A Conceptual Framework for Production Process Parameter Optimization with Modular Hybrid Simulations.

Optimizing the parameters of a manufacturing process is a time-consuming task requiring a series of experiments involving different parameter combinations. To alleviate this difficulty, we propose a conceptual framework for data-driven parameter optimization in production processes, which allows for virtual parameter tuning. To provide an insight into the practical application of our general method, we additionally explore its use on the example of lithium-ion battery (LIB) production. Our framework consists of two components: a modular hybrid simulation and an optimization tool. In the first component, the place of traditional process models is taken by a set of machine learning (ML) models which aim to imitate the behaviour of each process step. These individual models are trained on collected process data and connected in a modular simulation framework. While the resulting system already allows for manual exploration of parameter combinations, the introduction of an optimization tool unlocks further benefits. The proposed modular approach is independent of the production process type, therefore it can be applied to various manufacturing fields.