Fault detection and diagnosis for multi-rate descriptor systems using a combination of extended Kalman filter and support vector machines

Background

In modern engineering systems, accurate fault detection and diagnosis are crucial for reliability and efficiency. Multi-rate descriptor systems pose challenges due to measurements at various intervals and the consistent initialization required by equality constraints.

Methods

This paper addresses these challenges by proposing novel approach combining multi-rate Differential Algebraic Equation (DAE) based Extended Kalman Filter (EKF) and Support Vector Machines (SVM). The multi-rate DAE-EKF handles nonlinear dynamics and accounts for measurement noise, while SVM enhances fault detection by classifying system residues. The integration involves two stages: multi-rate DAE-EKF operates as the primary estimator, generating state information residues, and SVM uses these residues to distinguish between normal and faulty behaviour. This method enables isolating individual and simultaneous faults in multi-rate descriptor systems, improving accuracy and reliability.

Key Findings

The proposed approach exploits EKF's dynamic estimation strengths and SVM's classification robustness. Benchmark studies demonstrate its effectiveness on a Two-phase reactor condenser system with a recycle and a Reactive Distillation system. By combining multi-rate DAE-EKF and SVM, this methodology overcomes multi-rate challenges and achieves enhanced fault detection and diagnosis, contributing to operational reliability in complex systems. The results show improved performance in identifying faults and ensuring system stability, showcasing its potential in industrial applications.