Automation of load balancing for Gantt planning using reinforcement learning. (May 2021)

• Record Type: Journal Article
• Authors: Woo, Jong Hun; Kim, Byeongseop; Ju, SuHeon; Cho, Young In
• Is Part Of: Engineering applications of artificial intelligence. Volume 101(2021)
• Publication Date: 2021-05
• Subjects: Shipbuilding -- Production planning -- Workload balancing -- Reinforcement learning -- Deep neural networks; Engineering -- Data processing -- Periodicals; Artificial intelligence -- Periodicals; Expert systems (Computer science) -- Periodicals; Ingénierie -- Informatique -- Périodiques; Intelligence artificielle -- Périodiques; Systèmes experts (Informatique) -- Périodiques; Artificial intelligence; Engineering -- Data processing; Expert systems (Computer science); Periodicals; 620.00285
• DOI: 10.1016/j.engappai.2021.104226
• Languages: English
• ISSNs: 0952-1976

Abstract

Abstract: Typically, in the shipbuilding industry, several vessels are built concurrently, and a production plan is established through a hierarchical planning process. This process largely comprises strategic planning (long-term) and master planning (mid-term) aspects. The portion that requires the most manual work of the planner is the load balancing in the master planning stage. The load balancing of master planning is an area where optimization studies using mixed integer programming, genetic algorithms, tabu search algorithms, and others have been actively conducted in the field of operational research. However, its practical application has not been successful due to the complexity and the curse of dimensionality, which is dependent on the manual work of the planner. Therefore, a new method that can facilitate the efficient action of optimal decisions is required, replacing conventional production planning methods based on the manual work of the planner. With the advent of the 4th industrial revolution in recent years, machine learning technology based on deep neural networks has been rapidly developing and applied to a wide range of engineering problems. This study introduces a methodology that can quickly improve the load balancing problem in shipyard master planning by using a deep neural network-based reinforcement learning algorithm among various machine learning techniques. Furthermore, we aim to verify the feasibility of the developed methodology using the ship block production data of an actual shipyard. Highlights: Development of an automated workload balancing algorithm for activity planning based on Gantt chart. A3C (Asynchronous Advantage Actor–critic) reinforcement learning algorithm applied. Grid type environment model. Introducing the concept of a time window to apply consistent learning results to different planning data.