Scholarlink

<p>The study aims to determine how input parameters, such as shoulder plunge depth or penetration depth (0.1–0.4&#xa0;mm) and dwelling time (5–15&#xa0;s), affect mechanical properties in friction stir spot welding (FSSW) products. Various machine learning (ML) models, including Linear Regression, Decision Tree, Random Forest, SVM, and XGBoost, were used to estimate optimal ranges of input parameters. The current interpretability research assesses the model’s dependability to find the maximum shear load of FSSW finished joints. The XGBoost model outperformed all other ML models, particularly after hyperparameter adjustment, with an R<sup>2</sup> score of 0.96. To provide a mathematical representation with uncertainty quantification, a quadratic trend-based Gaussian Process Regression model with Bayesian optimization was developed and validated by Leave-One-Out cross-validation (LOO-CV), yielding a generalisation accuracy of R<sup>2</sup> = 0.78. Additionally, the SHAP analysis revealed that the effect of shoulder plunge depth and dwell time improved model interpretability and supported the experimental outcomes. In the current investigation, the maximum shear load was attained at 4.63 kN, with an optimum value of shoulder plunge depth of 0.3852&#xa0;mm and a dwelling time of 12.32&#xa0;s. The proposed approach not only captures the physical behaviour of the process but also reveals regions of high uncertainty and can therefore be used as a reliable tool to guide future optimisation and experimentation in FSSW.</p> Graphical abstract <p></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

A machine learning predictive model on friction stir spot welding: changing plunge depth and dwelling time

  • 期刊论文

  • 发表日期：

    2026年5月22日

查看全文
  • Suprim Sardar,
  • Sutanu Misra,
  • Partha Pratim Dey

摘要

The study aims to determine how input parameters, such as shoulder plunge depth or penetration depth (0.1–0.4 mm) and dwelling time (5–15 s), affect mechanical properties in friction stir spot welding (FSSW) products. Various machine learning (ML) models, including Linear Regression, Decision Tree, Random Forest, SVM, and XGBoost, were used to estimate optimal ranges of input parameters. The current interpretability research assesses the model’s dependability to find the maximum shear load of FSSW finished joints. The XGBoost model outperformed all other ML models, particularly after hyperparameter adjustment, with an R2 score of 0.96. To provide a mathematical representation with uncertainty quantification, a quadratic trend-based Gaussian Process Regression model with Bayesian optimization was developed and validated by Leave-One-Out cross-validation (LOO-CV), yielding a generalisation accuracy of R2 = 0.78. Additionally, the SHAP analysis revealed that the effect of shoulder plunge depth and dwell time improved model interpretability and supported the experimental outcomes. In the current investigation, the maximum shear load was attained at 4.63 kN, with an optimum value of shoulder plunge depth of 0.3852 mm and a dwelling time of 12.32 s. The proposed approach not only captures the physical behaviour of the process but also reveals regions of high uncertainty and can therefore be used as a reliable tool to guide future optimisation and experimentation in FSSW.

Graphical abstract