Unsupervised multimodal electrochemical-acoustic autoencoders for in-operando structural anomaly tracking in Lithium-polymer batteries

Advanced in-operando diagnostic tracking is critical to prevent catastrophic battery failures and to ensure the safety of high-energy-density lithium-polymer battery packs. Traditional machine learning diagnostics rely heavily on supervised classification or regression models, which require substantial amount of destructive failure datasets that are challenging to acquire in real-world applications. Furthermore, regular data normalization pipelines frequently obscure subtle electrochemical shifts, masking critical variations such as loop expansions in impedance data. To address these limitations, this study introduces an unsupervised multi-input deep convolutional autoencoder framework that detects out-of-distribution battery degradation by fusing concurrent acoustic and electrochemical data streams. This dual-branch architecture is trained on pristine operating data, and builds a shared latent space linking ultrasonic time-of-flight (ToF) waveforms with log-resampled electrochemical impedance spectroscopy (EIS) data. The framework demonstrates robust diagnostic sensitivity when validated against simulated structural degradation tracks. Macroscopic delamination and internal gas pockets trigger a distinct reconstruction error spike driven by wave scattering and phase velocity attenuation at internal solid-gas interfaces. Simultaneously, the model registers a localized error deviation that captures the geometric expansion of mid-frequency charge-transfer resistance arcs, tracing micro-scale binder contact degradation without requiring labeled anomaly data. Furthermore, this foundational non-destructive diagnostic approach offers scalable utility potential for the real-world scenarios where systems often face challenging environmental conditions like humidity and heat. By integrating early anomaly detection into intelligent Battery Management Systems (BMS), this physics-aligned multimodal strategy establishes an objective baseline to mitigate thermal runaway risks and secure demanding transport or renewable energy grids.