Chiller plant fault diagnosis considering fault propagation

Abstract

As a major subsystem of Heating, Ventilation and Air Conditioning systems (HVACs), a chiller plant provides chilled water to remove heat from buildings. Faults in a chiller plant can result in high energy consumption, and their early diagnoses, including failure modes and fault severities, will lead to significant energy savings. Fault diagnosis, however, is challenging since (1) measured variables are noisy, depending on many conditions, and may not be directly related to faults; (2) a fault in one module (e.g., chiller and cooling tower) may trigger a fault in another module (fault propagation); and (3) identifying both failure modes and fault severities accurately may require high computational efforts. In this paper, fault diagnosis of a chiller and a cooling tower is considered at the module level through a model-based and data-driven method. In particular, gray-box models are adopted, and model parameters, which characterize module performance, are used to supplement measured variables for fault diagnosis. To capture couplings among modules, a Coupled Hidden Markov Model (CHMM) capturing couplings among modules is synergistically integrated with Extended Kalman Filters (EKFs) to identify failure modes (CHMM) and fault severities (EKF). In this method, EKF rather than CHMM is used to estimate fault severities, thus accurate continuous estimates are obtained. Experimental results show that this method can accurately diagnose faults in a computationally efficient manner.