Neural Network Fault Diagnosis of Voltage Source Inverter under variable load conditions at different frequencies

Abstract

The major issue of open switch fault diagnosis in Voltage Source Inverters is false alarm generated as a result of load and frequency variations. The main objective of this paper is to solve such an issue by extracting minimum number of features from fault detection parameter. The fault diagnostic system (FDS) under variable load conditions requires more number of features to be extracted from detection parameter. Therefore stator currents are taken in the DQ coordinate that is Park’s Vector Transform (PVT). The PVT is used to normalize the currents without affecting nature of transients caused due to fault occurrence. The normalized currents are passed through Discrete Wavelet Transform (DWT) and features are extracted from detail coefficients of DWT under healthy and faulty conditions. As a result of normalized currents, the extracted features of three phase currents are same under different load conditions but have definite distinctive values under different faulty conditions. Hence, once features are extracted for single load conditions they remain same for all load conditions. An Artificial Neural Network is trained using these features. The results are presented for different fault configurations, single and multiple switch faults under variable load conditions at different frequencies. Additionally, the results are presented for the real-time diagnostic of faults, showing the instance of fault occurrence and the instance of fault isolation.