Intelligent control-a review of the basic concepts

To tackle real-world control problems using expert behaviour, an expert controller with learning capabilities must be built, so that the expert can transfer skill by a tutorial process of showing examples. That is, the machine must be able to build up its skill in the domain automatically in the form of appropriate control actions triggered by the sensed state variables. In machine learning adaptive control the control scheme is independent of the mathematical modelling of the object dynamics and the precise estimation of its physical parameters. The article presents the authors own work, and that of others, on rules for controlling dynamic systems, and for designing robust rule-based controllers, efficient and adaptable, that has been applied in numerous simulations and to physical process control. The work reported catalogues experiments into intelligent control by human-supervised learning, passive learning, and machine learning. The work started with observing human-supervised learning, these observations led to the development of generalised rules for an automatic controller. Later, when this controller was applied to the physical system, the sensed outputs were post-processed by the c4.5 induction rule generation algorithm to demonstrate passive learning. Passive learning of this type provided feedback and allowed a direct comparison to be made between human and machine generated rules