An intelligent agent-based architecture for flexible manufacturing systems having error recovery capability

This research proposes a hybrid intelligent agent-based architecture for manufacturing systems focusing at the workstation level. Chapter 2 describes a modeling approach for the architecture based on Color Petri Nets. Each agent is modeled with a subnet which represents the agent's methods. Agents are linked together and with the “environment” or physical system through communication channels also represented by Petri Nets.; Chapter 3 introduces extensions to previous work on modeling the physical system with Petri Nets. In particular, a new class of Petri Net termed as Sequence of Activities with Resources (SSAR) is proposed. In addition to serving as a modeling framework, the SSAR nets serve as a platform for optimization. A mathematical formulation and the structural properties of SSAR nets are also discussed.; Chapter 4 introduces an algorithm for optimization based on the classic A* search. The algorithm, called Zero-Backtracking A* Search (ZBAS), includes features that increase the speed of the search without a significant decrease of the solution quality. The ZBAS algorithm can be set to minimize either flowtime or mean tardiness. Extensive testing on the ZBAS algorithm is presented in chapter 5. The experiments show the validity of the approach when compared against optimal solutions, the Beam Search algorithm and dispatching rules.; Chapter 6 introduces the dynamic addition error recovery trajectories to the control logic of the system. Error recovery trajectories are modeled with Petri Nets and incorporated into the model of the physical system. Issues such as state explosion and deadlocks are discussed.; Chapter 7 introduces an extension of ZBAS (named ZBAS2) to perform “match-up” re-scheduling on systems modeled with Petri Nets. The purpose here is to study varying levels of re-scheduling ranging from taking no action to generating completely new schedules. The performance of ZBAS2 is tested through simulations in which errors are randomly introduced and recovered. An ANOVA analysis and pairwise comparisons show that higher levels of re-scheduling yield better results but at the expense of significantly longer computational times.