基于Transformer模型的IGBT剩余寿命预测

Residual Useful Life Prediction of IGBTs Based on Transformer Model

为了提前预测IGBT的剩余寿命(RUL),减小失效造成的损失并辅助维护,提出了一种基于Transformer模型的RUL预测方法,使用瞬态热阻曲线预测RUL。首先,使用不同的热循环参数进行老化试验,观察到IGBT模块结-壳瞬态热阻会随着老化而变化;然后,通过处理采集到的数据,计算出瞬态热阻并删除异常值;最后,训练一个Transformer神经网络来预测IGBT的寿命。试验结果表明,根据瞬态热阻的变化,神经网络能很好地预测IGBT剩余寿命。Transformer模型平均预测误差为0.188%,与长短时记忆网络模型进行对比,Transformer模型的预测准确度提高了0.126%。

In order to predict the residual useful life(RUL)of IGBTs,reduce the loss caused by failure and assist maintenance,a RUL prediction method based on the Transformer model was proposed,in which the transient thermal impedance curve was used to predict RUL.Firstly,different thermal cycle parameters were used to carry out aging tests.It is found that the junction-shell transient thermal resistance of the IGBT module changes with aging.Then,the transient thermal resistance was calculated and the abnormal values were deleted by processing the collected data.Finally,a Transformer neural network was trained to predict the IGBT life.The test results show that the neural network can predict the IGBT life well according to the change of transient thermal resistance.The average prediction error of the Transformer model is 0.188%.Compared with the long short term memory network model,the prediction accuracy of the Transformer model is improved by 0.126%.