Research On Key Technologies Of Open Circuit Fault Diagnosis Of Three-phase VSR Switching Devices Based On Current Characteristics

The converters developed based on the morden power electronics technology is the key application device in the fields such as national defense,energy,mining and metallurgy,transpotation,equipment manufacturing and aerospace engineerring.Among these converters,the PWM rectifier has been widely applicated as stable and green DC source supply due to its high efficiency,high power factor and the grid side current of which is sinusoidal waveform.However,because of the complex operation environment and control strategy,as well as the frequently on and off process of semiconductor power devices such as Insulated Gate Bipolar Transistor(IGBT),the faults and failures of the rectifier system is unavoidable.Statistics show that,the semiconductor power devices have the highest fault rate in the whole converter system,which consists of short-circuit fault and open-circuit fault.Since the open switch fault can lead to the voltage pulse in the DC side and the harmonic current in the AC side,thus resulting in a decline in performance of the system and even cause the secondary fault.Therefore,the reaserch of open switch fault diagnosis of switching devices in the PWM VSR system can confer the tolerance operation capability on the system,and it is an effective way to improve the reliability of the VSR system.Taking the three-phase two-level voltage source PWM rectifier as an example,this paper studies the influence of open-switch faults of the semiconductor power devices on the operation state of VSR system,and derives the fault model of VSR system under the different fault conditions.This paper summerises the distortion regulation of AC side currents in the VSR system when the open-switch faults occur and proposes a diagnosis method of open-switch fault based on the distortion current analysis.The proposed method detects the fault via the zero-crossing distorton of the faulty phase current and locates the faulty switch via the change regulation of the non-faulty phase currents.Aiming at the complex conditions that the open-switch fault occurs at two devices together,this paper proposes a diagnosis method for double-switch open faults based on a fault registor.This method employes the proposed single-switch open fault diagnosis method as a basic module and construct a fault logical using the current distortion regulation on the double-switch fault conditions,thereby reliazing the online fault diagnosis.The experimental results show that the open circuit fault diagnosis of switching devices in VSR system can be realized quickly and accurately based on the distortion current analysis.In order to improve the intellectualization of the fault diagnosis methods and reduce the dependence of diagnosis methods on the manual experience,an open circuit fault diagnosis method for VSR system based on the optimization deep learning technology and the fault pattern recognition model is proposed in this paper.This method employes the Deep Belief Network(DBN)to extract the fault features from the AC current signals of the VSR and then employes the Ensemble-Support Vector Machine(E-SVM)method to construct the fault identification model,thereby realizing the automatic diagnosis of open switch faults.Aiming at the problem that the traditional fault diagnosis methods based on machine learning algorithms need to collect data of at least one fundamental period of the current,which leads to a low diagnosis efficiency,an imporved method based on the Double Chain Quntaum Genetic Algorithm(DCQGA)is proposed to optimize the number of DBN's input layer nodes.In this method,the accuracy of fault identification and the length of sampled signals are taken into account comprehensively,and the length of sampled signals is reduced as much as possible on the premise of ensuring the recognition accuracy,thus significantly improving the diagnosis efficiency.Experimental results show that the feature vectors with high degree of separation can be obtained by using DBN technology,which effectively ensures the accuracy of fault identification.By using the DCQGA method to optimize the number of DBN's input layer nodes,the sampling time which usually exceeds one fundamental period in the traditional methods can be reduced to less than one-tenth of the fundamental period,effectively improving the fault diagnosis efficiency.In order to solve the problem that a large amount of historical fault data is required for existing methods based on machine learning algorithms,An open circuit fault diagnosis method for VSR switching devices based on the current prediction model and the residual vector is proposed.Firstly,the Ensemble-Support Vector Regression(E-SVR)method is used to establish the regression model of the AC current in the VSR system,then the residual vector is constructed using the residual currents.The fault detection and the fault location are realized through the amplitude and the phase of the residual vector.The experimental results show that the current regression model constructed based on the E-SVR method has high precision,high reliability and strong generalization capability,which can realize the accurate prediction of the AC current under the operating conditions of load transient,grid voltage fluctuation and non-unit power factor.The open circuit fault can be detected quickly based on the residual vector amplitude,and the fault switches can be located accurately based on the vector phase.