Integrated Approach to Reliable Leakage Diagnosis

Abstract

A generic and reliable diagnosis framework, for incipient leakages is proposed using an integrated approach combining two entirely different model-free (MF) and model-based (MB) schemes. The primary objective of this vital combination is to ensure a timely monitoring of the critical information about the presence/absence of leakages, and an accurate unfolding-in-time of the complete leakage status. Whereas the model-free approach provides the user with the possibility of an early incipient fault detection via limit checks and some form of knowledge-based (KB) analysis, the model-based approach allows for a Kalman filter (KF)-based detection followed by a parameter identification-based fault isolation. As such the MB approach will provide a confirmation of the detection of any fault flagged by the MF approach. A new model order selection scheme is proposed here to provide a reliable model. Two vital advantages accruing from this integrated approach are a substantial reduction in false alarms and an increased probability of leakage detection. The proposed integrated approach is tested on a physical fluid system, exemplified by a benchmarked two-tank system.