As one of the major modules of Heating, Ventilation and Air Conditioning systems (HVACs), the Air Handling Unit (AHU) conditions the air and delivers it to rooms to satisfy occupants' comfort requirements. Fault diagnosis for AHUs is challenging because the interactions among components are complex. A fault may cause variations in variables of different constituent components, thus making it difficult to localize. To overcome this difficulty, in this paper, a model-based and data-driven fault diagnosis method integrating the Dependency-matrix (D-matrix) and Principal Component Analysis (PCA) is developed, where the D-matrix is a compact representation of the complex interactions between failure modes and fault indicators. In this method, by using PCA, both model parameters and variables relating to faults are selected to obtain Squared Prediction Errors (SPEs) as fault indicators for D-matrices. In D-matrices, failure modes can be distinguished from each other if they have different signatures. This method has three benefits: (1) SPEs are sensitive to faults since the relationships between model parameters and failure modes are more explicit comparing to measured variables alone; (2) only one sequence of fault indicator outcomes corresponds to one failure mode, thus the number of fault indicators decreases; and (3) SPEs obtained by using PCA could contain most of the fault information; thus it is not necessary to make the effort at selecting the effective variables as fault indicators. If failure modes are still ambiguous, the variables which represent the unique features of failure modes are selected as fault indicators for further diagnosis. Numerical results show that our method can distinguish faults in AHUs accurately.
展开▼
