Multi-Information Fusion Diagnosis of High-Voltage Circuit Breakers via Newton–Raphson Optimised Transformer and Meta-Transfer Learning

Mechanical fault diagnosis of high-voltage circuit breakers (HVCBs) remains challenging due to the complex and nonstationary nature of vibration signals, scarcity of fault samples, and the limited feature-extraction capacity of existing few-shot learning models. To address these challenges, this paper proposes a multi-information fusion diagnostic framework that integrates a Newton-Raphson optimised Transformer with meta-transfer learning (MTL). Specifically, entropy-weighted fusion is introduced to suppress conflicting channels and aggregate multi-directional vibration measurements into an informative representation. To improve training stability and reduce sensitivity to manual trial-and-error under limited data, a Newton–Raphson-based optimiser is employed offline to select key Transformer hyperparameters. For data-scarce and cross-scenario diagnosis, a meta-transfer learning scheme with a lightweight scale-shift adaptation module enables fast adaptation while mitigating overfitting. The proposed framework is validated on a self-developed multimodal vibration acquisition platform and compared with representative baselines. Experimental results show that the proposed approach achieves the highest diagnostic accuracy (98.45%) and F1-score (98.26%) under 5-way 5-shot settings, outperforming conventional baselines by 2.0%–5.5%. The method exhibits strong interpretability and adaptability to variable operating conditions, providing a reliable solution for intelligent mechanical fault diagnosis of HVCBs.