Intelligent Fault Diagnosis for Battery Packs Using FCE-GCN: Addressing the Challenges of Class Imbalance and Feature Similarity

Lithium-ion batteries (LIBs) play an essential role in modern energy systems such as electric vehicles and renewable energy applications, where their long-term stability and reliability are crucial for ensuring system safety. However, traditional fault diagnosis methods often struggle to effectively address the issues of class imbalance and feature similarity in complex real-world scenarios. This paper proposes an innovative fault diagnosis method for battery packs based on Feature Contrastive Encoding and Graph Convolutional Networks (FCE-GCN), aiming to enhance the recognition of minority fault samples and overall classification performance. First, the feature contrastive encoder enhances the separability between healthy and faulty samples in the feature space by maximizing inter-class feature differences. Then, the graph convolutional network model, constructed based on the encoded features, further improves classification performance by capturing structural information among samples. Validation results on real-world battery pack data demonstrate that FCE-GCN achieves an accuracy of 94.5% when handling imbalanced data and feature similarity challenges, significantly improving the accuracy of fault diagnosis.