Automatic Facility Layout Design Using Reinforcement Learning and a Analytic Hierarchy Process

Abstract

The authors propose a novel solution to the facility layout problem (FLP). Facility layout aims to deploy a set of objects in logistics facilities and production facilities. The time to create layouts and determine the efficiency of the design layouts created significantly depends on a designer's ability, as consideration must be given to lead time, relations among functional objects to deploy, and material handling costs. Because the computational complexity of FLP is NP-hard, meta-heuristic methods that classify the problem and find an approximate solution are the major solutions. A layout design that reduces transportation processes and wasted space is very difficult to create. In this study, an automatic design system is proposed that can design relationship-oriented layouts for various unit groups and sites using reinforcement learning and the analytic hierarchy process (AHP). Reinforcement learning is a machine-learning method that realizes optimal system control through trial and error by the system itself. AHP is one of the multicriteria decision methods that create a hierarchy of options for the decision divided into levels. Reinforcement learning is used to create a near optimal layout. AHP is used to consider relationships among the units when the agents deploy in the site. The results of applying the system to benchmark problems show that the system can produce relationship-oriented layouts that successfully deploy units within a given one-floor site.