Deconvoluting silicon-graphite electrode degradation using acoustic emission and wavelet analysis

This study applies acoustic emission (AE) analysis to identify degradation mechanisms in silicon–graphite composite electrodes during electrochemical cycling. A Random Forest model, originally trained on completely unrelated graphite-based degradation data, was successfully transferred to the silicon-graphite system using wavelet-derived features. Our approach increased classification results by 25 % compared to conventional FFT-based analysis, with 65 % of AE events classified at a confidence above 75 %. The results show that the majority of detectable AE activity can be attributed to fractures of the silicon particles, particularly during first lithiation and subsequent delithiation. Only a minor proportion arises from gas evolution. Wavelet analysis revealed distinct spectral signatures for these mechanisms, confirming that we are able to separate the processes with our analysis. Further, the cross-compositional applicability of the model demonstrates the robustness of wavelet-based AE classification. These findings highlight the potential of AE sensing combined with machine learning as a non-invasive, interpretable diagnostic tool for detecting degradation in real time for many further material classes. Potentially, the tool can be integrated into battery management systems for predictive maintenance and health monitoring. • Wavelet-based AE features improve degradation classification results by 25 %. • Major contribution to AE activity by silicon particle fracture during lithiation/delithiation. • Trained model transfers reliably from graphite to silicon–graphite systems. • Undecided AE events arise from overlapping signals and system-specific noise limits. • Approach supports real-time, non-invasive monitoring of battery degradation.