Fault detection and diagnosis using unknown input observer for non-linear chemical processes

Abstract

Advanced automatic control technologies have brought significant benefits to the chemical industry. This is however, hampered by the inefficiency in providing effective detection and diagnosis of process faults that may emerge from various aspects of plant operation. Among the available techniques, unknown input observer (UIO) method has been highlighted as a potentially effective approach as it offers effective capability to deal with residuals between the model estimation and actual measured values of the process variables. UIO modeling strategy creates a specific residual signal that carries information of specific faults, as well as model uncertainties and exogenous disturbances decoupled from fault features. With this characteristic, process faults can be effectively detected, isolated, and identified. The UIO technique was tested on a multi-variable distillation system configured with multiloop feedback control. For this purpose, various scenarios of sensor faults were introduced, and a bank of unknown input observers was designed. Successful results were obtained to detect, isolate, and identify faults. The UIO based fault detection and diagnosis (FDD) system was further tested on case studies involving sensor faults, in open and closed-loop conditions in a non-linear exothermic continuous stirred tank reactor. The proposed FDD scheme was proven robust enough to deal with model uncertainties and exogenous disturbances introduced in the case studies. The results obtained in this study proved the suitability of the UIO modeling approach to be used in FDD system to provide effective early warning feature in process plant alarm management.