A robot scheduling method based on rMAPPO for H-beam riveting and welding work cell.

Area of Science:

• Robotics and Automation
• Artificial Intelligence
• Manufacturing Engineering

Background:

• Automated H-beam processing involves complex multi-agent coordination for tasks like riveting and welding stiffener plates.
• Existing H-beam riveting and welding work cells have untapped potential for productivity improvements within intelligent manufacturing frameworks.

Purpose of the Study:

• To address the multi-agent scheduling problem in H-beam processing using a novel reinforcement learning algorithm.
• To enhance the manufacturing efficiency and reduce operational delays in automated H-beam work cells.

Main Methods:

• Development of a recurrent multi-agent proximal policy optimization (rMAPPO) algorithm incorporating recurrent neural networks and action masking.
• Creation of a reinforcement learning environment for multi-agent scheduling of H-beam processing tasks.
• Validation of the rMAPPO algorithm on a physical work cell and its digital twin, comparing it against baseline multi-agent reinforcement learning methods.

Main Results:

• The rMAPPO-based scheduling method effectively reduces robot waiting times compared to baseline methods (MAPPO, MADDPG, MASAC).
• Demonstrated enhanced adaptability in managing diverse riveting and welding tasks.
• Achieved significant improvements in the overall manufacturing efficiency of stiffened H-beams.

Conclusions:

• The proposed rMAPPO algorithm offers a robust solution for multi-agent scheduling in automated H-beam manufacturing.
• This approach significantly enhances intelligent manufacturing by optimizing robotic task coordination and improving production efficiency.
• The method's effectiveness is validated on both physical and simulated platforms, highlighting its practical applicability.