Paper:
Views over last 60 days: 673
Reinforcement Learning for Online Industrial Process Control
James J. Govindhasamy*, Seán F. McLoone**, George W. Irwin*, John J. French***, and Richard. P. Doyle***
*Intelligent System and Control Research Group, Queen's University Belfast, Belfast BT9 5AH, Northern Ireland, United Kingdom
**Department of Electronic Engineering, National University of Ireland Maynooth Maynooth, Co. Kildare, Ireland
***Seagate Technology Media Ltd., 99 Dowland Road, Aghanloo Industrial Estate Limavady BT49 OHR, Northern Ireland, United Kingdom
Received:October 30, 2004Accepted:November 1, 2004Published:January 20, 2005
Keywords:reinforcement learning, action dependent adaptive critic, neural networks, nonlinear process control, intelligent control
Abstract
Reinforcement learning, in the form of Adaptive Critic Designs (ACDs), have the ability to analyse or evaluate a situation and respond to it accordingly. They offer an excellent alternative for adaptively controlling and optimising the highly nonlinear processes found in industry. Here, an enhanced implementation of the action dependent adaptive critic design (ADAC) of Si and Wang [9] is investigated for modelling and control of an industrial grinding process used in the manufacture of hard disk drive platters. This study, one of the first reported industrial applications of this emerging technology, shows that the proposed ADAC control scheme can achieve a 33% reduction in platter rejects compared to an existing proprietary controller.
Cite this article as:J. Govindhasamy, S. McLoone, G. Irwin, J. French, and R. Doyle, “Reinforcement Learning for Online Industrial Process Control,” J. Adv. Comput. Intell. Intell. Inform., Vol.9 No.1, pp. 23-30, 2005.Data files:
