Intelligent Product Agents for Multi-Agent Control of Manufacturing Systems

摘要

The current manufacturing paradigm is shifting toward more flexible manufacturing systems that produce highly personalized products, adapt to unexpected disturbances in the system, and readily integrate new manufacturing system technology. However, to achieve this type of flexibility, new system-level control strategies must be developed, tested, and integrated to coordinate the components on the shop floor. One strategy that has been previously proposed to coordinate the resources and parts in a manufacturing system is multi-agent control. The manufacturing multi-agent control strategy consists of agents that interface with the various components on the shop floor and continuously interact with each other to drive the behavior of the manufacturing system. Two of the most important decision-making agents for this type of control strategy are product agents and resource agents. A product agent represents a single product and a resource agent represents a single resource on the plant floor. The objective of a product agent is to make decisions for an individual product and request operations from the resource agents based on manufacturer and customer specifications. A resource agent is the high-level controller for a resource on the shop floor (e.g., machines, material-handling robots, etc.). A resource agent communicates with other product and resource agents in the system, fulfills product agent requests, and interfaces with the associated resource on the plant floor. While both product agents and resource agents are important to ensure effective performance of the manufacturing system, the work presented in this dissertation improves the intelligence and capabilities of product agents by providing a standardized product agent architecture, models to capture the dynamics and constraints of the manufacturing environment, and methods to make improved decisions in a dynamic system. New methods to explore the manufacturing system and cooperate with other agents in the system are provided. The proposed architecture, models, and methods are tested in a simulated manufacturing environment and in several manufacturing testbeds with physical components. The results of these experiments showcase the improved flexibility and adaptability of this approach. In these experiments, the model-based product agent effectively makes decisions to meet its production requirements, while responding to unexpected disturbances in the system, such as machine failures or new customer orders. The model-based product agent proposed in this dissertation pushes the fields of manufacturing and system-level control closer to realizing the goals of increased personalized production and improved manufacturing system flexibility.