Data-Driven Approach to Analyze Interdependencies between Electrode Manufacturing Parameters and Electrochemical Performance of the Lithium-Ion Battery Cell

The lithium-ion battery cell production is a complex process chain with a high number of process steps and manifold interdependencies. The majority of the properties determining the electrochemical performance of the battery cell are established during electrode manufacturing. An in-depth understanding of electrode manufacturing and its influence on cell performance can pave the way toward high-quality, cost-efficient cell production. Considering the complexity of the process chain and the high share of material costs in battery production, the classic one-factor-at-a-time experimentation is deemed inadequate and expensive. Machine learning (ML) approaches have been proven suitable for analyzing complex process chains. Design of experiment (DoE) methods can be used to generate a sufficient informative database for ML techniques. This study was based on the joint application of DoE and ML methods in electrode manufacturing. For this purpose, in the first step, the response surface methodology (RSM) was used to conduct experiments and analyze the effect of mass loading, drying temperature, and porosity on cell performance. The study was focused on graphite anodes produced at a pilot scale. The electrochemical performance of electrodes was assessed on coin cells at discharge C-rates from C/10 to 5C. Additionally, symmetric coin cells were used to analyze the electrochemical impedance spectra. Supervised machine learning models were adopted to predict the cell characteristics based on the electrode properties. Additionally, a factor analysis reflecting the contribution and importance of each factor on the final cell properties was conducted. The efficient experiment-based data generation using RSM combined with the applied ML techniques can support practitioners towards smart quality-oriented battery cell production.