Abstract:
Owing to the numerous benefits of process monitoring, the subject has attracted a
lot of attention in the last two decades. Process monitoring is an art of identifying
abnormal deviations in a process from the normal operating condition using various
techniques. Generally, the development of these monitoring techniques is geared towards
applying these techniques to industrial processes. In addition, most industrial
processes are dynamic and non-linear in nature. Therefore, in the development of
the monitoring algorithms, the dynamic as well as the non-linear properties of the
plant should be taken into consideration.
Process monitoring techniques like the Principal Component Analysis (PCA) and
Partial Least Squares (PLS) regression analysis were developed based on the assumption
that the process data is normally distributed. Nevertheless, this assumption
of normality is invalid for most industrial processes due to the non-linear nature of
these plants. For such processes, the distribution of the process variables in general
will be non-Gaussian, therefore making the widely applied PCA and PLS approaches
inappropriate for the monitoring of plants. To address this limitation of the PCA
and PLS for Dynamic processes, the Dynamic PCA (DPCA) and dynamic PLS
(DPLS) approaches were developed.
The challenge of efficiently monitoring process plants with dynamic and non-linear
characteristics is the motivation for this study. The overall aim of this study is to
develop process monitoring strategies that are able to take the dynamic and nonlinear
properties of the plant into account. With these strategies, more efficient
performance monitoring of the plant can be achieved. Cont/d.