Fusion of model-based and model-free approaches to leakage diagnosis

A diagnosis scheme for incipient leakage faults is proposed using a combination of two entirely different approaches, namely model-free and model-based ones, ensuring thereby that critical information about the presence or absence of leakage is monitored in the shortest possible time and the complete status regarding the leakage is unfolded in time. Model-free approaches include limit checks & knowledge-based analysis, while model-based approaches include the extended Kalman filter and a parameter identification scheme. The knowledge-based analysis indicates quickly a possible onset of leakage, the Kalman filter detects the presence/absence and finally the identification scheme isolates the detected fault. Further, the whole combined helps in designing an effective preventive maintenance strategy. The proposed scheme is evaluated on a physical fluid system exemplified by a bench-marked two-tank system.