Sagar S. Kamarthi
Date of Award
Master of Science
Department or Academic Unit
College of Engineering. Department of Mechanical and Industrial Engineering.
Disassembly, End of life products, Multi-Kanban System, Reinforcement Learning
Computer Engineering | Engineering | Mechanical Engineering
Disassembly line is the best way to disassemble products with similar components in large quantity. Similar to an assembly line, a disassembly line consists of a series of workstations. Factors such as multiple demand arrivals, multiple arrivals of EOL products, different state of EOL products, fluctuations in the inventory and levels make a disassembly line very complex and difficult to control. Earlier researchers have proposed a Multi-Kanban mechanism for a disassembly line, which involves dynamic routing of the Kanbans to various workstations. This research focuses on applying reinforcement learning to the disassembly environment.
We have designed the reinforcement learning agent with 19 state variables - 5 component buffers, 9 assembly buffers and 5 demand variables. The agent selects its action from 14 possible actions which are related to either pulling an assembly for disassembly or pulling a component for satisfying a demand. The agent then receives a scalar reward based on the selected action and the current state of the system. The weights of the neural network are then updated. In this way, the neural network gets trained as the simulation progresses. We observed that the number of demands satisfied by the intelligent system is greater than that satisfied by the Multi-Kanban System. In addition work-in-process inventory in case of the intelligent system is much smaller than in case of the Multi-Kanban System.
Harshawardhan Mukund Kulkarni
Kulkarni, Harshawardhan Mukund, "Reinforcement learning approach for disassembly" (2010). Computer Systems Engineering Master's Theses. Paper 5. http://hdl.handle.net/2047/d20002057
Click button above to open, or right-click to save.COinS