关键词: 深度聚类, 小波去噪超图卷积编码器, 多传感器故障识别, 无监督学习

Abstract: To address the challenges of insufficient labeled data in real-world industrial scenarios and complex high-order heterogeneous relationships among multi-sensor data,a multi-sensor fault identification method based on wavelet denoising hypergraph deep clustering network was proposed.The K-nearest neighbor algorithm was used to construct a hypergraph for each sample composed of multi-sensor data.The high-order heterogeneous relationships among sensors were modeled by the hypergraph.Then,the wavelet denoising hypergraph convolutional encoder based on the discrete hypergraph wavelet framework was designed to extract and fuse high-frequency detail components and low-frequency approximation components at multiple scales.Finally,through joint optimization of clustering loss and reconstruction loss,the deep fault features and fault cluster centroids were iteratively updated,achieving deep fault pattern clustering.To validate the effectiveness of the proposed method,extensive experiments were conducted on two public datasets.The results demonstrated that the proposed method exhibited significant superiority in unsupervised fault identification compared to benchmark methods,and demonstrated robust noise resistance.

Key words: deep clustering, wavelet denoising hypergraph convolutional encoder, multi-sensor fault identification, unsupervised learning