Contribution of deep reinforcement learning to solve reconfigurable facilities layout problems

Amine Chiboub, Julien Francois, Thècle Alix, Rémy Dupas,
Contribution of deep reinforcement learning to solve reconfigurable facilities layout problems,
Manufacturing Letters,
Volume 46,
2025,
Pages 16-20,
ISSN 2213-8463,
https://doi.org/10.1016/j.mfglet.2025.09.003.
(https://www.sciencedirect.com/science/article/pii/S2213846325002743)
Abstract: The Facilities Layout Problem involves arranging facilities within a given space to achieve specific objectives, such as minimizing transportation costs or reducing energy consumption. This issue arises in advanced manufacturing, particularly in Reconfigurable Manufacturing Systems (RMS), which allow layout adjustments based on changing product mixes, volumes, or processes. This paper compares the Double Dueling Deep Q-Network with traditional Q-learning and simulated annealing metaheuristic to assess the effectiveness of Deep Reinforcement Learning in addressing such challenges. Specifically, the study evaluates DDDQN performance in interactive environments where workstations are represented using a discrete approach, highlighting the role of reconfigurability in adjusting workstation implantation, orientation, and pickup/drop-off locations as required in RMS.
Keywords: Facilities Layout Problem; Reconfigurable Manufacturing Systems; Deep Reinforcement Learning; Q-learning; Simulated Annealing