THE DESIGN OF A DISTRIBUTED KNOWLEDGE-BASED SYSTEM FOR THE INTELLIGENT MANUFACTURING INFORMATION SYSTEM (ARTIFICIAL INTELLIGENCE, PLANNING, OPERATIONS MANAGEMENT)
Abstract
This thesis incorporates planning and distributed problem solving into the design of manufacturing systems. Such an information system is characterized by the hierarchical organization and distributed control. The environment consists of a group of manufacturing cells; each cell uses a knowledge-based planning system to manage the jobs within the cell while interacting with other cells through the communication network. First, a framework for using the knowledge-based method to perform the planning and control of manufacturing jobs is developed. The dynamic environment is represented by a world model, the operations are represented by state-changing transformations, and manufacturing steps are derived by the plan generation. The coordination of concurrent activities and the management of shared resources are emphasized by including the duration and the resource in the planning formalism. Plans are constructed in three steps. First, a linearly-sequenced plan is generated for each job independently by a search procedure. In the second step, a plan generator is used to establish necessary precedence relationships between operations by performing look ahead and avoiding any conflicts. Conflict detection can be achieved either by a "critic" mechanism or a "reasoning about resource" mechanism. In step three, a plan-revision scheme is used to improve the plan so that the final plan has the shortest duration. The planning system can function as a scheduler in a manufacturing cell, characterized by being goal-directed, event-driven, and able to perform both static and dynamic types of on-line scheduling. Because the control is decentralized, the whole information system can be viewed as a society of experts and each cell is a problem solving agent with predefined expertise. A negotiation protocol is used to regulate communication and task allocation among cells. This thesis attempts to use the market structure to organize the information system: tasks are viewed as commodities and cells as the bidders with varying preferences. The augmented Petri net model is used to represent the negotiation protocol and is implemented in a controlled rule-based system. The execution of the negotiation protocol is accordingly accomplished by an inference procedure in the rule-based system. Thus, this thesis in effect has adopted a unified approach to the planning process and the allocation process.
Degree
Ph.D.
Subject Area
Management
Off-Campus Purdue Users:
To access this dissertation, please log in to our
proxy server.