Improved PCA Models for Fault Detection Using Delay Adjustment.

MUSULIN, E.; BASUALDO, M.

PEM Fuel Cells with Bio-Fuel Processor System: A Multidisciplinar Study of Modelling, Simulation, Fault Diagnosis and Advanced Control

Springer

Año: 2011;

Principal Component Analysis (PCA) has been successfully used with process monitoring purposes. However, it has an important drawback as it does not account for time-delays present in data thus causing an inefficient dimensionality reduction of process variables and the subsequent poor monitoring and disturbance detection performance. In this work a new method, the Genetic Algorithm based Delay Adjusted PCA (GA-DAPCA), based on genetic algorithm optimization is proposed to improve the PCA performance in the presence of time delays. The optimization is performed in two loops. The first one finds the shift between variables that minimizes the number of principal components to be considered as common cause variance, while the second loop maximizes the variance contained in the previously selected principal components dimensions. An extension of DAPCA called MS-DAPCA that uses wavelet decomposition to deal with noise and multi-scale processes is also presented. The methodology is demonstrated in the Tennessee Eastman process benchmark. Results are analyzed used standard and combined statistics.