Batch-to-batch learning for process control with application to cooling crystallization.

Project M. Forgione, MSc (Marco), P.M.J. Van den Hof (Paul), X.J.A. Bombois (Xavier)
Keywords:Model-based control, Process technology, System identification, Learning and adaptive control
Sponsored by:ISPT

An industrial crystallizer
An industrial crystallizer

Crystallization can be defined as phase change in which a solid
product is obtained from a solution, melt or a gas. In the industrial
practice crystallization is utilized as a separation and purification
step in sectors such as pharmaceutical, food and fine chemicals.
The process is often operated in batch mode. The solution
is loaded into the vessel at the beginning of a batch and the solid
product is removed at the end. The batch time is of the order of
few hours. Multiple batches are performed in order to fulfil the
production demand.

In many cases, stringent requirements on the quality of the final product have to be met. The control of a number of variables
throughout the batch infuences the quality of the final product to
a large extent. For this reason, the industries in the field are interested into the introduction of advanced automation technology.

Batch operations bring both challanges and opportunities for
identification and control. From one hand, batch process system
are operated over a wide dynamical range. For this reason, strategies based on linearization around a single operating point lead often
to poor results. In some cases, first-principles models describing the
full nonlinear dynamic behavior are available. Unfortunately, these models are often either not very accurate or not suitable for
control purposes.

On the other hand, the repetitive nature of batch processing allows the use of iterative techniques based on the information
collected during the previous

In this project we are developing identification and control strategies
suitable for industrial batch processes such as
crystallization. We are particularly interested in

  • Batch-to batch Identification & Optimization

  • Iterative Learning Control

  • Experiment Design

We are collaborating with several industrial partners such as DSM, Albemarle,
Frieslandcampina, MSD and DotX Control Solutions.

© Copyright Delft Center for Systems and Control, Delft University of Technology, 2017.