尘土颗粒影响下电路板电化学迁移失效寿命建模探索  

Exploring the Life Modeling Methods for Electrochemical Migration Failure of Printed Circuit Board under Dust Particles

作　　者：周怡琳[1] 杨璐 鲁文睿 ZHOU Yi-lin;YANG Lu;LU Wen-rui(Automation School,Beijing University of Posts and Telecommunications,Beijing 100876,China)

机构地区：[1]北京邮电大学自动化学院,北京100876

出　　处：《北京邮电大学学报》2020年第3期11-18,31,共9页Journal of Beijing University of Posts and Telecommunications

基　　金：国家自然科学基金项目(61674017)。

摘　　要：针对离散尘土颗粒与温度、湿度、电场强度交互作用的复杂条件下,难以有效建立电路板电化学迁移的失效物理模型的问题,通过温湿偏置加速实验,模拟不同积尘密度下电路板的电化学迁移失效,分析颗粒分布密度对电路板绝缘失效时间的作用特性.采用正交实验获取不同尘土颗粒密度、温度、湿度、电场条件下电路板绝缘失效的寿命数据.基于数据驱动的方法,探讨电路板在尘土颗粒污染下的电化学迁移失效寿命建模.对比了多项式回归、机器学习中的梯度提升回归树和随机森林3种方法在尘土分布密度的高低区间内的寿命预测效果.讨论了尘土颗粒污染下利用机器学习建立电路板电化学迁移失效寿命模型的有效性.Facing the complex conditions that the discrete dust particles interact with the temperature,the humidity,and the electric field intensity,it is difficult to effectively establish the life model of electrochemical migration( ECM) of printed circuit board( PCB) based on failure physics. Through the temperature humidity bias tests,the ECM process under different dust density is simulated. The effect of particle distribution density on time to failure( TTF) of PCB is analyzed. The TTF data of PCB under different particle distribution density,temperature,relative humidity and electric field intensity are obtained by an orthogonal experiment. Based on the data driven method,the ECM life modeling of PCB under dust particle pollution is discussed. The life prediction effects of polynomial regression,gradient boosting regression tree and random forest in machine learning for high and low dust distribution density are compared. The effectiveness of machine learning to establish ECM life model of PCB under dust particle contamination is discussed.

关 键 词：尘土污染 电化学迁移 寿命模型 机器学习 

分 类 号：TN406[电子电信—微电子学与固体电子学]