated heat conductivity for a ctive heat
penetration experiment when a-priori information of = f (T ) is missing for x i =30 mm and
in case the spacing between the ctive thermocouples are x=10 mm and x=6 mm, respectively. The spacing between the ctive thermocouples x is 10 mm. The properties of the
packed bed and the initial and boundary conditions, with which the ctive heat penetration
experiment is simulated, are listed in table 4.1.

qls describing the squared difference of the measured and simulated temperatures over the whole
time domain in which the simulations and the measurements are performed is given by
n

qls = k1,i ca k2,i cb k3,i

(4.19)

i=1

in which ca and cb are the estimates for ca and cb and n is the number of time-periods at

which the temperature data are collected from the ctive heat penetration experiment. Estimates for the parameters ca and cb providing the smallest difference between measured and
calculated temperatures are obtained by minimization of the criterion q ls with respect to the unknown estimates ca and cb . This results in a set of two linear algebraic equations from which

146

Chapter 4. Heat conductivity of glass forming batches

50
40

Heat conductivity [W m1 K1 ]

30

20
10
0

True heat conductivity

-10
-20
-30
-40

PSfrag replacements

-50
400

600

800

1000

1200

1400

1600

Temperature [K]

Figure 4.12: True (dotted line) and estimated (solid lines) heat conductivity of a solid particle mixture
as function of temperature. A indicates the estimated heat conductivity for a ctive heat
penetration experiment in case a-priori information of = f (T ) is missing for x i =30 mm.
The spacing between the ctive thermocouples x is 10 mm. The properties of the packed
bed and c p and the initial and boundary conditions, with which the ctive heat penetration
experiment is simulated, are listed in table 4.1. In contrast to table 4.1, the heat conductivity
is given by =0.2 + 1 103 T.

estimates of the parameters ca and cb can be derived.


n
n
qls
= 2 k2,i k1,i ca k2,i cb k3,i = 0 k1, j k2,i ca

ca

i=1
i=1
n
n
q
= 2 k3,i k1,i ca k2,i cb k3,i = 0 k1,i k3,i ca

cb

i=1
i=1

i=1

i=1

k2,i cb k3,i k2,i = 0


n

i=1

i=1

(4.20)

k2,i k3,i cb k3,i = 0

(4.21)
This results in the following equations for the estimates for the parameters c a and cb , with
2
K1 = n k1,i k2,i , K2 = n k2,i , K3 = n k3,i k2,i , K4 = n k1,i k3,i , K5 = K3 and K6 =
i=1
i=1
i=1
i=1
n
2 .
i=1 k3,i
1
K4 KKK6
3
(4.22)
c a =

K2 K6
K3 K3

4.3. Estimation of the heat conductivity from heat penetration experiments

147

K1 ca K2

(4.23)
K3
The estimated heat conductivity of the solid particle mixture as function of temperature in case
x=10 mm and for xi =20 mm, xi =30 mm and xi =40 mm is shown in gure 4.13.
c b =

1.1

Heat conductivity [W m1 K1 ]

1.0

PSfrag replacements

0.9

0.8

0.7
0.6
0.5

True heat conductivity

0.4
0.3
0.2
300

400

500

600

700

800

900

1000 1100 1200 1300

Temperature [K]

Figure 4.13: True (dotted line) and estimated (solid lines) heat conductivity of the solid particle mixture
as function of temperature using the differential method with a-priori knowledge of =
f (T ). A, B and C indicate the estimated heat conductivity at x i =20 mm, xi =30 mm and
xi =40 mm, respectively. The spacing between the ctive thermocouples is x is 10 mm.
The properties of the solid particle mixture and the initial and boundary conditions, with
which the ctive heat penetration experiment is simulated, are listed in table 4.1. The true
heat conductivity is given by true =0.2 + 3.0 104 T. The estimated heat conductivity for
the different positions in the solid particle mixture are A =0.279 + 3.93 104 T, B =0.267
+ 4.41 104 T and C =0.228 + 6.79 104 T.

Although a-priori information of the temperature dependency of the heat conductivity is taken
into account, the minimum deviation between the true and estimated heat conductivity equals
34 %.
The integral method
The integral form of equation 4.14 is given by
c p

x+
x

T
t

x =

x+
x

x
x

x,

(4.24)

148

Chapter 4. Heat conductivity of glass forming batches

in which the spatial dependency of the temperature and the heating rate are described by a 2 nd
order polynomial given by T (x) = m1 x2 + m2 x + m3 and T /t = n1 x2 + n2 x + n3 . The values
for n1 , n2 , n3 , m1 , m2 and m3 at a certain time t and position x are determined by the method
described in Appendix C. The integration domain around position i is given by x, which is
dened by 0.5 (xi1 + xi ), and x+, which is dened by 0.5 (xi + xi+1 ). Integration of equation
4.24 results in
x+
n1 3 n2 2
cp
x + x + n3 x
= ca + cb T [2m1 x + m2 ]x+ .
(4.25)
x
3
2
x
Figure 4.14 shows both the estimated and the true heat conductivity of the solid particle mixture
as function of temperature for xi =20 mm and x=10 mm calculated with equation 4.25. It is
observed that the difference between the true and the estimated heat conductivity is only very
small, which indicates that the integral method provides an accurate estimation of the heat
conductivity in case the model parameters and c p , the temperatures and the positioning of the
ctive thermocouples are known exactly beforehand.

0.65

Heat conductivity [W m1 K1 ]

0.60

0.55
0.50
0.45
0.40
0.35
0.30

PSfrag replacements
400

600

800

1000

1200

1400

Temperature [K]

Figure 4.14: True and estimated heat conductivity of the solid particle mixture as function of temperature for xi =40 mm and x=10 mm with a-priori information on = f (T ) according to
the integral method. Line 1 indicates the true thermal heat conductivity, which is given by
= 0.20 + 3.0 104 W m1 K2 . Line 2 indicates the estimated heat conductivity, which
is given by = 0.18 + 3.2 104 W m1 K1 .

Next, the sensitivity of the estimated temperature dependent heat conductivity on errors in

4.3. Estimation of the heat conductivity from heat penetration experiments

149

the exactly known solid particle mixture properties, the measured temperatures and the positions
of the thermocouples is studied. According to Coleman and Steele [17], the uncertainty 9 in a
parameter Y , which is depending on n parameters X, viz. Y = Y (X1 , .., Xi , .., Xn ), is given by
UY =

Y
UX
X1 1

Y
UX
+ ... +
Xi i

Y
UX
+ ... +
Xn n

2 0.5

(4.26)

in which UXi is the uncertainty in parameter Xi . The effect of random errors on the estimated
temperature dependent heat conductivity is studied in order to identify
the effect of uncertainties in the model parameters10 on the uncertainty in the estimated
heat conductivity, and
the model parameter, which has the major contribution to the uncertainty in the estimated
heat conductivity.
The uncertainty in the estimated heat conductivity at position i in the solid particle mixture is
given by
U = f U ,Uc p ,UTi1 ,UTi ,UTi+1 ,Uxi1 ,Uxi ,Uxi+1 ,UT0 ,UTb .
(4.27)
To identify the major source for the uncertainty in the estimated heat conductivity, the percentual
contribution of the uncertainty of each parameter (, c p , Ti1 , Ti , Ti+1 , xi1 , xi , xi+1 , T0 and Tb )
to the squared total uncertainty of the estimated heat conductivity, which is given by equation
4.28, are derived by additional simulations with the numerical model.

UPCi =

Y
Xi

2
UY

2
UXi

100%.

(4.28)

During these additional simulation studies, the effect of small changes on the a-priori known
model parameters on the estimated value for the heat conductivity is determined. The small
change of the parameter values is dened by the uncertainty of the specic model parameter.
The contribution of the uncertainty of parameter i to the squared total uncertainty of the heat
conductivity is given by the so-called Uncertainty Percentage Contribution of a parameter i [17]
(see equation 4.28).
Figure 4.15 shows the true and the estimated heat conductivity of the solid particle mixture
as function of temperature in case x=10 mm for xi =20 mm and xi =40 mm. The uncertainty in
the model parameters which are used to calculate the uncertainty in the estimated heat conductivity are U =5 %, Uc p =5 %, UT =5 K and Ux =1 mm. The lines A indicate the condence limits
in case xi =20 mm, whereas the lines B indicate the condence limits in case x i =40 mm.
It is observed that the accuracy of the heat conductivity estimated from the temperature at
xi =20 mm is higher (i.e. the condence limits of the estimated heat conductivity are smaller)
than for xi =40 mm. This indicates that the uncertainty in the estimated temperature dependent
heat conductivity increases with decreasing temperature gradients in the solid particle mixture.
9 The

uncertainty of a measured parameter is described by the standard deviation of the measured parameter in
case the spread of measured values are normal distributed around the mean value.
10 With model parameters is ment the packed bed properties , c and the measured temperatures and the meap
sured positions.

150

Chapter 4. Heat conductivity of glass forming batches

1.4

Heat conductivity [W m1 K1 ]

1.2

1.0
0.8

vity
ducti

n
eat co
ated h
Estim

0.6
0.4

0.2

PSfrag replacements
0.0
400

600

800

1000

1200

1400

1600

Temperature [K]

Figure 4.15: True and the estimated heat conductivity of the packed bed as function of temperature in
case x=10 mm for xi =20 mm and xi =40 mm. The estimation is performed with a-priori
information of = f (T ) according to the integral method. Line 1 indicates the true thermal
heat conductivity, which is given by = 0.20 + 3.0 104 W m1 K2 . The lines A indicate
the condence limits in case xi =20 mm, whereas the lines B indicate the condence limits
in case xi =40 mm.

An increased accuracy of the heat conductivity is obtained in case the heat conductivity is estimated at a position in the solid particle mixture where large temperature gradients are present.
Figure 4.16 shows the contribution of the parameter (, c p , Ti1 , Ti , Ti+1 , xi1 , xi and xi+1 ) to
the squared total uncertainty of the heat conductivity as function of temperature. It is observed
that the major source for the inaccuracy in the estimated heat conductivity is the positioning of
the thermocouples. This indicates that for improving the accuracy of the estimated heat conductivity, the uncertainty in the positioning of the thermocouples should be increased. In the
following section, the third mathematical technique, i.e. the numerical-experimental technique
is discussed.
Numerical-experimental technique
For the estimation of the effective heat conductivity of a solid particle mixture from heat penetration measurements, also a numerical-experimental technique [18] was used. This technique
consists of three parts, i.e. an experiment, a mathematical simulation of the experiment and an

4.3. Estimation of the heat conductivity from heat penetration experiments

151

70
60

50

UPC [%]

40
30

B
C

20

PSfrag replacements

10

DH

0
400

600

800

1000

1200

1400

1600

Temperature [K]

Figure 4.16: The contribution of the parameter (, c p , Ti1 , Ti , Ti+1 , xi1 , xi and xi+1 ) to the squared total
uncertainty of the heat conductivity, indicated by the Uncertainty Percentage Contribution
as function of temperature. The lines A, B and C indicate the UPC values for the parameters
xi , xi1 and xi+1 , respectively. The lines D H indicate the UPC values for , c p , Ti1 , Ti ,
and Ti+1 , respectively.

estimation algorithm (see gure 4.17).


During the experiment, the temperatures at different positions x in the solid particle mixture
are measured as function of time, with xT = x1 , .., x j , .., xnrT , j is the thermocouple number and nrT is the total number of thermocouples in the solid particle mixture. The measured temperatures in the solid particle mixture at time ti are stored in the column mi , with
mT = Ti,1 , .., Ti, j , .., Ti,nrT . During the numerical simulation of the experiment, the temperai
tures at the same positions x in the solid particle mixture as in the experiment are calculated by
solving the one-dimensional thermal energy equation
cp

T
T

=
t
x
x

(4.29)

For the simulation of the heat penetration in the solid particle mixture, the nite element package
SEPRAN [16] is used. The thermal heat equation is solved in one-dimension, because in the
experimental set-up, which is discussed in section 4.4, the horizontal temperature gradient in
the solid particle mixture can be neglected with respect to the vertical temperature gradient in

152

Chapter 4. Heat conductivity of glass forming batches

mi

Experiment

PSfrag replacements

Numerical simulation of experiment

yi

Hi

Estimation algorithm

Figure 4.17: Scheme of the numerical-experimental technique according to Op den Camp et al. [18],
in which u contains the input data for the (numerical simulation of the) heat penetration
process, mi contains the measured temperatures at time i in the packed bed, y i contains the
calculated temperatures at time i, i contains the differences of the measured and calculated

temperatures at time i in the solid particle mixture, H i is the sensitivity matrix at time i,

contains the values for the estimates of the a-priori unknown parameters and 0 contains the
initial estimates for the unknown parameters.

the solid particle mixture. The calculated temperatures at the different positions in the solid
particle mixture at time ti are stored in the column yi , with yT = yi,1 , .., yi, j , .., yi,nrT . The xed
i
input data for the numerical simulation of the experiment, which are stored in the column u,
enclose:
the thickness dspm of the solid particle mixture,
the initial temperature of the solid particle mixture T0 ,
the time-dependent temperatures at the boundaries of the solid particle mixture Tb1 (t) and
Tb2 (t),
the a-priori known or estimated values for the density and the heat capacity c p of the
solid particle mixture, and
the positions xi of the thermocouples in the solid particle mixture.
Next to these xed input parameters, the numerical simulation of the heat penetration process
also requires information of the model parameters to be estimated. In the current study, rst estimates for the model parameters describing the temperature dependent heat conductivity given
by, for this case, = ca + cb T are required. The estimates for these model parameters are
T


stored in the column , with = c , c . The initial estimates of these model parameters is
a

4.3. Estimation of the heat conductivity from heat penetration experiments

153

stored in the column 0 .


Because the initial estimates for the model parameters ca and cb will deviate from the true

values of the model parameters (0 = true ), the calculated temperatures at time ti (yi ) differ
from the measured temperatures at time ti (mi ). The deviation column of temperatures at time ti
is given by
(4.30)
i = mi yi .

To obtain a more accurate simulation of the measured heat penetration in the solid particle mix
ture with respect to the estimation based on the initial estimates 0 , improved estimates for the
model parameters are required. The improved estimates are determined by the estimation algorithm, which minimizes the deviation matrix i with respect to the unknown model parameters
by a least squares approach. For this minimization procedure, a weighted square criterion q ls
is dened, which is given by
n

qls =

i=1

mi y i

W i mi yi

(4.31)

The matrix W i is a time dependent positive weighting matrix, with which differences in reliability of the different measured temperatures can be accounted for. According to Hendriks [19],
the matrix W i has to be chosen such that temperatures in the solid particle mixture, which are
measured with high accuracy have a higher contribution in the least square criterion q ls compared to temperatures which are measured with a lower accuracy. Because the thermocouples,
which are used in the experimental set-up to measure the temperature as function of position
in the solid particle mixture, are made from the same thermocouple wire, it is assumed that the
values indicated by the thermocouples have an identical accuracy. Now, the matrix W i equals
the unity matrix.
The temperatures at time t = ti in the solid particle mixture can be determined by

yi = hi (),

(4.32)

in which hi () characterizes the numerical calculation of the temperatures at the different posi
tions in the solid particle mixture at time t = ti . In this specic case, the column consists of
the estimates for the model parameters ca and cb , which describe the temperature dependent
effective heat conductivity of the solid particle mixture according to = c a + cb T . The calculated temperatures in the solid particle mixture at time t = t i depend in a non-linear way upon

the model parameters ca and cb . Therefore, the operation hi () is non-linear in the model
parameters and an iterative procedure is required to estimate .
The new estimates for the model parameters during the iterative estimation procedure are
derived by minimization of the criterion qls with respect to the unknown model parameters ,
which results in (see Appendix B):
n

k+1 = k +

i=1

HT Hi
i

in which
Hi =

HT
i

i=1

hi ()
,

mi hi (k ) .

(4.33)

(4.34)

154

Chapter 4. Heat conductivity of glass forming batches

and k and k+1 are the estimates of the model parameters after iteration k and k + 1. The iter

ative estimation procedure is repeated until k+1 k has become smaller than a predened
value 0 > 0.
In the following, the accuracy of the numerical-experimental technique is studied by estimating the model parameters from a ctive heat penetration experiment. The measured temperatures from the ctive heat penetration experiment are derived by numerical simulation of the
heat penetration process with xed input parameters.
Accuracy of the numerical-experimental technique in case no uncertainty in the input
parameters is present
The accuracy of the numerical-experimental technique is studied by estimating the model parameters from a ctive heat penetration experiment. The ctive heat penetration experiment is
obtained by numerical simulation of the heat penetration in a solid particle mixture with the
input data as described in table 4.1. The initial estimated values for the parameters c a and cb ,

which are stored in 0 , are 0.1 W m1 K1 and 1.0 104 W m1 K2 , respectively. Figure
4.18 shows the calculated temperature at 5 mm from the boundary of the solid particle mixture
as function of time.
1100
1000

900

Temperature [K]

800
700
600
500
400

PSfrag replacements

300
0

15

30

45

60

75

90

105

120

135

150

Time [min.]

Figure 4.18: Calculated temperature in the solid particle mixture at 5 mm from the hot side of the solid
particle mixture as function of time. The black dots indicate the experimental data, from
which the unknown material parameters ca and cb are estimated.

Because only two model parameters have to be estimated, only two measured temperatures

4.3. Estimation of the heat conductivity from heat penetration experiments

155

are required for estimation of these unknown model parameters. Figure 4.19 shows the estimation for the model parameters ca and cb as function of the iteration number in case the
temperature values 1 and 2 in gure 4.18 are used for the estimation procedure. It is observed
that for both model parameters only ve iterations are required to obtain the true value of these
model parameters.

PSfrag replacements

-4

3.0x10

-4

0.35

2.5x10

-4

0.30

2.0x10

-4

0.25

1.5x10

-4

0.20

1.0x10

-4

0.15

5.0x10

-5

0.10

0.0

0.05

-5.0x10
0

Model parameter cb [W m1 K2 ]

3.5x10

0.40

Model parameter ca [W m1 K1 ]

0.45

-5

Iteration number

Figure 4.19: Estimations of model parameter ca and cb as function of the iteration number. The true
values for ca and cb equal 0.2 W m1 K1 and 3 104 W m1 K2 , respectively. The
initial estimates for ca and cb equal 0.1 W m1 K1 and 1 104 W m1 K2 , respectively.

Accuracy of the numerical-experimental technique in case uncertainties in the input parameters are present
Similar to the integral method, the accuracy of the numerical-experimental technique in case
uncertainties in the input parameters u are present is calculated by equation 4.26. The uncertainties in the model parameters are U =5 %, Uc p =5 %, UT =5 K and Ux =1 mm. Figure 4.20
shows
The estimated heat conductivity at xi =20 mm in the packed bed (thick solid line), which
equals the true heat conductivity of the solid particle mixture, estimated with the numericalexperimental method.
The estimated heat conductivity at xi =20 mm in the packed bed (thick dotted line) estimated with the integral method.
The condence limits of the estimated heat conductivity at x i =20 mm in the solid particle
mixture (thin solid lines) estimated with the numerical-experimental method.

156

Chapter 4. Heat conductivity of glass forming batches

The condence limits of the estimated heat conductivity at x i =20 mm in the solid particle
mixture (thin dotted lines) estimated with the integral method.
It is observed that the accuracy of the estimated heat conductivity derived from the numericalexperimental technique is higher compared to the accuracy of the estimated heat conductivity
derived from the integral method. Further throughout this section, this mathematical technique
is used to estimate the heat conductivity of solid particle mixtures.

1.2
1.1

Heat conductivity [W m1 K2 ]

1.0
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
400

PSfrag replacements

600

800

1000

1200

1400

1600

Temperature [K]

Figure 4.20: The estimated heat conductivity at xi =20 mm in the solid particle mixture estimated with
both the numerical-experimental method (thick solid line) and the integral method (thick)
dotted line. Also the condence limits of the estimated heat conductivity at x i =20 mm in
the solid particle mixture derived from the numerical-experimental method (thin solid lines)
and the integral method (thin dotted lines) are indicated.

4.4

Experimental set-up for measuring the heat penetration


in solid particle mixtures

Figure 4.21 shows a cross-sectional view of the experimental set-up, which was developed for
measuring the temperatures as function of time at different positions in a mixture of solid particles. In the experimental set-up, the mixture of solid particles is heated in vertical direction by
conductive and radiative heat transfer from a radiative SiC-plate, which is in direct contact with

The heating section, which is a box with a square surface area of 600 mm x 600 mm with a
height of 300 mm, contains 1873 K insulation bricks. In the insulation bricks, a square aperture
is made of 250 mm x 250 mm and a depth of 150 mm. In this aperture, 6 SiC elements with a
maximum power of 5.4 kW each are positioned. The aperture is closed on top by a SiC plate. In
the aperture, a Pt/Pt-10% Rh (type S) thermocouple is positioned. To avoid overheating of the
SiC elements, the maximum temperature allowed temperature in this section is xed at 1673 K.
On top of the SiC-plate, a second SiC-plate is positioned, which separates the heating section
and the sample section. This additional SiC-plate is present to avoid damage of the SiC heating
elements if the SiC-plate which is in direct contact with the solid particle mixture breaks. The
heating elements in the heating section were controlled by a control unit based on the value of
the thermocouple, which is positioned between the two SiC-plates.
On top of the heating section, the sample section is located. The sample section is insulated
by ceramic board, with an aperture of 250 mm x 250 mm. The height of the sample section is
150 mm. The aperture is surrounded by highly insulating ceramic board of 170 mm thickness.
The width of the aperture and the thickness and properties of the ceramic insulation board are
chosen in such a way that the heat penetration in the mixture of solid particles can be regarded
as one dimensional (vertical) heat transport. Later in this section, heat transport in only vertical
Figure 4.21: Cross-sectional view of the experimental set-up, with which the temperature as function of
time at different positions in a mixture of solid particles is measured.
r

Thermocouple joints
Heating elements
1873 K
insulation bricks 5.4 kW each
600 mm

Top
section

Ceramic blocks for


tightening thermocouple wires

Sample
section

Heating
section

SiC plate
1873 K
insulation bricks
Ceramic board

250 mm
Furnace lid

frag replacements

Thermocouple wires

the solid particle mixture. The experimental set-up consists of 3 sections, i.e. a heating section,
a sample section and a top section.
4.4. Experimental set-up for measuring the heat penetration in solid particle mixtures 157

158

Chapter 4. Heat conductivity of glass forming batches

direction in the solid particle mixture is proven.


The sides of the aperture in the sample section is covered at the inside by a thin layer of insulation bricks through which 0.35 mm Pt/Pt-10% Rh (type S) thermocouple wires are guided.
The joints of the thermocouples wires are positioned at dened vertical distances from the
SiC-plate at the center of the aperture. The positioning of the thermocouple wires is xed by
tightening the wires in ceramic blocks located on top of the sample section. Five thermocouple
wire joints are positioned at the center of the sample section at about 10 mm, 20 mm, 30 mm,
40 mm and 50 mm from the SiC-plate. A sixth thermocouple joint is positioned at a distance
of 30 mm from the SiC-plate and at a horizontal distance of about 10 mm from the center of
the sample section. This sixth thermocouple is used for comparing the heat ow in horizontal
direction relative to the heat ow in vertical direction in the solid particle mixture.
The thermocouple wires, which are connected with compensation wires, are connected to
a data acquisition system. This data acquisition system monitors and records all thermocouple
readings from the experimental set-up. On top of the sample section a furnace lid, which is
composed of insulation material, is located.
Experimental procedure
First, the bulk density of the mixture of solid particles is determined by measuring the weight
of a dened volume of the solid particle mixture. After positioning the thermocouple wires and
measuring their vertical position inside the sample section, the sample section is slowly lled
with the raw material batch at room temperature via the walls of the sample section to avoid
vertical displacement of the thermocouple wires. After lling the sample section up to a height
of approximately 8 cm, the furnace lid is put on top of the sample section. The set point of the
thermocouple, which is located between the two SiC-plates, is set to the desired nal temperature.
The thermocouples in the sample section were prepared from the same wire. The uncertainty in the thermocouple values is estimated to be 5 K over the whole temperature range.
As mentioned in the previous section, the uncertainty of the thermocouple readings is not the
major source for the uncertainty in the estimated values for the heat conductivity of the mixture of solid particles. The measured uncertainty level of the positioning of the thermocouple
joints equals 1 mm. It is assumed that the uncertainty of the positioning of the thermocouple
joints is independent on temperature. However, especially at high temperatures, at which the
thermocouple wire is slightly expanded, and in case large volumes of batch gases are released
during heating, the positioning of the thermocouples may uctuate slightly due to the ascension
of batch gases.
The heat conductivity of the mixture of solid particles is estimated by applying the numericalexperimental approach described in the previous section. The temperatures indicated by the
thermocouples, which are positioned at about 10 mm and 50 mm from the SiC-plate, serve as
transient boundary condition for the numerical simulation of the heat penetration experiment.
The value for the temperature dependent heat capacity of the mixture of solid particles is derived from thermodynamic tables [20].
Evaluation of the radial temperature gradient
To evaluate whether it is allowed to consider the heat ow in the experimental set-up as onedimensional, the horizontal and vertical temperature gradient at about 30 mm from the SiC-plate
is calculated from the temperatures in a silica sand batch with a bulk density of 1487 kg m 3 .

4.4. Experimental set-up for measuring the heat penetration in solid particle mixtures 159

The vertical and horizontal temperature gradients, expressed in K per mm, are approximated by
T
x
and

T
x

=
vert,x=30mm

=
hor,x=30mm

Tx=40mm,z=0mm Tx=20mm,z=0mm
,
20

(4.35)

Tx=30mm,z=0mm Tx=30mm,z=10mm
.
10

(4.36)

Figure 4.22 shows both the horizontal and the vertical temperature gradient as function of the
temperature at the center of the sample section, at 30 mm from the SiC-plate.

2
0

Temperature gradient [K mm1 ]

-2
-4
-6
-8
-10
-12
-14
-16
-18
-20
400

PSfrag replacements

600

800

1000

1200

Temperature [K]

Figure 4.22: Horizontal (dotted line) and vertical (solid line) temperature gradient as function of the
measured temperature at the center of a silica sand batch at 30 mm from the SiC-plate.

Figure 4.22 clearly shows that the horizontal temperature gradient is negligible compared to the
vertical temperature gradient, which indicates that the heat transport process in the solid particle
mixture can merely be regarded as a one-dimensional process.
Accuracy and reproducibility
The accuracy and reproducibility of the estimation of the thermal heat conductivity of a silica
sand batch from the measured heat penetration in the silica sand batch is performed by estimating the heat conductivity from two subsequent heat penetration experiments with the same

160

Chapter 4. Heat conductivity of glass forming batches

silica sand type. Figure 4.23 shows both the calculated heat conductivity of the silica sand
batches and the 95 % condence limits of the calculated heat conductivity values in case the
uncertainty in the positioning of the thermocouples equals 1 mm. The uncertainty in the ef0.70
0.65

Heat conductivity [W m1 K1 ]

0.60
0.55

0.50

0.45
0.40
0.35
0.30
0.25
0.20
0.15

PSfrag replacements

0.10
300

400

500

600

700

800

900 1000 1100 1200 1300 1400

Temperature [K]

Figure 4.23: Heat conductivity of a silica sand batch estimated with the numerical-experimental technique from two heat penetration experiments with the same silica sand type.

fective heat conductivity of the silica sand batch equals approximately 18 % at 1373 K. It is
seen that the heat conductivity estimated from the two subsequent heat penetration experiments
lies well within their 95 % condence limits. It was also shown that two sequential performed
heat penetration measurements provided almost similar estimates for the heat conductivity of
the silica sand batch.

4.5

Experimental determination of heat conductivity in a particle bed

In this section, the heat conductivity of solid particle mixtures, in the solid-state regime is studied. The identication of the most suitable structural model describing the heat conductivity of
solid particle mixtures composed of industrial grade raw material components of glass batches,
is discussed and shown for a silica sand batch in section 4.5.1. In section 4.5.2, it is investigated whether the identied structural model is also able to describe the heat conductivity of
multi-component mixtures.

4.5. Experimental determination of heat conductivity in a particle bed

161

4.5.1 Heat conductivity of a silica sand batch


The most suitable network model describing heat conductivity of solid particle mixtures is identied by comparison of the predicted heat conductivity of a mixture of silica sand and air for the
different structural models described in section 4.2.3 with the estimated heat conductivity from
heat penetration experiments in well dened silica sand batches. The bulk density of the silica
sand batch, in which the heat penetration was measured, equals 1487 kg m 3 . The porosity of
the silica sand batch is given by
b
p = 1
,
(4.37)
SiO2
in which b is the bulk density of the silica sand batch, and the SiO2 is the true density of silica.
The true density of silica crystals equals 2650 kg m3 , which results in a value for the porosity
of the silica sand batch equal to 0.44. The temperature dependent heat conductivity of air is
given by Wakao and Kaguei [13] and is given by
air = 1.20 102 + 5.50 105 T,

(4.38)

in which T is the temperature in K. The intrinsic temperature dependent heat conductivity of


crystalline silica is tted from gure 4.3 and equals
silica = 2.81 + 8.60 102 T 1 + 2.11 105 T 2 ,

(4.39)

in which T is the temperature in K. Figure 4.24 shows the calculated heat conductivity of the
silica sand batch for the different network models. Next to these calculated values for the
heat conductivity, also the heat conductivity estimated from the heat penetration experiment is
shown.
From gure 4.24 it can be seen that neither of the four heat conductivity models predicts a
temperature dependent heat conductivity for the silica sand batch equal to the heat conductivity
estimated from the heat penetration experiment. Although the SN- and the DPN-model seem
to provide a reasonable prediction of the temperature dependent heat conductivity of the silica
sand batch, the difference between the measured and predicted heat conductivity of the silica
sand batch11 equals 43 % and 66 % at 1273 K for the DPN- and the SNM-model, respectively.
It appears that the heat ow through the silica sand batch is described by an intermediate form
of the four heat conductivity models presented in section 4.2.3. It is also observed that, similar
to the DPNM- and the SNM-model, the heat conductivity of the silica sand batch increases
with increasing temperature. Because the intrinsic heat conductivity of silica decreases with
temperature and the heat conductivity of air increases with temperature, it appears that the
conduction of heat in the silica sand bed is predominated by the heat conductivity of air.
Figure 4.25 shows the schematic representation of the intermediate form of parallel and
serial arranged thermal resistances, which will be used to describe the heat ow through the
silica sand batch.
The heat conductivity of the silica sand batch is now given by
=

PNM

1
+ SNM

11 The difference of the measured and predicted heat conductivity is dened as est pred ,
est

(4.40)

in which est is the heat


conductivity, which is estimated from the heat penetration experiments and pred is the heat conductivity, which is
predicted by either of the two models.

162

Chapter 4. Heat conductivity of glass forming batches

Heat conductivity [W m1 K1 ]

PNM
CPNM

PSfrag replacements
DPNM
SNM

0
200

400

600

800

1000

1200

1400

Temperature [K]

Figure 4.24: The calculated heat conductivity of a particle bed composed of silica sand grains and air
according to the PN-, CPN-, DPN-, and the SN-model is indicated by solid lines. The dotted
line indicates the heat conductivity of the particle bed estimated from a heat penetration
experiment.

in which is the parameter, which distinguishes between the contribution of the parallel and the
serial arrangement of thermal resistances to the overall heat ow through the silica sand batch.
For the silica sand batch out of the heat penetration experiments, the experimental derived value
for equals 0.29. This value is calculated from equations 4.10, 4.11 and 4.40 using the intrinsic
value of the heat conductivity of silica and air given by equations 4.38 and 4.39, respectively.
This indicates that the major part of the silica sand batch can be regarded as a serial arrangement of solid particles and enclosures of gas phase. Now, the heat conductivity of the silica
sand batch is not only determined by the relative amount of silica sand and air per unit volume,
which is characterized by the porosity, but also by the mutual contact between the silica sand
particles. The parameter can be regarded as a measure for the heat transfer area between the
silica sand particles in the bed. The value for depends on both the porosity and the shape of
the silica particles. A low porous bed will have a higher contact area between the silica particles compared to a high porous bed. This results in a higher value for . Prediction of the heat
conductivity of a particle bed requires knowledge of the porosity p , which can easily be measured, and , which is not known a-priori and can not easily be determined. In the following,
the accuracy of the prediction of the heat conductivity of a silica sand batch as function of the

4.5. Experimental determination of heat conductivity in a particle bed

163

PSfrag replacements
RTg
RTg

T1

RTs

T2

Q
RTs
parallel
connection

serial
connection
(1-)

Figure 4.25: Structural arrangement of thermal resistances.

porosity assuming a constant value for is evaluated based on heat conductivity data provided
by Eligehausen [11]. Eligehausen measured the heat conductivity of a silica sand batch for two
different bulk densities, i.e. 1505 kg m3 and 1605 kg m3 . The value for the shape factor
for the silica sand batch with a bulk density of 1505 kg m 3 was calculated using equations
4.10, 4.11 and 4.40 and equals 0.24. Figure 4.26 shows the predicted, using equation 4.40,
versus the measured heat conductivity of the silica sand batch with a bulk density of 1605 kg
m3 assuming that the shape factor equals 0.24. The largest difference between the predicted
and measured heat conductivity is observed in the low temperature range and equals 17 % at
273 K. In the temperature range from 673 K up to 1273 K, the difference between the predicted
and measured heat conductivity is less than 2 %.
The measured temperature dependent heat conductivity of a silica sand batch with a bulk
density of 1605 kg m3 is given by
sand/air,=1605,measured = 0.235 + 3.07 104 T,

(4.41)

while the predicted temperature dependent heat conductivity of this silica sand batch is given
by
sand/air,=1605,predicted = 0.173 + 3.77 104 T.
(4.42)

The effect of the difference in the measured and the predicted temperature dependent heat conductivity of the silica sand batch with the bulk density of 1605 kg m 3 (see gure 4.26) on the
heating of this batch is evaluated by simulating the heat penetration in a silica sand batch with a
thickness of 10 cm, a heat capacity of 1200 J kg1 K1 , a bulk density of 1605 kg m3 , an initial temperature of 293 K, and a constant boundary temperature at both sides of the silica sand
batch layer of 1673 K. Figure 4.27 shows the temperature at the center of the silica sand batch
as function of time for both measured and predicted temperature dependent heat conductivity.
The maximum temperature difference over the total temperature range is 27 K after 60 minutes
of heating up.
Resuming, in this section it is shown that the heat conductivity of a silica sand batch as
function of bulk density can be predicted with a reasonable accuracy in case the shape factor is determined from a heat penetration experiment in a silica sand batch with a known
porosity. However, it is remarked that for dense batches, the parameter is likely to increase.
Also in case a solid particle bed is remained a long time at a certain high temperature, the heat
conductivity may become time dependent due to sintering of the solid particles.

164

Chapter 4. Heat conductivity of glass forming batches

Predicted heat conductivity [W m1 K1 ]

1.0

PSfrag replacements

0.8

0.6

T
0.4

0.2

0.0
0.0

0.2

0.4

0.6

Measured heat conductivity [W

0.8

m1

1.0

K1 ]

Figure 4.26: Predicted versus measured [11] heat conductivity of a silica sand batch with a bulk density
of 1605 kg m3 .

4.5.2 Heat conductivity for multi-component mixtures


Based on the previous section, the prediction of the heat conductivity of a bed of a multicomponent mixture of solid particles in a gas phase, requires knowledge of the following properties
the intrinsic heat conductivity of the pure solid components,
the intrinsic heat conductivity of the gas phase,
the porosity of the multi-component mixture, and
the shape of the overall mixture of the individual mixture components.
However, data on the intrinsic heat conductivity of the pure solid components, which are used
in glass batches, are scarce. Also the a-priori estimation of the net effect of the shapes of the
different mixture components on the heat conductivity of the multi-component mixture is not
easy. Therefore, it is decided to characterize the heat conductivity of a solid component by
an apparent heat conductivity, which contains both the intrinsic heat conductivity of the pure
solid component and the contribution of the shape factor . To determine the apparent heat
conductivity of a solid component i, a heat penetration experiment with a particle mixture of
solid component i is required.
The value for the apparent heat conductivity of silica sand, estimated from a heat penetration
experiment in a silica sand batch, is determined both regarding the silica sand batch as a serial

4.5. Experimental determination of heat conductivity in a particle bed

165

30
1600

25

15
10

1000

5
800

600
400

PSfrag replacements

-5

Temperature difference [K]

20

1200

Temperature [K]

1400

-10

20

40

60

80

100

120

140

160

-15
180

Time [min]

Figure 4.27: Calculated temperature as function of time at the center of a silica sand batch with a thickness of 10 cm. The heat capacity and the density of the silica sand batch equal 1200 J kg 1
K1 and 1605 kg m3 , respectively. The initial temperature and the constant boundary temperature at both sides of the silica sand batch layer equal 293 K and 1673 K, respectively.
For the solid line, equals 0.173 + 3.77 104 T , whereas for the dotted line, equals
0.235 + 3.07 104 T .

and as a parallel arrangement of thermal resistances. In case of a serial arrangement of thermal


resistances, negative values for the apparent heat conductivity of silica sand were obtained.
Therefore, the apparent heat conductivity of a solid phase i, estimated from a heat penetration
experiment in a batch containing solid phase i as the only solid phase, is given by a parallel
arrangement according to
p g
i,app =
.
(4.43)
1 p
Assuming that the net apparent heat conductivity of a mixture of solid components is given by
n

s,app = xi i,app ,

(4.44)

i=1

in which n is the number of solid species in the mixture, x i is the weight fraction of solid phase
i, and i,app is the apparent heat conductivity of solid phase i, the heat conductivity of a multicomponent mixture is estimated by
= p g + (1 p )

xi i,app .

i=1

(4.45)

166

Chapter 4. Heat conductivity of glass forming batches

The porosity of the multi-component mixture is given by


p = 1 bulk

xi true,i

(4.46)

i=1

in which true,i is the true intrinsic density of component i.


The applicability of equation 4.45 describing the heat conductivity of multi-component particle mixtures, is evaluated and checked for a three component mixture containing industrial
grade silica sand, soda ash and limestone. The values for the effective heat conductivity of
these three components are listed in table 4.2. The values for both silica sand and limestone
have been determined with a heat penetration experiment, whereas for the data for soda ash is
referred to Eligehausen [11].
Table 4.2: effective heat conductivity of silica sand, soda ash and limestone in case the effective heat
conductivity is given by = ca + cb T The regression coefcients of this equation are determined with using equation 4.43.
Specie
Silica sand
Soda ash
Limestone

c a
[W m1 K1 ]

c b
[W m1 K2 ]

0.18
0.01
0.36

5.40 104
6.46 105
2.84 104

Figure 4.28 shows the measured temperatures as function of time at different positions in a
three component particle bed containing silica sand, soda ash and limestone. From these heat
penetration curves, the heat conductivity of the three-component mixture is estimated using the
numerical-experimental technique.
Figure 4.29 shows the predicted versus the measured effective heat conductivity of the threecomponent mixture. It appears that for the three-component mixture, with a bulk density of
1394 kg m3 , the predicted effective heat conductivity deviates less than 2 % from the measured effective heat conductivity. This indicates that accurate values for the heat conductivity
of glass batches in the solid-state regime can be predicted in case both the batch composition
and the porosity of the glass batch are known. However, it is remarked that this validation experiment is performed only once. It is recommended to perform a similar experiment in case an
other type of batch is used.

4.5.3 Heat conductivity of particle beds containing cullet


In contrast to opaque industrial glass batch components, such as silica sand, dolomite and limestone, it is expected that a glass batch containing cullet is (partly) transparent for radiative heat
transfer. In the following part, the contribution of radiative heat transfer to the net effective heat
conductivity of two solid partly transparent particle mixtures is discussed.

4.5. Experimental determination of heat conductivity in a particle bed

1000

x=0 mm
x=10 mm
x=20 mm
x=30 mm
x=40 mm

900

Temperature [K]

800

PSfrag replacements

167

700
600
500
400
300
0

50

100

150

200

250

300

350

400

450

Time [min]

Figure 4.28: Measured temperature as function of time at different positions in a batch layer containing
silica sand, soda ash and limestone.

Predicted app [W m1 K1 ]

0.40

PSfrag replacements

0.35

0.30

0.25

0.20

0.15
0.15

0.20

0.25

0.30

Measured app [W

m1

0.35
K1 ]

0.40

Figure 4.29: Predicted versus measured effective heat conductivity for a glass batch composed of silica
sand, soda ash and limestone.

168

Chapter 4. Heat conductivity of glass forming batches

One mixture is composed of oat glass cullet12 (48.6 wt.%) and silica sand (51.4 wt.%)
and a second mixture, contains only oat glass cullet. According to section 4.2.2, the photon conductivity is proportional to the absolute temperature to the power of 4. The effective
heat conductivity of the two mixtures is estimated with the numerical-experimental method
described in section 4.3.3. The temperature dependent effective heat conductivity is tted by
= ca + cb T + cc T 3 , in which ca + cb T represents the conduction of heat and cc T 3
represents the radiation of heat. The percentual contribution of radiative heat transfer to the net
effective heat conductivity of the mixtures, , is in this study estimated by
=

c c T 3
100%.
c a + c b T + c c T 3

(4.47)

Figures 4.30 and 4.31 show the estimated temperature dependent effective heat conductivity of
the two mixtures and the contribution of radiative heat transfer to the net effective heat conductivity. From gure 4.30 it is observed that the contribution of radiative heat transfer in the
mixture containing both silica sand and oat glass cullet in the temperature range up to 1173 K
is less than 14 %. This indicates that the major mode of heat transfer in this mixture (of a typical
batch layer thickness) is phonon conduction. Photon conduction (or radiative heat transfer) is of
minor importance, which is likely because the amount of non-radiative conducting silica sand is
sufcient to block radiative heat transport through the thick layer of the solid particle mixture.

PSfrag replacements

14

0.46
0.44

12

0.42
10

0.40
0.38

0.36
6

0.34
0.32

[%]

Effective heat conductivity [W m1 K1 ]

0.48

0.30

0.28
0.26
300

400

500

600

700

800

900

1000

1100

0
1200

Temperature [K]

Figure 4.30: Temperature dependent effective heat conductivity and the contribution of radiative heat
transfer to the net effective heat conductivity for a bed composed of oat glass cullet (48.3
wt.% ) and silica sand (51.7 wt.% ).

Figure 4.31 shows that for the 100 % oat glass cullet mixture a remarkable higher effective
12 The

particle size of the oat glass cullet ranges from 4 mm up to 8 mm

4.5. Experimental determination of heat conductivity in a particle bed

169

heat conductivity is estimated compared to the mixture containing both oat glass cullet and
silica sand. Also the contribution of radiative heat transfer to the net effective heat conductivity
of the 100 % oat glass cullet mixture is higher. For this mixture, radiative heat transfer contributes to a large extent to the effective heat conductivity of the mixture.

PSfrag replacements

2.5

90
80

2.0

70
60

1.5

50
40

1.0

[%]

Effective heat conductivity [W m1 K1 ]

100

30
20

0.5

10
0.0
300

400

500

600

700

800

900

1000

1100

0
1200

Temperature [K]

Figure 4.31: Temperature dependent effective heat conductivity and the contribution of radiative heat
transfer to the net effective heat conductivity for a 100 % oat glass cullet bed.

The effect of the higher effective heat conductivity from mixtures with transparent cullet
grains can easily be seen from gures 4.32 and 4.33, which show the temperatures as function
of time at different positions in both solid particle mixtures.
Resuming, these experiments indicate that radiative heat transfer in glass batches increase
by the use of cullet as raw material. However, at least below a weight fraction of cullet in a
glass batch of 50 %, the contribution of radiative heat transfer to the total heat transfer in the
glass batch is not or hardly noticeable, which agrees with the observation mentioned by Faber et
al. [5]. The benecial effect of glass cullet on the heating rate of glass batches is only apparent
for cullet fractions above 50 wt.%. As mentioned in section 4.2.4, it is expected that the photon
conduction in glass batches is not only dependent on the cullet fraction in the glass batch, but
because of photon scattering, also on the particle size and the surface roughness of the cullet. With the determination of the contribution of radiative heat transfer to the net effective
heat conductivity, the effect of these cullet properties on the effective heat conductivity can be
determined in future studies according to the here presented method.

170

Chapter 4. Heat conductivity of glass forming batches

1100
1000

PSfrag replacements

Temperature [K]

900

x=0 mm
x=10 mm

800
700
600

x=20 mm
x=30 mm

500
400

x=40 mm

300
0

10

20

30

40

50

60

70

80

90

100

Time [min]

Figure 4.32: Measured temperature as function of time at different positions in a 100 % oat glass cullet
batch. The cullet particle size varies between 4 mm and 8 mm.

1100

x=0 mm

1000

x=10 mm

PSfrag replacements

Temperature [K]

900
800
700
600

x=20 mm

500

x=30 mm

400

x=40 mm

300
0

10

20

30

40

50

60

70

80

90 100 110 120 130 140

Time [min]

Figure 4.33: Measured temperature as function of time at different positions in a mixture containing
silica sand (51.7 wt.%) and oat glass cullet (48.3 wt.%). The cullet particle size varies
between 4 mm and 8 mm.

4.6. Concluding remarks

4.6

171

Concluding remarks

The purpose of this section was three-fold, viz:


1. Identication of the most accurate mathematical method to estimate the temperature dependent heat conductivity in powder layers from heat penetration experiments.
2. The development of an experimental set-up to measure the temperatures at different positions in a solid particle mixture as function of time.
3. Modelling of the temperature dependent heat conductivity of solid particle mixtures based
on the estimated heat conductivity of single type particle mixtures.
The most accurate mathematical method to estimate the temperature dependent heat conductivity of solid particle mixtures from heat penetration experiments is identied by comparison
of the accuracy of the estimated heat conductivity derived from ctive heat penetration experiments with different mathematical models. The most accurate mathematical method appeared
to be the numerical-experimental method during which the heat conductivity is estimated from
heat penetration experiments by iteratively adapting the relation of the heat conductivity with
temperature in a numerical model simulating the heat penetration in a solid particle layer in a
ctive experiment.
With this mathematical method, the heat conductivity can be estimated using only one thermocouple value in contrast to the other discussed techniques for which at least three temperature
values are required. The numerical-experimental method appeared to be less sensitive for inaccuracies in the other a-priori known or estimated material properties compared to the other
described mathematical methods. A sensitivity analysis for the numerical-experimental method
showed that the accuracy of the positioning of the thermocouple(s) in the solid particle mixture
is the major source for the uncertainty in the estimated heat conductivity.
Next, structural models for the description of heat ow through a three dimensional arrangement of thermal resistances simulating the heat transport in a solid particle mixture are
examined. It is shown that the effective heat conductivity of mixtures of solid particles can
be predicted based on the (apparent) heat conductivity of the individual solid particle mixture
components and the particle bed porosity. Finally, a start was made with the determination of
the effect of glass cullet in the batch on the effective heat conductivity of solid particle mixtures.
It was shown by estimation of the contribution of radiative heat transfer to the net effective
heat conductivity, that the effect of cullet on the radiative heat conductivity is only observed in
case the weight fraction of cullet in the particle bed exceeds at least 50 %.

172

Chapter 4. Heat conductivity of glass forming batches

4.7

Nomenclature

Latin symbols
a
A
c
c,a
c,b
c,c
cp
d
dp
Hi
Hchem
i
l
L
mi
n
qls
q
Q
R
RT
T
t
U
UPC
v
yi

absorption coefcient
cross section
heat capacity
constant in the temperature dependent heat conductivity given by = c,a + c,b T + c,c T 3
constant in the temperature dependent heat conductivity given by = c,a + c,b T + c,c T 3
constant in the temperature dependent heat conductivity given by = c,a + c,b T + c,c T 3
heat capacity
diameter
pore diameter
sensitivity matrix at time ti
chemical energy demand
indicator
mean free path
length
column containing measured temperatures at
time ti
refractive index
least square criterion
heat ux
heat ow
universal gas constant
heat resistance
temperature
time
uncertainty
uncertainty percentage coefcient
velocity
column containing calculated temperatures at
time ti

[m1 ]
[m2 ]
[J m3 K1 ]
[W m1 K1 ]
[W m1 K2 ]
[W m1 K4 ]
[J kg1 K1 ]
[m]
[m]
[J mol1 ]
[m]
[m]
[K]
[-]
[W m2 ]
[W]
[J K1 mole1 ]
[K W1 ]
[K]
[s]
[%]
[m s1 ]
[K]

Greek symbols
p

porosity
emissivity
column containing the differences in measured
and calculated temperatures at time ti
contribution of radiative heat transfer to the net
effective heat conductivity
heat conductivity
frequency

[-]
[-]
[K]
[%]
[W m1 K1 ]
[s1 ]

4.7. Nomenclature

packed bed parameter


density
Stefan-Boltzmann constant
column containing the values for the parameters
to be estimated c,a and c,b

Sub- and superscripts


app
b
c
CPNM
d
DPNM
eff
g
int
mn
PNM
0
r
s
SNM
spm
z

apparent
boundary
continuous phase
continuous phase network model
dispersed phase
dispersed phase network model
effective
gas phase
intrinsic
mean
parallel network model
initial
radiative
solid phase
serial network model
solid particle mixture
vertical direction
estimates

173

[kg m3 ]
[W m2 K4 ]

174

4.8

Bibliography

Bibliography

[1] A. Ungan and R. Viskanta. Melting behavior of continuously charged loose batch blankets
in glass melting furnaces. Glastech. Ber., 59(10):279291, 1986.
[2] H.R.S. Jack and J.A.T. Jacquest. Heat transfer in glass batch materials. Symposium sur la
fusion du verre, Bruxelles, pages 339360, 1958.
[3] M. Daniels. Einschmelzverhalten von Glasgemengen. Glastechn. Ber., 46(3):4046, 1973.
[4] P. Costa. Untersuchung des Einschmelzverhaltens von pelletiertem Gemenge zur
Glasherstellung. Glastech. Ber., 50(1):1018, 1977.
[5] A.J. Faber, R.G.C. Beerkens, and H. de Waal. Thermal behaviour of glass batch on batch
heating. Glastech. Ber. Glass Sci. Technol., 7(7):177185, 1992.
[6] R. Conradt, P. Suwannathada, and P. Pimkhaokham. Local temperature distribution and
primary melt formation in a melting batch heap. Glastech. Ber. Glass Sci. Technol.,
67(5):103113, 1994.
[7] J. Krzoska, N. Samkham, and R. Conradt.
Bestimmung der lokalen
Temperaturleitf higkeit in aufschmelzendem Gemenge. In 73. Glastechnische Tagung,
a
pages 214217, Halle, Germany, 1999.

[8] C. Kr ger and H. Eligehausen. Uber das W rmeleitverm gen des einschmelzenden
o
a
o
Glasgemenges. Glastechn. Ber., 32(9):362373, 1959.
[9] W.D. Kingery, H.K. Bowen, and D.R. Uhlmann. Introduction to ceramics. John Wiley &
Sons, 2nd edition, 1976.
[10] A.J. Slifka, B.J. Filla, and J.M. Phelps. Thermal conductivity of magnesium oxide from
absolute, steady-state measurements. J. Res. Natl. Inst. Stand. Technol., 103:357363,
1998.

[11] H. Eligehausen. Uber die W rmeleitf higkeit von Gl sern und einschmelzenden Glasgea
a
a
mengen. PhD thesis, Rheinisch-Westf lischen Technischen Hochschule Aachen, 1957.
a
[12] D. Mann, R.E. Field, and R. Viskanta. Determination of specic heat and true thermal conductivity of glass from dynamic temperature data. Warme- und Stoff bertragung, 27:225
u

231, 1992.
[13] N. Wakao and S. Kaguei. Topics in chemical engineering: Vol. 1 - Heat and mass transfer
in packed beds. Gordon and Breach, Science Publishers, Inc., New York-London-Paris,
1st edition, 1982.
[14] P. Zehner and E.U. Schl nder. W rmeleitf higkeit von Sch ttungen bei m ssigen
u
a
a
u
a
Temperaturen. Chemie Ing. Techn., 42(14):933941, 1970.
[15] J.P. Holman. Heat Transfer. McGraw-Hill Book Company, 5 edition, 1981.

Bibliography

175

[16] A. Segal. SEPRAN users manual, standard problems and programming guide. Ingenieursbureau SEPRA, Leidschendam, The Netherlands, 1993.
[17] H.W. Coleman and W.G. Steele. Experimentation and uncertainty analysis for engineers.
John Wiley & Sons, Inc., 2nd edition, 1999.
[18] O.M.G.C. Op den Camp, C.W.J. Oomens, F.E. Veldpaus, and J.D. Janssen. An efcient algorithm to estimate material parameters of biphase mixtures. Int. J. Numer. Meth. Engng.,
45:13151331, 1999.
[19] M.A.N. Hendriks. Identication of the mechanical behaviour of solid materials. PhD
thesis, Eindhoven University of Technology, Eindhoven, The Netherlands, 1991.
[20] O. Knacke, O. Kubaschewski, and K. Hesselmann. Thermochemical Properties of Inorganic Substances. Springer-Verlag Berlin, Heidelberg, Germany, 2 nd edition, 1991.

176

Bibliography

Chapter 5
Complete simulation model for the heating
of glass forming batches
5.1

Energy conservation equation and temperature dependent glass batch properties

The simulation of the heating of glass batches requires the solution of the energy conservation
equation of the glass batch together with appropriate boundary conditions describing the heat
transport to the top and boundary of the glass batch. According to section 1.4, the Lagrangian
description of the energy equation of the two-phase glass batch is given by
Hchem
,
t
(5.1)
in which cp T mn represents the mean value of the enthalpy of the glass batch, g and cp,g
are the density and heat capacity of the gas phase, Tmn is the mean temperature of the glass
batch, t is the time, p is the porosity of the glass batch, vg,z is the vertical velocity of the gas
phase relative to the condensed phase1 , eff is the effective heat conductivity of the glass batch,

q r,eff is the effective radiative heat ux through the glass batch, and Hchem is the temperature
dependent energy per unit volume of the glass batch required for batch reactions. For a complete
description of the heating of glass batches, values for the parameters m , c p,m , p , g , c p,g , vg,z ,

eff , r,eff , and Hchem are required. The temperature dependent average heat capacity of a
q
reacting glass batch is given by

cp T
t

mn

q
= p g cp,g vg,z Tg (eff Tmn ) + r,eff + (1 p )

nb

cp,mn = wi c p,i ,

(5.2)

i=1

in which nb is the number of (intermediate) glass batch components (including the gas species),
wi is the time and temperature dependent weight fraction of component i and c p,i is the temperature dependent heat capacity of component i. The heat capacity of the individual glass batch
components can be derived from thermodynamic tables [2].
In chapter 2 of this thesis, the energy required for chemical reactions was studied. According to Madivate [3], the chemical reaction energy is mainly dependent on the energy required
1 It

is assumed that the horizontal velocity of the gas phase is similar to the horizontal velocity of the condensed
phase [1].

177

178

Chapter 5. Complete simulation model for the heating of glass forming batches

for calcination reactions. It was shown that for a oat glass batch, the CO 2 -release from the
complete batch can be described by three (almost) independent calcination reactions, i.e. the
thermal calcination of dolomite and limestone and the reactive calcination of soda ash. For the
estimation of the time and temperature dependent chemical energy demand Hchem of the oat
glass batch, this allows the determination of the kinetics of the individual calcination reactions.
Combination of the measured calcination kinetics of the individual decomposition reactions
with the calcination enthalpies, which were derived from thermodynamic tables [2], resulted in
an expression for the chemical energy required for complete calcination of the oat glass batch
as function of time, temperature and partial CO2 -pressure:
nc
Hr,i w0 c i
Hchem
i
,
=
t
Mi
t
i=1

(5.3)

in which nc is the total number of carbonates in the glass batch, Hr,i is the enthalpy required
for the calcination of carbonate i expressed in J mole1 , w0 is the initial weight fraction of
i
carbonate i in the condensed phase, c is the density of the condensed phase, Mi is the molar
mass of carbonate i, i is the time and temperature dependent degree of conversion2 of carbonate
i, and t is the time. The conversion rate of carbonate i is given by
Ea,i
i
= Ai e R T (1 i ) ,
t

(5.4)

in which A is a pre-exponential factor and Ea is an apparent reaction activation energy. The


reaction energy and the reaction kinetic parameters for the three calcination reactions occurring
during heating of the oat glass batch are listed in table 5.1.
Table 5.1: Reaction enthalpy and reaction kinetic parameters for the calcination reactions occurring during heating of the oat glass batch.
Reaction
Thermal decomposition of dolomite
Thermal decomposition of limestone
Reactive decomposition of soda ash

Hr,i
[kJ mol1 ]

Ai
[s1 ]

Ea,i
[kJ mol1 ]

129
172
118

2.58 1022
1.80 107
4.72 1021

513
191
500

Below 1200 K, the effective heat conductivity of solid particle mixtures such as glass batches
is in the order of 0.1-1.0 W m1 K1 [4]. Above 1200 K, the heat transfer process in glass
batches is relatively fast compared to temperatures below 1200 K, because the effective heat
conductivity of melting glass batches increases steeply above 1200 K [46]. Because of the low
effective heat conductivity below 1200 K, the residence time of a glass batch in this temperature
range is relatively long. To predict the heating process of a glass batch accurately, detailed data
on the low temperature effective heat conductivity of glass batches are required. In chapter 4,
the temperature dependent effective heat conductivity of solid particle mixtures containing glass
batch components such as silica sand, soda ash and limestone is determined up to about 1200
2 The

degree of conversion of a carbonate is dened as the ratio of the weight loss with respect to the total
weight loss when the carbonate has completely been dissociated.

5.1. Energy conservation equation and temperature dependent glass batch properties 179

K, quantitatively. It was shown that the temperature dependent heat conductivity of the solid
particle mixture can be estimated from
eff = p g + (1 p )

ns

wi i,app.

(5.5)

i=1

in which g is the true heat conductivity of the gas phase in the solid particle mixture, i,app is
the apparent heat conductivity of solid component i, n s is the number of solid particles in the
solid particle mixture and p is the porosity of the solid particle mixture which is given by
p = 1 bulk

ns

wi true,i

(5.6)

i=1

in which true,i is the true intrinsic density of component i. The true temperature dependent
heat conductivity values of different gas phases are given by Wakao and Kaguei [7]. The in this
study determined apparent heat conductivity of silica sand, soda ash and limestone is given by
app = c,a + c,b T , with c,a and c,b listed in table 5.2.
Table 5.2: Apparent heat conductivity of silica sand, soda ash and limestone in case the apparent heat
conductivity is given by app = c,a + c,b T .
Specie

c,a
[W m1 K1 ]
0.18
0.01
0.36

Silica sand
Soda ash
Limestone

c,b
[W m1 K2 ]
5.40 104
6.46 105
2.84 104

In this study, the effect of oat glass cullet on the effective radiative heat ux in solid particle
mixtures was studied. The effective radiative heat ux was estimated by

q r,eff = r,eff Tm ,

(5.7)

in which r,eff is the effective radiative heat conductivity, which is described by r,eff = c,c T 3 .
The contribution of radiative heat transfer to the net total effective heat conductivity in solid
particle mixtures containing cullet appeared to be dependent on the cullet fraction in the solid
particle bed. Below a cullet fraction of 50 wt.%, the effective radiative heat conductivity was
only about maximum 13 % at 1200 K. A more extensive study to the effect of cullet fraction and
cullet size on the heat penetration in solid particle mixtures is required to model the contribution
of the effective radiative heat transfer coefcient to the net effective heat conductivity in solid
particle mixtures as function of these cullet parameters.
In the solid particle mixture, the position dependent vertical gas velocity can be calculated
from a local gas balance describing the incoming gas ux from below and the generation of gas
due to the calcination reactions. The position dependent vertical gas velocity in the solid particle
mixture containing dolomite, silica sand and soda ash due to the calcination of the carbonates
can be calculated from
vg,z |z=z =

R Tg
p p As

z=z

m,g |z=zdz nc As dz w0 c i
i
+
Mi
t
Mg
i=1

(5.8)

180

Chapter 5. Complete simulation model for the heating of glass forming batches

in which vg,z |z=z is the vertical gas velocity at position z = z, R is the universal gas constant, p
is the pressure in Pa, As is the cross section of the solid particle mixture volume perpendicular
on the gas ow, m,g |z=zdz is the mass ow of the gas phase in kg s 1 at position z = z dz,
M g is the average molar weight of the gas phase and nc is the total number of carbonates in the
glass batch.
In this study, no quantitative description of the complete melting of glass batches is given.
To simulate this process, additional information is required concerning the change in mean
density c p,m and porosity p of the glass batch. The change in the porosity directly affects the
effective phonon conductivity as is shown by equation 5.5.

5.2

Heat transfer towards a batch blanket

The heat transfer towards the boundaries of a batch blanket in an industrial furnace is dependent on both (free and forced) convective and radiative heat transport. In glass tank simulation
models, no heat transfer coefcient at the bottom of the batch blanket is required, because
the heat transfer between the glass melt underneath the batch blanket and the batch blanket
is calculated by coupling the differential equations describing the energy conservation of both
sections.
In the following section, a description is given of the heat transfer process to the top layer
of the batch blanket and the major mode of heat transfer is identied.

5.2.1 Forced convective heat transfer


Approximating the batch blanket as a at plate, the forced convective heat transfer from the
hot combustion gases towards the relative cold batch blanket can be estimated by standard
empirically found heat transfer relations for gas ow over a at plate. These empirical heat
transfer relations are based on heat transfer over a thermal boundary layer, which is build up
when a free streaming gas ow meets the tip of the at plate as is indicated in gure 5.1. In
the case that a gas ow with uniform velocity reaches the tip of the at plate, the gas ow in
the vicinity of the surface of the at plate is retarded due to frictional forces. In case of no-slip
conditions at the surface of the at plate, the gas velocity equals zero at the at plate surface.
As a consequence of viscous forces, the gas velocity parallel to the at plate decreases in the
direction of the plate and a gaseous boundary layer with a velocity gradient is build up. Because
of differences in gas ow parallel to the at plate, also stronger temperature gradients are build
up resulting in a thermal boundary layer. The thickness of the thermal boundary layer increases
with increasing distance from the leading edge3 of the at plate for laminar ow. Near the
leading edge of the at plate the ow in the boundary layer is laminar. At a certain distance
from the leading edge, the ow in the boundary layer may become turbulent. The ow type in
the boundary layer, which is either laminar or turbulent, is dependent on the gas ow velocity,
the surface structure of the plate, the properties of the gas phase and the distance from the
leading edge. In general, for a smooth plate, the transition from the laminar ow regime into
the turbulent ow regime is characterized by a Reynolds number, calculated with equation 5.9,
equal to 5.0 105 [8].
3 The

tip of the at plate, which is the rst position where the gas ow hits the at plate

PSfrag replacements
5.2. Heat transfer towards a batch blanket

Transition
regime

Laminar regime
v , T

v=v

181

Turbulent
regime

T=T

Gas ow

Leading edge v=0

T=Ts

Flat plate

Figure 5.1: Schematic view of boundary layer in which v is the free stream gas velocity, T is the free
stream temperature, Ts is the surface temperature of the at plate, x is the ow direction of
the gas stream and is the boundary layer thickness for velocity which is dependent on x.

The Reynolds number at distance x from the leading edge is given by


Rex =

v x
,

(5.9)

in which is the density of the gas phase, v is the free stream gas phase velocity and is the
dynamic viscosity of the gas phase. Although the critical Reynolds number for the transition
from laminar to turbulent ow regime is about 5.0 10 5 , the practical transition value is strongly
dependent on the roughness of the surface layer and on the turbulence level of the free stream
gas ow. With a high degree of surface roughness and with large turbulences present in the free
stream, the Reynolds number for transition may shift down to values of 2.0 10 5 [8].
The heat transfer coefcient by forced convection from the hot combustion gases to the
batch blanket surface at position x, hfo,x , can be described by the local Nusselt number given by
Nux =

hfo,x x
,
g

(5.10)

in which g is the heat conductivity of the gas phase in the boundary layer. The heat ux per
unit batch area at position x is now described by
qfo,x = hfo,x (T Ts ) ,

(5.11)

in which Ts is the temperature of the surface of the batch blanket and T is the free stream
temperature of the gas. Characteristic values for the over length L averaged Nusselt number
NuL describing the forced convective heat transfer towards the batch blanket are given by
NuL,lam = 0.664 Pr1/3 Re1/2

(5.12)

for the laminar ow regime [9], and


NuL,turb = 0.664 Pr1/3 Re1/2 + 0.0359 Re0.8 Re0.8 Pr1/3
L
x,crit

(5.13)

for the turbulent ow regime [9], respectively. The Prandtl number Pr is described by
Pr =

cp
,

(5.14)

182

Chapter 5. Complete simulation model for the heating of glass forming batches

in which c p is the heat capacity of the gas phase in the thermal boundary layer.
To determine the forced convective heat transfer coefcient from the hot combustion gases
to the batch blanket in an industrial furnace, typical values for the free stream gas velocity and
temperature and the gas phase composition have to be known. However, the assumptions that
have been made for deriving the Nusselt numbers are a uniform gas ow (v ) and a constant
temperature for the batch blanket surface (Ts ), which are not full-lled in practice. To estimate
the forced convective heat transfer coefcient, characteristic values for the free gas stream velocity and temperature have to be estimated.
For a known gas phase composition, the gas phase properties can be calculated at the known
gas phase temperature. Because there is a temperature difference over the thermal boundary
layer, the gas phase properties will change with the distance to the batch blanket. To take this
variation in gas phase properties into account, the gas phase properties are evaluated at the
so-called lm temperature T f , which is the mean value of the surface temperature Ts and the
free-stream temperature T . Table 5.3 lists a typical gas phase composition for an oxy-red
glass melting furnace.
Table 5.3: Gas phase composition
Gas phase component

Molar fraction

CO2
H2 O
N2
O2

0.33
0.58
0.07
0.02

The density of the gas phase is calculated assuming ideal gas behavior in which p is the pressure, M is the mean molar mass of the gas phase which equals 27.6 g mol 1 for this gas phase
and T is the temperature expressed in K.
pM
.
(5.15)
RT
The dynamic viscosity of a gas phase mixture, mix , is calculated by the semi-empirical formula
of Wilke given in [10] and described by
=

ng

wi i
,
n
i=1 j=1 w j i j

mix =

(5.16)

in which ng is the number of gas phase components, i is the viscosity of the pure gas component
i, and wi and w j are the molar fractions of the components i and j in the gas phase. The constant
i j is given by
1
i j =
8

Mi
1+
Mj

1/2

i
1+
j

1/2

Mj
Mi

1/4 2

(5.17)

in which Mi and M j are the molar weights of components i and j in the gas phase mixture [10].
The viscosity of a pure gas phase component is estimated by

Mi T
5
,
(5.18)
i = 2.6693 10
2 i
i

5.2. Heat transfer towards a batch blanket

183

in which i is the Lennard-Jones collision diameter of gas species i, T is the temperature and i
is the collision integral. The collision integral i is dependent on the dimensionless temperature
i T /i . The ratio i /i is specic for each gas phase component and is tabulated for different
gas phase species in [10].
Similar to the description of the viscosity of the gas phase mixture, the thermal heat conductivity of a gas phase mixture, mix , is given by
n

wi i
.
n
i=1 j=1 w j i j

mix =

(5.19)

The thermal heat conductivity of a pure gas phase component is given by the Chapman-Enskog
formula given by
T /Mi
,
(5.20)
i = 1.9891 104 2
i i
in which i is the thermal heat conductivity of the pure gas phase component i. The general
equation describing the temperature dependency of the heat capacity of a gas phase component
is given by [11]
c p,i = c p,A,i + c p,B,i T + c p,C,i T 2 + c p,D,i T 3 + c p,E,i T 4 ,

(5.21)

in which c p,A , c p,B , c p,C , c p,D and c p,E are coefcients which are specic for each gas phase
component and can be derived from thermodynamic tables [11] and
n

c p,mean = wi c p,i .

(5.22)

i=1

Figure 5.2 shows the calculated Reynolds number of the gas ow over a at plate as a function
of the average lm temperature in case of a free-stream gas velocity of 10 m s 1 and a typical
plate length of 5 m. These values are representative for situations of batch blankets in glass
furnaces.
It is observed that for the current case, the Reynolds number is lower than the critical Reynolds
number at which the transition to the turbulent ow regime takes place for smooth plates. In
this ow regime, the heat transfer coefcient is in the order of 3.5 W m 2 K2 . However, as
mentioned above, the critical Reynolds number for the transition from the laminar ow regime
into the turbulent ow regime for a rough batch surface and gas ows with turbulence eddies
may be lower than the 5.0 105 . Assuming that the critical Reynolds number equals 1.0 10 5 ,
the forced convective heat transfer coefcient, which is calculated with equation 5.13, varies
between 7 - 10 W m2 K2 .

5.2.2 Free convective heat transfer coefcient


According to Holman [9], the average free convective heat transfer coefcient for the case of a
horizontal at plate with a gas phase above the at surface can be derived from
Nu = C (Gr Pr)m ,

(5.23)

Chapter 5. Complete simulation model for the heating of glass forming batches

4.5x10

4.0x10

3.5x10

3.0x10

2.5x10

2.0x10

Reynolds number [-]

5.0x10

3.75

3.70

3.65

3.60

3.55

PSfrag replacements
1100

1200

1300

1400

1500

1600

Average hc,fo [W m2 K1 ]

184

3.50
1700

Film temperature T f [K]

Figure 5.2: Reynolds number (solid line) and average forced convective heat transfer coefcient for laminar ow (dotted line) as function of the lm temperature.

in which C and m are constants which are both dependent on the geometry of the system and
Gr is the Grashof number. The Grashof number is dened by
Grx =

g (Ts T ) x3 2
2

(5.24)

in which g is the gravitational acceleration and is the volumetric expansion coefcient of the
gas phase mixture. For the composition given in table 5.3, the Rayleigh number (GrPr) ranges
from 2.4 109 down to 1.1 109 in the temperature range from 1073 K up to 1673 K. For these
Rayleigh numbers, the values for C and m are 0.15 and 0.33, respectively [9]. This results in a
heat transfer coefcient of approximately 1.5 W m2 K2 .

5.2.3 Radiative heat transfer coefcient


According to Holman [9], the radiative heat transfer coefcient h r between two parallel plates
with surface areas of A1 and A2 is given by
hr =

T12 + T22 (T1 + T2 )


1
1

1
1 + F + A1
A2

1
2

(5.25)

in which is the Stefan-Boltzmann constant, T1 and T2 are the temperatures of the plates, 1
and 2 are the emissivities of the plates and F is the fraction of energy which is emitted from
plate 1 and reaching plate 2. To estimate the order of magnitude of the radiative heat transfer
coefcient from the hot superstructure of a glass furnace (or a ame) towards the top layer of

5.2. Heat transfer towards a batch blanket

185

500

sup = 0.6

PSfrag replacements

Radiative heat transfer coefcient [W m2 K1 ]

450
400

sup = 0.5

350

sup = 0.4

300
250

sup = 0.3

200

sup = 0.2

150
100
50
0
400

600

800

1000

1200

1400

1600

Surface temperature batch blanket [K]

Figure 5.3: Radiative heat transfer coefcient as a function of the surface temperature of the batch blanket for different emissivity levels of the superstructure in case that the superstructure temperature equals 1773 K and the emissivity of the batch blanket equals 0.7.

the batch blanket, it can be assumed that the surface area of the superstructure perpendicular to
the surface of the superstructure and the batch blanket are similar and that all energy leaving the
superstructure reaches the surface of the batch blanket. The radiative heat transfer coefcient
for this situation is now given by
hr =

2
Tsup + Tb2 (Tsup + Tb )
1

1
sup + b 1

(5.26)

in which Tsu and Tb are the temperatures of the superstructure (or ame) and the surface of the
batch blanket. The emissivities of the surfaces are indicated by sup and b , respectively.
Figure 5.3 shows the radiative heat transfer coefcient hr as function of the surface temperature of the batch blanket for different emissivity levels of the superstructure in case of a
temperature of the crown of 1773 K. According to Ungan et al., a typical value for b is 0.7 [1].
From this gure it is observed that, for this idealized case, the radiative heat transfer coefcient is about two orders of magnitude larger than the convective heat transfer coefcient. This
indicates that heat transfer by radiation is the main mode for heat transfer towards the batch
blanket.

186

Chapter 5. Complete simulation model for the heating of glass forming batches

5.3

Nomenclature

Latin symbols
A
As
cp
c p,A
c p,B
c p,C
c p,D
c p,E
c,a
c,b
c,c
C
Ea
F
Gr
h
Hchem
Hr
i, j
m
M
M
nb
nc
ng
ns
Nu
NuL
p
Pr
q
Re
R

pre-exponential factor
[s1 ]
heat transfer area per unit volume
[m1 ]
heat capacity
[J kg1 K1 ]
heat capacity constant
[J kg1 K1 ]
heat capacity constant
[J kg1 K2 ]
heat capacity constant
[J kg1 K3 ]
heat capacity constant
[J kg1 K4 ]
heat capacity constant
[J kg1 K5 ]
constant in the temperature dependent heat con- [W m1 K1 ]
ductivity given by = c,a + c,b T + c,c T 3
constant in the temperature dependent heat con- [W m1 K2 ]
ductivity given by = c,a + c,b T + c,c T 3
constant in the temperature dependent heat con- [W m1 K4 ]
ductivity given by = c,a + c,b T + c,c T 3
geometry dependent constants in Nu =
C (Gr Pr)m
(apparent) reaction activation energy
[J mol1 ]
Fraction of radiative energy emitted from a sur- [-]
face, which reaches an opposite surface
Grashof number
[-]
heat transfer coefcient
[W m2 K1 ]
chemical energy demand
[J m3 ]
reaction enthalpy
[J mol1 ]
indicators
geometry dependent constants in Nu =
C (Gr Pr)m
molar mass
[kg mol1 ]
mean molar mass
[kg mol1 ]
number of (intermediately formed) glass batch [-]
components
number of different types of carbonates in a [-]
glass batch
number of different types of gases
[-]
number of different types of solid particles in a [-]
glass batch
Nusselt number
[-]
over length L averaged Nusselt number
[-]
pressure
[Pa]
Prandtl number
[-]
heat ux
[W m2 ]
Reynolds number
[-]
universal gas constant
[J K1 mole1 ]

5.3. Nomenclature

t
T
T
v
v
w
x

187

time
temperature
free stream temperature
velocity
free stream velocity
weight fraction
horizontal position

[s]
[K]
[K]
[m s1 ]
[m s1 ]
[-]
[-]

volumetric expansion coefcient


emissivity
porosity
Lennard-Jones potential parameter
heat conductivity
dynamic viscosity
collision integral
mass ow rate
constant in equation 5.17
density
bulk density
true density
Lennard-Jones collision diameter
degree of conversion

[m3 ]
[-]
[-]
[K]
[W m1 K1 ]
[kg m1 s1 ]
[-]
[kg s1 ]

Greek symbols

p
/

m
i j

bulk
true

Subscripts
app
b
c
crit
eff
fo
g
lam
mn
mix
r
s
su
turb
z

apparent
batch
condensed phase
critical
effective
forced
gas phase
laminar
mean
mixture
radiative
surface
superstructure
turbulent
vertical direction

[kg m3 ]
[kg m3 ]
[kg m3 ]
[m]
[-]

188

5.4

Bibliography

Bibliography

[1] A. Ungan and R. Viskanta. Melting behavior of continuously charged loose batch blankets
in glass melting furnaces. Glastech. Ber., 59(10):279291, 1986.
[2] O. Knacke, O. Kubaschewski, and K. Hesselmann. Thermochemical Properties of Inorganic Substances. Springer-Verlag Berlin, Heidelberg, Germany, 2 nd edition, 1991.
[3] C. Madivate, F. M ller, and W. Wilsmann. Thermochemistry of the glass melting process
u
- energy requirement in melting soda-lime-silica glasses from cullet-containing batches.
Glastech. Ber. Glass Sci. Technol., 69(6):167178, 1996.

[4] C. Kr ger and H. Eligehausen. Uber das W rmeleitverm gen des einschmelzenden
o
a
o
Glasgemenges. Glastechn. Ber., 32(9):362373, 1959.
[5] A.J. Faber, R.G.C. Beerkens, and H. de Waal. Thermal behaviour of glass batch on batch
heating. Glastech. Ber. Glass Sci. Technol., 7(7):177185, 1992.
[6] R. Conradt, P. Suwannathada, and P. Pimkhaokham. Local temperature distribution and
primary melt formation in a melting batch heap. Glastech. Ber. Glass Sci. Technol.,
67(5):103113, 1994.
[7] N. Wakao and S. Kaguei. Topics in chemical engineering: Vol. 1 - Heat and mass transfer
in packed beds. Gordon and Breach, Science Publishers, Inc., New York-London-Paris,
1st edition, 1982.
[8] W.S. Janna. Engineering heat transfer. PWS Publishers, Hong Kong, 1988.
[9] J.P. Holman. Heat Transfer. McGraw-Hill Book Company, 5th edition, 1981.
[10] R.B. Bird, W.E. Stewart, and E.N. Lightfoot. Transport phenomena. John Wiley & Sons,
Inc., New York, 1960.
[11] Gasunie. Physical properties of natural gases. N.V. Nederlandse Gasunie, 1980.

Appendix A
Calcination of a TV-panel glass batch
A.1

Description of the calcination mechanism of a TV-panel


glass forming batch

As discussed in section 2.1, the chemical energy demand of glass batches is to a large extent
dependent on the energy required for decomposition reactions, such as calcination reactions.
For the estimation of the time and temperature dependent chemical energy demand of a glass
batch, the kinetics of the different decomposition reactions are required. For the oat glass
batch, which was discussed in section 2.4, the calcination of dolomite, limestone and soda ash
occurred (almost) independently from each other. This allows the determination of the kinetics
of the individual calcination reactions to describe the calcination behavior of the complete oat
glass batch. Combination of the measured kinetics of the individual calcination reactions with
the calcination enthalpies, which were derived from thermodynamic tables [1], resulted in an
expression for the chemical energy required for complete calcination of the oat glass batch as
function of time, temperature and partial CO2 -pressure.
Kr mer [2] and Kawachi et al. [3] measured the release of batch and ning gases as function
a
of temperature for a TV-panel glass batch. It was observed that the release of CO 2 occurs in
a broad temperature range from 700 K up to 1500 K with a maximum CO 2 release between
1100 K and 1200 K. To determine the chemical energy required for complete calcination of a
TV-panel glass batch as function of time, temperature and partial CO 2 -pressure, similar to the
oat glass batch, the kinetics of the different calcination reactions occurring during heating of
the TV-panel glass batch need to be determined. In this study, the calcination behavior of a TVpanel glass batch, with as major constituents silica sand, soda ash, potash (K 2 CO3 ), potassium
nitrate (KNO3 ), nepheline (Na2 O Al2 O3 SiO2 ), strontium carbonate (SrCO3 ), barium carbonate (BaCO3 ), and zircon silicate (ZrO2 SiO2 ), is investigated.
In section A.2, it is shown by thermogravimetric analysis that, in contrast to the decomposition of the earth-alkaline carbonates MgCO3 and CaCO3 in the oat glass batch, the decomposition of the earth-alkaline carbonates SrCO3 and BaCO3 occurs via reactive calcination
with primary formed melt phases. In the oat glass batch, the primary formed melt phases are
mainly generated by the reactive calcination of soda ash with silica sand (see section 2.5.3). In
the TV-panel glass batch, the primary melt phases are (at least partly) formed by a combination
of reactive calcination of both potash and soda ash with silica sand. Thermogravimetric analy189

190

Appendix A. Calcination of a TV-panel glass batch

sis of a mixture of potash and silica sand1 shows a similar calcination behavior as mixtures of
soda ash and silica sand. The measured onset temperature for reactive calcination of potash in
a mixture with silica sand in a 1 bar CO2 -atmosphere is about 970 K, which equals the onset
temperature for reactive soda ash calcination with silica sand.
As mentioned above, the reactive calcination of the earth-alkaline carbonates SrCO 3 and
BaCO3 in the TV-panel glass batch is dependent on the presence of melt phases. After the
primary melt phases have been formed, the amount of melt phase in the TV-panel glass batch
increases due to the (reactive) dissolution of silica sand, nepheline, strontium carbonate, barium
carbonate (BaCO3 ), and zircon silicate. Now, the reactive calcination of the earth-alkaline carbonates SrCO3 and BaCO3 is not only dependent on the primary formed melt phases, but also
on the formation of the melt phases generating by the other batch reactions. The dependency
of the reactive calcination of SrCO3 and BaCO3 on different batch reactions does not allow the
easy determination of the kinetics of the individual calcination reactions in the TV-panel glass
batch. Therefore, in contrast to the oat glass batch, an expression for the chemical energy
required for complete calcination of the TV-panel glass batch as function of time, temperature
and partial CO2 -pressure cannot easily be determined.
Although detailed modelling of the kinetics of the different calcination reactions in the TVpanel glass batch is complex, the calcination of the complete TV-panel glass batch can be
approximated by identication of the temperature regimes in which the different calcination
reactions occur. Based on the results of the thermogravimetric analysis, both soda ash and
potash decompose under a 1 bar CO2 -atmosphere in the temperature range from 970 K up to
about 1120 K. The onset temperature for the reactive calcination of the earth-alkaline carbonates SrCO3 and BaCO3 equals the onset temperature for reactive calcination of potash and soda
ash. The decomposition of these carbonates may last up to about 1250 K. This maximum temperature for reactive calcination of the earth-alkaline carbonates is derived from phase analysis
on quenched TV-panel batch samples, which were ramp heated with 10 K min 1 in an ambient
atmosphere in the temperature range from 920 K up to 1720 K (see gure A.1 and [4]).
From gure A.1 it is also observed that at 920 K, all TV-panel batch components, except
KNO3 and K2 CO3 , are still present in the heat-treated batch. The reason for the absence of
crystalline KNO3 is the low melting temperature of KNO3 (607 K). This indicates that in the
TV-panel glass batch, already at 607 K energy is consumed, which is used for the KNO 3 melting. Although, based on the thermogravimetric analysis, it is expected that K 2 CO3 is still
present at 900 K, it is not identied with X-ray diffraction. The cause for not identifying crystalline K2 CO3 has not been studied in detail, but may be due to low diffraction peak intensities
of K2 CO3 and peak overlap of the diffraction lines of K2 CO3 with other batch components.
Because the detailed analysis of the calcination rate of the complete TV-panel batch is timeconsuming, the chemical energy demand of the TV-panel batch is not easy to determine. Before
the kinetics of the individual calcination reactions in the TV-panel glass batch is studied in detail, it should be known how important a detailed description of the chemical energy demand is
for an accurate prediction of the heating process of a glass batch.
A rst estimation of the chemical energy required for complete calcination of the TV-panel
glass batch can be obtained by combination of assumed degree of conversion of the different
carbonates in their specic calcination temperature range with the enthalpy of these calcination
reactions. Because melt phases are formed during the calcination of K 2 CO3 , Na2 CO3 , SrCO3 ,
1 These

thermogravimetric analysis are not reported in this thesis.

A.1. Description of the calcination mechanism of a TV-panel glass forming batch

191

SiO2 (cr)
Na2 O 2BaO 2SiO2
Na2 O 2SrO 2SiO2
PSfrag replacements
2SrO SiO2
Na2 O SiO2
SiO2 (q)
ZrO2 SiO2
Na2 O Al2 O3 SiO2
BaCO3
SrCO3
Na2 CO3
K2 CO3
KNO3
900

1000

1100

1200

1300

1400

1500

1600

1700

Temperature [K]

Figure A.1: Identied crystalline species during heating of a complete TV-panel glass batch containing silica sand, soda ash, potash, potassium nitrate, strontium carbonate, barium carbonate,
nepheline, zircon and some minor constituents. The TV-panel batch is ramp heated with 10
K min1 in an ambient atmosphere. Phase identication is performed between 1020 K and
1720 K with temperature steps of 50 K. SiO2 (q) and SiO2 (cr) indicate SiO2 in the quartzand cristobalite modication, respectively.

and BaCO3 , the determination of the enthalpy of the calcination reactions requires the knowledge of the thermodynamic properties of a glass melt containing SiO 2 , K2 O, Na2 O, SrO and
BaO. To estimate the thermodynamic properties of this glass melt, a thermodynamic model as
discussed by Shakhmatkin et al. [5] and Conradt [6] is used. Therefore, for a description of the
chemical energy demand of the TV-panel glass batch, rst the thermodynamics of the TV-panel
glass needs to be determined.
In the next section, a description is given of the calcination behavior of the TV-panel glass
batch carbonates SrCO3 and BaCO3 . The conditions for either thermal or reactive calcination
of SrCO3 - and BaCO3 -grains during heating is studied by
thermogravimetric analysis of (particle mixtures containing) SrCO 3 -grains heated in different atmospheres, and
phase analysis on quenched samples of heat-treated glass batch mixtures containing SrCO 3
or BaCO3 .

192

A.2

Appendix A. Calcination of a TV-panel glass batch

Calcination of SrCO3- and BaCO3-grains

This section describes the results of an experimental study into the calcination behavior of both
SrCO3 - and BaCO3 -grains.
Figure A.2 shows the standard Gibbs free energy for the thermal decomposition of strontium
carbonate and barium carbonate as function of temperature.

rag replacements

Standard Gibbs free energy of reaction [J mol1 ]

1.2x10

1.0x10

8.0x10

6.0x10

4.0x10

2.0x10

104

103

BaCO

3 (s)

SrCO

3 (s)

Ba

Sr

0.0

102

O(s)+
C

O2 (g)

O(s)+
C

101
100.3

O2 (g

100

-2.0x10

100.7

-4.0x10

101

-6.0x10

-8.0x10

102

1200

1300

1400

1500

1600

Temperature [K]

Figure A.2: Standard Gibbs free energy of spontaneous thermal decomposition of SrCO 3 and BaCO3 .
The dashed lines represent isobars for the partial CO2 -pressure in bar above the carbonates.

The calcination temperature for strontium carbonate and barium carbonate in a 1 bar CO 2 atmosphere is 1510 K and 1646 K, respectively. Figure A.3 shows the degree of SrCO 3 calcination measured with thermogravimetric analysis for the following four cases:
individual SrCO3 -grains in a N2 -atmosphere,
individual SrCO3 -grains in a 1 bar CO2 -atmosphere,
a mixture of SrCO3 - and sand grains in a 1 bar CO2 -atmosphere, and
a mixture of SrCO3 -, soda ash- and sand grains in a 1 bar CO2 -atmosphere.

A.2. Calcination of SrCO3 - and BaCO3 -grains

193

From gure A.3, it is observed that the measured onset temperature for thermal calcination of
SrCO3 in a 1 bar CO2 -atmosphere equals 1540 K (curve B), which is 30 K higher than the
calcination temperature of 1510 K in a 1 bar CO2 -atmosphere which is determined by thermodynamic calculation (see gure A.2). In gure A.3, the impact of the partial CO 2 -pressure on
the thermal calcination of SrCO3 is clearly shown by the shift of the temperature dependent
degree of SrCO3 calcination with about 300 K to lower temperatures when changing from a 1
bar CO2 -atmosphere (curve B) to a N2 -atmosphere (curve A). This indicates that the calcination
of individual SrCO3 grains is (at least partly) governed by thermodynamic driving forces.

PSfrag replacements

Degree of SrCO3 calcination [-]

1.0

0.8

0.6

0.4

0.2

0.0
900

1000

1100

1200

1300

1400

1500

1600

Temperature [K]

Figure A.3: Measured degree of the SrCO3 calcination in A) a N2 -atmosphere, B) a 1 bar CO2 atmosphere, C) when mixed with sand in a 1 bar CO2 -atmosphere and D) when mixed with
soda ash and sand in a 1 bar CO2 -atmosphere.

According to thermodynamics, a mixture of SrCO3 and SiO2 may form crystalline strontium
silicates via a solid-state calcination reaction according to
2 SrCO3 (s) + SiO2 (s)

2SrO.SiO2 (s) + 2 CO2 (g),

(A.1)

SrO.SiO2 (s) + CO2 (g).

(A.2)

and
SrCO3 (s) + SiO2 (s)

The calcination temperatures of these mixtures in a 1 bar CO2 -atmosphere, which are calculated
from thermodynamic tables [1], are 668 K and 825 K, respectively. This means that reactive
solid-state calcination of SrCO3 in a 1 bar CO2 -atmosphere is expected to start at temperatures
above 668 K. However, the measured onset temperature for SrCO 3 -calcination in a 1 bar CO2 atmosphere when mixed with silica sand is approximately 1470 K (curve C in gure A.3). This
indicates that, similar to a mixture of soda ash and silica sand, the formation of a crystalline

194

Appendix A. Calcination of a TV-panel glass batch

silicate by a solid-state calcination reaction of SrCO3 and silica does hardly contribute to the
calcination of the carbonate. This conclusion is supported by the observation that phase analysis on quenched samples of a heat-treated mixture of silica sand and SrCO 3 did not show the
presence of crystalline strontium silicates at temperatures up to 1200 K.
During thermogravimetric analysis of a mixture of soda ash, silica sand and SrCO 3 in a 1
bar CO2 -atmosphere, a sharp increase in weight loss for the mixture is observed at about 970
K (curve D in gure A.3), which is similar to the calcination onset temperature for a mixture
of soda ash and silica in a 1 bar CO2 -atmosphere (see section 2.5.3). Complete calcination2 of
the three-component mixture in a 1 bar CO2 -atmosphere is observed at about 1220 K, which is
approximately 330 K lower than the thermal calcination temperature of individual SrCO 3 grains
in a 1 bar CO2 -atmosphere. This indicates that, similar to reactive dissolution of silica sand (see
section 2.5.3), the decomposition of SrCO3 is enhanced in the presence of a silicate melt phase.
Figure A.4 shows the identied crystalline phases as function of temperature in quenched
samples of heat-treated mixtures of silica sand, soda ash and strontium carbonate. These glass
batch mixtures were ramp heated with 10 K min1 in an ambient atmosphere. Similar to a
mixture of soda ash and silica sand, the rst formed intermediate crystalline phase is sodium
metasilicate starting at about 1073 K. The SrCO3 grains in the mixture react with this primary
formed melt phase forming a sodium strontium silicate melt phase. At about 1123 K, the presence of crystalline 2SrO SiO2 is observed in the partly reacted three-component mixture. The
dissolution of SrO in the primary melt phase leads to an increase of the SiO 2 solubility in the
melt phase. Therefore, the reactive dissolution of SrCO3 is likely to affect the dissolution rate
of sand grains.
The determination of the residual amount of silica during ramp heating of a glass batch mixture containing silica sand, soda ash and SrCO3 with 10 K min1 in an ambient atmosphere up
to 1273 K shows that, for a given particle size of the silica sand and soda ash grains, a smaller
particle size of the SrCO3 grains resulted in a lower residual amount of undissolved silica in
the heat-treated glass batch mixture. In case of a particle size of the SrCO 3 grains between 250
m and 500 m, the percentage of the initial amount of silica sand in the ternary glass batch,
which is still present as crystalline silica equals 60 % at 1273 K. For a SrCO 3 grain size less
than 63 m this percentage equals 50 %. The amount of silica sand, which is dissolved in the
three components glass melt, increases with decreasing particle size of the strontium carbonate
grains, which indicates that decreasing the particle size of the SrCO 3 grains results in an enhanced reactive calcination of strontium carbonate.
Figure A.5 shows the identied crystalline phases as function of temperature in quenched
samples of heat-treated mixtures of silica sand, soda ash and barium carbonate. These glass
batch mixtures were ramp heated with 10 K min1 in an ambient atmosphere. Similar to mixtures of soda ash and silica sand and a three-component mixture of silica sand, soda ash and
strontium carbonate, the rst formed intermediate crystalline phase is sodium metasilicate starting at about 1073 K. At about 1123 K, both crystalline 2BaO SiO 2 and Na2 O 2BaO 2SiO2 are
observed. The crystalline Na2 O 2BaO 2SiO2 was also observed by Krol and Janssen [7] during heating of a mixture containing silica sand, soda ash and barium carbonate.

2 Calcination

of both Na2 CO3 and SrCO3 to Na2 O and SrO.

A.2. Calcination of SrCO3 - and BaCO3 -grains

195

2SrO SiO2
Na2 O SiO2
SiO2 (q)
SrCO3

PSfrag replacements

Na2 CO3
900

950

1000 1050 1100 1150 1200 1250 1300 1350 1400

Temperature [K]

Figure A.4: Identied crystalline species in quenched samples of a glass batch containing silica sand,
soda ash and strontium carbonate as function of temperature. The glass batch is ramp heated
with 10 K min1 in an ambient atmosphere. Phase identication was performed up to 1350
K with temperature steps of 50 K. SiO2 (q) indicates SiO2 in the quartz modication.

Na2 O 2BaO 2SiO2


2BaO SiO2
Na2 O SiO2
SiO2 (q)

PSfrag replacements
BaCO3
Temperature [K]
Na2 CO3
900

950

1000 1050 1100 1150 1200 1250 1300 1350 1400

Temperature [C]

Figure A.5: Identied crystalline species in quenched samples of a glass batch containing silica sand,
soda ash and barium carbonate as function of temperature. The glass batch is ramp heated
with 10 K min1 in an ambient atmosphere. Phase identication was performed up to 1350
K with temperature steps of 50 K.SiO2 (q) indicates SiO2 in the quartz modication.

196

A.3

Bibliography

Bibliography

[1] O. Knacke, O. Kubaschewski, and K. Hesselmann. Thermochemical Properties of Inorganic Substances. Springer-Verlag Berlin, Heidelberg, Germany, 2 nd edition, 1991.
[2] F. Kr mer. Gasprolmessungen zur Bestimmung der Gasabgabe beim Glasschmelzproze.
a
Glastechn. Ber., 53(7):177188, 1980.
[3] S. Kawachi, M. Kato, and Y. Kawase. Evaluation of reaction rate of rening agents.
Glastech. Ber. Glass Sci. Technol., 72(6):182187, 1999.
[4] R.S.A.H. Huijben. Modeling the melt behavior of tv-panel batch. Afstudeerverslag, Eindhoven University of Technology, Eindhoven, The Netherlands, 2001.
[5] B.A. Shakhmatkin, N.M. Vedishcheva, and C.A. Wright. Thermodynamic properties: A
reliable instrument for predicting glass properties. In Proc. Int. Congr. Glass, volume 1,
pages 5260, Edinburgh, Scotland, 1-6 July 2001.
[6] R. Conradt. A simplied procedure to estimate thermodynamic activities in multicomponent oxide melts. Molten Salt Forum, 5-6:155162, 1998.
[7] D.M. Krol and R.K. Janssen. Raman study of the chemical reactions during the melting of
a 15Na2 CO3 18BaCO3 75SiO2 batch. J. Physique, 43(12), 1982.

Appendix B
Estimation of model parameters by a least
squares approach
B.1

Description of the least squares approach for parameter


estimation

To estimate the model parameters stored in the column , a criterion of the measured and calculated temperatures in a solid particle mixture, that is dened by
n

qls = yi mi
i=1

yi mi ,

(B.1)

is minimized. Herein i is the time indicator, n is the total number of points of time at which
temperatures in the solid particle mixture are measured, m i is the column containing the measured temperatures at different positions in the solid particle mixture at t = t i and yi is a column
containing the calculated temperatures at these positions in the solid particle mixture at t = t i .

Given an estimate of the model parameters , temperatures in the solid particle mixture at
time ti can be determined by

(B.2)
yi = h(,ti ) = hi (),
in which hi characterizes the numerical calculation of the temperatures at the different positions
in the solid particle mixture at t = ti . In this specic case, the column consists of the model
parameters ca and cb , which describe the temperature dependent effective heat conductivity
of the solid particle mixture according to = ca + cb T .

The estimates for the model parameters, , are derived by minimization of the criterion qls
with respect to the unknown model parameters . This requires the derivative of q with respect
to each component of to be equal to 0. This derivative is written as
q
1
= lim
0

i=1

hi ( + ) mi

hi ( + ) mi hi () mi

hi () mi

(B.3)
The calculated temperatures in the solid particle mixture depend in a non linear way upon
the model parameters ca and cb . Therefore, the operation hi () is non linear in the model

parameters. When is chosen to be sufciently small, the model output hi ( + ) given in


197

198

Appendix B. Estimation of model parameters by a least squares approach

equation B.3 can be approximated by

hi ( + ) = hi () + H i ,

(B.4)

when second or higher order terms of are assumed to be negligible.


In case m is the number of positions at which for each point of time a temperature measurement is available in the solid particle mixture, the matrix H i is given by

Hi =

h1
ca

.
.
.

hm
ca

h1
cb

.
.
.

hm
cb

(B.5)

t=ti

For the solid particle mixture, as described in this thesis, the derivatives of the components of
h with respect to each of the model parameters can easily be determined in a numerical way.
For this purpose, the model parameters are slightly changed and the impact on the result of h is
computed.
q
Setting the derivative in equation B.3 to zero results in
lim

i=1

hT ( + ) hi ( + ) 2 mT hi ( + ) hT () hi () + 2 mT hi () = 0.
i
i
i
i
(B.6)

Combining equation B.4 with equation B.6 results in


lim

i=1

2 hT () H i + H i
i

H i 2 mT H i = 0.
i

(B.7)

Assuming that is chosen to be sufciently small and neglecting the second and higher order
terms of yields
n

i=1

2 hT () 2 mT
i
i

H i = 0.

(B.8)

Since the operation hi () is non-linear in the model parameters , an iterative procedure is re


quired to estimate . Suppose that a previous estimate k of the model parameters is available,
such that equation B.8 is not fully satised:

hi (k ) = mi ; for all i.

(B.9)

As k is the best available estimate for the model parameters, it may be assumed that a new and

better estimate k+1 of will not deviate much from the previous estimate k plus a correction

term according to

k+1 = k + .
(B.10)

The estimation algorithm is aimed at nding the correction term to improve the estimate
of the model parameters. For this reason, equation B.10 is substituted in equation B.8, which
results in:
n
T

hT (k ) + H mT H = 0
(B.11)

i
i
i=1

B.2. Bibliography

199

It must be denoted that H i is determined based on the previous estimate k .


According to Kreyszig [1], the transpose of a product equals the product of the transposed
factors, taken in reverse order, i.e. (A B)T = BT AT . Now, the change in estimation of the model
parameters can be calculated from rewriting equation B.11 to:
n

HT
i

i=1

Hi

HT
i

i=1

mi hi (k ) .

(B.12)

Once the correction has been computed from equation B.12, the estimate of underline can
be updated according to equation B.10. The new estimate of the model parameters is consequently written as
n

k+1 = k +

i=1

HT
i

Hi

HT
i

i=1

mi hi (k ) .

(B.13)

Because the calculated temperatures in the solid particle mixture depend nonlinearly on the

model parameters ca and cb , this iterative estimation procedure must be repeated until
has become smaller than a predened value 0 > 0.

B.2

Bibliography

[1] E. Kreyszig. Advanced engineering mathematics. John Wiley and Sons, Inc., 6 edition,
1988.

200

Bibliography

Appendix C
Derivation of 2nd order derivative of
position dependent temperature
Estimating the 2nd order spatial derivative of temperature requires at least a 2nd order spatial
dependency of temperature, viz. T = ax2 + bx + c. The determination of the 2nd order spatial
derivative of temperature, given by 2a, requires at each time i information of at least three
temperatures at three different positions j distributed in the solid particle mixture:
T j1,i = ai x2 + bi x j1 + ci
j1

(C.1)

T j,i = ai x2 + bi x j + ci
j

(C.2)

T j+1,i = ai x2 + bi x j+1 + ci
j+1

(C.3)

From equation C.1 an equation for the coefcient ci is derived:


ci = T j1,i ai x2 bi x j1
j1

(C.4)

Combining equation C.4 with equations C.2 and C.3 results in equations C.5 and C.6.
T j T j1 = ai x2 x2
j
j1 + bi x j x j1

(C.5)

T j+1 T j1 = ai x2 x2
j+1
j1 + bi x j+1 x j1

(C.6)

From equation C.5 an equation for the coefcient bi is derived.


bi =

T j T j1 ai x2 x2
j
j1
x j x j1

(C.7)

Combining equations C.6 and C.7 results in an expression for the coefcient a i .
ai =

T j+1 T j1
x2 x2
j+1
j1

x j+1 x j1
x j x j1
x j+1 x j1
x j x j1

T j T j1
x2 x2
j
j1

(C.8)

In case that x j+1 x j1 = x j x j1 = x than equation C.8 simplies to


ai =

T j+1 2T j + T j1
.
2x2
201

(C.9)

202 Appendix C. Derivation of 2nd order derivative of position dependent temperature


The 2nd order spatial derivative of temperature is now given by
T j+1 2T j + T j1
2 T
=2
.
2
x
2x2

(C.10)

Samenvatting
De kwaliteit van de glassmelt die geproduceerd wordt in continu bedreven glasovens wordt
voor een belangrijk deel bepaald door het temperatuur-tijd traject dat vers ingesmolten glas
doorloopt in de smeltwan van een glasoven. Tijdens het transport van het vers ingesmolten glas
door de glasoven dient de verblijftijd en de temperatuur van de glassmelt voldoende te zijn voor
het oplossen van nog ongesmolten materiaal, het verwijderen van de in de glassmelt opgeloste
gassen en het homogeniseren van de gevormde glassmelt. Berekeningen met glasovensimulatiemodellen1 tonen aan dat de dikte, lengte, reactiviteit en thermische eigenschappen van de
gemengdeken, bestaande uit glasvormend gemeng2 , in een industri le glasoven een grote ine
vloed hebben op de glassmeltstroming en de temperatuurverdeling in de glassmelt in de gehele
oven. Hiermee benvloedt de gemengdeken indirect de kwaliteit van het geproduceerde glas.

Om glasovensimulatiemodellen te kunnen gebruiken als hulpmiddel bij het begrijpen, beheersen


en optimaliseren van de glassmeltkwaliteit en bij het verbeteren van glasovenontwerp, zijn gedetailleerde mathematische modellen nodig die het thermisch en chemisch gedrag van glasgemeng
beschrijven.
Tijdens het opwarmen van glasgemeng treedt een verscheidenheid aan chemische reacties op, zoals ontwateringsreacties, kristalovergangen, vaste-stof reacties, ontledingsreacties,
smeltvormende reacties en oplosreacties. De opwarming van het uit drie fasen bestaande (reagerende) glasgemeng wordt bepaald door warmte-overdracht van zowel de verbrandingsruimte
boven de gemengdeken als van de hete smelt onder de gemengdeken naar de randen van de

gemengdeken en door het warmtetransport in het inwendige van de gemengdeken. Vanwege

de complexiteit van zowel het smelt- als het opwarmproces van een gemengdeken en vanwege
het gebrek aan voldoende sensoren en nauwkeurige analysetechnieken om de voortgang van het
opwarm- en smeltproces van glasgemeng te kunnen meten, is tot op heden geen kwantitatieve
beschrijving van het opwarmen en reageren van glasgemeng bekend.
De doelstelling van dit proefschrift is het kwantitatief beschrijven van het opwarmproces
van een glasgemeng en de omzettingssnelheid van het glasgemeng in de glassmelt. Het warmtetransport in een gemengdeken kan worden beschreven door:
(m c p,m Tm )
Hchem

= ( g c p,g vg,z Tg ) (eff Tm ) + r,eff + (1 )


q
, (C.11)
t
t
Het linkerlid van deze energievergelijking beschrijft de lokale ophoping van warmte in de
gemengdeken. De eerste, tweede en derde term in het rechterlid van de energievergelijking
beschrijven het transport van energie door de gemengdeken via convectief, diffusief en stralingswarmtetransport. De laatste term in de energievergelijking beschrijft de energie die nodig
1 Glasovensimulatiemodellen beschrijven onder andere het insmelten van glasgemeng, de stroming van de glass-

melt en andere thermische processen tijdens het industrieel glassmelten.


2 Het glasvormend gemeng wordt in het algemeen glasgemeng genoemd.

203

204

Samenvatting

is voor de chemische reacties die optreden tijdens het opwarmen van het glasgemeng. Voor
een gedetailleerde beschrijving van het opwarmen en smelten van een glasgemeng is kwantitatieve kennis nodig van de temperatuurafhankelijke eigenschappen van het glasgemeng. In
deze studie is de nadruk gelegd op het bepalen van de temperatuurafhankelijke chemische energiebehoefte en warmtegeleidbaarheid van glasgemeng. Voor het beschrijven van de omzetting
van glasgemeng tot een glassmelt is het oplossen van zandkorrels in het reagerende glasgemeng
onderzocht.
In dit promotiewerk zijn een drietal activiteiten te onderscheiden:
1. De identicatie van de belangrijkste gemengreacties voor een vlakglasgemeng en een
TV-schermengemeng.
2. Het ontwikkelen van zowel experimentele als mathematische methodieken waarmee de
snelheid van gemengreacties en de warmtegeleidbaarheid van een glasgemeng kwantitatief bepaald kunnen worden.
3. Het bepalen en modelleren van de temperatuurafhankelijke chemische energiebehoefte,
de warmtegeleidbaarheid en de reactiekinetiek voor het vlakglasgemeng, waarbij ondermeer gebruik gemaakt wordt van de in deze studie ontwikkelde technieken.
Voor een vlakglasgemeng bestaande uit zand, soda en dolomiet, kan de calcinering van het
complete gemeng beschreven worden door de thermische ontleding van dolomiet en kalk en
de reactieve ontleding van soda met zand. De reactieve ontleding van soda met zand bepaalt
de vorming van primaire smeltfasen beginnend bij circa 1070 K. De vorming van deze smeltfasen zorgt voor een versneld oplossen van het zand. In tegenstelling tot het vlakglasgemeng
kan de calcinering van een typisch TV-schermengemeng niet per afzonderlijke carbonaat in het
gemeng beschreven worden.
De chemische energiebehoefte van een glasgemeng wordt grotendeels bepaald door de energie
benodigd voor het calcineren van de carbonaten die aanwezig zijn in het glasgemeng. Daarom
is voor het bepalen van de chemische energiebehoefte van vlakglasgemeng de kinetiek van
de thermische ontleding van dolomiet en kalk en de kinetiek van de reactieve calcinering van
soda in deze studie bepaald. Combinatie van de gemeten calcineringskinetiek met de reactieenthalpi n van betreffende calcineringsreacties heeft geresulteerd in een beschrijving van de
e

chemische energiebehoefte van vlakglasgemeng als functie van tijd, temperatuur en parti ele
CO2 -druk in het gemeng.
Het warmtetransport in het inwendige van een gemengdeken wordt bepaald door drie thermische processen, te weten conductieve, convectieve en stralingswarmtegeleiding. De bijdrage
van deze afzonderlijke thermische processen op het totale warmtetransport in de gemengdeken
is bestudeerd met behulp van een speciaal ontwikkelde experimentele opstelling waarmee op
verschillende plaatsen en als funktie van tijd de temperaturen in een opwarmend glasgemeng
gemten kan worden. Om uit deze gemeten temperaturen de temperatuurafhankelijke warmtegeleidbaarheid te bepalen is een numeriek-experimentele techniek gebruikt.
Met deze numeriek-experimentele techniek is de temperatuurafhankelijke warmtegeleidbaarheid van individuele glasgemengcomponenten en van mengsels van glasgemengcomponenten bestudeerd. Het is aangetoond dat de warmtestroom door deze mengsels beschreven kan

Samenvatting

205

worden als een warmtestroom door een combinatie van serieel en parallel geschakelde thermische weerstanden.
De voorspelling van de effectieve warmtegeleidbaarheid van mengsels van glasgemengcomponenten op basis van gemengsamenstelling en de porositeit van het glasgemeng is mogelijk
door het glasgemeng te beschouwen als een parallele schakeling van thermische weerstanden.
Hierbij worden, in deze studie bepaalde, schijnbare warmtegeleidbaarheidswaarden van de
glasgemengcomponenten gebruikt in plaats van de intrinsieke waarden voor de warmtegeleidbaarheid. Het verschil in de gemeten en de voorspelde warmtegeleidbaarheid van een driecomponent mengsel met deze aanpak bedraagt ca. 0.01 W m1 K1 bij een waarde voor de
warmtegeleidbaarheid varierend tussen de 0.15 en 0.40 W m 1 K1 .
De invloed van de aanwezigheid van scherven in het glasgemeng op de netto effectieve
warmtegeleidbaarheid van glasgemeng in het temperatuurgebied tot circa 1200 K is bestudeerd
met de numeriek-experimentele techniek. De berekening van de bijdrage van stralingswarmtetransport op de netto effectieve warmtegeleidbaarheid van glasgemeng toont aan dat de verwachte toename in de warmtegeleidbaarheid bij schervenhoudend glasgemeng (met een scherfgrootte varierend van 4 tot 8 mm) alleen waargenomen wordt indien de glasscherffractie in het
glasgemeng groter is dan tenminste 50 %.
Het opwarmen van een glasgemeng wordt niet alleen bepaald door het warmtetransport in
het inwendige van een glasgemeng, maar tevens door de warmteoverdracht vanuit de verbrandingsruimte naar de toplaag van het glasgemeng. Het is aangetoond dat warmteoverdracht naar
de gemengdeken toe voornamelijk bepaald wordt door stralingswarmteoverdracht en in mindere
mate door warmteoverdracht via vrije en gedwongen convectieve stroming van hete gassen over
de gemengdeken.
In het algemeen wordt de omzettingsgraad van een glasgemeng gerelateerd aan de omzettingsgraad van zandkorrels in het reagerende glasgemeng. Voor de modelmatige beschrijving van
dit stofoverdrachtsgelimiteerde proces dienen processen op microschaal, zoals bijvoorbeeld het
bevochtigen van zandkorrels door laag visceuze smeltfasen, beschreven te worden samen met
complexe tijdsafhankelijke randvoorwaarden. Voor het gebruik van glasovensimulatiemodellen
als hulpmiddel in de glasindustrie is het gewenst om een eenvoudige uitdrukking te hebben
voor het tijd- en temperatuurafhankelijk oplossen van zandkorrels in plaats van het toepassen
van complexe oplosmodellen voor beschrijvingen op microschaal. Een voorwaarde voor het
gebruik van deze eenvoudige uitdrukking is dat deze uitdrukking voor het beschrijven van het
oplossen van zandkorrels in een reagerend gemeng het praktisch waargenomen oplosgedrag
goed beschrijft en dat de afhankelijkheid met korrelgrootte, opwarmsnelheid en scherffractie
goed weergegeven wordt. Het is algemeen gebruikelijk om benaderende theoretische modellen
te gebruiken voor het beschrijven van complexe processen zoals het oplossen van zand tijdens
gemengsmelten.
Het meest geschikte benaderende theoretische model dat de snelheid van een driedimensionaal diffusie gelimiteerd proces, zoals het oplossen van zandkorrels in een reagerend glasgemeng beschrijft, is het Ginstling-Brounstein model (GB-model). De toepasbaarheid van het
GB-model is bestudeerd door simulatieresultaten van het GB-model te vergelijken met resultaten verkregen met een gedetailleerder numeriek model. Het blijkt dat het onder bepaalde omstandigheden mogelijk is om met een aangepast GB-model het oplossen van zandkorrels in een
reagerend glasgemeng te voorspellen als functie van tijd en temperatuur. Dit geldt voornamelijk
voor grote zandkorrels die oplossen in relatief dikke smeltlagen. Voor dunne smeltlagen blijkt

206

Samenvatting

de voorspelling van het GB-model niet accuraat te zijn.


Het gebruik van het aangepaste GB-model is experimenteel gevalideerd door het restgehalte
zand te meten in een reagerend vlakglasgemeng. Het restgehalte zand in gedeeltelijk inge
smolten vlakglasgemeng is bepaald via kwantitatieve fase analyse met R ontgendiffractie. Voor
het vlakglasgemeng zijn de GB-model parameters bepaald als functie van zandkorrelgrootte,
opwarmsnelheid en schervenfractie.

Dankwoord
Het promotieonderzoek Thermal and chemical behavior of glass forming batches heeft plaatsgevonden in het kader van ontwikkeling van het TNO Glass Tank Simulation Model. Dit onderzoek werd genancierd door het Nationaal Comit van de Nederlandse Glasindustrie (NCNG),
e
Glaverbel, LG.Philips Displays, PPG Industries, Schott Glas, Thomson Multimedia en Visteon.
Bij deze wil ik mijn dank uitspreken voor de nanci le ondersteuning van dit promotieondere
zoek. TNO bedank ik voor de mogelijkheid die mij geboden is om mij via dit promotiewerk
zowel op technisch als op persoonlijk vlak verder te kunnen ontwikkelen.
Prof. Ruud Beerkens wil ik bedanken voor zijn initiatie van dit promotiewerk, zijn goede

adviezen en zijn grote betrokkenheid. Ich mochte Herrn Prof. Conradt f r die Ubernahme des
u

Korreferats danken, wodurch ich mich sehr geehrt fuhle.


Mijn collegas en afstudeerders bij TNO en het promovendi/post-doc groepje van de leerstoel Glastechnologie wil ik bedanken voor hun steun en bijdrage bij zowel het experimenteel
onderzoek als het modelleerwerk. In het bijzonder ben ik Olaf zeer dankbaar voor zijn altijd
luisterende oor, zijn goede adviezen en voor zijn geestelijke steun tijdens mijn promotiewerkzaamheden.
Familie en vrienden wil ik bedanken voor hun steun en voor hun begrip voor het feit dat
ik er de afgelopen tijd vaak niet was. Arjen wil ik bedanken voor het delen van alles wat een
promotie raakt en ook wat daarbuiten van belang is. De vele biertjes zullen niet snel vergeten
zijn!
Tot slot en in het bijzonder wil ik Roelie bedanken voor alle steun die zij mij gegeven heeft
tijdens mijn promotiewerkzaamheden bij TNO en vooral gedurende de lange tijd dat ik thuis
mijn proefschrift aan het schrijven was. Zeker ook door haar grote doorzettingsvermogen is het
promotiewerk tot een goed einde gekomen. De tijd van genieten breekt nu aan!

207

208

Curriculum Vitae
Oscar Verheijen werd op 3 augustus 1970 geboren te Arnhem. Hij behaalde in 1988 het VWOdiploma aan het Nederrijn College te Arnhem. In 1988 begon hij met de studie Chemische
Technologie aan de Hogeschool Eindhoven, alwaar hij in 1992 afstudeerde. Van 1992 tot
1995 volgde hij de verkorte opleiding Scheikundige Technologie aan de Technische Universiteit Eindhoven. In mei 1995 trad hij in dienst bij TNO TPD te Eindhoven als wetenschappelijk
medewerker bij de afdeling Glastechnologie. In de periode van juli 1997 tot en met december
2001 verrichte hij een promotie-onderzoek bij TNO TPD. Sinds januari 2002 is hij werkzaam
als projectleider bij de afdeling Glastechnologie van TNO TPD met als voornaamste werkgebieden smelttechnologie en thermochemie.

209

Você também pode gostar