A smart manufacturing paradigm for robotic welding process optimization through machine learning

Area of Science:

• Robotics
• Manufacturing Engineering
• Artificial Intelligence

Background:

• Robotic welding is crucial for structural integrity across industries.
• Conventional methods struggle with workpiece location and trajectory accuracy, limiting optimal welding outcomes.
• Advanced optimization methodologies are needed to address these limitations.

Purpose of the Study:

• To present a smart manufacturing paradigm using Reinforcement Learning for Robotic Welding Process Optimization (RL-RWPO).
• To enhance the accuracy, efficiency, and overall quality of robotic welding processes.
• To integrate Vision-Based and CAD-Based methodologies for a comprehensive understanding of workpiece placement and arrangement.

Main Methods:

• Developed the RL-RWPO framework combining Vision-Based and CAD-Based approaches.
• Implemented 3-D calibration and workpiece placement techniques for improved precision.
• Utilized Reinforcement Learning for trajectory optimization in robotic welding operations.

Main Results:

• Achieved an average Weld Seam Quality of 93.27%.
• Reached a Welding Speed of 50.51 mm/s and a Weld Penetration Depth of 5.25 mm.
• Demonstrated Trajectory Accuracy of 3.04 mm and Welding Efficiency of 91.52%.

Conclusions:

• The RL-RWPO method significantly enhances robotic welding effectiveness and excellence.
• The integrated approach improves precision through 3-D calibration and workpiece placement.
• This framework offers a viable solution for optimizing robotic welding in various industrial sectors.