Kalman Filter, FDD, HVAC
We propose to improve the energy efficiency of commercial HVAC systems by implementing a Kalman filter based fault detection and diagnosis (FDD) scheme to accurately identify certain categories of abnormal conditions that are the most prevalent in hot, humid climates like the UAE. The general approach used to deal with drift in system state includes the following tasks: 1) Determine possible model specifications, 2) Use a Kalman filter to determine time-varying parameters of the specified models, 3) Perform fault detection using a statistical process control and, 4) Perform fault diagnosis by formulating qualitative rules and comparing with those derived from decision trees. The model is built using from normal and faulty data that was generated during ASHRAE Project RP-1312 in which several Air Handling Unit (AHU) faults were artificially introduced and the resulting operational data recorded at 1-minute intervals. The results show that the Kalman filter is well-suited to detecting changes in mean level and the rules derived from the parameter estimates perform as well as those derived from decision trees.