马帅杰,郭鹏,李元,等.基于机器学习的键合丝退火性能预测研究[J].精密成形工程,2025,17(3):1-8.
MA Shuaijie,GUO Peng,LI Yuan,et al.Machine Learning-based Prediction of Bonding Wire Annealing Performance[J].Journal of Netshape Forming Engineering,2025,17(3):1-8. |
| 基于机器学习的键合丝退火性能预测研究 |
| Machine Learning-based Prediction of Bonding Wire Annealing Performance |
| 投稿时间：2025-01-20 |
| DOI：10.3969/j.issn.1674-6457.2025.03.001 |
| 中文关键词: 键合丝 退火 机器学习 拉断力 延伸率 |
| 英文关键词: bonding wires annealing machine learning tensile strength elongation |
| 基金项目:河南省重点研发专项(231111231500) |
|
| 摘要点击次数: 72 |
| 全文下载次数: 9 |
| 中文摘要: |
| 目的 通过机器学习方法优化键合丝的退火参数，调节其拉断力和延伸率，以适应电子封装中对键合丝性能的不同需求。方法 首先通过收集键合丝的生产试验数据构建退火参数数据集。其次，采用穷举法对特征进行筛选，以识别影响拉断力和延伸率的最优特征子集。使用交叉验证与网格搜索相结合的方法对机器学习模型的超参数进行优化，以提高预测准确性。在模型选择过程中，比较了多种机器学习算法的性能。结果 拉断力预测的最优特征子集包括丝径、退火前拉断力、退火温度和退火速度;而延伸率预测的最优特征子集则为丝径、退火前延伸率、退火温度和退火速度。支持向量回归(SVR)模型在拉断力预测中表现最佳，其最优超参数为{'C':400.0, 'gamma':1.0}，测试集上的均方根误差(RMSE)、平均绝对误差(MAE)和决定系数(R²)分别为1.12 mN、3.07 mN和0.982;随机森林(RF)模型在延伸率预测中表现优异，其最优超参数为{'min_samples_leaf':1, 'min_samples_split':2, 'n_estimators':5}，测试集上的均方根误差(RMSE)、平均绝对误差(MAE)和决定系数(R²)分别为0.551%、0.620%和0.972。结论 通过机器学习模型的建立与优化，成功实现了对键合丝退火参数的准确预测，所构建的模型满足了预期的精度要求，满足了预期的精度要求，为进一步提高键合丝的可靠性奠定了基础。 |
| 英文摘要: |
| The work aims to optimize the annealing parameters of bond wires by machine learning methods, and adjust their tensile strength and elongation, so as to meet the varying performance demands of bond wires in electronic packaging. First, a dataset of annealing parameters was constructed by collecting production test data from bond wires. Then, the exhaustive search method was employed to select features, to identify the optimal subset of features affecting the tensile strength and elongation. The hyperparameters of the machine learning models were optimized through combination of cross-validation and grid search to enhance prediction accuracy. During model selection, multiple machine learning algorithms were compared in terms of performance. The results revealed that the optimal feature subset for predicting tensile strength included wire diameter, tensile strength before annealing, annealing temperature, and annealing speed; while the optimal feature subset for predicting elongation consisted of wire diameter, elongation before annealing, annealing temperature, and annealing speed. The Support Vector Regression (SVR) model achieved the best performance in predicting tensile strength, with optimal hyperparameters of {'C':400.0, 'gamma':1.0} and a root mean square error (RMSE), mean absolute error (MAE), and coefficient of determination (R²) of 1.12 mN, 3.07 mN, and 0.982, respectively, on the test set. The Random Forest (RF) model excelled in predicting elongation, with optimal hyperparameters of {'min_samples_leaf':1, 'min_samples_split':2, 'n_estimators':5} and RMSE, MAE, and R² of 0.551%, 0.620%, and 0.972, respectively, on the test set. Through the establishment and optimization of machine learning models, this study successfully achieved accurate prediction of the annealing parameters for bond wires. The constructed models demonstrate good performance in predicting annealing properties, meeting the expected accuracy requirements and laying the foundation for further enhancing the reliability of bond wires. |
|
查看全文
查看/发表评论 下载PDF阅读器 |
| 关闭 |
|
|
|
渝公网安备 50010702501719号