Observer-based fault diagnosis in chemical plants

The main focus of this paper is on the study and application of existing methodology on unknown input observer (UIO) for on-line diagnosis of faults on input actuators, output sensors, model parameters and disturbances of complex chemical plants operating under model predictive control (MPC). Two industrial systems are studied by simulation: styrene polymerization reactor and fluid catalytic cracking (FCC) unit. For each of these cases, the development of the method is presented and the design of the fault diagnosis system is discussed. In the first case, a bank of reduced-order UIOs is used for fault diagnosis of process parameters and external disturbances. The design is based on the rigorous first principles model of the polymerization process. In the second case, a single bank of full-order UIOs is used for sensor and actuator fault diagnosis based on an input-output linear model of the FCC unit. In both cases, extensive simulation results are presented and discussed.