A Fault-Detection System Approach for the Optimization of Warship Equipment Replacement Parts Based on Operation...

Area of Science:

• Naval Systems Engineering
• Information Systems
• Machine Learning for Anomaly Detection

Background:

• Complex naval systems require robust engineering and specialized information systems for logistical support and operational status monitoring.
• Current information systems for warship maintenance may necessitate highly skilled operators with deep system knowledge.
• Failure detection systems using intelligent techniques offer a promising approach to identify anomalies without requiring expert input.

Purpose of the Study:

• To investigate the application of one-class anomaly detection techniques for complex naval information systems.
• To compare the effectiveness of statistical models, geometric boundaries, and dimensional reduction for anomaly detection in warship subsystems.
• To lay the groundwork for applying these techniques to entire warship systems.

Main Methods:

• Comparative analysis of one-class classification techniques.
• Application of statistical models for anomaly detection.
• Implementation of geometric boundary methods for anomaly identification.
• Utilizing dimensional reduction techniques for system behavior analysis.

Main Results:

• Demonstrated the feasibility of using one-class techniques for anomaly detection in specific warship subsystems.
• Provided a comparative overview of the performance of different anomaly detection methods in a naval context.
• Identified potential for reducing reliance on expert knowledge for system monitoring.

Conclusions:

• One-class anomaly detection techniques show significant promise for enhancing the maintenance and provisioning of naval vessels.
• The comparative study provides valuable insights for selecting appropriate methods for warship anomaly detection.
• Future work can focus on scaling these techniques for comprehensive ship-wide anomaly detection.