D18.9 - Convolutional Neural Network-Based Equivalent Circuit Model Classification from EIS Data for Battery Degradation Prediction

This paper presents a deep learning-based framework for the automated extraction and classification of equivalent circuit models (ECMs) from electrochemical impedance spectroscopy (EIS) data, aiming to enhance lithium-ion battery (LIB) diagnostics. EIS provides detailed insights into internal battery mechanisms, such as charge transfer, diffusion, and double-layer capacitance, by analyzing frequency-dependent impedance responses. Traditional interpretation methods are labour-intensive and limited in scalability. To address this, a one-dimensional convolutional neural network (1D-CNN) is employed to classify EIS spectra into four distinct ECM classes using features derived from real and synthetic datasets. The model architecture incorporates hierarchical convolutional layers, dropout, batch normalization, and global average pooling, achieving a classification accuracy of 98.25% on the test set. The predictions align closely with Nyquist plot characteristics of each ECM, validating the interpretability and robustness of the model.