Optimization of complex food formulations using robotics and active learning

Abstract

The creation and optimization of formulated products represents a major challenge for science and industry in the food sector. Thereby, different raw materials are mixed and processed to meet predefined and often competing targets. During this procedure, applied experimental campaigns not only require expert knowledge, but, depending on the complexity, also cause a high consumption of resources and costs. In the present work, a fully automized milli-fluidic laboratory driven by the Thomsen sampling efficient multiobjective optimization (TSEMO) algorithm was designed. The methodology was successfully applied to optimize the aggregation process of a liquid formulation consisting of whey protein isolate, NaCl and CaCl2. Within 48 h 90 experiments could be performed without human intervention, resulting in a Pareto front formed by a set of 18 optimal recipes. It is thus a successful demonstration of an actively learning, self-driving food formulation process.