Analysis of Welding Process Path Planning and Process Simulation Using Digital Twins and Reinforcement Learning

To address challenges in complex non-standard welding scenarios within shipbuilding and offshore equipment manufacturing—such as path planning reliant on experience, insufficient dynamic response capabilities, and a disconnect between simulation and physical processes—this paper proposes an intelligent welding process path planning and simulation method integrating digital twin (DT) technology and reinforcement learning. By constructing a high-fidelity DT model of the welding process, real-time perception and prediction of groove geometry, thermal deformation, and equipment status are achieved. A reinforcement learning controller based on the twin delayed deep deterministic policy gradient algorithm autonomously learns optimal welding paths and process parameters within the DT environment, enabling multi-objective optimization of weld bead quality, efficiency, and equipment wear. Experimental results demonstrate that this method achieves a 98.5% first-pass weld acceptance rate under V-groove abrupt transition conditions, reduces process development time by over 60%, and exhibits significant advantages in quality.