Data Mining Approach Using Cluster Analysis and Decision Trees for Optimizing Electrode Paste Quality in Lithium-Ion Battery Production

Lithium-ion batteries are crucial for the energy transition, especially for emission reduction in the automotive sector and energy storage solutions. Therefore, an efficient, sustainable, and economic battery cell production is needed. To ensure optimal cell quality and enable quality control, the highly complex inter-dependencies in the battery cell production must be understood and the potential for improvement must be discovered. In the context of complex processes and data structures, data mining techniques have been established in the industry. Here, a data mining approach consisting of a Cluster Analysis and Decision Trees is applied to a data set obtained by measuring process and quality data during the electrode paste mixing process. The aim is to analyse the influence of manufacturing parameters on selected quality characteristics (i.e. viscosity and particle size distribution of the electrode paste). The Cluster Analysis is used to create classes of similar quality characteristics. Decision trees map the dependencies between the formed clusters and process parameters. The results show that the density-based clustering algorithms achieve the best clustering outcomes for product quality. Based on the feature importance of different Decision Trees, the most important process parameters that affect the quality classes are determined. This approach is transferable to other process steps and offers the capability to analyse a series of process steps and the final product quality.