Intelligent regulation of laser welding deviation for skin-frame structures based on soft voting ensemble learning and PSO-optimized fuzzy control

This paper addresses the challenge of welding deviation control in laser welding of skin-frame structures by proposing an intelligent control system that integrates a soft voting ensemble learning model with a particle swarm optimization (PSO)-tuned fuzzy controller. The system employs high-speed imaging to capture weld pool images in real-time, from which geometric features are extracted. An ensemble model combining Support Vector Machines (SVM), XGBoost, and Random Forest via a soft voting mechanism accurately predicts both the direction and magnitude of welding deviations, achieving an AUC of 0.94 for classification and a Mean Squared Error (MSE) of 0.1887 for regression. A fuzzy logic controller, whose membership functions are automatically optimized by the PSO algorithm, enables real-time, precise adjustment of the welding process. Experimental results demonstrated the system's effectiveness, maintaining precise alignment across 12 consecutive circumferential welds without observable deviation or missed welds. Simulation tests further verified the controller's robustness and adaptability, confirming its ability to converge rapidly even under significant initial deviations. This study presents a high-precision, robust intelligent control solution for automated precision welding of complex structures.