Multi-agent cooperative swarm learning for dynamic layout optimisation of reconfigurable robotic assembly cells based on digital twin
Likun Wang1, Zi Wang1, Kevin Gumma1
1Centre for Aerospace Manufacturing, University of Nottingham, Advanced Manufacturing Building, Nottingham, Nottinghamshire NG7 2GX UK.
Related Experiment Videos
View abstract on PubMed
Summary
This study introduces a new multi-agent learning framework for optimizing the layout of reconfigurable robotic assembly cells. The system uses 3D digital twins to significantly improve layout compactness, reduce production time, and lower rearrangement costs in aerospace manufacturing.
Area of Science:
- Robotics and Automation
- Manufacturing Systems Engineering
- Artificial Intelligence in Manufacturing
Background:
- Reconfigurable manufacturing systems (RMS) are essential for meeting demands for product variety and short production cycles.
- Dynamic factory layout optimization is critical for adapting RMS to changing mechanical structures, software/hardware integration, and production capabilities.
- Current RMS layout design often simplifies autonomous devices to 2D shapes, leading to approximate solutions for issues like overlapping and transportation distance.
Purpose of the Study:
- To present a novel multi-agent cooperative swarm learning framework for dynamic layout optimization of reconfigurable robotic assembly cells.
- To utilize a 3D digital twin representation for more accurate facility modeling in the optimization process.
- To address layout challenges using a decentralized multi-agent approach instead of traditional centralized learning.
Main Methods:
- Development of a multi-agent cooperative swarm learning framework.
- Establishment of a digital twin in a learning environment (Visual Components and TWINCAT) using 3D facility models.
- Implementation of a decentralized learning approach for layout optimization.
Main Results:
- In aerospace use cases, layout compactness was reduced by 3.8 times, simulated production time by 2.3 times, and rearrangement cost by 33.4%.
- Ensured feasible robot paths without joint limits, reachability, or singularity issues for all manufacturing activities.
- Demonstrated flexibility in adjusting learning objectives (compactness, cost, time) by modifying weight parameters.
Conclusions:
- The proposed framework offers an effective solution for dynamic layout optimization in reconfigurable robotic assembly cells.
- The use of 3D digital twins and multi-agent systems enhances accuracy and flexibility in layout design.
- The framework is validated through aerospace manufacturing use cases, showing significant improvements in key performance indicators.