Machine learning approach with various regression models for predicting the ultimate tensile strength of the friction stir welded AA 2050-T8 joints by the K-Fold cross-validation method

This research work aims to develop the friction stir welding (FSW) technique to resolve the solidification issues when joining the AA2050-T8. The identification of the right process parameters is a major task in the FSW process. Consequently, the experimentation was carried out to optimize the process parameters for the AA2050 FSW joint. For this, an empirical relationship was developed using a statistical tool such as the design of experiments (DOE) with analysis of variance (ANOVA) to optimize the ultimate tensile strength (UTS) with the highest level of confidence. The desirable process parameter was identified to achieve the maximum UTS from the Response Surface Methodology through numerical optimization. The welding was conducted using the optimized process parameters and the tensile test was carried out. The obtained results (average UTS is 371 MPa) were extremely close to the optimized maximum UTS (378 MPa). Furthermore, the same input and output responses from DOE were utilized as input parameters for the machine learning (ML) approach. To predict the UTS based on process parameters, various regression models were used in the ML approach. As a result, all the regression models show very small variations except the polynomial regression model. So, the K-Fold crossvalidation method was implemented to achieve the highest accuracy of these regression models. The result, compared to other regression models, the Random Forest regression model is highly suitable to predict UTS for AA2050 FSW joints.