Research On IGBT Life Prediction Based On Data Drive

With the continuous progress and development of power electronics technology and industrial intelligent technology,the reliability issues of power electronic devices such as IGBT are gradually exposed to the public's vision.Generally speaking,reliability management issues often involve a wide range of factors,considering too many factors,and are usually restricted by various factors in the application process,making the analysis results difficult to meet industry standards.For the life prediction problem of IGBT,the principle of the device is first analyzed,on this basis,appropriate failure parameters are selected and appropriate data sets are selected for analysis.For the characteristics of the original sample,the original sample is first processed by the data preprocessing method,and then a series of deep learning models and their derivatives are introduced.By analyzing the data characteristics of the IGBT sample set,the model that meets the corresponding characteristics is selected and used Based on the expansion of time series samples,the number of samples can meet the needs of deep learning models,and finally the quality of the generated data is analyzed.After obtaining enough samples,use the statistical model and the deep learning model to train the life prediction model of the IGBT through the Tensorflow framework in the Colaboratory platform,and analyze and compare the corresponding accuracy of the model under different prediction tasks.In order to improve the training accuracy and learning ability of the model,the model introduces an attention mechanism,which improves the learning ability of the model for sequence correlation.Finally,in order to further improve the generalization performance of the model in the long-term prediction,the sample distribution is further optimized through the downsampling method,so that the model has more excellent performance.Finally,based on the trained model,the sample quality of the generated samples is evaluated to estimate the pros and cons of the generation algorithm.The prediction model is imported into the life prediction software,which enhances the application value of the model in the engineering environment.Figures:53,Tables:8,References:54.