Acquisition of Control Algorithms from Operation Data

Abstract

The article deals with automatic acquisition of control algorithms from process or device operation data. Our objective is to present a methodology for elimination of the mathematical modelling and programming stages in the control system development. In the presented approach these stages are replaced by a training stage followed by generation of decision rules equivalent to a Boolean network. The rules are produced from logged operation data obtained from experienced operators. The rules are subsequently converted into control program source code. The adapted training methodology has been developed within the framework of the theory of rough sets [1] and implemented as a commercial system for data analysis and rules extraction system by REDUCT Systems Inc., Regina, Canada. The presentation is illustrated with a comprehensive example demonstrating the generation of the control algorithms from simulated data representing movements of the robot arm.