Approach Research Of Sensor Fault Detection And Diagnosis In HVAC System

Fault detection and diagnosis in heating, ventilation,air-conditioning (HVAC) system is important in reducing power consumption and energy costs, maintaining and improving comfort in building , reducing wear on HVAC equipment and reducing emission of greenhouse gas. However, it has been beyond operator's power to detect and diagnose fault in HVAC system .rapidly and timely with building HVAC system getting complicated more and more. So, it made automation fault detection and diagnosis be indispensable.This thesis presents principal component analysis(PCA) approach for sensor fault detection, identification and reconstruction in HVAC system. PCA approach models the HVAC system with its measurement data of normal operation condition. So, complicated analysis model is evitable. It is crucial to choose the number of principal component in modeling system with PCA approach. In this thesis, the optimal number of principal component is decided by minimizing the total unreconstructed variance. The PCA approach partitions the measurement space into principal component subspace (PCS), in which normal data is included, and residual subspace(RS), in which noise or fault is included. Fault can be detected by detecting the projection of the measurement data in RS. In this paper, a statistic variable, squared prediction error (SPE), is defined to detect fault and calculating formula of confident limit for SPE is given.Essential of fault reconstruction is a process of seeking an estimation value for correct value corresponding to fault measurement data. Reconstruction via iteration is used for fault data recovery in this thesis. Iterative reconstruction is a process of sliding the measure to PCS along the direction of fault.Fault identification is an important composition of fault diagnosis. An index, SVI, is defined to identify fault. Research illustrates that SVI index can identify fault completely when fault occurs in temperature sensors and it can only find that fault occurs in flow meters but doesn't know which is when fault occurs in flow meters. This phenomenon is resulted from the collinearity between flow sensors. In order to identify flow sensor fault completely, we use wavelet to achieve it. Owe to the goodlocal time-frequency property of wavelet transformation, we can find fault by the variance of details of measurement signal. So, problem in identification of flow meter sensor is solved preferably.Finally, a summary of research and recommendation of fault detection and diagnosis research in HVAC system in future is given.