Optimization of Welding Parameters Using an Improved Hill-Climbing Algorithm Based on BP Neural Network for...

Area of Science:

• Materials Science and Engineering
• Artificial Intelligence in Manufacturing
• Robotics and Automation

Background:

• Achieving uniform weld surfaces is critical for mechanical performance in multi-pass welding.
• Complex nonlinear relationships between welding parameters and weld bead geometry challenge traditional optimization.
• Automated multi-bead overlay welding requires precise control for surface quality.

Purpose of the Study:

• To develop an intelligent framework for predicting and optimizing weld surface smoothness.
• To integrate a backpropagation (BP) neural network with an improved hill-climbing algorithm.
• To enhance automated multi-bead overlay welding processes.

Main Methods:

• Trained a BP neural network using experimental data on arc voltage, wire feed rate, and welding speed.
• Employed an improved hill-climbing algorithm to adaptively adjust BP model weights and biases.
• Validated the framework against conventional BP approaches for prediction accuracy and convergence.

Main Results:

• The proposed intelligent framework demonstrated significantly higher prediction accuracy and convergence efficiency compared to conventional BP methods.
• Optimal welding parameters identified by the model resulted in demonstrably smoother weld surfaces.
• Reduced need for post-weld surface finishing operations.

Conclusions:

• The integrated BP neural network and improved hill-climbing algorithm offer a novel and effective solution for weld surface optimization.
• This intelligent approach facilitates real-time control and optimization in advanced welding systems.
• The study advances automated welding by improving surface quality and reducing processing time.