This project has an overall goal of designing model-based control algorithms using multi-scale descriptions of particle attributes in high-medium shear granulators. The first objective of the second year of this project was to design a reduced order model that is appropriate for real time applications. The second objective involved a feasibility study of applying a reduced order model in a Model Predictive Control (MPC) algorithm. The methods used to reach these objectives have prepared us for the next stage, the objective for the third year: application of MPC algorithms on a pilot plant granulator. During the scope of this project four different granulation models have been considered (in order of complexity):
1. Transfer function model (Pottmann, et al., 2000; Gatzke and Doyle, 2001)
2. One dimensional well mixed discretized Population Balance Model (PBM) (Sanders, et al., 2005)
3. One dimensional three compartment discretized PBM (Wang, et al., 2006)
4. Three dimensional PBM (Immanuel and Doyle, 2005) The models were either used as a plant model to simulate a real plant or as the basis for a linear controller model. This report described the details of the research with the second model. The fourth model has been described in our first year report (IFPRI # ARR51-01). The third model has been developed by the group in Queensland and had been adjusted to model their pilot plant, which will be used in future studies in this project. The main differences between model 2 and model 4 are the recycle flow (+ screens and crusher) and the introduction of three separate regimes in the granulator. More details about the UQ model and pilot plant are described in section 6: Summary and Future Work.
The first part of this report (Chapter 3 and 4) present the framework that was used to setup and test an MPC controller on a simulated plant model. The plant model is based on earlier work done at The University of Sheffield and Organon, The Netherlands (Sanders, et al., 2005). The granulation kinetics were measured in a 10 liter batch granulator with an experimental design that included four process variables. The aggregation rates were extracted with a Discretized Population Balance model. Knowledge of the kinetics was used to model a continuous (well mixed) granulator. The controller model is a linearized state space model, derived from the nonlinear DPB model. It has the four process variables from the experimental design and a feed ratio as input variables. Since the DPB model describes the whole size distribution (GSD), different sets of output variables were chosen and compared. Preliminary results show successful implementation of Model Predictive Control on a continuous granulator. Further research is needed to test the linear model on an actual pilot plant.