Real-Time Predictive Condition Monitoring Using Multivariate Data.

Area of Science:

• Engineering
• Data Science
• Machine Learning

Background:

• Effective condition monitoring and state forecasting are crucial for operational efficiency and safety.
• High-dimensional data presents challenges in extracting meaningful insights for predictive maintenance.
• Existing methods often lack real-time applicability and robustness in complex systems.

Purpose of the Study:

• To develop a robust algorithmic framework for real-time condition monitoring and state forecasting using multivariate data.
• To improve the accuracy, efficiency, and robustness of condition monitoring and state predictions.
• To enable real-time anomaly detection and risk assessment.

Main Methods:

• Utilizes a combination of Proper Orthogonal Decomposition (POD) for feature extraction and dimension reduction.
• Employs Optimal Sampling Location (OSL) to identify the most informative data points.
• Integrates Dynamic Mode Decomposition (DMD) for state forecasting and Support Vector Regression (SVR) for data imputation.

Main Results:

• Demonstrated effectiveness on thermal imagery data of a ship's engine.
• Achieved significant dimension reduction and identified key system dynamics.
• Enabled real-time applicability with lower computational resource demand.
• Successfully implemented unsupervised anomaly detection coupled with state prediction.

Conclusions:

• The proposed framework offers a robust and real-time solution for predictive condition monitoring.
• The integration of POD, OSL, and DMD enhances the analysis of complex multivariate data.
• The system provides capabilities for real-time risk assessment and anomaly prediction.