"Material for Model-Enabled Asymmetric Autoencoder for Fault Detection in Li-Ion Battery Packs"

"Supplementary materials for the paper: \u201cModel-Enabled Asymmetric Autoencoder for Fault Detection in Li-Ion Battery Packs\u201d Abstract\u2014 Standard autoencoders, despite their potential for unsupervised fault detection in li-ion battery packs, struggle to capture early fault characteristics due to limited latent representations. In this paper, we propose a model-enabled asymmetric autoencoder that enhances feature extraction by aligning a subset of latent variables with model parameters and states. First, an asymmetric encoder-decoder architecture is designed to prioritize the high-fidelity reconstruction of cell voltage signals. The autoencoder is trained using normal data to learn health\u0002related features. Second, a quantitative fault metric is derived from the reconstruction error and cell voltage deviation. The metric is evaluated using the local outlier factor algorithm to detect anomalies. Finally, the proposed method is validated on a multi-year dataset from commercial battery packs, demonstrating effective fault identification and providing early warnings prior to failures."