Development of Soft Sensor Model Using Moving Window Approach

Soft sensors are used broadly in the industries to predict the process variables which
are not measurable by sensors. The objective of this project is to develop a datadriven
soft sensor using Moving Window approach with the selective regression
techniques and to evaluate and validate the advantages and performances of Moving
Window approach over the traditional soft sensor models. Time invariant and
stationary process conditions are those assumptions made in developing soft sensors,
and these assumptions causes degradations and limitations to the soft sensors in
estimating process variables. Degradations of soft sensors are caused by process
shift, catalyst performance lost and et cetera. Besides that, the restrictions of sensors
in estimating difficult-to-measure variables and the delays during the laboratory tests
have becomeone of the factors in developing soft sensor. This paper presents a study
regarding the multivariate statistical process control techniques that can be used in
developing soft sensors such as Least Square Regression method, Partial Least
Square Regression method and Principle Component Analysis. The scope of study
for the project includes understanding the concept andwhat are the adaptive schemes
available to construct the soft sensors. Besides that further research on Moving
Window approach together with MSPC techniques will be carried out which can be
adapted into the adaptive models to develop the soft sensors. Systematic approach
will be presented through this project in using Moving Window approach to
construct the soft sensors and this includes an analysis of an appropriate case study
where the approach can be implemented.
Keywords: Multivariate Statistical Process Control techniques, Least Square
Regression method, Partial Least Square Regression method and Principle
Component Analysis