Study On The Fault Diagnosis System Of Active Neutral Point Clamped Three Level Inverter Based On Neural Network

With the development of the inverter technology, three level inverter has been widely applied in many occasions. The power electronic devices need to withstand high voltage, high current and high frequency switch state. Therefore, the fault of inverter caused by the fault of power switch devices has already become the problem that could not be ignored. The active neutral point clamped three level inverter has been widely used because the switching losses of the power devices can be balanced. However, locating the fault of specific power switch devices fast and accurately is very difficult because of its complex internal topology structure. Therefore, the fault diagnosis problem of the active neutral point clamped three level inverter should to be solved urgently.The main research work of the thesis includes the following.First of all, the active neutral point clamped three level inverter was chosen as the diagnosis object in the thesis. The topology structure and the working principle of active neutral point clamped three level inverter were introduced and the fault reasons were analyzed. All of the open circuit fault conditions were classified and each fault condition was matched to a fault code differently. In order to reduce the influence of the load mutation for fault diagnosis, to the extent possible, the leg voltage signal was chosen as the test signal. The simulation and fault analysis of the inverter were completed by the Simulink module of MATLAB software.Secondly, the frequency-band energy values as the fault diagnosis input were extracted by the wavelet packet decomposition from the leg voltage signal under different fault conditions. In order to reduce the dimension of the fault information, the 3/2 transformation was used to convert the three-phase leg voltage signal to two-phase to output.Thirdly, the fault diagnosis method of ANPC three level inverter based on neural network optimized by genetic algorithm was researched. The global optimization performance of genetic algorithm was used to optimize the weights and thresholds of BP neural network because the training of traditional BP algorithm was easy to fall into local minimum value and the training speed was slow. It could improve the speed and accuracy of diagnosis. At the same time, in order to solve the difficulty that the fault feature values of Sx2(Sx3) single power tube opening were consistent with the fault feature values of Sx1Sx2(Sx3Sx4) two power tubes opening meanwhile, two BP neural networks were proposed to diagnose. All kinds of open circuit fault conditions could be diagnosed effectively by this method verified through the simulation.Finally, the fault diagnosis system of ANPC three level inverter based on wavelet packet decomposition and neural network optimized by genetic algorithm was developed using the mixed programming of Lab VIEW and MATLAB. The software possessed the functions including the preservation of fault information, the extraction of the fault feature by wavelet packet decomposition, training by genetic algorithm-neural network and fault diagnosis of inverter. A variety of open circuit faults of ANPC three level inverter could be diagnosed by the software fast and accurately.