Hypergraph construction based on Markov chain-assisted multi-source spatial-temporal data fusion for few-shot fault diagnosis of mechanical–hydraulic equipment

Abstract Mechanical–hydraulic equipment is generally characterized by nonlinearity, structural complexity and scarcity of fault samples, leading to poor performance of traditional fault diagnosis. With advancements in sensor technology, monitoring systems now collect time-dependent multivariate data, and effectively mining data correlations for intelligent diagnosis remains a challenge. This paper proposes a multi-source data fusion fault diagnosis method that accounts for temporal relationships, addressing issues of fault data scarcity and multi-source timing signal modeling in small sample scenarios. First, a simulation model of the device was constructed, and its operational reliability was validated through experiments. Second, a hypergraph structure for multi-channel data fusion was developed, leveraging Markov chain optimization to model higher-order dependencies between sensor data and extract fault information at a unified time scale. An efficient fault classification model was then developed based on a Long–Short-Term-Memory-Self-Attention neural network. Finally, the high accuracy of the proposed method in the field of mechanical hydraulic equipment fault diagnosis is verified through a subsea control module hydraulic system fault test, and a Case Western Reserve University (CWRU) bearing fault dataset is introduced to further verify the wide applicability of the proposed method across different signal types.