Research On Open Circuit Fault Diagnosis Of Three-level Inverter Based On Signal Analysis

Three-level inverters have been widely used in aerospace,industrial drives,and other fields due to their low output waveform distortion rate and low switching loss.Among them,the threelevel inverters with neutral point clamped(NPC)topology is the most commonly used.However,open circuit faults happen frequently due to large number of NPC type three-level inverter switches.Moreover,the normal operation of equipment is affected due to long-term exposure to harsh environments such as high voltage and high current.In addition,the accidents about safety property of people happen.The traditional fault diagnosis method for NPC type three-level inverters are based on signal analysis.The method has the advantages of fast diagnosis speed and good real-time performance.However,it has problems such as spectrum leakage and lack of local time-frequency feature analysis capabilities.Therefore,the thesis conducts research on the open circuit fault signal processing and diagnosis methods for NPC type three-level inverters.The specific research contents are as follows:Firstly,the topology and working principle of NPC type three-level inverters are introduced.The main fault types of three-level inverters are analyzed and classified.In MATLAB/Simulink,a simulation model of a NPC type three-level inverter system is established to simulate the open circuit fault of power switches.The open circuit fault current signal are then obtained.This provides a data basis for subsequent signal preprocessing and fault diagnosis experiments.Secondly,in view of the traditional fault diagnosis methods for NPC type three-level inverters based on frequency domain analysis,which have the problems of spectrum leakage and difficulty in analyzing nonlinear fault signals.Therefore,we propose a fault feature extraction method for three-level inverters using the Prony method.The extracted fault signal is denoised with wavelet denoising to address Prony’s sensitivity to noise.Then,the denoised signal is processed using the Prony method to obtain four dimensional fault characteristics such as frequency,amplitude,phase,and attenuation factor.Finally,Classification of the collected fault feature vectors and fault location is performed using support vector machines.The proposed method is validated on the MATLAB/Simulink platform,and the simulation results demonstrate high diagnostic accuracy and fast diagnostic speed..Then,aiming at the problem that Prony method lacks the ability of time-frequency analysis.A fault feature extraction method for NPC type three-level inverters based on generalized S transform is proposed.Firstly,the signal is subjected to a generalized S transform to obtain a twodimensional high order matrix.The singular value decomposition is applied to reduce the dimension of the feature matrix,and the main eigenvalue components of the non zero singular values are extracted.Then,limit learning machines are used to classify and locate faults.Finally,the method is validated in MATLAB/Simulink platform,and the simulation results show that compared to the fault diagnosis method based on Prony analysis,the fault diagnosis method based on GST has a higher diagnostic accuracy.Finally,an experimental platform for NPC three-level inverter fault diagnosis system based on STM32 controller is built.Complete the overall design,hardware design,and software design of the platform.The two proposed fault diagnosis methods based on signal analysis are verified on this experimental platform.The experimental results show that the proposed algorithm has the ability to quickly locate the position of fault switches and has high accuracy in fault diagnosis.