Deep-learning based parameter identification enables rationalization of battery material evolution in complex electrochemical systems

• Published in: Journal of computational science Vol. 66; p. 101900
• Main Authors: Sgura, Ivonne, Mainetti, Luca, Negro, Francesco, Quarta, Maria Grazia, Bozzini, Benedetto
• Format: Journal Article
• Language: English
• Published: Elsevier B.V 01.01.2023
• Subjects: Battery modeling; Convolutional Neural Networks (CNN); Deep Learning; Parameter estimation; Reaction-diffusion PDEs; Turing patterns; Reaction-diffusion PDEs; Parameter estimation; Convolutional Neural Networks (CNN); Battery modeling; Deep Learning; Turing patterns
• ISSN: 1877-7503; 1877-7511
• DOI: 10.1016/j.jocs.2022.101900

High-energy density metal anodes are a key solution for next-generation mobility batteries, but the difficulty of studying materials in real-life battery context leads to a methodological gap between theory and experiments, translating into poor device control. Imaging and spectroscopy are the ultimate tools for knowledge-based battery studies, but the capability of quantitatively linking the electrical device response to the material evolution, is essentially missing, at the moment. High-throughput, Deep-Learning based parameter identification for morphochemical PDE modelling, can enable this link, in principle allowing to extract the key information from hyperspectral imaging tools and to feed it into next-generation battery management systems: this work is the first step of this process. Here we have employed a CNN, trained with the solutions of an electrochemical phase formation model we recently developed, to carry out three enabling tasks: (i) automatic partitioning of the parameter space, according to the types of patterns generated by the model; (ii) assignment of experimental patterns, derived by imaging of from real electrodes to the class patterns and (iii) identification of the model parameters for experimental electrode images. •CNN enables rationalization of experimental battery data with electrochemical model.•CNN trained with model solutions can recognize and classify experimental battery data.•Deep-learning based parameter identification of electrode shape-change with PDE model.•Automatic partitioning of model parameter space with respect to electrode shape types.•DL parameter identification is robust with respect to map rotations and scale-changes.