Scholarlink

<p>The current research work introduces a cutting-edge approach for accurately predicting the fatigue life of fine-pitch ball grid array (FBGA) solder joints under harmonic vibration—a critical loading condition prevalent in electronic assemblies. While previous research has explored the mechanical fatigue behavior of lead-free solder alloys, the effects of harmonic excitation, particularly within FBGA architectures, remain largely undeciphered. The originality of this work resides in the seamless integration of a robust machine learning framework with finite element simulation data, allowing for a comprehensive comparative assessment of six state-of-the-art supervised regression models, namely XGBoost, gradient boosting, light gradient boosting machine (LightGBM), random forest, decision tree, and support vector regression. These models were rigorously assessed across four advanced solder materials: SAC105, SAC305, SAC405, and InnoLot. The finite element model, meticulously developed in ANSYS APDL, focused exclusively on mechanical loading conditions, deliberately excluding thermal effects, and underwent thorough experimental validation through modal analysis to ascertain simulation fidelity and reliability. A systematic design of experiments approach was employed to vary critical input parameters, including geometric dimensions, applied force, and solder alloy composition. Among the evaluated algorithms, XGBoost displayed outstanding predictive accuracy, achieving a determination coefficient of <i>R</i><sup>2</sup> = 0.9958, outperforming other models such as gradient boosting and LightGBM. This study not only underscores the transformative potential of machine learning within the framework of fatigue life prediction but also establishes artificial intelligence (AI)-enhanced modeling as a powerful, efficient, and reliable alternative to conventional numerical methods. The ability to deliver rapid and accurate estimations of solder joint durability across diverse configurations positions this approach as a pivotal tool for boosting the workability and design optimization of high-performance electronic packages subjected to dynamic loading.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Advanced Reliability Prediction of FBGA Solder Joints Under Harmonic Vibration: Harnessing Supervised Machine Learning Models

  • 期刊论文

  • 发表日期：

    2025年9月2日

查看全文
  • Jihen Rebai,
  • Ahmed Ghorbel,
  • Nabih Feki,
  • Abdelkhalek El Hami,
  • Mohamed Haddar

摘要

The current research work introduces a cutting-edge approach for accurately predicting the fatigue life of fine-pitch ball grid array (FBGA) solder joints under harmonic vibration—a critical loading condition prevalent in electronic assemblies. While previous research has explored the mechanical fatigue behavior of lead-free solder alloys, the effects of harmonic excitation, particularly within FBGA architectures, remain largely undeciphered. The originality of this work resides in the seamless integration of a robust machine learning framework with finite element simulation data, allowing for a comprehensive comparative assessment of six state-of-the-art supervised regression models, namely XGBoost, gradient boosting, light gradient boosting machine (LightGBM), random forest, decision tree, and support vector regression. These models were rigorously assessed across four advanced solder materials: SAC105, SAC305, SAC405, and InnoLot. The finite element model, meticulously developed in ANSYS APDL, focused exclusively on mechanical loading conditions, deliberately excluding thermal effects, and underwent thorough experimental validation through modal analysis to ascertain simulation fidelity and reliability. A systematic design of experiments approach was employed to vary critical input parameters, including geometric dimensions, applied force, and solder alloy composition. Among the evaluated algorithms, XGBoost displayed outstanding predictive accuracy, achieving a determination coefficient of R2 = 0.9958, outperforming other models such as gradient boosting and LightGBM. This study not only underscores the transformative potential of machine learning within the framework of fatigue life prediction but also establishes artificial intelligence (AI)-enhanced modeling as a powerful, efficient, and reliable alternative to conventional numerical methods. The ability to deliver rapid and accurate estimations of solder joint durability across diverse configurations positions this approach as a pivotal tool for boosting the workability and design optimization of high-performance electronic packages subjected to dynamic loading.