An online model-based fault diagnosis scheme for HVAC systems

Abstract

In this paper, a model based fault detection and isolation (FDI) scheme with online fault learning capabilities is proposed for HVAC systems. An observer comprising of an online approximator in discrete-time (OLAD) and a robust term is used for detection. A fault is detected if the generated detection residual, which is defined as the error between the observer outputs and HVAC system states, exceeds an apriori chosen threshold. The OLAD term in the FD observer learns the fault dynamics online while the robust term guarantees asymptotic estimation of the system states. Subsequent to detection, a fault isolation observer, which comprises of the model of fault functions and another robust term, is initiated to identify the root cause. A fault is identified if the isolation residual converges to zero, where the residual is obtained by comparing outputs of the isolation observer and the system. Additionally, we consider different fault scenarios in the system such as single or simultaneous multiple faults. Analytical results for the FDI scheme guarantee the robustness and stability of the proposed scheme. Finally, a simulation example is used to demonstrate the proposed FDI scheme.