In order to address the problem of mechanical degradation of filter group circuit breakers in UHVDC projects caused by the coupling of multiple factors such as mechanism jamming, lubrication failure, and micro-deformation of connecting rods, existing monitoring methods rely on single electrical features, making it difficult to effectively identify and decouple complex fault sources, seriously restricting equipment status perception and predictive maintenance capabilities. To this end, this paper proposes an intelligent diagnosis method that integrates multi-physics field mechanism modeling and multi-source heterogeneous sensing. First, an electromagnetic-mechanical-contact multi-field coupling simulation model is constructed to identify key observable features sensitive to various types of degradation. Second, four types of signals, namely operating current, vibration, acoustics, and displacement, are synchronously collected in a strong electromagnetic interference environment, and high-dimensional heterogeneous features are extracted by combining time-frequency analysis and graph convolutional networks. Finally, a multimodal fusion architecture based on a dual-branch attention mechanism is designed to achieve dynamic weighting and precise decoupling of complex degradation sources. Experimental results show that the method achieves a macro-average F1-score of 96.3% under 13 operating conditions and a triple-combination fault decoupling success rate of 89.3%. It also demonstrates excellent robustness in small sample sizes, noise interference, and historical field case backtesting. This method is beneficial to improving the state perception accuracy and fault warning capabilities of key switchgear in UHV converter stations. The research results provide strong technical support for building a safe, reliable, and intelligent new power system and have important engineering application value and social benefits.
Wang et al. (Fri,) studied this question.