| Due to the growing use of discrete, logic-based processes in industry, including various manufacturing, chemical/petrochemical, and paper-making operations, there is a need for a task-specific diagnostic architecture for handling the malfunction scenarios unique to these processes. Diagnosis of discrete, logic-based processes is highly knowledge-intensive, requiring knowledge of the control program, the physical process functionality, and process operating experience. This dissertation presents the following: characterization of the types of diagnostic scenarios associated with discrete processes along with the types of malfunctions that occur; a task analysis identifying the diagnostic problem-solving strategies; a conceptual framework for capturing the knowledge and the problem-solving strategies needed for diagnosis; a computer framework (expert system) that provides the necessary programming structures matching those used in diagnostic problem solving; demonstration of this framework via three diagnostic expert system examples (a process interlock, a batch chemical reactor, and a large automated paper roll-wrapping operation); and knowledge engineering guidelines for developing and maintaining the expert system.;Specifically, this research examines discrete processes controlled by programmable logic controllers (PLCs). There are two major diagnostic scenarios associated with discrete, PLC-controlled processes: (1) a dead state scenario where part of all of the process comes to a dead stop and (2) a degrading performance scenario where the process is operating but a noticeable deterioration in process operation has occurred resulting in off-spec product quality parameters and/or abnormal physical process events. The expert system framework handles both diagnostic scenarios as separate, hypothesis-driven, diagnostic classification hierarchies which are based on process functionality. When used on-line for performing diagnosis, the expert system determines which diagnostic scenario is present; obtains on-line internal PLC data direct from the controller; gathers additional symptomatic information from the user that is not available on-line (obervations, field test information); isolates the cause of the process malfunction; and presents a corrective action for alleviating the problem. "WRAPITUP", an on-line diagnostic system developed for diagnosis of a roll-wrapping machine, correctly identifies a malfunction in 1 or 2 minutes, which formerly required hours of a human expert. |
