Automatic knowledge acquisition for an expert system for distillation control

A process continuously produces data about the state of its variables. Knowledge about the process is embodied in these data. If data representing stable and recovery states are selectively used to induce and refine rules, these rules will contain the best knowledge of the operators and also knowledge intrinsically contained in the mathematical model of the process control computer. A system is presented which acquires knowledge about an industrial process in the form of rules about how monitored variables can be related to controlled variables to produce stability in dependent variables. The system uses a genetic rule induction algorithm on data specifically selected to contain knowledge about the setting of controlled variables