Structural Fault Detection Using Dynamic Principal Component Analysis (DPCA)

ABSTRACT
Principal component analysis (PCA) is a well-known data dimensionality
reduction technique that has been used to detect faults during the operation of
industrial processes. A modification to this is the Dynamic Principal Component
Analysis (DPCA) which takes into account serial correlations for rapid sampling and
detection of faults. This method, although being studied for its fault detection
capabilities, has not yet been widely tested and proved to detect a particular type of
faults called structural faults. In this paper, a dynamic model of a Continuous Stirred
Tank Reactor (CSTR) is built using the MATLAB software and used to generate a
sample of base data and another sample of data with structural faults present. Further
on in this project, the two data sets would be compared and tested using T2-statistics
and Q-statistics method to identify the faults occurring in the system and study the
difference in performances of these methods. Q-statistics which quantifies variations
in the residual space is more sensitive but less robust to the faults than the T2-
statistics quantifying the variations in the score or state space. Faster fault detection
via DPCA is also achieved in T2-statistics whereas results from Q-statistics are
inconclusive.