Control of manufacturing systems has been studied in a previous CVDI project. However, this work assumed that information was shared hierarchically but not between individual entities (e.g. machines, systems, lines).
Previous attempts to model large systems, such as manufacturing systems, can suffer from explosion of action and state space or fails to capture the dependencies among individual elements in the system.
This project will study and characterize improvement to automatic control of manufacturing systems when information, such as queues and health estimates, is shared.
Graphs can be used to model many systems such as social networks, traffic networks, and manufacturing systems. This project will model the manufacturing system as a graph.
(Graph Neural Networks) GNNs are a recently developed method of encoding information in graphs.
GNNs’ message passing between the modes of graphs captures the dependence of nodes. This provide us the potential to use GNNs as a framework for cooperative multi-agent RL.
We propose to investigate the use of GNNs to form a multi-agents reinforcement learning framework, provide insights to control large-scale manufacturing systems with Internet of Things capabilities.