Research On Performance Degradtion And Fault Prediction Technology Of IGBT Module

The core device of the high-speed train is the traction converter,and the IGBT module is the core component of the traction converter.Compared with industrial IGBT modules,the working environment of the IGBT modules inside the converters of high-speed train is worse.The power fluctuations,electrical-thermal stress and mechanical shocks in the application are more severe,so they are more prone to wear and failure,which bringing huge challenges to the safe operation of high-speed train.So it is very important to study the performance degradation mechanism and fault prediction technology of IGBT modules.The existing IGBT module performance degradation research have problems such as impractical simulation and overly complex modeling,and the research on IGBT fault prediction technology also have shortcomings such as insufficient accuracy and poor anti-interference ability.In order to solve the above problems,this thesis studies the performance degradation mechanism of IGBT modules based on multi-physics coupling analysis and the fault prediction algorithm of IGBT modules based on machine learning.The main contents are summarized as follows:(1)Aiming at the problem of how to accurately obtain the power consumption and junction temperature distribution of the IGBT module during operation,a method based on electricthermal co-simulation is studied.Firstly,the working principle and working characteristics of the IGBT are introduced.On this basis,the high-frequency application scenarios of the module are considered,and the mathematical model of the power loss of the module is established from the two aspects of the on-state loss and switching loss.Then the foster thermal network is introduced to build the module’s electrical-thermal coupling simulink model and then the model solves the junction temperature and power dissipation distribution of IGBT module.The simulation results show that the model considering the switching loss can more accurately calculate the power consumption and junction temperature distribution of the IGBT module during operation,which provides a basis for finite element analysis.(2)Aiming at the complex problem of traditionally relying on experiments to analyze the performance degradation mechanism of IGBT modules,a multi-physics coupling analysis method based on finite element model is studied.Firstly,the thermal field,stress field and electro-thermal-mechanical coupling effect involved in the working process of the module are analyzed,and a finite element model of the electro-thermal-mechanical multi-physics coupling of the module is established.Aiming at the problem that the single heating mode of the previous model makes the simulation results inaccurate,it is improved by loading the power consumption on the chip and passing the current into the bonding wire.At the same time,setting a variety of boundary conditions for the model to simulate the actual working conditions,and set up multiple groups of failure control experiments.The experimental results show that the bonding wires and the solder layer are the most vulnerable parts of the module to be damaged by thermal stress,but they have different mechanisms that lead to the failure of the entire module: the bonding wires fall off one by one,so that the current cannot flow,which leads to the failure of the module;the delamination of voids in the solder layer greatly reduces the heat dissipation performance of the module,and the heat accumulation changes the material properties and causes the module to gradually fail.(3)Aiming at the problem of how to accurately predict the fault of IGBT module,adopt a data-driven approach,and select the IGBT collector-emitter saturation voltage drop as the fault characteristic parameter.At the same time,a fault prediction method of IGBT module based on wave neural network is proposed.Then,aiming at the problem that the wave neural network weight correction algorithm(gradient descent method)is easy to fall into local optimum and the prediction accuracy is insufficient,a genetic algorithm is proposed to improve it.The experimental results show that the two fault prediction models enable failure prediction of IGBT modules,but the prediction accuracy and precision of the improved genetic algorithm model are significantly improved.