Battery production design using multi-output machine learning models

Abstract

The lithium-ion battery (LiB) is a prominent energy storage technology playing an important role in the future of e-mobility and the transformation of the energy sector. However, LiB cell manufacturing has still high production costs and a high environmental impact, due to costly materials, high process fluctuations with high scrap rates, and high energy demands. A lack of a profound knowledge of LiB cell production processes and their influence on the quality and the performance of the LiB cells makes it difficult to plan, control and execute the production. Therefore, a systematic approach is necessary to establish an in-depth understanding of the interlinkage of processes and products’ quality and performance. This paper presents a multi-output approach for a battery production design, based on data-driven models predicting final product properties from intermediate product features. The given concept shows how the approach can be deployed within the framework of a cyber-physical production system for continuous improvement of the underlying data-driven model and decision support in production.