Optimizing and development of friction stir welding using AI-supported prediction method and digital twin technology

Abstract This study proposes an integrated approach for optimizing the friction stir welding (FSW) process by combining sensor-based data acquisition, machine learning (ML), and digital twin (DT) technologies. Real-time sensor data including rotational speed, welding speed, axial force, torque, and temperature were collected during FSW operations. These parameters were correlated with weld quality indicators, such as surface appearance, internal defects, and tensile strength. A dataset of 132 weld samples was used to train supervised and unsupervised ML models, achieving a defect classification accuracy of 95%. In parallel, a COMSOL-based digital twin was developed to simulate thermo-mechanical aspects of the welding process. The model incorporated temperature-dependent material properties, frictional heat generation, and plastic deformation behavior to predict stress, strain, and temperature distributions. Model predictions were validated against experimental sensor data, confirming accuracy in peak temperature and torque estimation. The integrated ML-DT system functioned as a decision-support tool, enabling real-time process monitoring, virtual experimentation, and predictive defect detection. When implemented in an industrial environment, the system dynamically adapted welding parameters to maintain optimal conditions. This approach enhances process stability, reduces material waste, and improves weld integrity, offering a scalable solution for intelligent manufacturing and Industry 4.0 applications.