Fault-tolerant chemical reactor control

A supervisory Model-Object Based Expert Control System (MOBECS) has been developed to provide fault tolerant control of a chemical reactor. This supervisory process control expert system is capable of initializing the control system based upon the current operating conditions and monitoring the process and control system for faults or deteriorating performance. If a fault is detected, MOBECS can formulate and implement corrective actions ranging from automatic controller tuning through complete restructuring of the process control algorithm. If a primary sensor fails, the expert system can replace the missing measurement with a state estimator.;The architecture of the expert system is based upon a hybrid paradigm that uses a class-object structure to represent the process and control system with a rule based inference mechanism. The process control system uses a similar architecture based upon process control function blocks. This commonality of data structures gives MOBECS the ability to modify every facet of the control system in real time. An expert system design formalism, based upon the Entity-Relationship relational data base model was used to develop the expert system. MOBECS has been generalized to any type of process control system or unit operation. All of the process specific knowledge is confined to the objects in the knowledge base and the procedures called by the rules and metaslot methods. The expert system design was validated by applying it to a tubular, fixed bed, autothermal reactor in which carbon monoxide is oxidized to carbon dioxide over a platinum catalyst. MOBECS successfully tailored the Internal Model Control algorithms, used to control the reactor, in response to changes in the operating conditions and faults in the process control system.