Digital Twin-Enabled Predictive Analytics for Friction Stir Welding: Multi-Physics Simulation, Sensor Synchronization, and Defect Detection

Abstract Friction stir welding (FSW) is a solid-state joining process capable of producing high-quality, high-strength joints without material melting. However, the strong sensitivity of FSW to process parameters, material variations, and operational disturbances often leads to inconsistent weld quality, creating challenges for reliable industrial quality assurance. Although finite element simulations and experimental optimization studies have improved understanding of underlying thermo-mechanical behavior, most existing approaches remain offline and lack the ability to adapt to real-time process variations. This study presents a comprehensive digital twin framework for FSW by integrating multi-physics finite element simulations, real-time multi-sensor data acquisition, and machine-learning-based predictive analytics. The finite element model captures key physical phenomena, including heat transfer, material flow, and residual stress evolution during welding of AA6061-T6 aluminum alloy, providing a robust physics-based baseline. Real-time synchronization is achieved through thermal, force, and vibration sensor data, enabling continuous updating of the virtual model to reflect process disturbances such as tool wear and clamping variability. Machine learning models are embedded within the digital twin to enhance predictive capability. An artificial neural network predicts mechanical properties with a coefficient of determination (R²) exceeding 0.9, while a support vector machine classifier detects weld defects with accuracy above 92%. Experimental validation confirms that predicted tensile strength and hardness deviate by less than ±4% from measured values. The proposed framework enables real-time monitoring, predictive quality assurance, and proactive defect prevention, advancing friction stir welding toward intelligent, self-adaptive manufacturing aligned with Industry 4.0.