Performance evaluation of friction stir welding using machine learning approaches.

Area of Science:

• Materials Science
• Mechanical Engineering
• Computational Science

Background:

• Friction stir welding (FSW) is a critical joining process for various metals.
• Accurate prediction of ultimate tensile strength (UTS) is essential for ensuring the integrity of FSW joints.
• Evaluating advanced machine learning models can enhance predictive capabilities.

Purpose of the Study:

• To assess the efficacy of Gaussian process regression (GPR), support vector machining (SVM), and multi-linear regression (MLR) for predicting the UTS of FSW joints.
• To compare the performance of these machine learning models using experimental data.
• To identify the most suitable regression model for UTS prediction in FSW.

Main Methods:

• Developed three regression models based on GPR, SVM, and MLR.
• Utilized rotational speed and feed rate as input parameters, with UTS as the output.
• Employed Pearson VII (PUK) and radial basis kernel functions (RBF) for GPR and SVM models.
• Trained models on 19 experimental readings and tested on the remaining 6.

Main Results:

• GPR demonstrated superior performance compared to SVM and MLR in predicting UTS.
• The models were evaluated based on their ability to predict experimental outcomes.
• GPR successfully predicted the UTS of friction stir welded joints with high accuracy.

Conclusions:

• Gaussian process regression (GPR) is the most effective method among the evaluated techniques for predicting the UTS of friction stir welded joints.
• The GPR approach provides a reliable tool for material property prediction in FSW.
• This study highlights the potential of advanced machine learning in optimizing welding processes.