What Can Be Gained via Model Order Reduction of Bread-Baking Simulations?

Bread baking is among the oldest and most important processes in the food industry. The transformation of dough into the final product involves complex coupled chemical and physical phenomena, which make the baking process challenging to accurately simulate. At the same time, recent increases in energy prices have forced industrial bakeries to seek opportunities to reduce energy consumption. In the long term, our aim is to develop a methodology for simulation-based optimization of energy use in industrial bread baking. As a first step, we have already developed a high-fidelity numerical model that describes the transformation of the dough during baking. However, the computational cost of this model currently limits its direct use in optimization workflows. To overcome this limitation, model order reduction is employed. Specifically, an a posteriori approach is adopted, in which the high-fidelity model is evaluated for a selected set of parameter combinations. Proper orthogonal decomposition is then combined with artificial neural networks to construct a reduced-order model with significantly lower computational cost. The practical utility of the reduced-order model is demonstrated through its application to the identification of otherwise difficult-to-detect simulation parameters.