A Large Language Model-based multi-agent manufacturing system for intelligent shopfloor

Zhen Zhao,Dunbing Tang,Haihua Zhu,Zequn Zhang,Kai Chen,Changchun Liu,Yuchen Ji

DOI: https://doi.org/10.48550/arXiv.2405.16887

2024-05-27

Abstract:As productivity advances, the demand of customers for multi-variety and small-batch production is increasing, thereby putting forward higher requirements for manufacturing systems. When production tasks frequent changes due to this demand, traditional manufacturing systems often cannot response promptly. The multi-agent manufacturing system is proposed to address this problem. However, because of technical limitations, the negotiation among agents in this kind of system is realized through predefined heuristic rules, which is not intelligent enough to deal with the multi-variety and small batch production. To this end, a Large Language Model-based (LLM-based) multi-agent manufacturing system for intelligent shopfloor is proposed in the present study. This system delineates the diverse agents and defines their collaborative methods. The roles of the agents encompass Machine Server Agent (MSA), Bid Inviter Agent (BIA), Bidder Agent (BA), Thinking Agent (TA), and Decision Agent (DA). Due to the support of LLMs, TA and DA acquire the ability of analyzing the shopfloor condition and choosing the most suitable machine, as opposed to executing a predefined program artificially. The negotiation between BAs and BIA is the most crucial step in connecting manufacturing resources. With the support of TA and DA, BIA will finalize the distribution of orders, relying on the information of each machine returned by BA. MSAs bears the responsibility for connecting the agents with the physical shopfloor. This system aims to distribute and transmit workpieces through the collaboration of the agents with these distinct roles, distinguishing it from other scheduling approaches. Comparative experiments were also conducted to validate the performance of this system.

Artificial Intelligence,Multiagent Systems,Robotics

What problem does this paper attempt to address?

The problem that this paper attempts to solve is the deficiencies of traditional manufacturing systems in meeting the demands of multi - variety and small - batch production. With the development of productivity, customers' demands for personalized products are increasing day by day, which leads to frequent changes in production tasks, and traditional manufacturing systems are often unable to respond in a timely manner in such situations. Specifically, the paper aims to solve the following problems:

1. **Insufficient flexibility of traditional manufacturing systems**:
- Traditional manufacturing systems rely on pre - defined heuristic rules for resource scheduling and task allocation. These rules are not intelligent enough and difficult to adapt to the dynamic changes of multi - variety and small - batch production.

2. **Limitations of existing scheduling methods**:
- Current scheduling methods (such as meta - heuristic algorithms and deep reinforcement learning algorithms) perform well in small - scale resource scheduling, but in large - scale manufacturing systems, their performance will decline significantly due to the increase in complexity and computing time.

3. **The negotiation mechanism in multi - agent manufacturing systems is not intelligent enough**:
- Although the existing multi - agent manufacturing systems provide a flexible architecture, the negotiation mechanism between their agents still relies on pre - defined heuristic rules and cannot make optimal decisions according to the real - time workshop situation.

To solve these problems, the paper proposes a multi - agent manufacturing system based on large language models (LLM) for intelligent workshops. This system enhances the intelligence of agents by introducing large language models, enabling them to select the most appropriate machines according to the real - time workshop situation, and negotiate and collaborate through natural language. The main contributions include:

- **Improving scheduling flexibility and intelligence**: By using large language models, the system can make intelligent decisions according to the current workshop situation without relying on pre - defined rules.
- **Reducing scheduling complexity and cost**: Compared with traditional AI methods, this system avoids the data collection and training process, reducing complexity and cost.
- **Directly controlling manufacturing resources**: Through the machine server agent (MSA), the system can directly interact with physical manufacturing resources to achieve automated production and task allocation.

Therefore, this research aims to improve the intelligence level of multi - agent manufacturing systems by introducing large language models to better meet the challenges of multi - variety and small - batch production.