An isolation enhanced PCA method with expert-based multivariate decoupling for sensor FDD in air-conditioning systems

Principal component analysis (PCA) has been found to be powerful in detecting sensor faults in multivariate processes, but it is inefficient in isolating faults due to its pure data-driven nature, especially when dealing with processes with strongly coupled multiple variables, such as the air-handling processes in typical variable air volume air conditioning systems. This paper presents an expert-based multivariate decoupling method to enhance the capability of the PCA-based method in fault diagnosis by taking advantage of expert knowledge about the process concerned. The decoupling method develops unique fault patterns of typical sensor faults by analyzing the physical cause-effect relations among variables. Through comparing fault symptoms reflected by the residual vectors of the PCA models with fault patterns, a sensor fault can be successfully isolated. The isolation enhanced PCA method is implemented and validated in a typical air-handling process. The test results show that the joint approach to enhance the fault isolation ability of the PCA-based fault detection and diagnosis method is effective. The robustness of the PCA-based sensor FDD method against component faults is also proved to be improved because the fault symptoms of sensor faults are unique.