Sensor-based monitoring and machine learning modelling for predicting the mechanical properties of friction stir welded Al 6061-Cu 11000 dissimilar joints

The engineering properties of dissimilar metals like Aluminum and Copper have led to their increased use of Frictional Stir Welding (FSW) of both materials in electrical, automotive, and energy-storage applications. However, it is still a challenge to obtain good-quality FSW joints with good mechanical properties because of the complicated thermal and mechanical interactions that occur during the welding process. This study presents a low-cost sensor-based monitoring and machine-learning framework for the prediction of parameters for FSW of dissimilar metals. The integration of real-time sensor-based monitoring for data acquisition enabled the development of machine learning models for predicting the mechanical properties of FSW Al 6061-Cu 11000 joints. A modified milling machine equipped with an integrated Arduino-based acquisition system was developed to continuously record key welding parameters, including rotational speed, travel speed, plunge depth, tilt angle, axial force, torque, and peak temperature. Experimental welding trials were conducted to generate the dataset linking process parameters with ultimate yield strength (YS), tensile strength (UTS), and hardness (HV). To capture the changing dynamics of the system, four machine learning models (linear regression, random forest, gradient boosting, and XGBoost) were used to fit the dataset for mechanical property prediction that maps the input and output variables. The predictive model with the best performance had a root mean square error (RMSE) of 4.865 MPa and an R 2 value of 0.744, indicating a strong prediction capacity for UTS. The forecasts for yield strength and hardness had R 2 values of 0.384 and 0.478, respectively. Among the predictors, peak temperature was found to be the most significant factor influencing weld performance in the feature importance analysis, underscoring the crucial role that thermal management plays in dissimilar Al-Cu FSW. This outcome provides a basis for future intelligent process optimization and real-time adaptive control of friction stir welding operations by demonstrating the viability of both real-time sensor-based monitoring and machine learning modeling of FSW parameters for the prediction of quality assessment. The novelty of the study lies in combining multi-sensor data acquisition towards the prediction of dissimilar aluminum-copper FSW parameters, while explicitly evaluating the extent to which process-level and sensor-derived variables can explain properties having different levels of microstructural sensitivity. The framework provides a basis for future online quality estimation and adaptive control, although the predictions in the present study were performed offline.