Interpretable Sensitivity Analysis and Electrode Porosity Classification for Li-ion Battery Smart Manufacturing

Lithium-ion batteries have become one of the most promising sources for accelerating the development of sustainable energy, where effective cell manufacturing plays a direct role in determining battery qualities. Due to the highly complicated process and strongly coupled interdependencies of battery manufacturing, a data-driven approach that can evaluate the sensitivity of manufacturing parameters and provide the effective classification is urgently required. This paper proposes a boosting tree-based ensemble machine learning framework to analyze and predict how the battery electrode porosity varies with respect to the key parameters of both mixing and coating stages for the first time. Three boosting models including the AdaBoost, LPBoost, and TotalBoost are established and compared. Illustrative results demonstrate that the proposed ensemble machine learning framework is able to not only give effective quantification of both importance and correlations of parameters of interest but also provide satisfactory early-stage prediction. These kinds of information could benefit the monitoring and analysis of battery manufacturing chain, further help to produce high quality batteries for wider sustainable energy applications.