IGBT疲劳下基于BP神经网络修正的结温监测

Calibrated Junction Temperature Monitoring for Fatigue-coupled IGBT Using BP Neural Network

结温T_(j)是绝缘栅双极型晶体管(IGBT)的重要状态参数。大电流注入结温监测方法只需测量IGBT模块的饱和压降U_(CE)、集电极电流I_(C),然后根据数据手册的U_(CE)-l_(C)-T_(j)模型即可计算结温,便于工程在线应用。但是,随着IGBT模块疲劳,实际的U_(CE)-l_(C)-T_(j)关系会逐渐偏移,产生明显的结温监测误差。为此,在大电流结温监测方法中引入反向传播(BP)神经网络的自学习映射建模能力,提出一种由BP神经网络修正的结温监测方法,它根据部分历史数据将影响模块疲劳的工况累积量映射为一个疲劳因子,用于对大电流法的结温计算结果进行修正。搭建了疲劳加速功率循环试验平台,实现了对IGBT结温监测误差的量化观测,随后将已循环次数作为工况累积量,对所提结温修正方法进行了功率循环条件下的试验验证。结果表明,对象IGBT模块在1100h疲劳后期,标准大电流法的结温监测误差达到约20℃,修正后的误差小于0.5℃,验证了方法的有效性。

Junction temperature T_(j)is an important state parameter of insulated gate bipolar transistor(IGBT).The large-current junction temperature monitoring method only needs to measure the saturation voltage U_(CE)and collector current I_(C)of an IGBT module,and then calculate junction temperature according to the U_(CE)-l_(C)-T_(j)model in the data manual,which is convenient for engineering online application.However,with IGBT module fatigue,the actual U_(CE)-l_(C)-T_(j)rela-tionship will gradually shift,resulting in obvious junction temperature monitoring errors.Therefore,the self-learmning mapping modeling ability of back propagation(BP)neural network is introduced into the large-current junction tem-perature monitoring method,then a calibrated junction temperature monitoring method using BP neural network is proposed.Based on partial historical data,the cumulative working condition affecting module fatigue is mapped as a fa-tigue factor.It is then used to calibrate the junction temperature calculation results of the large-current method.A fa-tigue acceleration power cycling test platform is built to realize the quantitative observation of IGBT junction temper-ature monitoring errors.Then,the number of cycles is taken as the cumulative working condition to verify the proposed junction temperature calibraton method under power cycling conditions.The results show that the junction temperature measurement error of the standard large-current method reachs 20℃,while the error of the calibrated method is less than 0.5℃at the end of the 1100 h fatigue,which verifies the effectiveness of the method.