Reinforcement Learning for Disassembly Task Control

doi:10.1016/j.cie.2024.110044

Reinforcement Learning for Disassembly Task Control

2024-04-01

0

OA

AI

DOI：10.1016/j.cie.2024.110044

Original

Save

View PDF

Abstract

En 中文

Inventory management is an integral part of systems that assemble or disassemble products. However, unlike assembly operations, disassembly operations can accumulate soaring work-in-process inventory if not carefully managed. Inventory management in disassembly systems is particularly challenging due to unexpected fluctuations in the inventory levels of subassemblies and salvage components caused by factors such as multiple demand arrivals, multiple core (used product or part) arrivals, uncertainty in core and demand arrivals, uncertainty in core condition, and varying processing times. Complexity and complications in disassembly systems lead to high operating costs and customer dissatisfaction, which may ultimately discourage remanufacturing. This article proposes an innovative approach employing Deep Reinforcement Learning (DRL) to control tasks in disassembly systems with the aforementioned complexities and complications. This approach presents a viable alternative to conventional disassembly lot sizing methods. The effectiveness of the approach is explored through experiments conducted on disassembly systems for Quantum -dot LED (QLED), Organic LED (OLED), and Quantum Dot OLED (QD-OLED) TVs. We analyze the potential impacts of the RL-driven policies on inventory accumulation and unfulfilled demand in the disassembly systems and compare them with the Multiple Elman Neural Networks (MENN) mechanism of controlling disassembly tasks. Further, we illustrate the generalizability of the proposed approach across three Markov Decision Process (MDP) configurations under Poisson arrivals, demands, product obsolesces, and processing rates. Simulation results indicate a 21% reduction in inventory accumulation, a 12% reduction in unfulfilled demand for less complex setups, and substantial performance improvements in intricate setups compared to the performance of MENN. This work bolsters the concept of remanufacturing by controlling disassembly tasks to make component recovery from EOL products cost-efficient.

Keywords:

Disassembly

Remanufacturing

Lot sizing

Machine learning

Reinforcement learning

Reinforcement Learning for Disassembly Task Control

Abstract

AI Summary

Journal

Organization

Cited Papers