Investigating the Construction, Training, and Verification Methods of k-Means Clustering Fault Recognition Model for Rotating Machinery

Considerable studies have been carried out in recent years regarding fault diagnosis and prediction for the rotating machinery in industrial plants. However, few works present the use of clustering approaches applied to time series to diagnose machine faults. With the increasing practical requirement of safety, reliability, availability and maintainability of machinery running, predictive maintenance based on the technologies of fault diagnosis and prediction has also received significant attention in recent years. In the present study, under Cyber-physical systems (CPS) condition, k-means clustering analysis based on the fault case big data machine learning is applied to investigate the fault identification of the rotating machinery without external expert support. K-means cluster-based fault identification model, which includes the k-means cluster analysis module, fault mode - fault cluster centroid knowledge base module and fault identification module has been constructed. Moreover, the fault feature extraction and fault eigenvectors screening are studied in detail. The vibration data of surge, rubbing, misalignment and normal status of the centrifugal compressor in industrial plants are utilized to train and verify the effectiveness of the k-means cluster fault recognition model. The obtained result shows that recognition accuracy rates of the surge, rubbing and misalignment faults reach 94%, 100% and 80%, respectively. However, the effectiveness of the cluster analysis of vibration data for five or more operating states should be studied in the future.