Real-time direct expert control using progressive reasoning

• Published in: Engineering applications of artificial intelligence Vol. 2; no. 2; pp. 109 - 119
• Main Authors: Broeders, H.M.T., Bruijn, P.M., Verbruggen, H.B.
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 1989
• ISSN: 0952-1976; 1873-6769
• DOI: 10.1016/0952-1976(89)90025-0

To investigate the capability of knowledge-based real-time control a number of experiments have been performed. The knowledge-based system should combine the advantages of numerical control algorithms implemented in process computer systems with the capabilities of human operators to adjust or tune these algorithms or to take over the operations of the algorithms by manual control. The implemented system combines algorithmic calculations with symbolic reasoning. Starting with a simple, empty shell expert system (Expert-2), written in Forth, a more sophisticated expert shell (Expert-3) has been developed. This shell extended the backward-chaining inference engine of Expert-2 with a number of speed optimizing facilities, such as agenda control and forward chaining. To realize real-time direct expert control this expert system was embedded inside a real-time environment (RETEX) based on real-time Forth. The progressive reasoning concept requiring knowledge divided into different knowledge layers (rule bases), was used to capture varying inference times within the framework of a constant sampling period. In order to control arbitrary processes with unknown parameters, control knowledge has been elicited and divided into five rule bases. The control strategy in the first three layers is based on classification of symbolic states of the process into the phase plane, and in the fourth on ‘Model Reference Expert Control’. The fifth knowledge layer is the supervisor of the underlying layers. The five rule bases contain 170 production rules. Inferencing is performed by a combination of forward and backward chaining. This application of a direct expert control strategy was realized on a 68000-microprocessor system (ATARI 1040ST). It proved to be a robust strategy with a minimum sampling interval of 55 ms for the lowest layer and 430 ms using all the rule bases.