Abstract
In multi-pass welding processes, achieving a uniform and smooth weld surface is crucial for mechanical performance and dimensional accuracy. However, the complex nonlinear relationships between welding parameters and weld bead geometry present significant challenges for traditional optimization methods. This study proposes an intelligent prediction and optimization framework that integrates a backpropagation (BP) neural network with an improved hill-climbing algorithm to enhance weld surface smoothness in automated multi-bead overlay welding. Experimental data collected under varying arc voltages, wire feed rates, and welding speeds were used to train the neural network. The improved hill-climbing algorithm adaptively adjusts weights and biases in the BP model to overcome issues of local minima and slow convergence. Comparative results demonstrate that the proposed method significantly outperforms conventional BP approaches in terms of prediction accuracy and convergence efficiency. Furthermore, optimal welding parameters identified by the model yield smoother weld surfaces, reducing the need for post-processing. This work provides a novel solution for intelligent control and real-time optimization in advanced welding systems.