摘要
为获取老旧电动汽车中拆解的绝缘栅双极型晶体管(IGBT)模块在再制造时的剩余寿命,提出经典循环神经网络的变体,即GRU神经网络的IGBT模块剩余寿命预测模型。分析IGBT模块的内部结构及老化失效机理,明确老化失效的具体形式,结合IGBT模块功率循环试验的老化数据,确定通态饱和压降作为模块老化失效特征量。通过试验构建最优参数的GRU神经网络剩余寿命预测模型,完成对老化失效特征量的预测,并与同样是经典循环神经网络另一种变体LSTM网络预测模型进行对比。结果表明:经过优化参数的GRU网络模型的均方根误差为0.0046,平均绝对误差为0.0041,决定系数为99.96%,相对LSTM网络精度更高,更适合所选IGBT模块的剩余寿命预测,同时检测的时间成本更低,更能提高IGBT模块再制造时的检测与生产效率。
In order to obtain the remaining useful life of the disassembled insulated gate bipolar transistor(IGBT)module during remanufacturing in used electric vehicles,a variation of the classical cyclic neural network,namely the remaining useful life prediction model of IGBT module of GRU neural network,is proposed.The internal structure and aging failure mechanism of IGBT module are analyzed,and the specific form of aging failure is defined.Combined with the aging data of IGBT module power cycle test,the on-state voltage drop is determined as the aging failure characteristic quantity.The remaining useful life prediction model of GRU neural network with the best parameters is constructed through experiments,and the prediction of aging failure characteristic quantity is completed.The results are compared with the LSTM network model,another variant of the classical cyclic neural network.The results show that the root-mean-square error of the optimized GRU network model is 0.0046,the average absolute error is 0.0041,and the coefficient of determination is 99.96%.Compared with LSTM network,it is more accurate and more suitable for the residual life prediction of the selected IGBT module.At the same time,the detection time cost is lower,and it can improve the detection and production efficiency of the IGBT module remanufacturing.
作者
李新宇
孟子民
盛光鸣
刘志峰
LI Xinyu;MENG Zimin;SHENG Guangming;LIU Zhifeng(School of Mechanical Engineering,Hefei University of Technology,Hefei 230009,China;Key Laboratory of Green Design and Manufacturing of Mechanical Industry,Hefei University of Technology,Hefei 230009,China)
出处
《中国测试》
CAS
北大核心
2024年第11期25-32,共8页
China Measurement & Test
基金
国家重点研发计划资助项目(2019YFC1908002)。
关键词
绝缘栅双极型晶体管
GRU
神经网络
剩余寿命预测
insulated gate bipolar transistor
GRU
neural network
remaining useful life prediction
