Unsupervised anomaly detection for critical components in industrial manufacturing via multi-stage clustering

To address the challenges of confidentiality in industrial field data and the scarcity of labeled samples, this study proposes an unsupervised fault diagnosis framework that achieves autonomous discovery of fault modes through multi-stage feature optimization and robust clustering analysis. The method consists of four core components: simulation dataset construction, feature extraction and selection, outlier anomaly detection, and clustering-based common anomaly detection. The generalization ability of the proposed method has been validated through CWRU cross-load testing, which accurately identifies 10 fault states with an accuracy of 95.2%. The collaborative design of feature engineering and anomaly cleaning effectively enhances clustering stability and the ability to distinguish complex faults. The research provides an interpretable and low-cost solution for intelligent diagnosis in privacy-sensitive industrial environments, offering an innovative technological pathway for health management in intelligent manufacturing systems.