Fault Diagnosis Method of Rotating Machinery Based on Graph Convolutional Network

In industrial production, rotating machinery is essential and malfunctions in crucial parts like rotors, bearings and gears can result in significant equipment damage and safety hazards. Early detection depends on intelligent fault diagnosis, however existing technologies have problems with deep fault feature extraction, multi-source heterogeneous data fusion, excessive background noise in sensor data and a lack of assessment techniques for sensor architecture. This chapter investigates the use of Graph Convolutional Networks (GCN) for rotating equipment failure diagnostics in order to overcome these issues. In particular, (1) vibration signals with Euclidean structure are first transformed into non-Euclidean graph data using algorithms such as K-nearest neighbor and GCN is then applied to mine deeper fault information and establish correlations between sensors; (2) same-dimensional features are extracted from various sensor types and fused into a unified graph structure, resolving the heterogeneity problem before being input into the GCN model for diagnosis; experimental studies validate both approaches: a helical gearbox setup combines time series and image data to demonstrate multi-source data fusion and a rotor bearing test rig is used to classify 11 different health conditions using vibration data. The suggested multi-source heterogeneous fusion approach maintained 87.6% accuracy under noise and reached 99.92% diagnostic accuracy. Lower accuracies of 75.92% and 77.33% were obtained when time series or infrared images were the only data used. After 15 iterations, the proposed method achieved a standard deviation of 0.65%, an average accuracy of 99.00% and a maximum accuracy of 100%. Significantly better diagnostic performance was demonstrated by the suggested method in comparison to CNN (82.84%) and LSTM (82.00%). Overall, the results demonstrate that GCN effectively improves intelligent fault diagnosis performance rotating machinery.