Prognostics-Driven Optimal Control for Equipment Performing in Uncertain Environment

Abstract

This paper discusses the problem of optimal control for systems performing in uncertain environments, where little information is available regarding the system dynamics. A reinforcement learning approach is proposed to tackle the problem. A particular method to incorporate Prognostics and Health Management information derived on the system of interest is proposed to improve the reinforcement learning routine. The ideas behind reinforcement learning-based search for optimal control strategies are outlined. A numerical example illustrating the benefits of using PHM information is given. 1 2

Reinforcement Learning for Scheduling of Maintenance

“…There have already been some initial attempts to explore reinforcement learning for restricted tasks in scheduling, routing, and network optimisation. [23,24,25,26,27,28,29,30,31,32] Our approach differs from these since it offers a practical application for RL in a real-world online environment. In this application RL will not only adapt to the broad properties of the problem but also to the individual properties of the equipment used.…”