PMSLM Demagnetization Fault Detection Based on Multisensor Signal Fusion and Enhanced Graph Neural Network

An innovative diagnosis method for detecting demagnetization faults (DFs) in permanent magnet synchronous linear motors (PMSLMs) is proposed in this study. The developed approach leverages the combined strengths of the external three-line measurement (ETM) method and the adaptive edge coefficient graph attention network (AECGAT). The approach begins with the deployment of a tunnel magnetoresistance signal acquisition array to collect external stray magnetic flux signals (ESMFSs) for fault diagnosis. Compared with other signal acquisition strategies, the multisensor array offers advantages, including enhanced spatial resolution and the ability to capture comprehensive magnetic field distribution characteristics. Subsequently, the demagnetization three-line graph (DTG) is utilized to map the three-line signals and effectively capture the spatial relationships among sensors. The adaptive edge coefficient (AEC) prior mechanism is integrated into graph attention network (GAT) to further enhance the diagnostic capabilities. The AEC prior mechanism can reduce redundant connections and increase sparsity in the graph structure, thus improving the model’s efficiency and performance. Experimental validation on simulation reaches 98.75%, and prototype PMSLMs achieves 97.50% classification accuracy. The proposed method offers a highly accurate and reduces dependence on internal sensors, lowering maintenance costs and operational downtime.