| In HVAC system of intelligent buildings, unit faults greatly affectenergy-saving and comfort performance index. Analyses find that HVAC systemfault detection has four difficult problems (parameter drift, multiple modes, incipientfailures and pre-existing failures). Traditional fault detection methods are difficult tomeet requirements. Considering characteristics of HVAC system and these problems,these existing HVAC system fault detection methods were described and a newapproach was proposed for fault detection in this research.This paper considers the fault detection of the three-way valve for example.Slide regression model for each mode of the three-way valve was established byusing Eviews software according to the real experimentally measured data. AMarkov chain model was established with taking unit mode switching into account.IMM filter was chosen for filtering with taking into account the existing ofconversion modes. The IMM filtering was implemented in Matlab. We can getsliding autoregressive parameters of each unit. And the impact of changes in theexternal environment on fault detection can be effectively avoided by using theseparameters.Parametric statistical characteristics of similar units could be got by processingthese slide autoregressive parameters. Analyzing these characteristics we can detectwhether there are failures in one or more units. A variety of hypothesis testingmethods were compared in this paper. And residual chi-square test was chose torealize the fault detection. This method was realized in Matlab.Effects of the method we proposed were analyzed by ROC curve. Resultsdemonstrate that the fault detection method can effectively detect sudden failures,incipient failures and pre-existing failures of the HVAC system with taking unitmode switching, parameter drift, system noise and measurement noise into account.And the number of similar units, the detection window size and the size of thresholdcan affect failure detection results. |
