Impedance-Spectrum-Based Internal Short-Circuit Diagnosis of Two Lithium-Ion Battery Types via Transfer Learning with Sample Imbalance Consideration

Internal short circuit (ISC) diagnosis is vital for ensuring the safety and reliability of battery systems. Electro-chemical impedance spectroscopy (EIS), which provides comprehensive insights into internal battery dynamics, is a promising tool for ISC diagnosis. However, existing EIS-based methods often lack generalizability, as models trained on one chemistry cannot be readily applied to others. In particular, the imbalance between healthy and faulty samples further exacerbates the data acquisition burden required for both training reliable diagnostic models and achieving cross-chemistry transfer. To address these challenges, this work presents a preliminary exploration of a potential ISC diagnostic approach for cross-system application, using two representative lithium-ion battery chemistries, namely lithium iron phosphate (LFP) and nickel manganese cobalt (NMC). Diagnostic features are extracted from impedance spectra via the distribution of relaxation times (DRT), and a deep neural network is trained on a sufficiently large dataset from one chemistry. The model is then fine-tuned with limited data from another chemistry, enabling robust adaptation across material systems. To mitigate data imbalance, a variational autoencoder generates synthetic healthy samples, yielding a more balanced dataset. Experimental results indicate that the proposed method achieves a certain level of diagnostic accuracy and transferability between LFP and NMC cells, suggesting a possible pathway for EIS-based ISC diagnosis toward cross-system applications.