车用功率模块热疲劳应力的高精度在线提取方法与实证

High-precision Online Extraction Method and Demonstration of Thermal Fatigue Stress of Automotive Power Modules

随着车用功率模块的功率密度和电压等级不断提高,热可靠性问题日益突显。其中,温度应力是功率模块老化的主要因素之一。为了评估车用功率模块的热疲劳,需要获取完整的载荷历程,并采用应力循环计数方法进行损伤统计,但该方法难以在电动汽车驾驶过程中实时评估损伤信息,预测性维护存在滞后性。文中基于标准离线雨流计数法进行改进,针对不同材料以及所处不同寿命阶段,设计可变的滤波窗宽对实时结温历程进行筛选和除噪;通过基于“三点双变程”法改进的实时雨流计数算法对有效结温极值点进行高效实时处理,即可实时得到温度应力全循环与半循环数据;通过离散标准化程序对原始热应力计数结果进行聚类存储,满足边缘计算的需要,实现低成本应力循环统计;最后,基于乘用车测试该在线算法在电动汽车实际驾驶路谱下的可行性。实验结果表明,与标准离线计数法相比,累积损伤度误差小于1%。进一步地,讨论温度采样周期对应力计数和损伤计算的影响,表明温度采样周期为500 ms可以兼顾算力与精度,支撑电动汽车的实时可靠性评估。

As the power density and voltage levels of automotive power modules continue to rise,thermal reliability challenges have become increasingly significant..Among them,temperature stress is one of the main factors contributing to power module aging.To evaluate the thermal fatigue of vehicle power modules,it is necessary to obtain a complete load history and use the stress cycle counting method for damage statistics.However,this method poses challenges in assessing the damage information in real time during driving,and introduces delays in predictive maintenance.Based on the improvement of the standard off-line Rainflow counting method,a variable filter window width is designed for different materials and different life stages to screen and denoise the real-time junction temperature.Through the improved counting algorithm based on the“three-point double-variable range”,the effective junction temperature extreme points are processed efficiently,and the full cycle and half cycle of temperature stress could be obtained.The original thermal stress counting results are clustered and stored through a discrete normalization program to meet the needs of edge computing and achieve low-cost stress cycle statistics.Finally,based on the passenger car,the feasibility of the online algorithm is tested under the actual driving road map of electric vehicles.Experimental results show that the cumulative damage degree error is less than 1%compared with the standard off-line counting method. Furthermore, the influence of temperaturesampling period on stress counting and damage calculation isdiscussed, indicating that a sampling period of 500ms canbalance both computing power and accuracy, and supportreal-time reliability evaluation of electric vehicles.