On the Structure Determination of a Dynamic PCA Model using Sensitivity of Fault Detection

Abstract

This work proposes a dynamic PCA modeling method for dynamical non-linear processes. This method uses fault free data to construct data matrix used to compute the correlation matrix and faulty system data in order to fix the dynamic PCA model parameters (the backshift and the number of principal components). It is shown that the sensitivity of dynamic PCA-based fault detection depends on the parameters used in the model. This method is tested on a three serial interconnected tanks and subject to fluid circulation faults in its pipes.