(5) Current address: School of Chemical Engineering, Purdue University, West Lafayette, IN 47907-2100, USA

Applications of control to granulation processes in industry are rare. This paper details a methodology for a model predictive controller (MPC) that can be applied in a straightforward manner to an industrial granulation plant. Granulation is a complex process whereby fine particles agglomerate together due to adhesive forces. As Iveson et al. described in a review paper [1], the understanding of the fundamental processes that determine granulation behavior and product properties have increased in recent years. This knowledge can be used during process design as well as for choosing the right formulation and operating conditions. Furthermore it can be used to improve process control via closed-loop strategies. Wang et al. [2] showed that a simplified and discretisized one-dimensional population balance model (1D-PBM) is suitable for optimal control studies on a continuous drum granulator. In our work we will investigate the same continuous drum granulation process with the focus of deriving a model predictive controller for closed-loop control. Previously Sanders et al. [3] showed the benefits of MPC compared to a standard PID controller. An actual granulation pilot plant, which is located at the University of Queensland, Brisbane, Australia, was used for step response tests in order to analyze the dynamics of the granulation process. In particular the process of granulating agricultural limestone (Limestone, Aglime, Landmark, Australia) with water and 2.5% polyvinyl alcohol (PVA, Elvanol T66, Du Pont, USA) binder mix has been under investigation. The aim of this work is to design an MPC in order to control the process. With this aim in focus, three steps have been applied.

1. Step change experiments have been undertaken to gain input-output data sets which contain the plant dynamics.

2. A PBM developed by Wang et al. [2] has been fitted to the measurement data and controllability of the process was analyzed. Furthermore this yields a simulation test-bed to test the performance of the derived model predictive controller without the need to run labor and capital intensive experiments at the pilot plant.

3. System identification techniques have been used to derive a linear time invariant (LTI) model which was used as a prediction model for the model predictive controller.

The schematic of the pilot plant which is located at the Particle and System Design Center at the University of Queensland, Brisbane, Australia can be seen in Figure 1. The structure of the 1D-PBM which has been used for the numerical simulations is pictured in Figure 2 and is based on the Wang model [2].

Figure 1

Figure 2

Although multiple inputs in general allow an improved controllability of a plant, a single input strategy has been followed here for simplicity. Since the moisture to solid ratio is the most effective value [4, 5] to influence the particle size distribution, the powder feed-rate has been chosen to be the manipulated variable (MV) of our single input system. The output particle size distribution is described with the mass based mean (d₅₀) particle size, resulting in a single input single output (SISO) system. The experimental step response of the d₅₀ around a chosen steady state (SS) showed the shape of a first order transfer function (TF) with delay (t_d), even though the process is nonlinear. Because of this a simple first order TF model (PT_tdT₁) has been fitted to the experimental data. This has been done by deriving the three LTI model parameters: dc gain k, the time constant τ and the delay t_d from the step response data. Simulations show that satisfactory MPC tunings can be found for regulatory control. The flexibility of process constraint handling, the intuitive tuning of MPC, and the capability to handle multiple input and multiple output (MIMO) systems make MPC the first choice for closed-loop control of granulation.

Acknowledgments

Special thanks to David Page, University of Queensland, St. Lucia, Australia, for insightful conversations pertaining to the experimental setup of the pilot plant. The authors also acknowledge financial support by the International Fine Particles Research Institute (IFPRI), from the Australian Research Council (Discovery Grant DP0345777) and from the Engineering and Physical Sciences Research Council (Grant EP/C511301/1).

References

[1] S. M. Iveson, J. D. Lister, K. P. Hapgood, B. J. Ennis. Nucleation, growth and breakage phenomena in agitated wet granulation processes: a review. Powder Technology 117 (2001) 3-39.

[2] F. Y. Wang, X. Y. Ge, N. Balliu, I. T. Cameron. Optimal control and operation of drum granulation processes. Chemical Engineering Science 61 (2006), 257-267.

[3] C. F. W. Sanders, M. J. Hounslow, and F. J. Doyle III. Model Predictive Control of Wet Granulation Using an Experimentally Validated Population Balance Model. AIChE Annual Meeting (2006), San Francisco.

[4] A. A. Adetayo, J. D. Lister, S. E. Pratsinis, B. J. Ennis. Population balance modeling of drum granulation of materials with wide size distribution. Powder technology 82 (1995), 37-49.

[5] J. Zhang, J. D. Lister, F. Y. Wang, I. T. Cameron. Evaluation of control strategies for fertilizer granulation circuits using dynamic simulation. Powder Technology 108 (2000), 122-129.