Fault diagnosis based on independent component analysis and Fisher discriminant analysis

A large number of process variables are measured in a chemical process, but this process is usually driven by fewer essential variables, which may or may not be measured. The process fault diagnosis performance will be improved by extracting such essential variables. A mixture method of fault diagnosis is proposed by using independent component analysis (ICA) and Fisher discriminant analysis (FDA). The major purpose of ICA is to find underlying components from the data under normal operating condition, so transformation matrix is found from the origin data space to the independent component space for extracting the essential variables. FDA model is constructed based on various known fault data that have been projected to the independent component space in order to diagnose faults. The performance of the proposed diagnosis method is tested using the Tennessee Eastman Process. The results of fault diagnosis show that the proposed method outperforms other FDA-based diagnosis methods.