DMOPP: A Deformation-Informed Multi-Objective Path Planning Method for Multi-Seam Robotic Welding

Abstract

Robotic multi-seam welding path planning for box-type thin-walled structures faces significant challenges due to multiple constraints, multiple objectives, and its high-dimensional discrete combinatorial nature. To address these issues, a deformation-informed multi-objective path planning method for multi-seam welding (DMOPP) is proposed, comprising a multi-objective formulation and an optimization algorithm. At the modeling level, welding path length and maximum structural deformation are defined as the optimization objectives. Collision-free path planning is used to calculate the path length, while an XGBoost-based surrogate model is developed to establish the mapping between welding path variables and maximum deformation, enabling rapid deformation prediction without computationally expensive finite element simulations. At the optimization level, a modified discrete artificial lemming algorithm (MODALA) is proposed to improve global search capability and convergence stability. Experimental results show that MODALA outperforms the comparison algorithms on benchmark functions and discrete optimization problems, demonstrating its superior optimization performance. The XGBoost surrogate model achieves an R2 of 0.8447 and an RMSE of 0.0841 mm, indicating good predictive accuracy. In welding path planning simulations, the proposed method effectively reduces the path length and welding deformation while achieving a favorable trade-off between path efficiency and structural quality. Robotic welding experiments further validate its practical effectiveness.