Early Detection of Plant Equipment Failures: A Case Study in Just-in-Time Maintenance

The development and testing of a model-based fault detection system for electric motors is briefly presented. The fault detection system was developed using only motor nameplate information. The fault detection results presented utilize only motor voltage and current sensor information, minimizing the need for expensive or intrusive sensors. Dynamic recurrent neural networks are used to predict the input-output response of a three-phase induction motor while using an estimate of the motor speed signal. Multiresolution (or wavelet) signal-processing techniques are used in combination with more traditional methods to estimate fault features for use in winding insulation and motor mechanical and electromechanical failure detection.