Deep-Compact-Clustering Based Anomaly Detection Applied to Electromechanical Industrial Systems.

Area of Science:

• Industrial Engineering
• Machine Learning
• Fault Diagnosis

Background:

• Industrial growth necessitates reliable machinery, leading to complex systems.
• Automatic detection of unknown machinery faults is challenging.
• Existing anomaly detection methods struggle with complex industrial systems.

Purpose of the Study:

• To develop a novel fault diagnosis methodology for anomaly detection.
• To present an unsupervised anomaly detection framework, DAECC-OC-SVM.
• To improve anomaly detection performance using deep neural networks.

Main Methods:

• Combines a deep-autoencoder with a clustering compact model.
• Integrates a one-class support-vector-machine for outlier detection.
• Applies the methodology to rolling bearing and multi-fault experimental test benches.

Main Results:

• The DAECC-OC-SVM methodology accurately detects unknown defects.
• The proposed method outperforms existing state-of-the-art anomaly detection techniques.
• Demonstrated effectiveness on both single and multiple fault scenarios.

Conclusions:

• The DAECC-OC-SVM framework offers a robust solution for anomaly detection in complex machinery.
• This unsupervised approach enhances the reliability and productivity of industrial systems.
• The methodology shows significant potential for real-world industrial applications.