A reformative PCA-based fault detection method suitable for power plant process

Because the operating condition changes frequently, it's difficult to describe the statistical property of the power plant process with single principal component model (PCM). So the application of traditional PCA-based fault detection method can bring many misdiagnoses. A reformative PCA-based fault detection method suitable for power plant process is proposed. First K-mean cluster analysis is used to classify the process data and obtain the data sets under the various stable operating condition. Then the PCM group is established using the classified data sets to describe the entire process. Finally the detecting sample is carried on fuzzy partition during fault detecting and the PCM suitable for current operating condition is dynamically calculated and used for fault detection. The field data is used to contrast the application of traditional method with reformative method in the fault detection of boiler process. The results indicate that the reformative method can adapt the operating condition change, reduce the misdiagnosis and enhance the detection sensitivity.