Você está na página 1de 5

Chemometrics and Intelligent Laboratory Systems 73 (2004) 67 71

www.elsevier.com/locate/chemolab

The use of Design of Experiment and sensory analysis as tools for the
evaluation of production methods for milk
Gunilla Wormbs *, Annika Larsson, Josefine Alm, Christina Tunklint-Aspelin,
Olof Strinning, Eva Danielsson, Henrik Larsson
Arla Foods Innovation, SE-105 46, Stockholm, Sweden
Received 17 June 2003
Available online 10 March 2004

Abstract
Milk and milk-based products are heat treated at the dairy to eradicate pathogenic microorganisms and to prolong the products shelf life.
There are differences of opinion about which process is the most gentle towards milk and milk-based products when the product is heat
treated at high temperatures.
Before making an investment decision on new production equipment for milk and milk-based products, Arla Foods was given the
opportunity to compare two different production methods.
It was decided to use Design of Experiment (DoE), as this is an effective method for obtaining maximum information with a minimum of
experiments. Its a method to determine which factors significantly influence the measured variables.
The effects of the two production methods for milk were investigated. Variable factors comprised the processes (A and B), fat content of
the milk and temperature. This was a full factorial design with three centre points for each process. The responses were different sensory
attributes.
The results showed that there were no significant differences regarding sensory analysis for the production methods. This facilitated an
investment decision based on other criteria such as price, maintenance costs, service agreement, etc.
D 2004 Elsevier B.V. All rights reserved.
Keywords: Experimental design; Milk production; Sensory analysis; Heat treatment process

1. Introduction
Arla Foods is one of the largest dairy companies in
Europe and produces a variety of milk and dairy products at
different dairies.
Milk and milk-based products, such as milk drinks and
infant formulas, are heat treated to eradicate pathogenic
microorganisms and to prolong the products shelf life. The
intensity of the heat treatment is determined by the combination of temperature and holding time. From a microbiological viewpoint, an intense heat treatment is desirable to
prolong shelf life. The more intense the heat treatment,
however, the more off-flavours are generated in the milk and
milk-based products and the more changes occur in the
milks nutritional value, e.g., protein denaturation at high
temperatures. Therefore, the choice of the time/temperature
* Corresponding author. Tel.: +46-8-677-32-10; fax: +46-8-20-33-29.
E-mail address: gunilla.wormbs@arlafoods.com (G. Wormbs).
0169-7439/$ - see front matter D 2004 Elsevier B.V. All rights reserved.
doi:10.1016/j.chemolab.2003.12.013

combination is a compromise between the microbiological


aspects and quality aspects.
To make this heat treatment possible, various categories
of heat treatment have been developed over the years [1]
(Table 1). Each heat treatment generates its own flavour
profile due to the process different thermal load, typically
different levels of cooked flavour, sterilized caramellized
flavour, UHT ketone flavour [2].
For processes with high-temperature heat treatment, such
as ultrapasteurisation and UHT, different methods can be
used. Heat treatment can be applied with indirect heating,
such as plate heat exchangers, tubular heat exchanges or
scraped surface heat exchangers, or with direct heating, such
as steam injection systems or steam infusion systems.
There are several of the abovementioned heat treatment
processes available within Arla Foods at different dairies.
For some years now, there has been a discussion which
process is generating the fewest off-flavours in the milk
compared to the HTST pasteurisation process normally

68

G. Wormbs et al. / Chemometrics and Intelligent Laboratory Systems 73 (2004) 6771

Table 1
Main categories for the heat treatment of milk
Process

Temperature (jC)

Time

Thermisation
LTLT pasteurisation of milk
HTST pasteurisation of milk
HTST pasteurisation of cream, etc.
Ultra pasteurisation
UHT (flow sterilisation) normally
Sterilisation in container

63 65
63
72 75
>80
125 138
135 140
115 120

15 s
30 min
15 20 s
24 s
24 s
A few seconds
20 30 min

Source: Ref. [1]

used. Products from different dairies are not easily comparable, due to the different process solutions rendering it
difficult to compare the products. Several years ago, work
was carried out at Arla Foods to compare different processes
at different dairies. There were no clear results from this
study.
When Arla Foods faced a decision about which process
to invest in for a new plant, the question of which process to
buy was raised again. Before making the investment decision, the company was given the opportunity to compare
two heat treatment processes at the same plant and in the
same production line. It was then decided to use Design of
Experiment (DoE) to achieve a good comparison of the two
processes, as DoE is an effective way of obtaining maximum information with a minimum of experiments [3]. This
provides enhanced information about the effects and noise
of the variable factors than by varying one factor at a time.
The aim of this study was to compare the two processes
regarding the sensory characteristics of milk under different
processing conditions.

2. Experimental
The investigation to compare the two heat treatment
processes comprised of pilot plant experiments to prepare
milk samples, to train the sensory panel, large-scale experiments at a production plant and analyses of the milk.

Natural milk was chosen as the product to be tested, as it


is more sensitive to off-flavours than other milk-based
products.
2.1. Large-scale experiments
Several experiments were made in large-scale production. The experiments were planned according to Design of
Experiment (DoE) as a full factorial design with three
factors (fat content of the milk, processes A and B, temperature), including three centre points for each process.
Holding time was kept constant throughout the experiments.
This resulted in 14 experiments being performed over a 4day period at a production plant.
The experiments were evaluated using Partial Least
Squares/Projection to Latent Structures (PLS) and Multiple Linear Regression (MLR) using the MODDE 6.0
software package, supplied by Umetrics AB, Umea,
Sweden.
2.2. Sensory analysis
The milk was analysed using Quantitative Descriptive
Analysis (QDA) [4,5] and a trained panel specialised in
assessing beverages. The sensory analysis was performed at
Arla Foods Innovation Centre Stockholm. The panel was
trained in sensory attributes for white milk. Presented in this
paper are milk flavour and off-flavour of chalk.
The sensory analysis was performed with a nine-point
scale for each given attribute. The trained panel consisted of
at least six trained assessors. The milk was assessed at + 8
jC on the expected use-by date. The assessors registered
their scores on computers in individual booths. The serving
order was random for all assessors. The assessors results
were evaluated using Tukeys test [6] at a 95% significant
level, looking for assessor/assessor interaction and assessor/
sample interaction. The mean score for each attribute was
then used as responses in the experimental design. The
collection of sensory data and evaluation was implemented

Fig. 1. Replicate plot of raw data for milk flavour.

G. Wormbs et al. / Chemometrics and Intelligent Laboratory Systems 73 (2004) 6771

69

Fig. 2. Replicate plot of raw data for chalk flavour.

using FIZZ 1.30 the software package, supplied by Biosystems, Dijon, France.
2.3. Chemical analysis
The milk from the different experiments was analysed
according to the fat content with IR Milko Scan FT120 to
control the settings for the milk fat content in the design.

3. Results and discussion


Looking at replicate plots of the sensory analysis (Figs. 1
and 2), it is clear that the trained sensory panel has given the
experiments with centre points fairly equal scores with the
exception of experiment no. 12. This milk sample received
numerous different off-flavour comments. Something had
gone wrong during the production of this milk, resulting in a
recontamination of microorganisms, and this experiment
was subsequently excluded from further evaluations. As
the centre points for the other process were performed in a
good way and as it was hard to make a new experiment

within a reasonable time frame from the other experiments,


it was decided not to redo this experiment, leaving the
investigation with two centre points from one process and
three centre points from the other process.
It was unknown whether the process would interact with
temperature or fat content during the planning phase of the
project. Therefore, it was considered interesting to make a
full factorial design to support interaction terms. This
opportunity to make trials with different processes under
the same conditions doesnt arise very often and it was
considered important to get the most information possible
about the varied factors in the design without too many
experiments. Four extra experiments, going from a fractional factorial design to a full factorial design, were
considered useful and allowed within the budget frame of
the project.
All responses were modelled using PLS, providing a
model with four principal components and fairly poor
quality, R2/Q2 0.74/0.28. The score plot for the first two
principal components shows a split among the experiments
according to the fat content of the milk (Fig. 3) and a split
among the experiments according to the process (Fig. 4).

Fig. 3. PLS model R2/Q2 0.74/0.28. Experiments labelled according to the fat content of the milk.

70

G. Wormbs et al. / Chemometrics and Intelligent Laboratory Systems 73 (2004) 6771

Fig. 4. PLS model R2/Q2 0.74/0.28. Experiments labelled according to the process.

After this preliminary evaluation, the experiments were


evaluated using MLR to obtain the best possible model for
each sensory attribute, as different sensory attributes
showed differing dependency on the factors from the PLS
model.
The full model with interaction terms was tested for all
responses using MLR. However, no interaction terms were
significant for any sensory attribute. Consequently, the
models were pruned down to linear models.
As the models for the different sensory attributes turned
out to be linear, it would have been possible to use a
fractional factorial design to make the experiments. During
the evaluation, this was tried and resulted in the same
models as with the full factorial design. However, as
mentioned earlier, it was decided during the planning phase
of the project to make the full factorial design using
knowledge at that time.
The model quality using MLR is not very good either
(R2/Q2 0.74/0.54 for chalk flavour, R2/Q2 0.81/0.54 for milk
flavour). In our experience, this model quality is typically
found using sensory attributes that are difficult to assess,

even when using a trained panel with expertise in the tested


product. Flavour attributes are more difficult to assess than
texture attributes, for example. In addition, chalk flavour in
all experiments was quite low (0 0.47 on a nine-point
scale) making it harder to assess that attribute.
Off-flavour of chalk shows significant influence by the
milk fat content (Fig. 5). High fat content lowers the offflavour. The fat content helps to conceal the off-flavour of
chalk. The process temperature also influences chalk flavour. A higher process temperature generates more chalk
flavour. It is a well-known fact that higher temperatures
generate more off-flavour in the milk [2], so this finding is
not surprising. The choice between the two tested processes
does not seem to influence the off-flavour of chalk. Looking
at the raw data, the level of the off-flavour of chalk in the
experiments is quite low. Even if fat content and temperature seem to influence the amount of off-flavour, the highest
levels measured in these experiments are still acceptable.
Another sensory attribute measured was milk flavour.
This attribute is only influenced by the fat content of the
milk (Fig. 6). Higher fat content generates more milk

Fig. 5. Influence of factors on chalk flavour, MLR model R2/Q2 0.72/0.54.

G. Wormbs et al. / Chemometrics and Intelligent Laboratory Systems 73 (2004) 6771

71

Fig. 6. Influence of factors on milk flavour, MLR model R2/Q2 0.81/0.54.

flavour. Neither the process nor the temperature significantly influences the milk flavour. There are indications regarding curvatures in the system, but as the main purpose of this
study was to investigate the influence of the processes on
various sensory attributes, this aspect was not investigated
further.
The process does not influence any sensory attribute
significantly. In this case, it was a desired result to see that
the choice between these processes does not influence the
off-flavours of milk. This can be seen as a robustness test,
where the aim is to find out that the varied factors do not
influence the system. In a robust system, it is hard to model
the investigated system with high model quality. Without
using Design of Experiment, it would have been difficult to
evaluate the influence of the different processes at different
temperatures on the sensory attributes measured.

4. Conclusion
The two processes for milk tested during these experiments show little, if any, differences concerning the evaluated sensory attributes. As there were no significant

influence from the processes regarding the sensory analysis


of milk, it was possible to make an investment decision
based on other criteria such as price, maintenance costs,
service agreement and so forth. Design of Experiment and
sensory analysis were useful tools for investigating the
processes prior to the investment and provided a good basis
for the decision making process.

References
[1] G. Bylund, Dairy Processing Handbook, Tetra Pak Processing Systems
AB, Lund, Sweden, 1995.
[2] P. Walstra, T.J. Geurts, A. Noomen, A. Jellema, M.A.J.S van Boekel,
Dairy Technology: Principles of Milk Properties and Processes, Marcel
Dekker, New York, USA, 1999.
[3] L. Eriksson, E. Johansson, N. Kettaneh-Wold, C. Wikstrom, S. Wold,
Design of Experiments: Principles and Applications, Umetrics AB,
Umea, Sweden, 2000.
[4] M.C. Gacula Jr. (Ed.), Descriptive Sensory Analysis in Practice, Food
and Nutrition Press, Connecticut, USA, 1, (1997) 1 146.
[5] M. Meilgaard, G.V. Civille, B.T. Carr, Sensory Evaluation Techniques,
2nd edition, CRC Press, Florida, USA, 1991.
[6] G.E.P. Box, W.G. Hunter, J.S. Hunter, Statistics for Experimenters,
Wiley, USA, 1978.

Você também pode gostar