Research On Fault Prediction Of IGBT Based On Terminal Characteristics

Insulated Gate Bipolar Transistor（IGBT）is popular because of its high input impedance and low on-off voltage.It is widely used in transportation,electric power,energy and other fields.In industrial applications,IGBT often runs in complex working conditions or even harsh working environment for a long time,suffering from a wide range of temperature,humidity changes and mechanical impact,gradually accumulating fatigue damage and eventually aging failure,which may lead to system shutdown and economic losses.Therefore,the research on IGBT reliability is of great significance.As a key part of reliability research,fault prediction technology can timely and effectively evaluate the device status,predict the aging trend and make early warning at the threshold through the changes of terminal parameters,so as to avoid serious safety problems caused by IGBT aging failure.In this paper,the IGBT fault prediction technology based on terminal characteristics is studied.The main research contents include:Firstly,the working principle and characteristics of IGBT are described.On this basis,the failure causes of IGBT are deeply analyzed,and the common failure modes and failure mechanisms are classified and explained.Because the aging status of IGBT can be reflected in its terminal parameters,the appropriate terminal parameters(Collector Emitter conduction voltage drop,V_CE（ON）)are selected as the aging characteristic parameters for fault prediction.Secondly,in the aging process of IGBT,the terminal parameters change continuously with time,so fault prediction can be regarded as the problem of time series prediction of terminal parameters of IGBT.Numerical algorithm can be used to solve this problem.Through the analysis of terminal parameters V_CE（ON）,it can be seen that this time series has the characteristics of segmentation,the former is gentle,and the latter has an obvious upward trend,and the accuracy of the single algorithm is not high.Therefore,this paper proposes to use the combined algorithm for fault prediction,the former uses Grey Verhulst Model（GVM）,and the latter uses Unscented Kalman Particle Filter（UPF）algorithm.UPF is an improved algorithm for the particle degradation problem of PF,which has high prediction accuracy.Thirdly,the application of deep learning in artificial intelligence in time series prediction has the advantages that traditional numerical algorithm does not have.Therefore,this paper further proposes a fault prediction method based on deep learning.Recurrent Neural Network（RNN）has great advantages in dealing with time series problems,but traditional RNN has the defects of gradient disappearance and gradient explosion.As an improved network of RNN,Long Short Term Memory（LSTM）network avoids this problem and has stronger robustness,so it is used for fault prediction.Based on the in-depth study of RNN and LSTM network,the LSTM network is designed and established from six different aspects.Finally,this paper designs and builds the IGBT accelerated aging test platform.On the premise of keeping the aging mechanism of IGBT unchanged,accelerating the aging process of devices,and then the aging characteristic parameter V_CE（ON）was collected in the experiment.The experimental data are used to verify the proposed fault prediction method based on combined algorithm and deep learning.Through comprehensive comparison,it is found that the combined algorithm and the deep learning fault prediction method can better predict the IGBT fault.The accuracy of the combined algorithm is higher than that of the single numerical algorithm,and the accuracy of the deep learning method is higher than that of the combined algorithm.More importantly,the deep learning method does not need the guidance of prior probability.But the combined algorithm has a short training time and low requirements for the computer hardware environment.Both of them have their advantages and disadvantages.The appropriate method can be selected according to the working conditions to complete the fault prediction of IGBT.In summary,this article has completed the IGBT fault prediction research based on the terminal characteristics,and verified it through experimental data,which has some guiding significance for engineering applications.