Efficient Analysis of Interdependencies in Electrode Manufacturing Through Joint Application of Design of Experiments and Explainable Machine Learning

Abstract

Battery cell production is a key contributor to achieving a net‐zero future. A comprehensive understanding of the various process steps and their interdependencies is essential for speeding up the commercialization of newly developed materials and optimizing production processes. While several approaches have been employed to analyze and understand the complexity of the process chain and its interdependencies – ranging from expert‐ and simulation‐based to data‐driven – the latter holds significant potential for real‐time application. This is particularly relevant for inline process control and optimization. To streamline the development and implementation of data‐driven models, a holistic framework that encompasses all necessary steps – from identification of relevant parameters and generation of data to development of models – is imperative. This article aims to address this objective by presenting a comprehensive and systematic methodology, demonstrated for efficient cross‐process analysis in electrode manufacturing. Through the combined utilization of design of experiments methods, data‐driven models, and explainable machine learning methods, the interdependencies between production parameters and the physical, mechanical, and electrochemical characteristics of the electrodes were uncovered. These actionable insights are essential for enabling informed decision‐making, facilitating the selection of appropriate process parameters, and ultimately optimizing the production process.

Interdependencies in Electrode Manufacturing: A Comprehensive Study Based on Design Augmentation and Explainable Machine Learning

“…The details concerning the experimental setup, machinery, and materials align with those documented in Ref. [17] and are available in the supplementary materials.…”

“…The historical data consists of two datasets: the first one is based on the data published in Ref. [17] while the other one includes variations in mass loading, porosity, and the temperature of the first dryer. Within each dataset, the rest of the drying conditions-drying web speed and temperature of the other two dryers-remain consistent.…”

“…While ML has received significant attention over the last five years in the battery community, the potential of innovative XML methods remains largely untapped, particularly within the battery cell production field. [16] In the previous publication presented by Haghi et al, [17] an effort was made to address some of the outlined gaps. The study aimed for an effective analysis of interdependencies between the coating, drying, and calendering processes.…”

“…The results indicated that the temperature of the second dryer, particularly at lower mass loadings, had a negligible impact on both the adhesion and the rate capability of the battery. [17] A plausible rationale for this observation was based on the primary stages of the drying process, and the assumption that, within the analyzed ranges, the phase critical to the electrode quality predominantly occurred in the first dryer. To provide further clarity, a brief overview of the key stages of the drying process is presented below.…”

“…[18,19] Identifying the specific time window and the dryer at which the film shrinkage phase ends can be extremely beneficial, as it allows for optimal conditions during the pore emptying stage, leading to a more efficient drying process and improved electrode quality. [20] Building upon the findings of the previous study and the formulated assumption, [17] this article aims to present a comprehensive analysis by incorporating additional parameters involved in the drying process, specifically the temperature of the first dryer and the drying web speed. In this work, a systematic methodology is presented, emphasizing the critical evaluation of historical data, especially in the context of data consolidation.…”