On-line fault diagnosis of multi-phase drives using self-recurrent wavelet neural networks with adaptive learning rates

In this paper, a robust fault diagnosis strategy for open switch faults isolation in multiphase drives using machine learning techniques is designed. An adaptive self-recurrent wavelet neural network as a nonlinear system identifier provides estimate of a nonlinear model to generate appropriate fault symptoms based on the gate signals and actual motor currents. The significant contribution of this work is combining component-based and system-based fault diagnosis methods. A component-based signal is defined as the input of the identifier, while a system-based signal is used as the output. Advantage of the proposed method is the ability of detecting inverter faults in less than one millisecond without deploying extra hardware. This method is applicable in current controlled, speed controlled, and speed sensorless systems. The fault detection scenario is followed by a classifier to locate the fault. Discriminant Analysis and Support Vector Machines have been implemented to identify the fault location. The evaluations are supported by a laboratory setup.