Optimization of Rotary Friction Welding Parameters Through AI-Augmented Digital Twin Systems

In this study, Artificial Neural Networks (ANN) were employed to develop a Digital Twin (DT) of the Rotary Friction Welding (RFW) process. The neural network models were trained to predict the peak temperature generated during the welding process of dissimilar Ti Grade 2/AA 5005 joints over a temperature range of 20–640 °C. This prediction was based on a parametric numerical model of the RFW process constructed using the Finite Element Method (FEM) within the ADINA System software. Numerical simulations enabled a detailed analysis of the temperature distribution within the weldment. Accurate temperature predictions are essential for assessing the mechanical properties and microstructural integrity of the welded materials. Artificial Intelligence (AI) models, trained on historical data and real-time inputs, dynamically adjust critical process parameters—such as rotational speed, axial force, and friction time—to maintain optimal weld quality. A key advantage of employing AI-augmented DT systems in the RFW process is the ability to conduct real-time (less than 0.1 s) optimization and adaptive control. By integrating a Genetic Algorithm (GA) with the DT algorithm of the RFW process, the authors developed an effective tool for analyzing parameters such as axial force and rotational speed, in order to determine the optimal welding conditions, which translates into improved joint quality, minimized defects, and maximized process efficiency.