Neural network based cycloconverter fault detection using wavelet decomposition

Power electronic devices may fail to function normally due to unexpected breakdown of various power electronic component. This fact motivates to find an effective but simple technique for real time diagnosis of component failure in cycloconverter, which is commonly used in industry to vary the speed of an
_{AC motor}
by means of controlling its frequency. Wavelet decomposition method is employed for detecting faulty component path by extracting the substantial signatures from the output waveforms across the load end of cycloconverter. Backpropagation multilayer perceptron(BPMLP) based artificial neural network(ANN) and probabilistic neural network(PNN) are used to correctly distinguish the faulty path including two power MOSFET in the cycloconverter. Results indicate high classification accuracy at most of 99.9%. The proposed method will also minimize the fault removal time and hence maintain the production consistency in the industry.