Research On Key Technology Of Condition Monitoring And Fault Diagnosis Of IGBT Modules

Insulated Gate Bipolar Transistor(IGBT)has become the most commonly used device in power electronic systems due to its advantages of low conduction voltage drop and fast switching speed.However,the difference of material properties in IGBT internal structure will gradually cause aging process during the electric-heating cycle,and eventually lead to the device failure.Therefore,how to effectively monitor the aging process of IGBT devices has become one of the focused research problems.On the other hand,the change of external equipment may also lead to the occurrence of over current or over voltage of IGBT,resulting in its sudden failure,which occurs instantaneously and cannot be predicted in advance.Therefore,how to diagnosis the faults timely and accurately for IGBT’s sudden failure has also become one of the research focuses.It is of great significance to improve the stability of the system to evaluate the reliability of IGBT module in its whole life cycle,especially for the condition monitoring of aging failure and fault diagnosis of sudden failure.In this paper,the condition monitoring and fault diagnosis technology of an IGBT module is studied.The research covers two failure modes consists of gradual failure and sudden failure.The main contents include:(1)A device-level junction temperature evaluation method of multi-chip IGBT module based on improved Cauer thermal network model is proposed.As the key parameter of lifetime prediction,junction temperature also become the most important factor of aging failure.In view of the trend of module integration,this paper firstly constructs a simplified heat conduction equation by cutting a cross section to study the thermal coupling effect in a multi-chip module.Considering the equivalent concept of heat spreading angle introduced in the traditional model,the results of the equation are fitted numerically in this paper.By the parameter fitting,the traditional Cauer model can describe the junction temperature of a multi-chip module.The experimental results show that the proposed method overcomes the shortcoming of the traditional method that it does not need finite element method co-simulation,which will greatly reducing the workload.And has the advantages of high precision.(2)A board-level condition monitoring method of parallel-connected IGBT modules based on the current distribution is proposed.Considering the common power expansion method,the aging state of the device is monitored by analyzing the electrical characteristics of IGBT modules in series and parallel.Among them,to solve the problem that the on-state voltage drop of IGBT module after parallel connection can no longer reflect the aging characteristics of the single module,the condition monitoring method of parallel connected modules is mainly studied.This paper selects three typical stages in its operation process to analyze the parallel current distribution,and proposes a condition monitoring method based on the current distribution rate.This method combines the current distribution characteristics of the parallel IGBT module with the traditional method of monitoring the on-state voltage drop to monitor the aging process of the parallel IGBT module.Experimental results show that this method can effectively monitor two typical failure modes,which is bond wires aging and solder layer aging.(3)A system-level fault diagnosis method based on convolutional neural network(CNN)is proposed.the fault analysis is firstly carried out,combined with the existing three-level inverter topology,aiming at the sudden failure of IGBT device during operation.Since CNN are sensitive to curve shape features,a fault diagnosis method based on CNN for three-level inverters is proposed.Firstly,all fault types are divided into half-cycle detection fault(type I fault)and full-cycle detection fault(Type II fault)according to the diagnosis difficulty.The first step of the diagnosis process is to identify the fault type,and the second step is to diagnose the corresponding fault type based on the identification results.Accordingly,three different CNN algorithm models are constructed for step diagnosis.Thus,aiming at the slow convergence of training process,an improved Adamod parameter updating method was proposed to accelerate the convergence process.Furthermore,full connection and maximum flexibility functions are used to achieve fault classification.At the same time,a three-level inverter fault diagnosis system is built for experiment.The experimental results show that the CNN technique has good accuracy and efficiency in fault diagnosis of NPC inverter.40% of these faults can be diagnosed within half a current cycle,while the remaining 60% faults can be diagnosed within one cycle.Finally,according to the requirements of the project and the research results of this paper,IGBT multi-level reliability system is built,and a condition monitoring and fault diagnosis software platform based on C# and MATLAB hybrid programming is designed.The test results show that the software runs reliably and can complete the corresponding functions.