Internal short-circuit diagnosis for lithium-ion batteries using autoencoder with temporal convolutional network and self-attention mechanism

The internal short-circuit (ISC) faults in lithium-ion batteries pose a significant threat to battery safety and can lead to thermal runaway. Detecting ISC faults is critical for the safe operation of batteries. This study proposes an ISC fault diagnosis method based on an autoencoder (AE). To address the challenge of limited high-quality ISC fault data, a fault detection strategy utilizing AE reconstruction residuals is introduced, with the AE trained exclusively on battery voltage data from normal modules. ISC faults are identified by evaluating the mean square error between the reconstructed and measured data. Notably, the AE incorporates a temporal convolutional neural network and a self-attention block to enhance reconstruction performance, thereby improving fault detection speed. Experimental results demonstrate that the proposed method effectively diagnoses ISC faults of varying severity. Furthermore, its robustness is validated through experiments conducted under various conditions, including cell inconsistencies, different battery types, and diverse operating conditions. This method is effective in identifying and diagnosing ISC faults, thereby benefiting the safety and reliability of batteries.