A Recursive Method for Variable Selection Using Principle Component Analysis and Factor Analysis for Identification of System Status in a Commercialized 300kW MCFC Power Plant

In a commercialized 300kW MCFC Power Plant, a univariate alarm system is usually employed to monitor system status. However, this is limited in extend monitoring system into fault diagnosis system. To overcome the limitation of the present monitoring system, a multivariable monitoring system based on PCA(Principle Component Analysis) has been developed. In progress of development, a recursive method for variable selection has been developed. This method is based on the PCA but has been modified to reduce the time for variable selection. Because PCA model that contains all variables in the process could not isolate process fault, factor analysis used for hierarchical variable grouping. The proposed recursive method is implemented in the SAS environment, and the results of fault isolation are analyzed and compared to trip event in a commercialized power plant. The estimation of type 1,2 errors show that this recursive method works well when the system fault occurs.