Online diagnostics of time-varying nonlinear chemical processes using moving window kernel principal component analysis and Fisher discriminant analysis

A combined method of diagnostics is proposed for a class of nonlinear multivariate processes where fast and slow faults are observed with essentially different rates of their development. Because fast faults occur against the background of the slow ones and have many common symptoms, the process turns out nonstationary. That is why it is proposed to use kernel principal component analysis with a moving window for monitoring purposes. The faults are identified with Fisher discriminant analysis started up when the fault is detected, and through some moments it is blocked to reduce blurring of classes. The efficiency of the method is demonstrated by the example of diagnostics of hydrocarbon pyrolysis process.