Fault diagnosis using SWT and Neyman Pearson detection tests

Abstract

This paper presents a new methodology for fault detection and identification in power system drives. The stationary wavelet transform is the tool used through the analysis of the three phase stator current signals measured at the stator of an induction machine fed by a three phase voltage inverter. Fault scenarios with one open-switch are considered because they are the most likely to occur. Several signals are analysed simultaneously in order to perform the diagnosis. The currents signals are filtered using the SWT performed with the DB4 wavelet to extract the detail and approximation coefficients up to level 6. Then, the approximation at level 6 is examined to detect changes in the mean. This is achieved with statistical hypothesis techniques. In this work, a Neyman Pearson change in the mean detection test is used. Finally, a signature table is deduced to isolate the faulty switch. The whole diagnostic procedure can perform on line because of its low computational cost. Real data recorded from a benchmark feed the proposed diagnostic tool. Presented results confirm the effectiveness of the proposed methodology.