Metal-Metal Bonding Process Research Based on Xgboost Machine Learning Algorithm

Jingpeng Feng; Lihua Zhan; Bolin Ma; Hao Zhou; Bang Xiong; Jinzhan Guo; Yunni Xia; Shengmeng Hui

doi:10.3390/polym15204085

Metal-Metal Bonding Process Research Based on Xgboost Machine Learning Algorithm

Polymers (Basel). 2023 Oct 14;15(20):4085. doi: 10.3390/polym15204085.

Authors

Jingpeng Feng^{1

2}, Lihua Zhan^{1

2}, Bolin Ma^{1

2}, Hao Zhou^{1

2}, Bang Xiong^{1

2}, Jinzhan Guo^{1

2}, Yunni Xia^{1

2}, Shengmeng Hui^{1

2}

Affiliations

¹ State Key Laboratory of Precision Manufacturing for Extreme Service Performance, Central South University, Changsha 410083, China.
² Light Alloys Research Institute, Central South University, Changsha 410083, China.

Abstract

Conventionally, the optimization of bonding process parameters requires multi-parameter repetitive experiments, the processing of data, and the characterization of complex relationships between process parameters, and performance must be achieved with the help of new technologies. This work focused on improving metal-metal bonding performance by applying SLJ experiments, finite element models (FEMs), and the Xgboost machine learning (ML) algorithm. The importance ranking of process parameters on tensile-shear strength (TSS) was evaluated with the interpretation toolkit SHAP (Shapley additive explanations) and it optimized reasonable bonding process parameters. The validity of the FEM was verified using SLJ experiments. The Xgboost models with 70 runs can achieve better prediction results. According to the degree of influence, the process parameters affecting the TSS ranked from high to low are roughness, adhesive layer thickness, and lap length, and the corresponding optimized values were 0.89 μm, 0.1 mm, and 27 mm, respectively. The experimentally measured TSS values increased by 14% from the optimized process parameters via the Xgboost model. ML methods provide a more accurate and intuitive understanding of process parameters on TSS.

Keywords: Xgboost machine learning algorithm; finite element models; interpretation toolkit SHAP; process parameter optimization; single-lap joints.

Grants and funding

This work is supported by the National Natural Science Foundation of China (Grant No. 52175373, 52005516), the National Key Research and Development Program (Grant No. 2018YFA0702800), the Project of State Key Laboratory of Precision Manufacturing for Extreme Service Performance, Central South University (No. ZZYJKT2021-03).