基于变可信度近似模型的BGA焊点可靠性预测与优化方法  

Reliability prediction and design optimization of BGA solder joint based on multi-fidelity surrogate model

作　　者：于敬丹 王儒[1] 吴文志[2] 胡子翔 张楚雷 王国新[1] 阎艳[1] YU Jingdan;WANG Ru;WU Wenzhi;HU Zixiang;ZHANG Chulei;WANG Guoxin;YAN Yan(Beijing Institute of Technology,Beijing,100081,China;The 38th Research Institute of China Electronics Technology Group Corporation,Hefei,230088,China)

机构地区：[1]北京理工大学,北京100081 [2]中国电子科技集团公司第三十八研究所,合肥230088

出　　处：《焊接学报》2024年第1期10-16,I0003,I0004,共9页Transactions of The China Welding Institution

基　　金：国家自然科学基金项目(52105241);部委科研项目(ZQ2020D210005);国防基础科研项目(JCKY2020210B007)。

摘　　要：焊点可靠性预测研究多采用有限元仿真与单一精度近似模型相结合的模式,存在仿真时间长、效率低、准确性差等问题,为此,提出了一种基于变可信度近似模型的球栅阵列(ball grid array,BGA)焊点可靠性预测与优化方法.首先,对不同网格细化方案进行收敛性验证,分别设计高/低精度样本点进行有限元仿真;其次,基于CoKriging模型融合高/低精度仿真数据进行焊点可靠性预测;最后,将预测结果与单一精度近似模型进行对比分析,并采用遗传算法优化模型获得对应结构参数.结果表明,在更少的仿真成本下,变可信度模型的预测效果更好,在同等预测精度下,变可信度模型高精度样本点数量仅为单一精度模型的1/4,相比高精度神经网络预测模型,在寻优过程中收敛更快.At present,the reliability prediction of solder joints is mostly based on the combination of finite element simulation and single precision surrogate model,which has some problems such as long simulation time,low efficiency and poor accuracy.Therefore,a reliability prediction method for BGA(ball grid array)solder joints based on multi-fidelity model is proposed.Firstly,the convergence of different meshing schemes was verified,and then the high and low precision sample points were designed respectively for finite element analysis(FEA).Secondly,the reliability of solder joints was predicted based on the Co-Kriging model based on multi fidelity FEA data.Finally,the prediction results were compared with the single precision surrogate model,Under the same cost constraints,the multi-fidelity model demonstrates significantly higher prediction accuracy.and NSGA(nondominated sorting genetic algorithm)was used to optimize the model to obtain the corresponding process parameters.The results show that with less simulation cost,the prediction result of multi-fidelity model is better.Under the same prediction accuracy,the number of high-precision sample points of the variable reliability model is only 1/4 of that of the single precision model.At the same time,compared with the neural network prediction model,it converges faster in the optimization process.This paper provides some reference for the research of reliability prediction of solder joint with multi-fidelity model.

关 键 词：焊点可靠性 寿命预测 变可信度近似模型 遗传算法 

分 类 号：TG407[金属学及工艺—焊接]