IT TAKES KNOWLEDGE TO APPLY NEURAL NETWORKS FOR CONTROL

Until very recently, process control applications of neural networks h ave been limited to theoretical studies and fairly small experimental projects on somewhat simple processes. Early problems identified with using these systems in an on-line environment included appropriate han dling of raw process data, extending the knowledge base of the network to include more broad operating conditions, transforming the network results into useful controller setpoints, and providing for graceful d egradation of the controller functions. This paper describes a practic al approach for implementing artificially intelligent process control functions based on a unique combination of rule-based expert systems a nd neural network technology. The system has been successfully applied to a complex pulp and paper process and new applications are currentl y under development for other industries.