ProADD: Proactive battery anomaly dual detection leveraging denoising convolutional autoencoder and incremental voltage analysis

The use of lithium-ion batteries is increasing fast in many fields such as electric vehicles (EVs) and energy storage system (ESS). However, the number of accidents caused by thermal runaway of lithium-ion batteries is also increasing. Hence, it is critical to diagnose lithium-ion batteries proactively for safe operation. This paper considers both electrochemical model and deep learning model to capture the intrinsic characteristics of battery and diagnose its state from complementary perspectives. First, the denoising autoencoder (DAE) is leveraged to detect outliers in latent space clustering. Second, the traditional incremental capacity analysis (ICA) is revisited and incremental voltage analysis (IVA) is proposed to make it suitable for real-time ESS operation. Then, a method is proposed that jointly considers the DAE error and the IVA peak to proactively detect anomaly battery modules of ESS. Specifically, one-class support vector machine (OCSVM) is leveraged as well as the transformed Z-score. Our results confirm that the proposed framework named ProADD clearly identifies and quantifies anomaly modules, which provides a guideline for safe ESS operation in real fields.