Signal-based versus model-based fault diagnosis-a trade-off in complexity and performance

Early detection and diagnosis of incipient faults is desirable not only for on-line condition assessment but also for product quality assurance and improved operational efficiency of induction motors. At the same time, reducing the probability of false alarms increases the confidence levels of equipment owners in this new technology. In this paper, a model-based fault detection and diagnosis system that has been proposed is tested for its effectiveness in minimizing the probability of false alarms. The proposed system is compared to the more traditional signal-based motor fault estimator. In addition to nameplate information required for the initial set-up, the proposed model-based fault detection and diagnosis system uses measured motor terminal currents and voltages, and motor speed. The motor model embedded in the diagnosis system is empirically obtained using dynamic recurrent neural networks. Receiver operating characteristic (ROC) curves are constructed to demonstrate the performance trade-offs of the two estimators, while observing their relative complexity.