The multi-objective data-driven approach: A route to drive performance optimization in the food industry

Although standardized, food processing is subject to many sources of variability resulting from compositional and structural variabilities of raw materials and/or ingredients, human perception and intervention in the process, capabilities of processing tools and their wear and tear, etc. . Altogether, they affect the reproducibility of final product characteristics representing deviations to standard, the production yield impacting the economic performance of the food manufacturing process, and many other performance indicators. They are grossly classified as economic, quality and environmental indicators and their simultaneous consideration can be used to define the overall performance of a manufacturing process. Optimizing the overall performance of food processing requires the use of multi-objective optimization methods. Multi-objective optimization methods include five steps: defining the objectives, modelling performance indicators, formulating the problem and constraints, solving the multi-objective problem, and finally identifying an ideal solution. The integration of data-driven approach, particularly machine learning, into the multi-objective optimization offers new perspectives for optimizing and controlling food processes. The potential of this approach is still underestimated by the food industry sector.