Model based monitoring of large scale processes in process industry

Project members: R. Bos, P.M.J. Van den Hof, X.J.A. Bombois

Sponsored by: TNO-TPD

In process industry a lot of time and effort is spent on modelling large scale processes using first principles relations. The resulting models usually consist of a large set of non-linear partial differential equations. These equations usually cannot be solved analytically, which means they have to be solved numerically on a fine spacial grid. The models can be converted to the familiar state-space form by assigning one or more states to each point on the spatial grid. Solving the equations on this grid tends to be computationally intensive.

Before we can use these first principles models for monitoring purposes, we face three main problems. The first problem is that the number of states in the model is often very large. A second problem is that the state-equations are very computationally intensive. The final obstacle is that these first principle models tend to be non-linear. These difficulties cause that standard solutions to monitoring, such as Kalman filtering cannot be used. In this project we will attempt to find alternative strategies for monitoring process variables.

The methods and techniques developed in the project are tested in a case study. The case study in this project consists of a dynamic model of the dryer section of a paper mill, see Figure 2.

This research is being done in cooperation with the Control Engineering and Process Physics groups of TNO-TPD in Delft.

Figure 2: Paper mill.