Hierarchical diagnosis of multiple fault types for lithium-ion batteries based on fine-tuned large language model

Abstract Fault diagnosis in lithium-ion batteries is of paramount importance for ensuring the operational safety of electric vehicles and energy storage systems. However, it faces significant challenges due to the diversity of fault types and the subtle nature of early-stage faults. Conventional diagnosis methods can identify a limited range of fault types and are prone to fault omission in complex operating conditions, rendering them inadequate for comprehensive safety assurance in practical applications. To address these challenges, this paper proposes a multi-fault diagnosis method based on fine-tuned large language models (LLMs). First, a distortion degree pair (DDP) feature is introduced, enabling effective downsampling of battery signals to less than 1/25 of their original length while preserving critical fault-related information. Subsequently, the processed data is input into a 7-billion-parameter LLM along with meticulously crafted prompts. The Low-Rank Adaptation (LoRA) technique is employed to fine-tune the LLM in a two-stage process, allowing it to accurately detect the fault and identify the fault types. Experimental results demonstrate that the proposed method can effectively diagnose five types common fault under complex operating conditions, achieving accuracy of 97.3% for fault classification.