Process fault detection and diagnosis in CSTR system using on-line approximator

This paper investigates the process fault detection and diagnosis in a continuous stirred tank reactor (CSTR) using artificial neural networks as on-line approximator. The results of the simulation show that in case of the full state is measurable, the process faults can be detected and diagnosed during the transient period. However, in case of that one state is not measurable, the unmeasurable state should be first estimated before process faults can be detected and diagnosed. In this latter case the final result can only accomplished after a certain period of time, required for the settling time, has elapsed.