IGBT Health Status Assessment And Remaining Useful Life Prediction Based On Deep Learning

Insulated gate bipolar transistor(IGBT)plays a pivotal role in new energy,aerospace and other fields,the effective evaluation of its health status and the accurate prediction of its remaining useful life are of great significance to ensure the stable operation of the power system.Due to the complexity and variability of IGBT application scenarios,the traditional remaining useful life prediction model based on shallow neural network can no longer meet the accuracy requirements of practical engineering applications.Therefore,this thesis uses the method based on deep neural network to effectively evaluate and accurately predict the health status and remaining useful life of IGBT.Based on the failure mechanism analysis and accelerated aging simulation experiment of IGBT,the experimental data obtained,and the method proposed in this thesis is used to predict its failure characteristic parameters in the time domain,the health status of IGBT is investigated with the model prediction results as parameters,and the remaining useful life of IGBT is accurately predicted.The main research content of the thesis is as follows:(1)Firstly,in order to verify the relationship between IGBT failure characteristic parameters and its remaining useful life,six IGBTs were selected as the experimental body,and an IGBT accelerated aging simulation model was established to collect key parameters in the IGBT aging process in real time.Through the analysis of the working principle and failure mechanism of IGBT and the review of relevant references,the validity of the simulation experiment data is verified,and the collector-emitter saturation voltage drop is selected as the failure characteristic parameter in the aging failure process of the IGBT,and its value increases by 15% with the aging of the device as the failure threshold of IGBT in this thesis.(2)Secondly,in order to make an accurate assessment of the health status of the IGBT,a long short-term memory(LSTM)neural network model was constructed,and empirical mode decomposition(EMD)was introduced on the basis of it.By using the EMD algorithm to decompose the time series data and then predict them sequentially,the accurate prediction of the IGBT failure characteristic parameters is realized,that is,the accurate evaluation of the health status of the IGBT is made.Finally,the experiments on the IGBT accelerated aging test data set released by the NASA PCo E laboratory prove that the EMD-LSTM model proposed in this thesis has higher prediction accuracy in IGBT health status assessment applications than other comparison models.(3)Thirdly,aiming at the low efficiency of model training,the difficulty of manual parameter adjustment of depth network model and the low accuracy of current prediction methods in the process of IGBT remaining useful life prediction,a bidirectional long short-term memory(Bi-LSTM)neural network model based on Bayesian optimization algorithm(BOA)with attention mechanism is proposed.In the process of building the model,based on the BiLSTM network,the difficulty of manual parameter adjustment in the deep network is solved through the addition of BOA;The employment of attention mechanism makes the network model better use of historical feature information in the training process.Finally,the performance of the prediction model is evaluated by using the commonly used model evaluation indicators in the process of remaining useful life prediction: root mean square error and average absolute percentage error.In the end,the experimental results of the data set obtained from the accelerated aging simulation experiment of IGBT show that the BOA-Attention-Bi-LSTM model proposed in this thesis can accurately predict the remaining useful life of six IGBTs under different experimental conditions,and the prediction results are better than the prediction results of the comparison model.The results show that the method proposed in this thesis can be effectively applied to the remaining useful life prediction of IGBT,and it has certain practical engineering application value.