IGBT Life Prediction Based On Machine Learning Algorithm

Insulate-Gate Bipolar Transistor?IGBT?has the advantages of low driving power,low switching loss,high operating frequency,and good thermal stability.Therefore,it is used in new energy power generation,aerospace,smart grid,high-voltage flexibility Scenarios such as DC transmission systems,communications,and portable equipment are widely used,and the reliability of IGBT modules has an important impact on the entire power system.Due to the wide variety of IGBT modules and the complicated usage scenarios,the traditional life prediction model has not reached the accuracy requirements.Therefore,this paper conducted a study on the life prediction model of IGBT based on the machine learning algorithm.The main research contents include:First,starting from the analysis of IGBT failure mechanism and life evaluation method, gate-emitter turn-off voltage spike V_GE-np and collector-emitter turn-off voltage spike V_CE-p are selected as the characteristic parameters of IGBT life prediction,Feature extraction is performed on the IGBT accelerated aging data published by the NASA PCo E Research Center.After obtaining the failure features selected in this paper,a data smoothing algorithm based on Least-Squares fitting is used to smooth the feature data;Second,use ANSYS Simplorer software to establish a dynamic model of IGBT model IRG4BC30KD,simulate the dynamic process of the model,and verify and compare with the IGBT accelerated aging data published by the National Aeronautics and Space Administration?NASA?PCo E Research Center;Third,apply artificial neural network and support vector machine to the field of IGBT life prediction respectively,and optimize it to establish IGBT life prediction model,use aging experimental data published by NASA to predict IGBT life,analyze and compare the prediction of the model accuracy;The prediction results show that the smoothing of the sample data can improve the prediction accuracy of the model.In the case of small-capacity samples,the use of support vector machines can achieve better prediction accuracy,while in the case of large-capacity samples,the artificial neural network is more accurate in predicting the life of the IGBT.In summary,this article establishes the IGBT life prediction model based on the machine learning algorithm and predicts the life of the IGBT model IRG4BC30KD.The prediction results show that the method proposed in this paper can be applied to the life evaluation of IGBT and has certain engineering applications value.