On-line fault detection and diagnosis for chiller system

Fault detection and diagnosis for heating, ventilation and air conditioning system (HVAC) significantly impact energy efficiency and human thermal comfort. A model-based fault diagnosis method is presented in this paper to provide reliable estimation of multiple and simultaneous fault conditions at the component level, in the presence of measurement noisy and time-varying operation conditions. The method is based on particle filtering (PF), a Bayesian nonlinear joint-state-and-parameter estimation technique, with physical health parameters that describe the operation status of HVAC components as input. A reference model is established first under nominal situation to express the health parameters as functions of operation conditions, through a kernel-based partial least square method, which demonstrates strong capability on revealing the underlying connections between system outputs and inputs. Subsequently, the residual for each health parameter is generated by comparing the measured value to its benchmark provided by reference models. Variations of these residuals reveal the fault location and type. In addition, trending of residuals enabled by PF evaluates fault severity and predict fault deterioration rate in the future. The developed method is applied to a chiller plant, which is a critical sub-system of the HVAC, and is shown to be effective in identifying and tracking faults, such as condenser fouling, refrigerant leakage, and reduced evaporator water flow.