Multi-objective knowledge-based strategy for process parameter optimization in Micro-fluidic chip production

We present an effective optimization strategy for industrial batch processes that is centered around two computational intelligence methods: linear and non-linear predictive mappings (surrogate models) for quality control (QC) indicators and state-of-the-art multi-objective evolutionary algorithms (MOEAs). The proposed construction methodology of the linear and neural network-based mappings integrates implicit expertbased knowledge with a new data-driven sample selection strategy that hybridizes several design of experiments paradigms. Using a case study concerning the production of micro-fluidic chips and 26 QC indicators, we demonstrate how incorporating modeling decisions like cross-validation stability analyses and objective clustering into our optimization strategy enables the discovery of well-performing surrogate models that can guide MOEAs towards high-quality Pareto non-dominated solutions