|M.Sc Student||Mitnovitsky Michael|
|Subject||Operation and Control of Manufacturing Systems by Agents|
with Local Intelligence
|Department||Department of Mechanical Engineering||Supervisors||Professor Emeritus Moshe Shpitalni|
|Ms. Miri Weiss Cohen|
|Full Thesis text|
Optimizing manufacturing systems operations can be converted into optimizing a job shop problem which is known to be a difficult and complex problem. This thesis examines a dynamic, flexible and stochastic job shop problem that considers random events, such as random job arrivals, uncertain processing times, unexpected machine break downs, various shop utilization levels and the possibility of processing flexibility. In addition, it studies the influence of various parameters over the shop performance.
To achieve this goal, a new agent-based adaptive control system has been developed. The basic concept of the proposed algorithm is that each job and each resource of the shop is accompanied by an agent. Those agents make the decisions regarding the navigation of their jobs during the process in the shop and about the usage of their resources. The agents make their decisions in real time based on the available data (local optimization). Each agent’s goal is to minimize the cost of the job under its supervision. The behavior of the above mentioned agents is supervised by a higher brain mechanism, which communicates with the agents, and has the knowledge about each of the agents’ location and current assignment. The higher brain is also responsible of limiting the agents’ degree of freedom in order to ensure the preservation of the shop’s global agenda over each agent’s egoistic agenda.
The developed system is an advanced decision-making environment with built in strategies for responsive factories. The system conveys adaptation and reconfiguration capabilities and advanced complementary scheduling. It facilitates operational flexibility and increases productivity, as well as offering strategic advantages such as analysis of factory development by simulation.
To demonstrate the algorithm’s performance a virtual factory was created and serves as the working environment for the algorithm’s simulation process. Simulations were conducted under various experimental settings such as shop utilization levels, due date tightness, machine breakdown level, various process flexibility routes and more.
As a result of the simulation, data was gathered regarding the performance of the factory and the decision making process. The data was analyzed and used for evaluation of the quality of the algorithm’s decision making process. The results conclusively demonstrate that the algorithm yields very good and qualitative solutions in all the tested scenarios. In addition, the combination of the developed simulation with the algorithm also provides the factory management with valuable administrative tools that can serve for evaluating various development and efficiency alternatives that can dramatically enhance the factory’s profitability.