Dynamic Multi-Scale Hypergraph Wavelet Neural Network for Mechanical Fault Diagnosis of Consumer Technology

With the increasing variety and widespread adoption of consumer electronic products, their operational failures have an increasingly significant impact on daily life. Traditional fault diagnosis methods suffer from limited applicability, poor robustness to noise, and inadequate capability to extract weak features from compact devices. To address these challenges, this paper proposes Dynamic Multi-Scale Hypergraph Wavelet Neural Network (DMHWN) for mechanical fault diagnosis in consumer electronics. Specifically, dynamic hypergraph learning module is designed to enhance the adaptability of the hypergraph structure to variations in measured signals and noise. In addition, hybrid hypergraph structure is constructed by integrating static and dynamic hypergraph, which improves the model’s ability to capture subtle correlations. By combining spectral graph wavelet operators with scaling function, multi-scale hypergraph wavelet convolutional layer is developed to strengthen the models capacity in representing both global and local fine-grained fault features. Finally, the proposed method is validated on two fan fault datasets with different signal types for consumer electronics. Experimental results demonstrate that DMHWN achieves excellent performance in both fault identification accuracy and noise robustness, making it well-suited for mechanical fault diagnosis of consumer electronics under complex working conditions.