An expert system for real-time control of chemical processes provides an environment for coordination of process fault diagnosis, assessment of process behavior, automated controller retuning and/or reconfiguration and consequently enables fault-tolerant process control. A supervisory expert system with object-based knowledge representation and heuristic (shallow) and model-based (deep) knowledge is presented. The expert system resides on a PC/386 and communicates with the control system developed in the form of control blocks residing on a PC/286. The prototype of the expert system is developed for retuning model-based controllers to improve the behavior of a packed-bed tubular CO oxidation reactor under autothermal operation. MOBECS is a supervisory expert system designed to tailor the process control system for a tubular autothermal reactor in response to process or control system faults, and changes in the process behavior. Specific functions include process fault diagnosis, control system performance monitoring and trouble-shooting, controller tuning and control system restructuring. As this work progresses, we will also address the problems of sensor placement, state estimation and process identification. Because of the high degree of coupling and interaction between process and control system entities, objects are the best knowledge representation for a process control expert system. The class—object structures of MOBECS is divided into three orthogonal trees representing the process and control system hierarchies and the topology of the system. Using this structure, knowledge is divided into general knowledge applicable to any process or control system and knowledge that is domain specific. The general knowledge is stored in the class structures and, together with generalized rules, forms a knowledge base kernel. This permits rapid prototyping of other applications of MOBECS. Properties are defined as high as possible in the class tree structure and inherited downwards. Knowledge specific to the process is stored in the objects at the leaves of the trees. Rules form the reasoning portion of the knowledge base and contain the causal relationships, meta-knowledge and performance monitoring functions. Rules that perform general monitoring and troubleshooting are applied at the class level through pattern matching. Rule sets are linked through context relationships to identify indirect relationships to the inference engine. These relationships are used with the process topology tree to reduce the search space when a fault is detected. The process monitoring and instability detection rules have been tested by simulating the inputs to the expert system. At the present time, experimental validation of the MOBECS prototype is being undertaken. Once the MOBECS prototype has been fully tested, the knowledge base will be extended to incorporate more complex fault diagnosis, state estimation, sensor placement and controller restructuring.