Você está na página 1de 83

Air Quality Research Program (AQRP)

Project # 10-022

Development of Speciated Industrial Flare


Emission Inventories for Air Quality
Modeling in Texas

Final Report

Prepared by:

Daniel Chen, PI
Helen Lou, Co-PI
Kuyen Li, Co-PI
Christopher Martin, Co-PI
X. Chang Li, Co-PI

Lamar University
Beaumont, TX 77710

December 14, 2011


Distribution List

Jim MacKay
TCEQ Project Liaison
Air Quality Division
Texas Commission on Environmental Quality
Austin, Texas

Mr. Vincent M. Torres


Project Manager
Air Quality Research Program
The University of Texas at Austin,
Austin, Texas

Dr. Daniel H. Chen


Principal Investigator
Lamar University,
Beaumont, Texas

General Requirements

This project is a secondary data project. A secondary data project involves the gathering and /or
use of existing environmental data for purposes other than those for which they were originally
collected. This document is based on EPAs National Risk Management Research Laboratory
(NRMRL) QAPP for secondary data project and satisfies a Category III level of QA.

Page 1
AQRP Project 10-022

Development of Speciated Industrial Flare Emission Inventories for Air


Quality Modeling in Texas

Executive Summary

In this project, computational fluid dynamics (CFD) methods based on CHEMKIN-CFD

and FLUENT are used to model low-Btu, low-flow rate propylene/TNG/nitrogen flare tests

conducted during September 2010 in the John Zink test facility, Tulsa, Oklahoma. The flare test

campaign was the focus of the TCEQ Comprehensive Flare Study Project (PGA No. 582-8-862-

45-FY09-04) in which plume measurements using both remote sensing and direct extraction

were carried out to determine flare efficiencies and emissions of regulated and photo-chemically

important pollution species for both air-assisted and steam-assisted flares under open-air

conditions. This project (1) models and predicts the performance of Tulsa testing flares at low

heating value and low jet velocity conditions using the CFD approach, and (2) further compares

with the measured flare performance data. This modeling tool has the potential to help TCEQs

on-going evaluation on flare emissions and to serve as a basis for a future State Implementation

Plan (SIP) revision and the effect of these relations on flare performance was studied.

The 50-species combustion mechanism was reduced from the combined GRI and USC

mechanisms with the goal of allowing NOx formation and handling light hydrocarbon

combustion. This optimized Lamar mechanism has been validated against methane, ethylene,

and propylene experimental data. Further, NO2 was added to the existing mechanism and was

shown in good agreement with the full mechanism. FLUENT models (species, turbulence-

chemistry, viscous flow and Numerical algorithms), model parameters, and boundary conditions

have been selected.

Page 2
The main operating, design, and meteorological data of the flare test campaign were

provided by the University of Texas (UT) including Combustion Efficiency (CE), Destruction

and Removal Efficiencies (DRE). Both Probability Density Function (PDF) and Eddy

Dissipation Concept (EDC) turbulence-chemistry interaction approaches have been adopted to

run Tulsa flare test cases. Twelve air-assisted flare test cases and nine steam-assisted flare test

case have been run and compared with the measured DRE/CE data.

In general, the EDC model under- predicts DRE by 6% to 19% with an average of 12%.

It under- predicts CE by 12% to 39% with an average of 25%. Comparing the EDC results with

measured results, DRE is within the uncertainty limits (19%). While CE is on average

reasonably close to the tentative uncertainty limit (23%); predicted CEs in many occasions go

beyond the stated uncertainty limit. The potential causes may be the low jet velocity, low heating

values, high air/steam assists, complexity of geometry, placement of the pilots, choice of

turbulence intensity, and the difference between local sampling and the full surface integration

(CFD post processing).

Even though the PDF approach was verified with University of Alberta wind tunnel data

and was shown in good agreement for certain high DRE/CE cases; the more simplistic PDF

model tends to over-predict flare efficiencies than the measured ones. Contrary to the EDC

model, the PDF model over- predicts DRE by 0.1% to 72% with an average of 16%. It over-

predicts CE by 0% to 78% with an average of 18% (except in one case where PDF under-

predicts by 3%). Since the assumption of fast reactions (or reaching chemical equilibrium) used

by the PDF approach is true only at high temperatures (2100K-2400K), the PDF model may not

be valid for many low heating value and high air/steam assisted flare cases.

In view of the significant differences between the PDF and EDC model results, further

investigations involving other flare test cases and geometries with pilot flames in the EDC model

Page 3
and more exploration on the PDF/EDC model parameters are warranted. It appears, however,

that measured DREs/CEs from the 2010 comprehensive flare study fall somewhere between the

EDC and PDF model predictions under low-jet velocity, low BTU conditions.

Page 4
Table of Contents

Page No
Executive Summary 2

1 Introduction 6
1.1 Project description 6
1.2 Project objectives 9
1.3 List of Project tasks 9

2 Methodology 11
2.1 Data collection and high performance CFD cluster 11
2.2 Combustion mechanism development 14
2.3 CFD model development 31

3 Results 51
3.1 Modeling air-assisted flare cases 51
3.2 Modeling steam-assisted flare cases 53
3.3 Parametric study for air and steam-assisted flares 59
3.4 Comparison of CFD prediction and flare test data 64

4 Discussion 75

5 Future work 79

References 81

Page 5
1. INTRODUCTION

1.1 Project Description

Current methodologies for calculating speciated and total VOC (Volatile organic

compounds) emissions from flaring activities generally apply a simple mass reduction to the

VOC species sent to the flare [1]. In most cases 98% is used as the destruction and removal

efficiency (DRE) for the flare without any intermediate VOC species generated or emitted by the

combustion process. Basic combustion chemistry demonstrates that many intermediate VOC

species are formed during combustion process. While it is assumed that a flare operating under

its designed conditions and in compliance with 40 CFR 60.18 may achieve 98% DRE or higher,

a flare operating outside of these parameters may have a DRE lower than 98% [2]. Other factors

that may affect DRE and the combustion efficiency (CE) include environmental factors such as

cross wind, ambient temperature, and humidity [3-5].

In this project, computational fluid dynamics (CFD) methods based on CHEMKIN CFD

and FLUENT are used to model low-Btu, low-flow rate propylene/TNG/nitrogen flare tests

conducted during September 2010 in the John Zink test facility in Tulsa, Oklahoma [6]. In these

flare performance tests, plume measurements using both remote sensing and direct extraction

were carried out to determine flare efficiencies and concentration/location of regulated and

photo-chemically important pollution species for air-assisted and steam-assisted flares. Various

combinations of fuel BTU and flow rates were performed under open-air conditions. This

research project primarily used CFD modeling as a predicting tool for the Tulsa flare

performance tests. The CFD modeling was further compared with the flare performance data,

i.e., flare efficiencies, reported in the TCEQ Comprehensive Flare Study Project Final Report

Page 6
[22]. This modeling tool has the potential to help TCEQs on-going evaluation on flare

emissions and to serve as a basis for a future SIP revision [7].

Lamar University modeled the data collected from the TCEQ Comprehensive Flare Study

Project (PGA No. 582-8-862-45-FY09-04) [6] using computational fluid dynamics (CFD)

programs. The modeling programs used by Lamar University were CHEMKIN and FLUENT.

GRI-Mech 3.0 is an optimized mechanism designed to model natural gas (C1) combustion,

including NO formation and reburn chemistry while the USC mechanism, optimized for C1-

C3 combustion, lacks chemistry needed to define NO formation for flaring in air. So the

inclusion of NOx formation chemistry from the GRI mechanism will make the USC mechanism

suitable for modeling Tulsa test flares (that burn fuel C1 and C3 and NOx emission was

measured). The 50-species combustion mechanism used in this project, LU 1.0, was reduced

from the combined GRI and USC mechanisms with the goal of allowing NOx formation and

handling light hydrocarbon combustion. This optimized Lamar mechanism has been tested

against methane, ethylene, and propylene experimental data such as laminar flame speed,

adiabatic flame temperature, and ignition delay [8]. Further, NO2 was added to LU 1.0

mechanism to become LU 1.1. Please note the name LU 1.0 and LU 1.1 are introduced in this

report for clarity. A comparison with the lab data was completed and shown in good agreement

[9].

Lamar University acquired the operating and design data of the flare tests conducted at

the John Zink facility in Tulsa, OK from the University of Texas. These input data include the

geometry of the steam-assisted and air-assisted flares used in the tests, meteorological data

(cross-wind speed/direction), and the operating data (aeration, steaming, exit velocity, waste

gas/pilot fuel species) available from the data acquisition system. The flare performance data

Page 7
provided include Combustion Efficiency (CE)/ Destruction and Removal Efficiencies (DRE)

[10].

Combustion efficiency is defined as the percentage of flare emissions that are completely

oxidized to CO2. It can be written mathematically as:

CO 2
% CE 100
CO 2 CO THC Soot (1.1)

Where:

CO2 - parts per million by volume of carbon dioxide

CO - parts per million by volume of carbon monoxide exiting from the flare

THC - parts per million by volume of total hydrocarbon exiting from the flare

Soot - parts per million by volume of soot as carbon

Soot is eliminated from industrial flares by adding appropriate amounts of steam or air and that is

the reason it can be set to zero in the above equation.

The destruction and removal efficiency is given as (using propylene as an example):

(1.2)

Page 8
1.2 Project Objectives

The objectives of the proposed project are to:

1) Model the low-BTU, low-flow rate Propylene/TNG/Nitrogen flare tests conducted during

September 2010 in Tulsa, Oklahoma for the TCEQ Comprehensive Flare Study Project,

using the detailed reaction mechanisms and Fluent CFD software.

2) Predict the test results: flare efficiencies (DRE/CE) and emissions using the CFD

modeling.

3) Compare the CFD prediction results with the flare performance data (efficiencies).

1.3 List of Project Tasks

To achieve the proposed objectives, specific tasks have been proposed and accomplished

accordingly. The tasks are outlined below and the accomplishments are detailed in the sections

followed.

Task 1: Work Plan

Task 2: Flare Test Operation/Design/Performance Data

Task 3: Hardware/Software Acquisition and Data Storage

Task 4: Combustion Mechanism Development

4A Combustion Mechanism Generation

4B Combustion Mechanism Validation

Task 5: CFD Model Development

5A Geometry Creation and Boundary Conditions Setup

5B Physical/Turbulence Model Selection and Parameter Evaluation

5C Model Development Presentation

5D CFD Model Calibration

Page 9
Task 6: CFD Modeling and Post Processing

6A Modeling Base Case

6B Base Case Presentation

6C Modeling Rest of the Cases

Task 7: Comparison CFD Prediction and Flare Test Data

Task 8: Reports

Page 10
2. METHODOLOGY

2.1 Data Collection and High performance CFD cluster

Data Collection

Lamar University acquired the operating and design data of the flare tests conducted at

the John Zink facility in Tulsa, OK from the University of Texas. These input data include the

geometry of the steam-assisted and air-assisted flares used in the tests, meteorological data

(cross-wind speed/direction), and the operating data (aeration, steaming, exit velocity, waste

gas/pilot fuel species) available from the data acquisition system. Tables 2.1 and 2.2 show cases

with different parameters used for CFD simulations.

On page 9 of the work plan submitted by Lamar University, a total of 10 cases were

proposed. Among these 10 cases, two bases cases + 2 (additional air flow rates) + 2 (additional

steam flow rates) + 2 (LHVs in air-assist flare) + 2 (LHVs in steam-assist flare) were proposed.

But on May 11th, Lamar University was given two sets of test cases named Appendix E Tables

E-1 with DRE for Development and Appendix D Tables D-1 with DRE for Development.

Each table had a set of 15 cases (Appendix E for air-assist test cases and Appendix D for steam-

assist test cases) to be run. Since the flare test data were provided, the initial work plan was

modified to run only those cases that were given to Lamar University.

Page 11
Table 2.1: CFD cases for steam-based flares

Vent
Test Actual Vent Gas (VG) Flow Rates Gas Actual Steam Flow Rates
Wind
No. Propylene TNG Nitrogen Total LHV* Vel Center Upper Total Vel
lb/hr lb/hr lb/hr lb/hr Btu/scf fps lb/hr lb/hr lb/hr Mph
S1.5 2337.48 0.00 0.00 2337.48 2145.11 1.52 525.87 3794.01 4319.88 8
S1.8 2338.4 0.00 0.00 2338.4 2145.96 1.5 505.87 7044.07 7549.94 8.6
S 1.9 2336.64 0.00 0.00 2336.64 2144.34 1.5 504.91 7939.33 8444.24 8.8
S 2.2 937 0.01 0.00 937.01 2103.14 0.95 541.62 7769.53 8311.14 10.8
S 2.3 937 0.00 0.00 937 2122.5 0.94 539.31 4761.25 5300.56 7.6
S 3.7 191.29 18.95 715.51 925.74 345.54 0.6 0 227.83 227.83 7.1
S 4.1 490.5 44.96 1799.74 2335.2 349.58 2.0 559.94 536.28 1096.22 5.6
S 4.3 485.21 45.05 1801.62 2331.88 346.37 2.0 567.28 1879.5 2446.77 5.2
S 4.3 500.38 46.62 1802.14 2349.14 355.19 2.0 627.77 2447.42 3075.19 6.6
S 7.3 296.76 29.92 1083.58 1410.26 352.77 1.4 515.59 537.91 1053.51 7.9
S 5.2 319.81 33.78 584.67 938.26 595.36 1.0 454.47 1580.38 2034.85 10.2
S 5.3 312.28 31.73 577.98 921.99 589.58 1.0 481.52 782.52 1264.04 9.3
S 5.4 317.61 32.17 579.44 929.22 595.36 1.0 483.73 1220.85 1704.57 10.9
S 5.6 312.17 31.82 577.17 921.16 590.08 1.0 490.6 462.92 953.53 9.6
S 6.1 826.42 79.13 1455.62 2361.16 608.89 1.9 517.78 1002.85 1520.64 8.8

Both the air and steam based cases are broadly divided into 3 sets, based on the 3 different

Lower Heating Values (2100, 600 and 350 BTU/SCF) of the fuel used. Each set further has five

cases, with different vent gas velocity, crosswind and other conditions. These CFD cases are

setup upon the data provided by AQRP to Lamar University. Table 2.3 shows the composition

of the Tulsa natural gas.

Page 12
Table 2.2: CFD cases for air-assisted flare

Test Actual Vent Gas (VG) Flow Rates Vent Air Flow Wind
No. Propylene TNG Nitrogen Total LHV Gas Vel rate Vel
lb/hr lb/hr lb/hr lb/hr Btu/scf fps lb/hr mph
A1.1 918.88 0 0 918.88 2107.71 1.4 149173 12.7
A2.1 355.02 0 0 355.02 2125.45 0.5 83818 12.8
A2.3 352.14 0 0 352.14 2108.22 0.5 88791 10.1
A2.4 352.87 0 0 352.87 2112.57 0.5 148799 10
A2.5 354.71 0 0 354.71 2123.55 0.5 119580 13.3
A3.1 181.23 18.77 702.55 902.55 338.67 1.9 19387 10.3
A3.3 181.23 18.37 700.6 900.2 333.86 1.9 60121 11.1
A3.6 181.23 18.76 704.18 904.17 337.55 1.9 47494 11.9
A5.2 72.29 7.69 274.41 354.39 342.86 0.8 75139.77 2.1
A5.3 71.26 7.55 271.37 350.18 341.87 0.8 32876.17 2.5
A4.3 298.74 30.3 591.1 920.14 562.91 1.9 66471.69 10.7
A6.1 117.8 11.86 221.21 350.87 583.73 0.7 11403.53 15.9
A6.4 118.06 12.08 221.25 351.4 584.89 0.7 40583.88 14.1
A6.5 117.85 12.08 221.11 351.04 584.44 0.7 56593.85 15.5
A6.6 118.55 12.44 220.68 351.66 588.07 0.7 146294.6 15

* LHV: Lower heating value; TNG: Tulsa natural gas

Table 2.3 Composition of TNG (Tulsa natural gas)

Tulsa Natural Gas (Volumetric Composition) [11]

CH4 93.40% C2H6 2.70%

C3H8 0.60% C4H10 0.20%


CO2 0.70% N2 2.40%

High Performance CFD Cluster

Lamar University purchased a new high performance cluster (HPC) in order to enhance

computational capability of the CFD lab. The use of newly acquired high performance cluster

greatly reduced the computational time. The capacity of the cluster is described below.

Page 13
Head Node (Dell Power Edge R710) Qty 1:

The Dell PowerEdge R710 is a cutting edge enterprise level rack server. Lamar

Universitys customized R710 server has dual Intel Xeon X5650(@2.67 GHz) processors with

24 CPUs and 48GB RAM. This server mainly act a head or master node.

Compute Nodes (Dell PowerEdge R410) Qty 2:

Along with the R710, Lamar University purchased two Dell PowerEdge R410 rack

servers. These two servers act as compute nodes and together they have 48 CPUs. The R410s

were customized with dual Intel Xeon X5670(@2.93 GHz) processors and 24 GB RAM.

Both the R410 units can either be run separately (for two different jobs) or can be

employed together to run a single job using parallel computing. The above units, R710 and

R410s, are interconnected through CISCOs networking switch, CATALYST 3560G which

delivers data transfer speed up to 10 GBs/sec. A 10 GB switch is adopted, which makes sure that

communication among the servers is not the bottleneck and hence maximizes the computing

performance.

ANSYS Fluent HPC Licenses:

To engage more cores or CPUs in solving a single or multiple CFD jobs, more licenses

are required. With the support from Lamar University, 28 HPC licenses were purchased, in

addition to the basic 5-sead research licenses.

2.2 Combustion Mechanism Development

Reduced Mechanism Without NO2

CFD simulation of combustion requires a comprehensive reaction kinetics mechanism,

which takes care of all the reaction pathways and the species that are produced during and at the

Page 14
end of combustion. CHEMKIN, a reaction engineering software package, was used to develop

the reaction mechanism for the combustion of ethylene.

Two widely used mechanisms, GRI 3.0 and USC (75 species), are available for the CFD

simulation of flaring. The GRI-Mech 3.0 performs well for an extensive range of combustion

conditions, which has been evaluated and shown on their website [12]. The USC mechanism

consisting of 75 species is a comprehensive kinetic model for representing ethylene and

acetylene combustion. It has been evaluated for predicting combustion properties of both C2 and

C3 fuels. However, these two mechanisms are not satisfactory for the combustion of ethylene for

the following reasons: 1. GRI-3.0 mechanism with 53 species was developed and optimized for

the combustion of methane not ethylene. 2. USC mechanism containing 75 species was

optimized for ethylene combustion reactions, but the absence of NOx producing species in the

mechanism does not reflect the reality for flaring in air.

To overcome the problems, above the two reaction mechanisms were combined to create

a more comprehensive mechanism that includes the chemistry of the NOx species and offers the

benefits of optimized USC ethylene combustion mechanism. The combined GRI-USC

mechanism consists of 93 species; and has to be further reduced to 50 species to satisfy the

maximum species limit set by FLUENT 6.3 for the pre-mixed model [13]. The reduction of

mechanism [14] was performed by sensitivity and rate of reaction analyses with a slight

emphasis on aldehyde reactions. This optimized mechanism for the combustion of C1-C3

hydrocarbons has been tested by the LU team against experimental results such as laminar flame

speeds, adiabatic flame temperature, and ignition delay [15]. Table 4.1 shows the list of species

involved in the full and the 50 species mechanisms. The detailed mechanism had 93 species and

600 reactions which were reduced in a step wise manner to 50 species and 337 reactions.

Page 15
Table 2.4: List of species involved in the mechanisms
Mechanism No. of Species Species List

H2, H, O, O2, OH, H2O, HO2, H2O2, C, CH, CH2, CH2*, CH3,
CH4, CO, CO2, HCO, CH2O, CH2OH, CH3O, CH3OH, C2H,
C2H2, H2CC, C2H3, C2H4, C2H5, C2H6, HCCO, CH2CO,
HCCOH, C2O, CH2CHO, CH3CHO, CH3CO, C3H2, C3H3,
pC3H4, aC3H4, cC3H4, aC3H5, CH3CCH2, CH3CHCH, C3H6,
Full mechanism 93
C2H3CHO, C3H7, nC3H7, iC3H7, C3H8, C4H, C4H2, H2C4O,
n-C4H3, i-C4H3, C4H4, n-C4H5, i-C4H5 C4H6, 1,2-C4H6, C4H7,
1-C4H8, C6H2, C6H3, l-C6H4, c-C6H4, A1, A1-, C6H5O,
C6H5OH, C5H6, C5H5, C5H4O, C5H4OH, C5H5O, N, NH, NH2,
NH3, NNH, NO, NO2, N2O, HNO, CN, HCN, H2CN, HCNN,
HCNO, HOCN, HNCO, NCO, Ar, N2
H2, H, O, O2, OH, H2O, HO2, CH, CH2, CH2*, CH3, CH4, CO,
Reduced mechanism CO2, HCO, CH2O, CH2OH, CH3O, C2H2, H2CC, C2H3, C2H4,
50 C2H5, C2H6, HCCO, CH2CO, CH2CHO, CH3CHO, C3H3,
(without NO2) pC3H4, aC3H4, aC3H5, C3H6, C3H8, C4H2, n-C4H3, i-C4H3,
C4H4, N, NH, NH2, NO, N2O, HNO, CN, HCN, HNCO, NCO,
Ar, N2

Figures 2.1-2.3 show the validation results of laminar flame speed vs. equivalence ratio

for methane, the adiabatic flame temperatures for various ethylene and air mixtures at STP

conditions [16]. The ignition delay vs. temperature for propylene combustion is also illustrated.

As can be seen, good agreements were obtained between the simulations and the experimental

data [17]. The inlet experimental conditions for the CHEMKIN simulation are listed in Table

2.5.

Page 16
Figure 2.1: Laminar flame speed vs. equivalence ratio for different fuels

Page 17
Figure 2.2: Adiabatic flame temperature vs. equivalence ratio for different fuels

Page 18
Page 19
Figure 2.3: Ignition delay vs. temperature for different fuels

Table 2.5: Inlet experimental conditions for the model


Inlet Composition of Fuel Equivalence Initial Pressure
Species
Mixture (vol%) Ratio (atm)

Methane CH4/O2/Ar (9.1/18.2/72.7) 1 1.8


C2H4/O2/Ar
Ethylene 1 1
(1/3/96)

Propylene C3H6/O2/Ar (3.17/7.83/89) 1 7.9

Using the software package CHEMKIN 4.1.1, the fidelity of the mechanism was further

validated against Burner Stabilized flame laboratory data reported in the literatures. The

experimental data is obtained from the work of Bhargava et al. [18]. The fuel is a mixture of

ethylene, oxygen and argon with ethylene and argon at equivalence ratio of one. A low-pressure

laminar premixed flame stabilized on a 6.0 cm diameter burner was used in the experiment. The

Page 20
CHEMKIN model was supplemented with the measured temperature profile. The simulated

species mole fractions along the length of the flame were extracted and compared with

experimental results. Figure 2.4 shows the comparison between simulation and experimental

data of the mole fraction of major species, such as C2H4, CO2 and O2. The experimental mole

fractions have an uncertainty of 10% for the stable intermediates, and a factor of 2 for radicals

[19]. The USC/GRI mechanism has an uncertainty of 10% for CO, 4% for CO2, 10% for C2H4,

0.005 (mole fraction) for CH4, and 0.004 (mole fraction) for O2 [20-21]. Therefore, a good

agreement among the major species is observed. Figure 2.5 shows that the reduced mechanism

is even capable of predicting the generation of formaldehyde (a radical producing species in

atmospheric chemistry), which may be important from environmental aspect, with sufficient

accuracy. This comparison thus validates the reduced mechanism against an important aspect of

validation, burner stabilized flame, for ethylene.

Figure 2.4: Comparison of the molar fraction of major species in burner stabilized flame for
C2H4/O2/Ar (phi = 1.9)

Page 21
Figure 2.5: Comparison of formaldehyde mole fraction data for burner stabilized flame

Reduced mechanism with NO2

In order to improve the emissions profile of CFD Simulation, the team has modified the

current reduced 50-species mechanism [15]. The species NO2, which plays an important role in

atmospheric chemistry, is incorporated in the reduced mechanism. Two new mechanisms are

developed: one with NO2 but without Ar; the other one includes both NO2 and Ar but CN is

removed.

Reduced mechanism with NO2 and without Ar

The existing mechanism optimized for C1-C3 light hydrocarbons was used first. In the

second version of C1-C3 combustion mechanism, additional NOx species (NO2 ) was added to

the existing mechanism. Then the full combined USC-GRI mechanism was subject to the

reduction process and the validation process. In addition to sensitivity and rate of reaction

analyses, the Reduced Dimension Mixture feature offered by Fluent will also be tested to handle

this version of combined mechanism [22]. The full mechanism was reduced based on the strategy

of removing the species of least interest. The species to be removed were identified depending on

their effect on mole fractions of the species of interest. Initially, the mechanism did not have

Page 22
NO2, but NO2 happens to be one of the important species to be studied for the emissions

analysis. To incorporate NO2 in the mechanism, we analyzed the mole fractions of Ar and its

effect on other species. Since Argon is an inert gas and was in relatively low concentration, it

had least effect on the concentrations of species of interest. We replaced argon with NO2 and the

corresponding mechanism data was also incorporated. Table 4.3 lists the species involved in the

reduced mechanism with NO2.

Table 2.6: List of species involved in the reduced mechanism with NO2 and without Ar
Mechanism No. of Species Species List
H2, H, O, O2, OH, H2O, HO2, CH, CH2, CH2*, CH3, CH4, CO,
CO2, HCO, CH2O, CH2OH, CH3O, C2H2, H2CC, C2H3, C2H4,
Reduced mechanism with C2H5, C2H6, HCCO, CH2CO, CH2CHO, CH3CHO, C3H3,
50
NO2 pC3H4, aC3H4, aC3H5, C3H6, C3H8, C4H2, n-C4H3, i-C4H3,
C4H4, N, NH, NH2, NO, N2O, HNO, CN, HCN, HNCO,
NCO, NO2, N2

The comparison of different type of product species was done with three different

equivalence ratio values 0.5, 1.0, 1.5. The results were studied in terms of actual error and %

error. It was found that at equivalence ratio = 1.0 the mole fractions were close enough to be

considered as matching. (Except for the main fuel since the fuel was defined as ethylene).

Further comparison was carried out at new values of residence times (0.8 and 1.0 s). The plots

of mole fractions of species, at various equivalence ratio values are given in Figure 2.6. The plot

of percentage errors of species mole fraction at equivalence ratio 1.0 is given in Figure 2.7.

Figure 2.8 shows a similar comparison but with different fuel.

Page 23
Figure 2.6: Mole fraction of different species from two mechanisms with C2H4 as fuel

Figure 2.7: Percentage errors of species mole fraction with C2H4 as fuel

Page 24
Figure 2.8: Mole fraction of different species from two mechanisms with C3H6 as fuel
Reduced mechanism with NO2 and Ar

As a part of the validation process, the following three tests are planned.

1) Laminar Flame Speed test [19]

2) Adiabatic Flame Temperature test [16]

3) Ignition Delay test [17]

All the three tests, the laminar flame speed test, the adiabatic flame temperature and ignition

delay test were performed by the team and are discussed below.

Page 25
1) Laminar Flame Speed

Laminar flame speed is the speed at which a laminar flame propagates through a pre-

mixture of fuel and air. Flame speed is a fundamental property of a fuel-air mixture which

strongly influences design parameters of combustion equipment.

Laminar flame speed was validated for propylene fuel. The reduced mechanism with 50

species including NO2 species was tested for flame speed in CHEMKIN. This was compared

with experimental data, Table 2.7 [19]. The mechanism with NO2 and Ar was also compared

with the mechanism without NO2 species.

For validation in CHEMKIN, the pressure was taken as 1 atm, and the temperature was

298 K. The equivalence ratio was varied between 0.6 and 1.5. The model considered was Flame

Speed Calculation model. The results obtained were as follows:

Table 2.7: Laminar flame speed- comparison of simulation and experimental results [19]
Experimental With NO2 w/o With Ar w/o NO2 With NO2 and with
Equivalence ratio
(cm/sec) Ar(cm/sec) (cm/sec) Ar( cm/sec)
0.6 - 14.74 14.52 14.34
0.7 22 23.80 27 23.35
0.8 29.5 32.21 35.02 32.38
0.9 37 38.24 40.14 37.79
1 42 42.54 42.60
1.1 44.5 43.76 46.68 43.73
1.2 44 42.03 41.63 41.66
1.3 41.5 35.39 37.59 35.40
1.4 34 26.33 26.09
1.5 - 17.82 17.70

Figure 2.9 shows the comparison of the experimental and simulation results of laminar

flame speed and equivalence ratios for propylene air mixtures with NO2 species and without NO2

species. The maximum laminar flame speed of 42-45 cm/sec is observed at an equivalence ratio

in the range of 1.0 to 1.1. It can be concluded that there is good agreement between the

experimental and simulation results except for those at very high equivalence ratio (1.3-1.4).

Page 26
Figure 2.9: Comparison of the experimental and the simulation results for laminar flame speed
with different equivalence ratios

2) Adiabatic Flame Temperature

Adiabatic flame temperature is the temperature that the flame would attain if the net

energy liberated by chemical reaction that converts the fresh mixture into combustion products

were fully utilized in heating those products.

Adiabatic Flame Temperature was tested for ethylene fuel. The reduced mechanism with

50 species, including NO2 species and Ar, was tested for adiabatic temperature in CHEMKIN.

This was compared with experimental data from [16] and with the old mechanism (with Ar and

without NO2).

For validation in CHEMKIN, the pressure was taken as 1 atm, and the initial temperature

was taken as 298 K. The equivalence ratio was varied between 0.5 and 2.0. The model

considered was Equilibrium reactor model. The results obtained are shown in Table 2.8 and

Fig.2.10.

Page 27
Table 2.8 Adiabatic Flame Temperature Comparison of Simulation and Experimental Results [16]
Adiabatic Flame Temperature (K)
Equivalence
Sr.No. Experimental with Ar and With NO2 and
ratio
results without NO2 Ar
1 0.5 1610 1580 1599
2 0.6 1815 1790 1806
3 0.7 2000 1980 1995
4 0.8 2155 2150 2161
5 0.9 2310 2280 2290
6 1 2390 2360 2368
7 1.1 2400 2380 2391
8 1.2 - 2350 2364
9 1.3 2325 2300 2312
10 1.4 - 2240 2252
11 1.5 2200 2170 2191
12 1.6 - 2110 2131
13 1.7 2090 2060 2072
14 1.8 - 2000 2014
15 1.9 - 1940 1958
16 2 1905 1890 1904

Figure 2.10 compares the experimental and simulation results with adiabatic flame temperature

for both the mechanisms.

Figure 2.10: Comparison of the experimental and the simulation results for adiabatic flame temperature at
various equivalence ratios

Page 28
The maximum adiabatic flame temperature was observed at an equivalence ratio range of 1.0-

1.1. The maximum adiabatic flame temperature was 2391 deg K for simulation with both NO2

and Ar, 2380 deg K for simulation with Ar and without NO2 and 2400 deg K for experimental

results. [16] In all the cases, the maximum temperature was observed slightly at the leaner side of

fuel air mixtures.

3) Ignition Delay

The ignition delay time can be defined as the period between the creation of a

combustible mixture when the fuel is injected in an oxidizing environment, and it sustains onset

of the rapid reaction phase leads to the rise of temperature and pressure.

Ignition delay was tested for propylene fuel. The reduced mechanism with 50 species

including NO2 species and Argon was tested for ignition delay in CHEMKIN. This was

compared with experimental data, Table 2.9 [17].

For validation in CHEMKIN, the pressure was taken as 4 atm, and the temperature was

varied between 1200 K and 1600 K. Fuel composition was C3H6/O2/Ar (3.17% - 7.83% - 89% by

volume). The model considered was closed homogenous reactor. The results obtained were as

follows:

Page 29
Table 2.9: Ignition delay comparison of simulation and experimental results [17]
With NO2 and without
With Ar and without NO2 Experimental Ar With Ar and NO2
Ignition
Ignition Delay Ignition Delay Ignition Delay
104/T -6 104/T Delay 10-6 104/T -6 104/T
10 sec 10 sec 10-6 sec
(K) (K) sec (K) (K)
6.25 57 6.38 100 6.25 71.5 6.25 56.85
6.45 90 6.52 140 6.45 114 6.45 90.41
6.66 145 6.64 185 6.67 184 6.67 145.12
6.89 235 6.82 375 6.9 298 6.9 234.62
7.14 376 6.96 410 7.14 478 7.14 376.64
7.4 596 7.11 550 7.41 758 7.41 596.04
7.69 928 7.24 615 7.69 1180 7.69 929.25
7.4 870 8.33 2680 8.33 2176.21
7.57 1285

Figure 2.11 shows the comparison of the experimental and simulation results between the

ignition delay time and 104/T for propylene air mixtures with NO2 species and Ar and without

NO2 and Ar species. Ignition delay is directly proportional for inlet temperature so the maximum

ignition delay occurs at the maximum inlet temperature considered.

Page 30
Figure 2.11: Graphical representation of the experimental and the simulation results ignition
delay time at various temperatures

The above mechanism has been developed but not used so far. This chemical mechanism

will be used in the future for the simulation of air- and steam- assisted flares if time permits.

2.3 CFD Model Development

FLUENT is a state-of-the-art computer program for modeling fluid flow and heat transfer

in complex geometries. Its solver has a wide span of modeling capabilities, for example, steady-

state or transient flows; heat transfer, including forced, natural, and mixed convection, conjugate

(solid/fluid) heat transfer, and radiation; inviscid, laminar, and turbulent flows; and volumetric

sources of mass, momentum, heat, and chemical species. Furthermore, FLUENT can simulate

the mixing and reaction of chemical species, including homogeneous and heterogeneous

combustion models and surface deposition/reaction models. Basically, the Navier-Stokes

Page 31
equations as well as equations for mass, energy and species transport are needed to be solved.

The conservation equations of mass, momentum and energy in a time-averaged steady-state

format are given as follows.


u i S m
x i (2.1)


x i

u i u j g j
P
x j


x i

ij - u'i u' j Fj (2.2)

T

x i

c p u i T

x i

x - c p u'i T' S h
(2.3)
i

Where, ui, T and P are the velocity components, temperature and pressure, respectively. ij is

the symmetric stress tensor defined as

u j u i 2 u k
ij . (2.4)
x x 3 ij x
i j k

The source terms (Sm, Fj and Sh) are used to include the contributions of mass, momentum

and energy from the other phases. is the viscous dissipation and is the heat conductivity.

The equation for species transport is


u i C j D j C j - u'i C' j S j , (2.5)
x i x i x i

where Cj is the mass fraction of the species (j) in the mixture, and Sj is the source term for this

species. Dj is the diffusion coefficient. The terms of u' u' , cpu' T' , and u'
i j i i C' j represent the

Page 32
Reynolds stresses, turbulent heat fluxes and turbulent concentration (or mass) fluxes; each of

them should be modeled properly if the flow is turbulent.

The CHEMKIN mechanism can be interfaced with the software FLUENT to perform

chemistry and flame calculations at both steady and unsteady state. The output file of

CHEMKIN can be exported to Fluent 6.3.26 to further model the combustion.

Geometry Creation and Boundary Conditions

Air-Assisted Flare Geometry

The detailed structure of flare stack and flare tip is shown in Fig. 2.12A. The geometry

of the air-assisted flare as well as the computational domain is constructed. As seen in Fig.

2.12B, the computational domain has a width of 30 m and a height of 30 m. The flare has a stack

of 10 m and is located at 5 m from the upstream of the crosswind. In this way, a sufficient space

can be applied to examine the effect of crosswind on the flare profile.

Great effort has been made to create the geometry of the flare burner. Due to the extreme

complicity of the actual structure, it is impossible to simulate the detailed flow from many small

jet holes. Simplification is made to introduce the waste gas and air flow without sacrificing the

major feature of the burner. Nine spider legs are created for waste gas outlet. Figure 2.12B

shows the tip of the flare burner, and both flow rate and the jet velocity will be matched to the

actual test.

In order to match the waste gas flow rate, the geometry of flare tip has been modified.

However, no change was taken place in computational domain. The finalized geometry contains

the following key features. The domain is made up of 30 m 30 m enclosed box. The flare stack

Page 33
is placed 5 m away from the left of the domain and its height is 10 m. The big domain has been

chosen to consider the entire flame structure.

At the left side of domain, the velocity inlet boundary condition is applied, which

considers the effect of cross wind in the computation. At the bottom of the domain, slip wall

boundary condition is applied to simulate a smooth flow. The boundary conditions on all other

sides are given as pressure outlet. In this simulation the spider shaped burner is considered as

given in the comprehensive flare study document. Rectangular slit is created for waste gas flow

to match the exact waste gas outlet area. The flare tip is divided in three parts:

1. Fuel/waste gas outlet

2. Air outlet

3. Spider wall

Velocity inlet boundary condition is applied at the flare tip for fuel and air flows and the

rest of the portion is defined as spider wall.

Figure 2.12A: Air-Assisted Flare stack and flare tip [6, 23]

Page 34
Waste Gas Inlet Air Inlet

Figure 2.12B: Computational domain

After the computational domain was created, the next step was to generate a mesh. In

this study, Gambit 2.3.16 was used for the meshing. Different size functions were used to create

the mesh. The final meshed geometry contains 0.77 million cells and 0.70 million nodes. The

number of the grids is a result after balancing the computational time and the simulation

uncertainty.

The geometry and boundary conditions have been described in the previous section. The

modeling to study the behavior of air-assisted flare tested was done in Gambit. Effort has been

made to create similar geometry to that one tested, but to reduce the computational time and

meshing complexity we have created simple geometry for air-assisted flare. The geometries has

been revised slightly many times in order to make the simulation more accurate.

In this study, Gambit 2.3.16 is used for the meshing. Firstly, the base of the domain is

meshed. Different size functions and boundary layers are used to create structured and linked

mesh. Then the meshed base is extended up to the tip of the flare. During creating the mesh on

Page 35
each face skewness is monitored and kept skewness below 0.45. The entire volume is meshed

using cooper algorithm. The tip of flare is meshed using very refined mesh. Meshing is done in

such a way that the aspect ratio will be equal to one at tip of flare. Total nine spiders are created

for fuel outlet. Three pilots are placed in front of spider. There was no fuel burning through the

pilot due to the high speed of the assisted air. In order to avoid that flame bridge (Wall) has been

created between spider and the pilot. The meshed tip of flare is shown as below. The complete

meshed domain contains one million cells.

Steam-Assisted Flare Geometry

Lamar university was provided with the approximate layout of the steam assisted flare

geometry, as shown in Figure 2.13. The main difficulty in realization of the steam assisted flare

was in the creation of steam nozzel and related mesh generation. A closeup of the steam nozzel

CAD is shown in Figure 2.14.

Page 36
Figure 2.13: Schematic of steam-assisted flare

Figure 2.14 CFD realization of steam-assisted flare

Page 37
Physical/Turbulence Models

In general, the flare flow is turbulent due to the relatively large exit diameter and high jet

velocity. The type of flow can be predicted more accurately with the Reynolds number which is

related to the flow velocity and system dimension. Turbulent flows are characterized by high-

frequency fluctuating velocity fields, which result in the Reynolds stresses. The turbulence

enhances the mixing of transported quantities such as momentum, energy, and species

concentration, and affects the chemical reaction process. Therefore, a turbulence model is

needed to simulate the flare with the chemical kinetics playing a dominative role in flare

reactions.

There are many turbulence models available in the commercial software package

FLUENT. The most common one is the standard k- model. The turbulent models available

also include the RNG k- model, k- model, and the shear-stress transport (SST) k- model.

The RNG k- model was derived using renormalization group theory. It uses the effective

viscosity to account for low-Reynolds-number effects. Theoretically, this model is more

accurate and reliable than the standard k- model. The standard k- model is an empirical model

based on transport equations for the turbulence kinetic energy (k) and the specific dissipation rate

(). The low-Reynolds-number effect is accounted for in the k- model. The SST model is

mixture of the k- model and the k- model.

The standard k- model is considered the most robust for a wide range of applications.

The standard k- model, which, based on the Boussinesq hypothesis, relates the Reynolds

stresses to the mean velocity as

Page 38
u u j 2
- u'i u'j t i kij
x j x i 3 (2.6)

where k is the turbulent kinetic energy, and t is the turbulent viscosity given by

t C k 2 / (2.7)

where C is a constant and is the dissipation rate. The equations for the turbulent kinetic energy

(k) and the dissipation rate () are:


ui k t k G k . (2.8)
x i x i k x i


ui t C1G k C2 .
2

x i x i x i k k
(2.9)

The term Gk is the generation of turbulent kinetic energy due to the mean velocity gradients. The

constants C1, C2, C, k, and used are: C1 = 1.44, C2 = 1.92, C = 0.09, k = 1.0, and =

1.3. Note that the constants adopted in the turbulence model may not be the most appropriate

values for the current application. Usually these constants need to be tuned for different flow

physics such as separated flow and low-Reynolds number flow, etc. Since a better knowledge is

needed on what values these turbulence constants should be for the current application, the

values of these constants will be kept unchanged unless there are strong experimental evidences

to justify a change.

The above k- model is mainly valid for high Reynolds number fully turbulent flow.

Special treatment is needed in the region close to the wall. The enhanced wall function is one of

several methods that model the near-wall flow. In the enhanced wall treatment, a two-layer

Page 39
model is combined with the wall functions. The whole domain is separated into a viscosity-

affected region and a fully turbulent region by defining a turbulent Reynolds number, Rey,

Re y yk1/2 / (2.10)

where k is the turbulent kinetic energy and y is the distance from the wall. The standard k-

model is used in the fully turbulent region where Rey > 200, and the one-equation model is used

in the viscosity-affected region with Rey < 200. The turbulent viscosities calculated from these

two regions are blended to make the wall functions applicable throughout the entire near-wall

region.

One of the more successful recent developments is the realizable k- model. As detailed

below, this model contains a new transport equation for the turbulent dissipation rate (). In

addition, a critical coefficient of the model, C, is expressed as a function of mean flow and

turbulence properties, rather than assumed to be constant as in the standard model. This allows

the model to satisfy certain mathematical constraints on the normal stresses consistent with the

physics of turbulence (realizability). The realizable k- model is substantially better than the

standard k- model for many applications. In flare simulation, the cross wind can impose a

significant effect on the shape as well as the efficiency of flare. Furthermore, the downwash of a

flare jet can be very important to sustain the flare. To predict the effect of cross winds and

downwash of flame, the realizable k- model will be adopted in this study, and it is expected this

model will produce more reasonable results. The transport equations of the realizable k- model

are given as below.

k
k ku j t G k G b YM Sk (2.11)
t x j x j k x j

Page 40
2
u j t C1S C 2

C1 C3 G b C (2.12)
t x j x j x i k k

In the above equations

(2.13)

where Gk is the generation of turbulent kinetic energy due to the mean velocity gradients,

Gb is the generation of turbulence kinetic energy due to buoyancy

YM is the contribution of the fluctuating dilatation in compressible turbulence to the

overall dissipation rate.

The constants C1, C2, k and are 1.44, 1.92, 1.0, and 1.2, respectively.

Several turbulence models have been tested in the preliminary study. The temperature

contours with the realizable k- model are more reasonable than other turbulence models. The

realizable k- model can also be successful to show the phenomena of flame downwash.

However, the simulation results should be eventually validated with experimental data. When

the field data become available, comparison between the numerical simulation and actual

measurement needs to be made. In case the simulation results do not agree with the field data

well, more turbulence models, such as the realizable k- model with enhanced wall treatment

should be tested. Note that the flare simulations can be remarkably affected by chemical kinetics

and the turbulence model is only partially contributing to the results. Therefore, the final

solution will be a proper combination of turbulence modeling and chemical kinetics.

Chemistry Model Selection and Model Parameters

Page 41
Two types of combustion/chemical reaction models are being considered: Eddy-

dissipation finite-rate model and non-premixed combustion (PDF) model.

PDF vs. EDC CFD Models

Two different models, EDC and PDF were used to run the test cases: (1) Non Premixed

Model (PDF Transport and Mixture Fraction); and (2) EDC (Eddy Dissipation Model- Species

Transport). A comparison of the two modeling approaches is further summarized below.

Table 2.10: Comparison of the two modeling approaches


Turbulence- Numerical
Species
Chemistry Mechanism Viscous Model Solution
Model
Interaction Method
EDC Species k-Epsilon Green Gauss
model Transport EDC 50 Species (realizable) Cell based
PDF Non k-Epsilon Green Gauss
model Premixed PDF 50 Species (realizable) Cell based

To run the cases in non-premixed combustion model we need to create pre-PDF which

contains the required mean mixture fractions of each species involved in combustion. PDF table

is made up from number of flamelets. These flamelets are embedded together to generate PDF

table. The fuel composition is given during the generation of flamelets. However, in EDC model

the species equations are solved for each species. The mesh used in both the simulation case is

same. The 50 Species reduced mechanism is used in simulation. The time required to obtain the

converged results using the PDF model is much less as compared to the EDC model.

Eddy-dissipation finite-rate model

When the user chooses to solve conservation equations for chemical species, FLUENT

predicts the local mass fraction of each species, Yi, through the solution of a convection-

diffusion equation for the ith species. This conservation equation takes the following general

form:

Page 42
(2.14)

where Ri is the net rate of production of species by chemical reaction (described later in this

section) and Si is the rate of creation by addition from the dispersed phase plus any user-defined

sources. The reaction rates that appear as source terms in Eq. 5.14 are computed in FLUENT by

one of three models:

Laminar finite-rate model: The effects of turbulent fluctuations are ignored, and reaction

rates are determined by Arrhenius expressions.

Eddy-dissipation model: Reaction rates are assumed to be controlled by the turbulence,

so expensive Arrhenius chemical kinetic calculations can be avoided. The model is

computationally cheap, but, for realistic results, only one or two step heat-release

mechanisms should be used.

Eddy-dissipation-concept (EDC) model: EDC model is an extension of the Eddy-

dissipation model. Detailed Arrhenius chemical kinetics can be incorporated in turbulent

flames. However, typical reaction mechanisms are invariably stiff and their numerical

integration is computationally costly. Hence, the model should be used only when the

assumption of fast chemistry is invalid, such as modeling the slow CO burnout in rapidly

quenched flames, or the NO conversion in selective non-catalytic reduction (SNCR).

The generalized finite-rate formulation is suitable for a wide range of applications

including laminar or turbulent reaction systems, and combustion systems with premixed, non-

premixed, or partially-premixed flames.

Non-premixed combustion (PDF) model

Page 43
Non-premixed modeling involves the solution of transport equations for one or two

conserved scalars (the mixture fractions). Equations for individual species are not solved.

Instead, species concentrations are derived from the predicted mixture fraction fields. The

thermo-chemistry calculations are preprocessed and then tabulated for look-up in FLUENT.

Interaction of turbulence and chemistry is accounted for with an assumed-shape Probability

Density Function (PDF).

The non-premixed modeling approach has been specifically developed for the simulation

of turbulent diffusion flames with fast chemistry. This approach is valid whenever non-

equilibrium effects such as extinction, reignition, lift-off and blow-out are not important, and it

greatly simplifies the chemistry modeling of any combustion system. For such systems, the

method offers benefits over the eddy-dissipation formulation. The non-premixed model allows

intermediate (radical) species prediction, dissociation effects, and rigorous turbulence-chemistry

coupling. The method is computationally efficient in that it does not require the solution of a

large number of species transport equations. This model is implemented in Fluent such that

chemistry calculations are also preprocessed and tabulated. When the underlying assumptions are

valid, the non-premixed approach is preferred over the eddy-dissipation formulation.

In a non-premixed combustion model, the mixture fraction concept plays a vital role.

Considering certain assumptions, the instantaneous thermo-chemical state of the fluid is related

to a conserved scalar quantity known as mixture fraction f. First, the mass fraction of species can

be defined as

(2.15)

Consider a mixture of pure waste gas, so mass fraction of waste gas propylene C3H6 (w) = 1. The

mass fraction of carbon C (ZC) = 0.86, and the mass fraction of hydrogen H (ZH) = 0.14. The

Page 44
elemental mass fractions remain constant throughout all the reactions. In non-premixed

combustion model, flame is considered as co-flow of fuel and oxidizer. In such mixture, the

mixture fraction f for element n at a specific point can be given as:

(2.16)

When two equations need to be solved in non-premixed combustion model: Mean mixture

fraction equation , and mixture fraction variance equation . The conservation equation for

the mean mixture fraction is given below:

(2.17)

The conservation equation for the mixture fraction variance is given below:

(2.18)
where

- User defined source term

- Source term due solely to transfer of mass into the gas phase from reacting particles

The constants , , and are 0.85, 2.86, and 2.0, respectively

After that, the probability density function (PDF), written as can be considered as

the fraction of time that the fluid spends in the vicinity of the state f.

(2.19)

where is the time scale, and is the amount of time that f spends in the band. The

shape of the function depends on the nature of the turbulent fluctuations in f. In

Page 45
practice, is unknown and is modeled as a mathematical function that approximates the

actual PDF shapes that have been observed experimentally. Figure 2.15 shows the graphical

description of the probability density function.

Figure 2.15: Graphical description of the probability density function,

Non-premixed model parameters

The parameters used for the non-premixed combustion simulation are given below:

Table 2.11: Parameters and settings for non-premixed combustion model:


State Relation Steady Flamelet
Energy Treatment Non-adiabatic
Number of Grid Points in Flamelet 100
Maximum number of flamelets 30
Initial Scalar Dissipation (1/s) 0.01
Scalar Dissipation Step (1/s) 5

Table 2.12: Parameters for PDF table generation in non-premixed combustion model:
Number of Mean Mixture Fraction Points 100
Number of Mixture Fraction Variance Points 50
Maximum number of Species 50
Number of Mean Enthalpy Points 41
Minimum Temperature (K) 288

Page 46
CFD Model Calibration

The CFD modeling was checked against the wind tunnel data from the University of

Alberta [24]. The schematic of a closed-loop wind tunnel facility (all dimension in meters) at the

University of Alberta is given in Fig. 2.16. The computational domain is simplified as in Figure

5.6. The diameter of the jet is taken as 0.0221 m. The Flare stack is surrounded by a cylinder of

0.5 m diameter to predict the smooth flame phenomenon through the Jet. Velocity inlet

boundary condition is applied at cross wind inlet (Left of the Box) and Jet Inlet. Pressure outlet is

applied at the extreme right surface of the box to predict the outlet phenomenon. Remaining

surfaces are defined as wall. Jet velocity is 2.11 m/s however the wind velocity is 8.27 m/s.

Mean mixture fraction is given 1 at the fuel inlet. Table 2.13 listed the important parameters in

simulation.

Fig. 2.16: Schematic of a closed-loop wind tunnel facility (all dimension in meters) at the
University of Alberta.

Figure 2.17: Computational domain for validation with wind tunnel experiments.

Page 47
Table 2.13: Parameters used in wind tunnel simulation
Jet Cross wind
Intensity = 15% Hydraulic Dia. = 0.0221 Intensity =0.2% Hydraulic Dia. = 1.94

Structured mesh has been created using cooper algorithm and different size functions.

The meshed geometry contains 0.62 million cells and 0.65 million nodes. To study the

downwash phenomenon of the flame, dense mesh has been used near the flare stack. The wind

tunnel geometry (top) and the structured mesh used in CFD (bottom) are shown in Fig. 2.18.

(a) Front View of Meshed Geometry

(b) Meshed Geometry


Figure 2.18: Mesh for simulation of flare in wind tunnel

Page 48
Table 2.14 lists the components of fuel as well as the flow rate. Table 2.15 shows the

major emissions of flare in wind tunnel simulation. The flare combustion efficiency calculations

are listed in Tables 2.16. To understand the flare combustion, Fig. 2.19 shows the temperature

profile in the central plane. The contours of different species are given in Fig. 2.20.

Table 2.14: Parameters used in wind tunnel simulation


Inlet Species
Species Flow Rate(Kg/s) Carbon In
CH4 4.82E-04 3.62E-04
CO2 1.07E-05 2.91E-06
C2H4 1.85E-06 1.58E-06
C2H6 1.92E-05 1.54E-05
Total 5.14E-04 3.81E-04

Table 2.15: Major emissions of flare in wind tunnel simulation


Emissions
Normalized Carbon Normalized
Species Flow Rate (Kg/s)
Flow Rates(Kg/Kg) In(Kg/s) Carbon
CH4 1.42E-06 2.76E-03 1.07E-06 1.06E-06
CO2 1.28E-03 2.49E+00 3.50E-04 3.47E-04
CO 7.95E-05 1.55E-01 3.41E-05 3.38E-05
CH2O 3.42E-17 6.66E-14 1.37E-17 1.36E-17
NO 2.68E-06 5.21E-03 0 0.00E+00
HO2 4.03E-10 7.84E-07 0 0.00E+00
OH 4.50E-08 8.76E-05 0 0.00E+00
Total 1.37E-03 2.66E+00 3.85E-04 3.81E-04

Table 2.16: Flare efficiency calculation


Total Carbon In 38.14
Total Carbon Out 38.47
Combustion Efficiency 91.65%
Mass Balance Error -0.858%

Page 49
Figure 2.19: Temperature profile of flare in wind tunnel

1. Contours of Mass fraction of 'NO' 2. Contours of Mass fraction of 'C2H4'

3. Contours of Mass fraction of 'OH' 4. Contours of Mass fraction of 'CH2O'

5. Contours of Mass fraction of 'CH4' 6. Contours of Mass fraction of 'CO'

7. Contours of Mass fraction of 'CO2' 8. Contours of Mass fraction of 'HO 2'

Page 50
Figure 2.20: Contours of different species of flare simulation in wind tunnel
3. RESULTS

3.1 Modeling Air-assisted Flare Cases

A new geometry was created to simplify the tip geometry so that the computation time

can be reduced. Unlike previous geometry, which had an assembly of jet spiders for fuel flow, a

concentric ring was used in the new geometry. This helps in keeping the fuel distributed over the

total flare area, and at the same time simplifies the geometry and hence the simulation. Figure

3.1 shows the computational domain and the flare with pilot tip, fuel inlet and air inlet.

Figure 3.1: Computational domain and the flare with pilot tip, fuel inlet and air inlet

Page 51
Tulsa air-assisted flare test cases were run using the geometry described in Section 6.3

[22]. The conditions are as provided in Tables 6.16 taken from [6, 23]. The boundary conditions

are given in Table 3.1. The different flow velocities, LHV and excess air factor used to simulate

each case is listed in Tables 3.2 and 3.3.

Table 3.1: Conditions used for air-assisted flare cases

Actual Vent Gas (VG) Flow Rates Air Flow Cross


Test Rate Excess Air
Propylene TNG Nitrogen Total LHV Wind
Point Factor
(MPH)
lb/hr lb/hr lb/hr lb/hr Btu/scf lb/hr

A1.1 918.88 0.00 0.00 918.88 2107.71 149173.00 11.00 12.70

A2.1 355.02 0.00 0.00 355.02 2125.45 83818.00 15.90 12.80

A2.3 352.14 0.00 0.00 352.14 2108.22 88791.00 17.00 10.10

A2.4 352.87 0.00 0.00 352.87 2112.57 148799.00 28.50 10.00

A2.5 354.71 0.00 0.00 354.71 2123.55 119580.00 22.80 13.30

A5.3 71.26 7.55 271.37 350.18 341.87 32876.17 28.00 2.50

Table 3.2: Boundary Conditions, flow velocities, LHV and excess air factor for Air-Assisted
Flare Cases
Turbulent
Velocities
Intensities Excess
Test Fuel Cross Air
Pilot Air Assist, LHV Air
Point Jet Wind Assist Cross
Pilot Factor
Wind
m/s m/s m/s m/s and Jet Btu/scf
A1.1 0.43 0.19 5.69 23.12 10.00% 5.00% 2107.71 11.00
A2.1 0.17 0.19 5.72 12.99 10.00% 5.00% 2125.45 15.90
A2.3 0.17 0.19 4.51 13.76 10.00% 5.00% 2108.22 17.00
A2.4 0.17 0.19 4.49 23.06 10.00% 5.00% 2112.57 28.50
A2.5 0.17 0.19 5.94 18.53 10.00% 5.00% 2123.55 22.80
A5.3 0.23 0.19 1.12 5.10 10.00% 5.00% 341.87 28.00

Page 52
Table 3.3: Boundary Conditions Air-Assisted Flare Cases
Boundary Conditions
Assisting Air Fuel (Pure Propylene) Cross wind
Velocity (m/s) 12.99 0.1656 5.74
Turbulence Intensity 5% 5% 5%

The modeled DRE/CE results and comparison with the measured values are given in Section 3.4.

As seen in Table 3.4, even at a relative high CE (e.g., 77%), significant amounts of

formaldehyde, ethylene, and HOX are also predicted.

Table 3.4: Normalized Emission Rates and Carbon Distribution


Normalized Emission Rates (kg/kg C3H6 in)
Species A1.1 A2.1 A2.3 A2.4 A2.5 A5.3
-01 -01 -01 -01
CO 2.86 10 1.90 10 1.25 10 1.29 10 4.03 10-01 1.29 10-01
THC 2.72 10-01 2.78 10-01 1.71 10-01 2.15 10-01 2.93 10-01 2.71 10-01
CH2O 1.72 10-02 1.21 10-02 7.16 10-03 4.06 10-03 1.99 10-02 1.03 10-02
C2H4 1.08 10-02 1.27 10-02 1.18 10-02 1.17 10-02 1.77 10-02 6.62 10-03
HOX 8.41 10-03 8.07 10-03 4.97 10-03 4.71 10-03 1.48 10-02 7.18 10-03
Carbon Distribution (%)
CO2 59.98% 63.97% 77.18% 72.42% 52.54% 66.91%
THC 25.74% 26.55% 16.58% 21.11% 27.31% 26.48%
CO 14.28% 9.48% 6.24% 6.47% 20.15% 6.61%

3.2 Modeling Steam-assisted Flare Cases

Theory and concept

Steam assisted flares are designed to dispose of heavier waste gases which have a greater

tendency to smoke. In order to prevent incomplete combustion, steam is injected into the waste

stream using peripheral steam rings, center steam sprayers, and/or inner induction tubes. The

injection of steam has two principal effects:

High-pressure steam flow causes turbulence in the waste stream which improves mixing

and therefore improves combustion efficiency.

Page 53
Additional air is induced into the waste gas providing the oxygen necessary for

augmented smokeless capacity.

Steam flares are typically used in applications where the customer has high-pressure

steam available on site [25]. CFD modeling will be done in three parts, which are mentioned

below.

1. Simulation of all steam assisted flare test with varying LHV, without any center or upper

steam.

2. Second phase would include introduction of center steam to cover all the cases tested in

Tulsa.

3. Third phase would include introduction of both center and upper steam.

Simple Geometry

Simple geometry consists of a flare stack, enclosed in a box and the mixture of fuel

(propylene) and Steam is passed through the flare. This geometry was made because when we

are using PDF transport model. We can define only two inlet streams fuel and air. There is no

way to define a separate stream for steam in the PDF model so centre steam and upper steam is

mixed with the propylene and passed through the flare. Total flow passed through flare is 6657

lbs/hr. The simple geometry is as shown in Fig.3.2.

Page 54
Flare Stack

Fuel + Steam Inlet

Fig. 3.2 Steam Assisted Flare- Simple Geometry

Complex Geometry

This geometry consists of a 20 upper steam nozzles, centre steam nozzle and flare stack.

This geometry is used to simulate the flow through the upper steam nozzles. Mixture of fuel and

steam is passed through upper steam and flare separately. Total flow rate is divided in such a

way that 43% mixture will go through flare and remaining 57% mixture will pass through upper

steam nozzles. 2863 lbs/hr flow rate of mixture of fuel and steam is passed through flare and

3794 lbs/hr flow rate of mixture of fuel and steam is passed through upper steam nozzles. The

Page 55
complex geometry is as shown in Fig.3.3. The operating conditions for Case S1.5 are given in

Tables 3.5 and the PDF model parameters are shown in Tables 3.6 and 3.7.

Figure 3.3: Steam Assisted Flare with Complex Geometry

Table 3.5: Operating conditions for case S1.5


Actual vent gas Actual Actual steam flow Assist
Wind Ambient
Test (VG) flow rates vent rates ratio
Geometry CZHV
point gas Steam
C3H6 TNG N2 Center Upper Total Speed Temp
LHV /Air
VG flow
lb/hr lb/hr lb/hr Btu/scf Btu/scf lb/hr lb/hr lb/hr MPH Deg. F
rate
Simple/
S1.5 2337 0.00 0.00 2145 405 526 3794 4320 1.85 8.0 79.6
Complex

Page 56
Table 3.6: Non-premixed (PDF) model parameters
State Relation Steady Flamelet
Energy Treatment Non-adiabatic
Max number of flamelets 50
Initial Scalar Dissipation 0.01
Scalar Dissipation Step 5
Initial Number of Grid points in Flamelet 4
Maximum Number of Grid points in Flamelet 60
Maximum Change in value ratio 0.5
Maximum change in slope ratio 0.5

Table 3.7: PDF table generation parameters


Number of Mean Mixture Fraction Points 100
Number of Mixture Fraction Variance Points 40
Maximum number of Species 50
Number of Mean Enthalpy Points 41
Minimum Temperature (K) 288

The comparison of DRE/CE among the simple and complex geometry predictions as well

as measured data are given in Table 3.8. Speciated emission rates are shown in Table 3.9.

Table 3.8: Comparison of data and predicted DRE and CE using the PDF Approach

S1.5 S1.5
Case #
Simple Geometry Complex Geometry
CFD 100 99.9
CE (%)
Experimental 99.9 99.9
CFD 100 100
DRE (%)
Experimental 99.99 99.99

Page 57
Table 3.9: Speciated Emission Rates and Carbon Distribution
S1.5 S1.5
Case #
Simple Geometry Complex Geometry
Calculated Emission Rates (lbs/lbs C3H6 in)
CO2 3.14 10 3.14 10
CO 1.51 10-09 2.50 10-04
THC 1.78 10-16 1.61 10-08
CH2O 0.00 10 1.23 10-11
C3H6 0.00 10 1.05 10-13
C2H4 0.00 10 5.33 10-11
HOX 0.00 10 1.46 10-04
Carbon Distribution
CO2 99.9% 99.987%
C3H6 0.00% 0.000%
CO 0.01% 0.013%
THC 0.00% 0.000%

Temperature Contours Contours of Mass Fraction of CO2

Contours of Mass Fraction of C3H6 Contours of Mass Fraction of CO

Figure 3.4: Case S1.5 contours with simple geometry

Page 58
Temperature Contours Contours of Mass Fraction of CO2

Contours of Mass Fraction of C3H6 Contours of Mass Fraction of CO

Figure 3.5: Case S1.5 contours with complex geometry

Even though the DRE and CE are almost same for both geometries, the shape and

deflection are different, Figures 3.4- 3.5.

3.3 Parametric Study for Air and Steam-Assisted Flares

Parametric Study for Air-Assisted Flare Using the EDC Approach

As turbulent intensity (TI) of vent gas, air assist etc. decreases, DRE and CE increase

(See Table 3.10). The use of very low values of TI (e.g., 0.1%) may not be appropriate for

assisted air; but it may still be appropriate for the fuel. That will be the subject of the next

parametric study.

Page 59
Table 3.10: Case A1.1 Turbulence intensity parametric study
Turbulence
5% (Jet) 0.1% (Jet) 5% (Jet) Measured
Intensity
10% (Cross wind) 10% (Cross wind) 2% (Cross wind)
DRE 78.65% 85.12% 79.10% 98.05%
CE 59.98% 71.66% 60.25% 96.72%

Fig. 3.6: Comparison of temperature contours with different turbulent intensities

Change in TI values for crosswind has negligible effect on DRE/CE. No visible change in

flame profile or deflection with change in crosswind TI values (Fig. 3.6). However, in the follow

up parametric study, the TI for cross wind can be further reduced.

Base Case and Testing of Effect of Turbulence Intensity (TI) on Flare Efficiency Using the
PDF Approach

Case A2.1 from the Tulsa flare test plant was used as the baseline case and to test the

effect of turbulence intensity on flare efficiency. The parameters used for the non-premixed

Page 60
combustion simulation are given in Tables 3.11-3.13. The other boundary conditions and

turbulence parameters were kept same as for the EDC model simulations.

Table 3.11: Parameters and settings for non-premixed combustion model:


State Relation Steady Flamelet
Energy Treatment Non-adiabatic
Number of Grid Points in Flamelet 100
Maximum number of flamelets 30
Initial Scalar Dissipation (1/s) 0.01
Scalar Dissipation Step (1/s) 5

Table 3.12: Parameters for PDF table generation in non-premixed combustion model:
Number of Mean Mixture Fraction Points 100
Number of Mixture Fraction Variance Points 50
Maximum number of Species 50
Number of Mean Enthalpy Points 41
Minimum Temperature (K) 288

Table 3.13: Operating conditions for Case A2.1


Fuel velocity 0.1656 m/s
Air velocity 12.99 m/s
Cross wind velocity 5.74 m/s
Fuel Propylene
LHV value 2125.45 BTU/scf

While simulating complex combustion process in a turbulent environment, parameter like

Turbulent Intensity (TI) plays an important factor. The effect of TI of fuel can easily affect the

extent of combustion and hence the flare efficiencies. A parametric study was run to study the

effect of TI of the fuel and the assisting air over the flare efficiencies. For this purpose Test Case

A2.1 was taken as the base case and then 5 different values of TI were used. The values used

were given in Table 3.14.

Page 61
Table 3.14: Turbulence Intensity Used for Parametric Study
Turbulent Intensities (%)

Cases 0.10 1.00 5.00 10.00 15.00

Figures 3.7-3.9 show the contours of temperature, mass fraction of CO2 and N2O at

different values of TI. From the three contours, no major difference in the flame temperature or

composition of the flame can be observed.

Fig. 3.7: Contours of static temperature at different TI values (K)

TI = 0.1% TI = 1.0% TI = 5.0%

TI = 10.0% TI = 15.0%
Fig.3.8: Contours of mass fraction of CO2 at different TI values

Page 62
TI = 0.1% TI = 1.0%

TI = 5.0% TI = 10.0% TI = 15.0%


Fig.3.9: Contours of mass fraction of NO2 at different TI values

To further observe any noticeable effect of the TI value on the flare, the emissions data

and the flare efficiencies (Combustion Efficiency and Destruction and Removal Efficiency) were

calculated, and Tables 3.15 and 3.16 list the results.

Table 3.15: Emissions (kg/s) of Species at Different TI Values Using the PDF Approach (Case
A2.1)
Species TI = 0.1 TI = 1.0 TI = 5.0 TI = 10.0 TI = 15.0
-06 -06 -06 -06
OH 5.54 10 6.93 10 7.28 10 7.02 10 4.45 10-06
HO2 7.57 10-09 9.47 10-09 9.95 10-09 9.59 10-09 6.08 10-09
CO 2.01 10-05 2.51 10-05 1.63 10-05 2.54 10-05 1.61 10-05
-01 -01 -01 -01
CO2 1.41 10 1.43 10 1.51 10 1.61 10 1.32 10-01
CH2O 2.88 10-15 3.57 10-15 3.75 10-15 3.61 10-15 2.30 10-15
C2H2 8.88 10-15 1.15 10-1 1.21 10-14 1.17 10-14 7.22 10-15
CH2CO 7.11 10-18 8.97 10-18 9.43 10-18 9.09 10-18 1.18 10-17
-18 -18 -18 -18
C3H6 5.19 10 6.57 10 6.91 10 6.66 10 4.18 10-18
N 1.28 10-11 1.66 10-11 1.75 10-11 1.69 10-11 2.19 10-11
NH 3.04 10-13 3.83 10-13 3.86 10-13 3.72 10-13 4.85 10-13
-05 -05 -05 -05
NO 5.15 10 6.13 10 6.47 10 6.36 10 4.34 10-05
N2O 3.24 10-09 4.04 10-09 4.24 10-09 4.09 10-09 2.60 10-09
HNO 7.29 10-11 0.00 10-11 9.56 10-11 9.21 10-11 5.96 10-11

Page 63
Table 3.16: Calculated efficiencies using the PDF approach (Case A2.1)
Calculated Efficiencies
TI (%) 0.1 1 5 10 15 Experimental
DRE 100.00% 100.00% 100.00% 100.00% 100.00% 97.15%
CE 99.98% 99.97% 99.98% 99.98% 99.98% 95.54%

The parameter study of TI from 0.1 to 15% in the PDF model does not show any significant

impact on DRE and CE.

3.4 Comparison of CFD Prediction and Flare Test Data

The report summarizes the comparison with analytical results generated from field

sampling at John Zink facility, Tulsa, Oklahoma. Modeled CE, DRE values were compared vs.

Aerodyne, IMACC, Telops and TRC data collected during the TCEQ Flare tests in Tulsa, OK.

Comparison of the EDC model with Measured Data for Air-Assisted Flares

Figs. 3.10 -3.11 show the DRE and CE values as a function of assisted air flow rate.

Series A2 was simulated using EDC model; however series A3 and A5 were simulated using

PDF transport model.

Page 64
Fig. 3.10: DRE as a function of air assist flow rate

Fig. 3.11: CE as a function of air assist flow rate

Fig.3.12- Fig.3.13 present DRE and CE as a function of lower heating value (LHV). In

general, the EDC model under- predicts DRE by 6% to 19% with an average of 12%. It under-

predicts CE by 12% to 39% with an average of 25%. Comparing the EDC results with measured

results, DRE is within the uncertainty limits (19%). While CE is on average reasonably within

Page 65
the tentative uncertainty limit (23%); predicted CEs in many occasions go beyond the stated

uncertainty limit [6, 23, 26, 27].

Fig. 3.12: DRE as a function of LHV

Fig.3.13: CE as a function of LHV

Page 66
Table 3.17: Comparison of predicted DRE/CE with measured data using the EDC approach

Destruction and
Test Combustion Efficiency
Removal Efficiency
Point
Measured CFD Measured CFD
A1.1 98.06% 78.65% 96.72% 59.98%
A2.1 97.15% 84.14% 95.54% 63.97%
A2.3 96.19% 87.42% 94.02% 77.18%
A2.4 92.26% 81.82% 87.64% 72.42%
A2.5 95.09% 82.95% 91.84% 52.54%
A5.3 82.32% 76.11% 78.84% 66.91%

Comparison of the PDF model with Measured Data for Steam-Assisted Flares

Nine steam-assisted flare cases were simulated using the PDF approach. The conditions are

given in Table 3.18. The geometry used for simulating steam flare cases is described in Figure

3.3 (complex geometry). Case S1.5 was simulated using both simple and complex geometries.

Table 3.18: Conditions used for steam-assisted flare cases

Test Actual Vent Gas (VG) Flow Rates Actual Steam Flow Rates Assist Ratio Wind
Point C3H6 TNG N2 Total LHV Center Upper Total Steam / Speed
lb/hr lb/hr lb/hr lb/hr Btu/scf lb/hr lb/hr lb/hr VG Flow Rate MPH
S1.5 2337 0.0 0 2,337 2,145 526 3,794 4,320 1.85 8.0
S1.8 2338 0.0 0 2,338 2,146 506 7,044 7,550 3.23 8.6
S3.7 191 18.9 716 926 346 0 228 228 0.25 7.1
S4.1 491 45.0 1,800 2,335 350 560 536 1,096 0.47 5.6
S4.3 485 45.0 1,802 2,332 346 567 1,879 2,447 1.05 5.2
S5.2 320 33.8 585 938 595 454 1,580 2,035 2.17 10.2
S5.3 312 31.7 578 922 590 482 783 1,264 1.37 9.3
S6.1 826 79.1 1,456 2,361 609 518 1,003 1,521 0.64 8.8
S7.3 297 29.9 1,084 1,410 353 516 538 1,054 0.75 7.9

The results for Case S1.5 with different geometries are reported in Table 3.19. It can be seen that

the PDF approach predicts same efficiencies when different geometries are used.

Page 67
Table 3.19: Comparison of data and predicted DRE and CE using the PDF Approach
S1.5 S1.5
Case #
Simple Geometry Complex Geometry
CFD 100.00 99.99
CE (%)
Experimental 99.99 99.99
CFD 100.00 100.00
DRE (%)
Experimental 99.99 99.99

In order to define the two streams (centre and upper steam) separately, the complex geometry

was used for simulating additional eight steam-assisted cases. In total, the PDF approach was

used to simulate nine steam assisted cases. The comparison of simulation versus experimental

results is shown in Table 3.22.

Temperature Contours Contours of Mass Fraction of C3H6

Fig 3.14: Contours of Case S4.1 using the PDF model

Page 68
Contours of mass fraction of CO Contours of mass fraction of CO2

Contours of mass fraction of CH2O Contours of mass fraction of NO

Fig 3.15: Contours of Case S4.1 using the PDF model

Figures 3.14-3.15 show various contours predicted by the CFD simulation. The contours show

the profiles of temperature and mass fractions of various species for Case S4.1 with the complex

geometry.

Figures 3.16 and 3.17 show DRE and CE as a function of total steam flow rate. All the

cases plotted in these charts are simulated using PDF transport approach.

Page 69
Fig. 3.16: DRE as a function of steam flow rate

Fig. 3.17: CE as a function of steam flow rate

Figure 3.18 and 3.19 present DRE and CE as a function of lower heating value. We had

tried to simulate steam flare cases using EDC model but flame was unable to sustain. So reported

all the cases of steam flare are simulated using PDF model.

Page 70
Fig. 3.18: DRE as a function of LHV

Fig. 3.19: CE as a function of LHV

Page 71
Comparison of the PDF model with Measured Data for Air-Assisted Flares

Along with the steam cases, an additional air based case was also simulated using the

PDF approach. Fig.3.20 below shows the geometry used for this case. Similar to the flare used

during the Tulsa air based tests, the CFD geometry consisted of a spider shaped fuel source.

Fig.3.20: Air assisted flare Spider shaped geometry

Page 72
Table 3.20: Conditions used for air-assisted flare cases

Test Actual Vent Gas (VG) Flow Rates Air Flow Excess Wind
Point Propylene TNG Nitrogen Total LHV rate Air Speed
lb/hr lb/hr lb/hr lb/hr Btu/scf lb/hr Factor MPH
A1.1 918.88 0.00 0.00 918.88 2107.71 149173.00 10.96 12.72
A2.1 355.02 0.00 0.00 355.02 2125.45 83818.00 15.94 12.84
A2.3 352.14 0.00 0.00 352.14 2108.22 88791.00 17.03 10.10
A2.4 352.87 0.00 0.00 352.87 2112.57 148799.00 28.48 10.04
A2.5 354.71 0.00 0.00 354.71 2123.55 119580.00 22.77 13.28
A3.1 181.23 18.77 702.55 902.55 338.67 19387.00 6.51 10.32
A3.3 181.23 18.37 700.60 900.20 333.86 60121.00 20.23 11.09
A3.6 181.23 18.76 704.18 904.17 337.55 47494.00 15.94 11.88
A5.2 72.29 7.69 274.41 354.39 342.86 75139.77 13.57 10.72
A5.3 71.26 7.55 271.37 350.18 341.87 32876.17 63.07 2.15
A4.3 298.74 30.30 591.10 920.14 562.91 66471.69 28.01 2.50
A6.1 117.80 11.86 221.21 350.87 583.73 11403.53 5.91 15.92

Table 3.21: Comparison of predicted versus measured DRE and CE values for EDC model

EDC Model Results


DRE (Destruction and Removal Efficiency) CE (Combustion Efficiency)
Test
Point Simulation Experimental Deviation Simulation Experimental Deviation
A1.1 78.65% 98.06% -19.41% 59.98% 96.72% -36.74%
A2.1 84.14% 97.15% -13.01% 63.97% 95.54% -31.57%
A2.3 87.42% 96.19% -8.77% 77.18% 94.02% -16.84%
A2.4 81.82% 92.26% -10.44% 72.42% 87.64% -15.22%
A2.5 82.95% 95.09% -12.14% 52.54% 91.84% -39.30%
A5.3 76.11% 82.32% -6.21% 66.91% 78.84% -11.93%
Mean % Deviation = -11.66% Mean % Deviation = -25.27%

Page 73
Table 3.22: Comparison of predicted versus measured DRE and CE values for PDF model

PDF Model Results


Test DRE (Destruction and Removal Efficiency) CE (Combustion Efficiency)
Point Simulation Experimental Deviation Simulation Experimental Deviation
A3.1 100.00% 99.56% 0.44% 100.00% 99.08% 0.92%
A3.3 100.00% 88.13% 11.87% 100.00% 84.68% 15.32%
A3.6 100.00% 91.73% 8.27% 100.00% 88.88% 11.12%
A4.3 100.00% 93.77% 6.23% 100.00% 91.50% 8.50%
A5.2 100.00% 69.18% 30.82% 100.00% 62.48% 37.52%
A5.3 100.00% 82.32% 17.68% 100.00% 78.44% 21.56%
A6.1 100.00% 99.71% 0.29% 100.00% 99.32% 0.68%
S1.5 100.00% 99.90% 0.10% 99.90% 99.90% 0.00%
S1.8 100.00% 97.80% 2.20% 100.00% 95.70% 4.30%
S3.7 100.00% 99.50% 0.50% 100.00% 99.20% 0.80%
S4.1 100.00% 96.40% 3.60% 100.00% 95.00% 5.00%
S4.3 100.00% 27.30% 72.70% 100.00% 21.60% 78.40%
S5.2 100.00% 38.10% 61.90% 100.00% 32.20% 67.80%
S5.3 99.97% 89.20% 10.77% 93.05% 86.60% 6.45%
S6.1 99.98% 99.50% 0.48% 95.85% 99.20% -3.35%
S7.3 99.99% 71.30% 28.69% 99.62% 71.30% 28.32%
Mean Deviation = 16.03% Mean Deviation = 17.71%

Table 3.20 summarizes all the conditions used for simulation of Air assisted cases. Tables

3.21 and 3.22 summarize all the air-based and steam-based cases simulated during the project. A

comparison of the DRE and CE values predicted by the CFD simulation and that of the

experimental values along with the type of model (EDC or PDF) used has been provided.

In the EDC modeled cases, the mean deviation of the predicted DRE values from

experimental DRE values is about 12%. Here, the negative sign implies that the CFD

simulation under predicts the experimental DRE. On the other hand, CE has a mean deviation of

around 25% from the experimental values.

Page 74
And in the PDF modeled cases, the mean deviation of the predicted DRE and CE values

from the experimental values are around 16% and 18%, respectively. Tables 3.21 and 3.22

clearly show that the EDC model under-predicts the CE/DRE values but the PDF model over-

predicts the same (except in one case where PDF under-predicts by 3%).

4. DISCUSSION

CFD Mechanisms

The 50-species combustion mechanism LU 1.0 developed by Lamar University has been

validated against methane, ethylene, and propylene experimental data. Further, NO2 was added

to LU 1.0 mechanism, dubbed as LU 1.1, and was shown in good agreement with the full

mechanism. FLUENT models (Species, Turbulence-Chemistry, Viscous, and Numerical

Solution), model parameters, and boundary conditions have been studied and selected for the

flare simulation.

Uncertainty Analysis

The USC and GRI mechanisms used in conjunction with CHEMKIN CFD and FLUENT

are all in good agreement with the laboratory experimental data [28]. Even so, the uncertainty of

measurements (listed for individual species and CE/DRE in Comprehensive flare test QAPP and

report) are somewhat smaller than the uncertainty of the combined USC/GRI mechanism [20,

29], which ranges from a factor of 1.2 to 5 for trace species and in the order of 4% (for CO2) to

15% (for NO) for major species, Table 4.1. The mechanism uncertainty factor depends on

residence, temperature, fuel composition, species, etc. Fortunately, these mechanism parameters

have been optimized with lab data, mostly within experimental errors. For the sake of simplicity,

a typical uncertainty factor of 2 will be used for trace species (in ppm range). A factor of 2

means the true value could lie from (0.5model value) to (2.0model value). For major species,

Page 75
in general, these mechanisms can reflect lab data within 10% for CO, within 15% for NO

based on GRIs assessment and are generally accepted in the combustion chemistry community.

Recent simulations suggest that the uncertainty in transport coefficients may be

significant. As a result, the uncertainty factor of 3 will be used for the trace species [29].

For relation A C / D , uncertainty is calculated by using

dA / A dC / C dD / D (4.1)

The above formula states that the uncertainty (in percentage) of A is the sum of the uncertainty

of C (in percentage) and the uncertainty of D (in percentage). For relation B b1 b2 b3 ,

uncertainty is calculated using

dB db1 db2 db3 (4.2)

The above formula states the uncertainty (in value) of B is the sum of the uncertainty of C (in

value) and the uncertainty of D (in value).

Table 4.1: Uncertainties in species prediction and measurement [6, 20-21]


Mechanism Prediction Uncertainties AQRP Measurement
Species
Uncertainties (Mechanism + CFD) Uncertainties

CO2 4% 6.00% 10%


CO 10% 15.00% 10%
CH4 0.005 0.0075 10%
C3H6 10% 15.00% 10%
H2O 4% 6.00% 15%
H2 10% 15.00% 15%
OH 10% 15.00% 15%
NO 15% 25.00% 5%
O2 0.004 0.006 2%
Note: CH4 and O2 prediction uncertainties are given in mole fractions

Page 76
Using the uncertainties from Table 4.1 and a fairly typical flare mole fraction data (CO2:

0.06; CO: 0.01; CH4: 0.03; C3H6: 0.03), the CFD prediction and Tulsa flare test uncertainties for

DRE and CE are estimated as shown in Table 4.2.

Table 4.2 Uncertainty estimates for prediction and measurement [23, 26, 27]
Uncertainties Prediction Measurement
Destruction and Removal Eff
15.00% 4.00%
(DRE)
Combustion Efficiency (CE) 19.00% 4.00%*
*Private Communication [27]

It should be noted that the expected discrepancy between the predicted and experimental

DRE and CE values are +/- 19% and +/-23%, respectively.

Summary for CFD modeling

The main operating, design, and meteorological data of the flare test campaign were

provided by the University of Texas (UT) including Combustion Efficiency (CE), Destruction

and Removal Efficiencies (DRE). Both Probability Density Function (PDF) and Eddy

Dissipation Concept (EDC) turbulence-chemistry interaction approaches have been adopted to

run Tulsa flare test cases. It was difficult to sustain the flame in cases with LHVs (lower heating

values) in the range of 350-400 BTU/Scf. Many attempts were made to keep the flame from

extinguishing but only Case A5.3 was successfully simulated. Very high amount of Air Assist (in

cases A3 andA5) coupled with extremely low LHVs are seen as the possible causes for keeping

the flame from burning.

In steam cases, with the addition of 21 steam jets (for Upper Steam) around the flare, the

number of cells, in the already complex geometry, exceeded one million. With such a complex

geometry and more than a million cells, using EDC model was not only difficult but also

computationally expensive. It was decided to run these cases using the non-premixed PDF

Page 77
model. The results of all the steam cases (simulated using the PDF model) have been added to

the final report. In addition to that, some steam cases with EDC model using a very simple

geometry is planned in the future for comparison purposes.

Twelve air-assisted flare test cases and nine steam-assisted flare test case have been run

and compared with the measured DRE/CE data. In general, the EDC model under- predicts DRE

by 6% to 19% with an average of 12%. It under- predicts CE by 12% to 39% with an average of

25%. Comparing the EDC results with measured results, DRE is within the uncertainty limits

(19%). While CE is on average reasonably close to the tentative uncertainty limit (23%);

predicted CEs in many occasions go beyond the stated uncertainty limit [6, 23, 26, 27]. The

potential causes may be the low flow rates, low heating values, high air/steam assists, complexity

of geometry, placement of the pilots, choice of turbulence intensity, and the difference between

local sampling and the full surface integration (CFD post processing).

Even though the PDF approach was verified with University of Alberta wind tunnel data

and was shown in good agreement for certain high DRE/CE cases; the more simplistic PDF

model tends to over-predict flare efficiencies than the measured ones. Nearly complete

combustion (over 99% DRE and CE) was seen when using the non-premixed combustion model.

The fast reaction or nearly equilibrium assumptions used by the PDF model may not be true for

the low BTU and high air/steam-assisted flares due to the lower flame temperatures (1600-

1950K). As a result, the PDF model appears to predict little intermediate species and radicals

like formaldehyde, OH, NO etc.

In view of the significant differences between the PDF and EDC model results, further

investigations involving other flare test cases and geometries with pilot flames in the EDC model

and more exploration on the PDF/EDC model parameters are warranted. It appears, however,

Page 78
that measured DREs/CEs from the 2010 comprehensive flare study fall somewhere between the

EDC and PDF model predictions under low-jet velocity, low BTU conditions.

Simulation of steam assisted cases using complex geometry with EDC model is

computationally expensive. Practically, simulation with EDC model using complex geometry is

very difficult. In order to simulate steam cases, we had created a simple geometry and ran

simulation but the flame was unable to sustain.

Parametric Study

The parameter study of TI from 0.1 to 15% in the PDF model does not show any significant

impact on DRE and CE.

5. FUTURE WORK

Future endeavors should include:

1. Work on all air- and steam-assisted flare cases for which DRE/CE data are available from

2010 flare study [23, 26, 27] using both EDC and PDF models

2. Fine-tune the geometries/ model parameters and investigate the difference between local

sampling and full surface integration as used in the FLUENT post processing to address

the discrepancy and uncertainty issues in DRE and CE to make the CFD models a useful

tool for flare operations.

3. Use new C1-C3 mechanism with NO2, LU 1.1, and newly developed optimized LU 2.0

mechanism for the simulation of air and steam assisted flares.

4. CFD modeling for other laboratory flames including the McKenna Burner for which

detailed, speciated lab data are available for incomplete combustion products (e.g.,

formaldehyde) [18].

Page 79
5. Model industrial flares with relative high jet velocity to simulate upset/start up scenarios

with various cross wind, air and steam-assisted conditions, and vent gas species.

6. Develop DRE/CE correlations based on experimental data and CFD simulations with jet

velocity, cross wind speed, heating value, air-assist, steam-assist, fuel composition, and

flare-tip diameter as input variables [30].

Abbreviations Used:

CFD: Computational Fluid Dynamics

PDF: Probability Density Function (Ansys Fluent species transport model)

EDC: Eddy Dissipation Concept (Ansys Fluent species transport model)

TNG: Tulsa Natural Gas

DRE: Destruction and Removal Efficiency

CE: Combustion Efficiency

VOC: Volatile Organic Carbon

TI: Turbulence Intensity

THC: Total Hydrocarbons

SST: Shear Stress Transport

SNCR: Selective Non-Catalytic Reduction

LU: Lamar University

Page 80
REFERENCES

1 Quality Assurance/Quality Control Plan, Quality Assurance Project Plan (QAPP) for Gulf Coast Hazardous
Substance Research Center (GCHSRC), Project Number 027LUB0599, June, 1998.

2 Flare efficiency study,EPA-600/2-83-052, Marc Daniel, July 1983

3 J.H. Pohl, Evaluation of the Efficiency of Industrial Flares, 1984/1985, EPA600-2-85-95 and 106

4 David Castieira and Thomas F. Edgar, Computational Fluid Dynamics for Simulation of Wind-Tunnel
Experiments on Flare Combustion Systems, Energy and Fuels 2008, 22, 16981706

5 Passive FTIR Phase I Testing of Simulated and Controlled Flare Systems, FINAL REPORT, (URS/UH/TCEQ,
2004)
(http://www.tceq.state.tx.us/assets/public/implementation/air/am/contracts/reports/oth/Passive_FTIR_PhaseI_Fl
are_Testing_r.pdf)

6 Quality Assurance Project Plan, Texas Commission on Environmental Quality Comprehensive flare Study
Project, PGA No. 582-8-862-45-FY09-04, Tracking No. 2008-81 UT/TCEQ/John Zink)

7 Sharma, R., State of the Ozone State Implementation Plan Proceeding of Ethylene Producers Conference, pp
392-397, AIChE Spring Houston Meeting 2007

8 Daniel J.Serry, C.T.Bowman, "An Experimental and Analytical Study of Methane Oxidation behind Shock
Wavess",Combustion and Flame ,14,37-48(1970).

9 Seinfeld, John H. ; Pandis, Spyros, 2006. Atmospheric Chemistry and Physics - From Air Pollution to Climate
Change, 2nd edition John Wiley and Sons

10 US EPA, Office of Air Quality Planning and Standards, Quality Assurance Handbook for Air Pollution
Measurement Systems, EP-454/R-98-004, August 1998.

11 Baukal, C.E and Schwartz, R. E., The John Zink Combustion Handbook, p. 163, CRC Press: Boca Raton, 2001

12 http://me.berkeley.edu/gri_mech/version30/text30.html#performance

13 Ferziger, J.H.; Peric, M. Computational Methods for Fluid Dynamics; 3rd edition; Springer: Berlin, 2002

14 Tomlin, A.; Pilling, M.; Turnyi, T.; Merkin, J.; Brindley, J. Mechanism Reduction for the Oscillatory
Oxidation of Hydrogen: Sensitivity and Quasi-Steady-State Analysis. Combust. Flame 1992, 91, 107-130

15 Lou, H., Martin, C., Chen, D., Li, K., Li, X., Vaid, H., Vaid, H. Tula, A., Singh, K., A Reduced Reaction
Mechanism for the Simulation in Ethylene Flare Combustion, Clean Technologies and Environmental Policy,
online edition, June 16, 2011.

16 C.K.Law , A.Makino , and T.F.Lu, "On the Off-Stoichiometric Peaking of Adiabatic Flame Temperature ", The
4th J. Meeting of the U.S. Sections of the Combustion Institute,#1844(Mar 2005)

17 Qin.Z, Yang.H, C.Gardiner, " Measurement and modeling of shock-tube ignition delay for propene ",
Combustion and Flame ,124, 246-254 (2001)

18 Anuj Bhargava and Phillip R. Westmoreland, Measured Flame Structure and Kinetics in a Fuel Rich Ethylene
Flame, COMBUSTION AND FLAME 113: 333-347, 1998

19 Davis, S. G. and Law, C. K. (1998), "Determination of and Fuel Structure Effects on Laminar Flame Speeds of
C1 to C8 Hydrocarbons", Combustion Science and Technology, 140(1), 427- 449.

Page 81
20 Hai Wang, Xiaoqing You, Ameya V. Joshi, Scott G. Davis, Alexander Laskin, Fokion Egolfopoulos and Chung
K. Law, USC Mech Version II. High-Temperature Combustion Reaction Model of H2/CO/C1-C4 Compounds.
http://ignis.usc.edu/USC_Mech_II.htm, May 2007

21 R.S.Barlow, A.N.Karpetis, J.H.Frank, J.Y. Chen,Scalar Profiles and NO formation in laminar opposed flow
partially premixed methane/air flames Combustion and flame(2001),

22 ANSYS Inc., ANSYS 13, Users Guide. Fluent Inc (2010)

23 2010 TCEQ Flare Study Final Report, The University of Texas at Austin, The Center for Energy and
Environmental Resources, TCEQ PGA No. 582-8-86245-FY09-04 and Task Order No. UTA10-000924-
LOAT-RP9, Aug. 1, 2011.

24 Kostiuk, L, Johnson, M and Thomas, G. (2004). University of Alberta Flare Research Project Final Report

25 Flare Industries, INC.


http://www.flareindustries.com/products/products/pdf/3%20Steam%20Assist%20Flares%201.pdf

26 Quality Assurance Project Plan, Lamar University, "Development of Speciated Industrial Flare Emission
Inventories for Air Quality Modeling in Texas", Project 10-022

27 Private communications with Vincent M. Torres, 2010 Flare Study Project manager, 10/07/2011.

28 David Castieira and Thomas F. Edgar, CFD for Simulation of Steam-Assisted and Air-Assisted Flare
Combustion Systems, Energy and Fuels 20, 1044-1056 (2006)

29 David A. Sheen, Xiaoqing You, Hai Wang, Terese Lovas, Spectral uncertainty quantification, propagation and
optimization of a detailed kinetic model for ethylene combustion, Proceedings of the Combustion Institute,
32(1), 2009, 535-542

30 M. R. JOHNSON and L. W. KOSTIUK, A PARAMETRIC MODEL FOR THE EFFICIENCY OF A FLARE


IN CROSSWIND, Proceedings of the Combustion Institute, Volume 29, 2002/pp. 19431950

Page 82

Você também pode gostar