# Modelling techniques

The application of statistical methods of analysis is common in many areas of food manufacture. They can be used in problem solving and quality optimisation, though in the manufacturing environment modelling methods tend to be confined to the plotting of trends using simple graphs as discussed in the example above for a laminated product. More sophisticated statistical and modelling techniques can play their part in helping to build up the information base on what the critical ingredient and process factors are which determine changes in product quality. Once identified these critical factors can be logged and matched with problems when they occur.

To develop such predictive models it will be necessary to carry out experiments in the test bakery or trials on the plant. While trials on the plant are preferred they can be wasteful of raw materials, energy and time so that the most common practice is to carry out evaluations in the test bakery and 'translate' the results to the plant. It is very important to establish any clear changes that are relevant when translating test bakery results to a plant environment. A simple example encountered by the authors was the development of a sponge cake recipe in a test bakery using a planetary-style mixer while the plant used a continuous mixer to prepare the same recipe batter. In this case it is necessary to remember that less carbon dioxide gas will be lost during continuous mixing than with a planetary mixer so that baking powder levels should be adjusted downwards to compensate for this difference. A typical adjustment would be to reduce the baking powder level for a continuous mixer to be about 75% of that used on a planetary mixer in order to keep sponge cake volume constant (Cauvain and Cyster, 1996).

There are a number of examples of modelling techniques which might be applied to bakery products. Street (1991) provides a review of suitable techniques that may be applied to baked products and there are many examples in the scientific and technical literature. The concept behind the development of such mathematical models is that a relatively limited number of experiments may be used to build models that can be used to predict changes in bakery product quality as a consequence of changes in combinations of ingredients and processes.

Once a predictive model has been established then the information can be used for problem solving. For example, suppose that we show by experimentation how loaf volume varies as a result of an interaction between the level of ascorbic acid in the dough and mixing time. At some later stage we may encounter a problem with low bread volume and then we would be able to use the output from our model to help decide whether the problem was associated with the level of added ascorbic acid or mixing time, or both. Furthermore we might use our model to show which changes were most likely to restore our bread volume to its original level.

Baking is a complex food process with many ingredient and process interactions. These interactions lead to complicated models that are often difficult to apply. For example, for a given set of mixing conditions we would observe that bread volume increases with increasing levels of ascorbic acid reaching a maximum and thereafter there will be little change in volume for increasing additions of ascorbic acid. This occurs because the oxidation effect of ascorbic acid is limited by the availability of oxygen from the air incorporated during dough mixing. The availability of oxygen is affected by yeast activity, so that yeast level becomes an influencing factor. Both yeast and ascorbic acid activity are temperature sensitive and proceed at a greater rate when the temperature increases. Dough temperature is a function in part of ingredient temperatures and in part the energy imparted to the dough during mixing. Energy transfer in turn is related to the mixing time. So, too, is gas occlusion to a lesser degree because during mixing an equilibrium point is reached when the entrainment process is balanced by the disentrainment process. This equilibrium may occur before the end of the mixing time.

So for the given example while we set out to study the effects of the level of ascorbic acid and mixing time we must also ensure that we measure:

• ingredient temperatures;

• final dough temperature;

• gas occlusion in the dough;

• energy transferred to the dough.

This is necessary because we cannot independently control some of the properties concerned, e.g. mixing time, energy and dough temperature. Whenever we do work during mixing we must expect there to be a temperature rise. This relationship also holds true if a water jacket has been fitted to the mixer and in this case we must remember that the water temperature in the jacket will also rise.

There tends to be greater variability in product quality for products manufactured on a plant than one sees in many test bakery environments. This process 'noise' in the data can mask some of the critical issues that control product quality and therefore weaken the value of any models which have been developed. There are a number of statistical techniques that can be used to help separate such noise from underlying effects, trends and relationships. In many manufacturing processes the specified product characteristics can be achieved by many different combinations of formulation and process conditions. Taguchi methods use experiments to search systematically and efficiently for combina tions of 'control' factors that minimise product variability in the face of variations in ' noise factors' such as ambient temperature. Taguchi methodology has been applied to the manufacture of bakery products, in particular in a study of the factors that affect the quality of puff pastry (DTI, 1993).

Continue reading here: The information sources