Fault isolability analysis based on steady state fault signatures

Once a fault has occurred, it is highly desirable to quickly detect and reliably isolate the fault. Reliable fault isolation is a key to ensure correct remedial action; therefore fault isolation module must determine the current fault scenario unambiguously and uniquely. However, it may be difficult to differentiate between all possible fault situations due to various practical and physical limitations. The task of a fault isolation module is to generate a set of candidate faults, which might have occurred such that misclassification is minimized while no candidate fault scenarios are missed. Fault isolation schemes essentially rely on fault signatures, either during transient or at steady state. Our main focus is on methods relying on steady state fault signatures, such as clustering based algorithms. These fault isolation schemes may sometimes misclassify if another historical event match the current fault scenario. It is therefore necessary to i) analyze the process, ii) analyze how different faults manifest themselves and iii) determine which faults can exhibit similar steady state fault signatures. Specifically, it may be very important and useful to determine which faults can be uniquely isolated from given process input-output data. In this paper we propose a simple fault isolability analysis based on steady state process model. A steady state process model can be obtained either based on first principles model or based on process input-output data. The proposed fault isolability analysis has been validated by obtaining simple steady state process model of the benchmark quadruple tank process for normal and fault operation. Fault isolability predictions, obtained using the proposed fault isolability analysis, are found to be accurate.