Supervised Machine learning models for predicting mechanical properties of dissimilar friction stir welded AA7075-AA5083 aluminum alloys

The study investigates the supervised machine learning regression-based method to find the ultimate tensile strength of friction stir welded dissimilar AA7075 (Al-Zn) and AA5083 (Al-Mg) joints. The machine learning (ML) models were trained using data from 20 experimental runs involving three input friction stir welding (FSW) process parameters like tool tilt angle 1 - 3 & ring;, welding speed 25---75 mm/min, and rotational speed 900---1300 rpm. Of these, 80 % of the dataset is used for training, and the remaining 20 % is considered for testing the models. Five ML algorithms, Decision Tree (DT), Random Forest, XGBoost, CatBoost, and AdaBoost, were evaluated using dataset 20 of experimental runs to assess their predictive capabilities. The DT model showed outstanding performance, achieving a mean absolute error (MAE) of 14.75, a mean square error (MSE) of 408.63, and an R2 score of 0.97. This means that 97 % of the variance in the ultimate tensile strength (UTS) values was accurately represented by the model, highlighting its robustness and reliability in predictions. The study demonstrates that ML models can effectively predict the UTS with accuracy. The results indicate that the predictive model marks a significant advancement in this field. It allows for more accurate and dependable predictions of essential welding process parameters, which greatly improves the efficiency and quality of welding operations.