Using Monte Carlo simulations to formulate analysis strategies for food processes.

Monte Carlo simulations and Design of Experiments (DoE) have been investigated as tools for developing applications, alongside multivariate calibration models for estimating quality attributes. Due to several constraints, most studies tend to use small data sets which do not span the process profile space, with batches sharing similar trajectories for all the experiments used to build and test the models. This leads to apparently promising models yielding potential pitfalls in their rigorous assessment. We attempt to mitigate this issue by using Monte Carlo simulation as a strategy for improving the robustness of our model performance.
In most cases in industrial plants, it is costly or impractical to do all the experiments in the DoE. In order to capture process trends and sample variations, we employ Monte Carlo simulation based on physical and chemical principles for the interpolation /extrapolation of a limited set of experiments from the DoE space, hence completing the required DoE. We discuss two case studies, dough blending for baking and wort production for brewing. The goal is to access quality attributes, namely the completeness of the dough mixing process, and the fermentable sugar content, respectively. The simulated data is processed using standard chemometric techniques, in order to assess their performance as predictors of real experiments.
In this work calibration models for estimating the amount of fermentable sugars during the course of the wort production, and the dough mixing process, have been developed using NIR. In the future we will also use multi-block data fusion with UV-Vis reflectance spectra and temperature profiles to maximize model performance.