Manufacturing improvement capabilities of machine learning algorithms: Evaluation using an electrode-separator compound handling demonstrator

Abstract

Machine learning based analyses of the production data promise direct improvements of manufacturing conditions and offer a way to justify involved investment costs for the data provision. In the current literature, explanatory and predictive capabilities of machine learning algorithms are described. The explanatory capabilities focus on root cause and structural analyses, while the predictive capabilities concentrate on real-time predictions. The aim of this contribution is to evaluate the different capabilities of data analyses using a handling device responsible for positioning and transporting electrodes for the lithium-ion cell assembly. The evaluation shows that the root cause analyses offer tools to improve especially simple use cases. The real-time prediction promises a reduction of production costs by classifying the final position of the electrodes and enables an early outlier detection. Additionally, an approach is presented to detect relevant production parameters of the resulting quality of produced cells.