Control of polystyrene batch reactors using neural network based model predictive control (NNMPC): An experimental investigation

https://doi.org/10.1016/j.conengprac.2011.01.007Get rights and content

Abstract

Controlling batch polymerization reactors imposes great operational difficulties due to the complex reaction kinetics, inherent process nonlinearities and the continuous demand for running these reactors at varying operating conditions needed to produce different polymer grades. Model predictive control (MPC) has become the leading technology of advanced nonlinear control adopted for such chemical process industries. The usual practice for operating polymerization reactors is to optimize the reactor temperature profile since the end use properties of the product polymer depend highly on temperature. This is because the end use properties of the product polymer depend highly on temperature. The reactor is then run to track the optimized temperature set-point profile. In this work, a neural network-model predictive control (NN-MPC) algorithm was implemented to control the temperature of a polystyrene (PS) batch reactors and the controller set-point tracking and load rejection performance was investigated. In this approach, a neural network model is trained to predict the future process response over the specified horizon. The predictions are passed to a numerical optimization routine which attempts to minimize a specified cost function to calculate a suitable control signal at each sample instant. The performance results of the NN-MPC were compared with a conventional PID controller. Based on the experimental results, it is concluded that the NN-MPC performance is superior to the conventional PID controller especially during process startup. The NN-MPC resulted in smoother controller moves and less variability.

Introduction

The favorable properties of polymeric products such as its usage, flexibility, lightweight, low cost and its ease of processing, resulted in rapid soar on their demand. However, polymerization processes involve complex reaction mechanisms and their process dynamics is highly nonlinear in nature. Practicing engineers face considerable pressure to operate smooth, low cost and safe polymer production lines. Advanced control techniques can be used as a viable solution for controlling and improving the efficiency and productivity of such nonlinear processes. The target here is to upgrade the control system in order to reduce energy consumption as well as to minimize production cost. However, the control of polymerization reactors in general and particularly batch polymerization reactors is very difficult due to its complex characteristics. In the usual practice of designing a good process controller, an accurate model is required for the chemical reactor and process.

Nowadays, vast variety of computational and control algorithms are widely implemented in the chemical industry. However, the application towards polymerization engineering is still very limited (Kiparissides, 1996). As the polymer industry became more global and competitive production pressures intensifies, the future is even brighter and more exciting for computer based algorithms and control software packages, aiming at safer and more stable plant operation, productivity improvement, quality improvement, cost-reduction, energy conservation, waste reduction and manpower reduction. Thus, by implementing advanced control techniques there are significant incentives for improving reactor operation and the efficiency of quality monitoring, which can translate to improved product quality, safer operation, and as the ultimate consequence, improved profits.

Model predictive control (MPC) is a generic term for a widely used class of process control algorithms (Cueli and Bordons, 2008, Harnischmacher and Marquardt, 2007, O'Brien et al., 2011). This type of control offers a flexible and powerful solution to the dynamic optimization and control of polymerization reactors. Different MPC schemes have been recently developed for polymerization processes (Causa et al., 2008, Henson, 1998, Karer et al., 2008a, Perea et al., 2010, Zhu et al., 2000), examples are: linear/nonlinear model predictive control (MPC) (Chew et al., 2007, Karer et al., 2008b, Konakom et al., 2008, Nagy, 2007, Salau et al., 2009, Shafiee et al., 2008), dynamic matrix control (DMC) (Altinten, Erdogan, Hapoglu, Aliev, & Alpbaz, 2006), generalized predictive control (GPC) (Ekpo and Mujtaba, 2008, Özkan et al., 2006) and internal model control (IMC) (Mujtaba, Aziz & Hussain, 2006). Table 1 shows the recently published research effort done to control batch polymerization reactors utilizing the MPC technology. These control schemes greatly improve the control action in terms of accuracy and robustness. However, the linear MPC offers poorer performance in controlling the polymerization processes since these processes are highly nonlinear. Therefore, an MPC based on a nonlinear model is more desirable (Atuonwu, Cao, Rangaiah, & Tadé, 2010). Neural networks offer an alternative nonlinear modeling approach for MPC (Ekpo and Mujtaba, 2007a, Willis et al., 1992). Nonlinear predictive control algorithms based on neural network (Ekpo and Mujtaba, 2008, Yu and Zhang, (2006)) imply the minimization of a cost function using computational methods for obtaining the optimal controller move command (Ou & Rhinehart, 2003). However, most of the previous studies related to MPC and neural network based MPC were limited to simulation based theoretical studies and very few papers exist that deal with the experimental verification of these controllers (Özkan et al., 2006, Zeybek et al., 2006). It is worth mentioning that the experimental results of Özkan et al. (2006) and Zeybek et al. (2006) for setpoint tracking of polystyrene polymerization reactors shows large oscillation throughout the batch run. Hence, there is considerable opportunity and need to improve the design and tuning of the nonlinear MPC for the control of batch polystyrene polymerization reactors.

In this work, a neural network based model predictive controller (NN-MPC) was developed for the control of a free radical polystyrene batch reactor experimentally. The NN-MPC was tuned and then implemented online to control the process in real-time experiments. The experiments were also conducted using a conventional PID controller and the performance of these two controllers were analyzed and compared.

Section snippets

Batch polystyrene reactor system

A schematic diagram of the experimental batch polystyrene polymerization reactor is shown in Fig. 1. A 2000 ml jacketed glass reactor was used with 12 cm inside diameter and 20 cm depth. The working volume capacity of the reactor is 1.5 l. Thermocouples were used for measuring the reactor, jacket inlet and outlet temperature. The mixture inside the reactor is stirred using a 25 mm diameter turbine agitator located 7 cm above the base of the reactor. The motor speed of the agitator can be adjusted in

Hybrid modeling

The free radical polymerization of monomer styrene, solvent (toluene) and a monomer soluble initiator benzyol peroxide (BPO) is modeled using a combined artificial neural network (ANN)-mechanistic modeling strategy. Traditionally, this process is modeled using a kinetic model derived from mass balances of the polymerization system with the reaction temperature. Further improvement of this model involves the effects of energy balance around the reactor and cooling jacket. In this work, the

Model predictive control

MPC can be defined, as a control scheme in which the control algorithm computes a manipulated variable profile which optimizes an objective function subject to a number of plant model and constraint functions over a finite future time horizon. The first move of this open-loop optimal manipulated variable profile is then implemented until a new plant measurement becomes available (Eaton & Rawlings, 1992).

Fig. 3 shows the basic strategy of the model predictive control. Based on measurements

Minimum time optimal temperature profile

The original minimum time optimal temperature profile policy was obtained from the work of Ponnuswamy, Shah, and Kiparissides (1987). Recently, the offline or open loop minimum time optimal control policies were applied to the solution of styrene polymerization in batch reactors (Altinten et al., 2008, Özkan et al., 1998, Özkan et al., 2001, Zeybek et al., 2004, Zeybek et al., 2006). In this work, the optimization problem of minimum time optimal temperature policy has been formulated and solved

Designing neural network for MPC

The polymerization first principle-NN kinetic model described above was simulated using Matlab software. The reactor operating conditions and the design parameters used in the simulations are given in Table 3. The prediction capability of this model is demonstrated by plotting the experimental values of the reactor temperature verses the first principle-NN kinetic model predictions as shown in Fig. 7.

The process of obtaining the neural network dynamic model for the batch polystyrene reactor is

Conclusion

The NN-MPC control algorithm was simulated and implemented for the control of a batch polystyrene polymerization reactor in real time. This is to cover the gap between theoretical and real time control studies for this reactor. The performance of the NN-MPC for the optimal setpoint tracking on the polystyrene reactor was compared with conventional PID controller. The experimental results showed that the NN-MPC was able to track the optimal reactor temperature profile efficiently and without a

References (61)

  • E.E. Ekpo et al.

    Evaluation of neural networks-based controllers in batch polymerisation of methyl methacrylate

    Neurocomputing

    (2008)
  • S. Erdoğan et al.

    The effect of operational conditions on the performance of batch polymerization reactor control

    Chemical Engineering Journal

    (2002)
  • M.A. Greaves et al.

    Neural-network approach to dynamic optimization of batch distillation—Application to a middle-vessel column

    Chemical Engineering Research & Design

    (2003)
  • G. Harnischmacher et al.

    Nonlinear model predictive control of multivariable processes using block-structured models

    Control Engineering Practice

    (2007)
  • M.A. Henson

    Nonlinear model predictive control: Current status and future directions

    Computers & Chemical Engineering

    (1998)
  • Y. Hiltunen et al.

    Quantification of human brain metabolites from in vivo 1H NMR magnitude spectra using automated artificial neural network analysis

    Journal of Magnetic Resonance

    (2002)
  • G. Karer et al.

    Model predictive control of nonlinear hybrid systems with discrete inputs employing a hybrid fuzzy model

    Nonlinear Analysis: Hybrid Systems

    (2008)
  • G. Karer et al.

    Self-adaptive predictive functional control of the temperature in an exothermic batch reactor

    Chemical Engineering and Processing: Process Intensification

    (2008)
  • C. Kiparissides

    Polymerization reactor modeling: A review of recent developments and future directions

    Chemical Engineering Science

    (1996)
  • K. Konakom et al.

    Batch control improvement by model predictive control based on multiple reduced-models

    Chemical Engineering Journal

    (2008)
  • I.M. Mujtaba et al.

    Neural network based modelling and control in batch reactor

    Chemical Engineering Research and Design

    (2006)
  • Z. Nagy et al.

    Model predictive control of a PVC batch reactor

    Computers & Chemical Engineering

    (1997)
  • Z.K. Nagy

    Model based control of a yeast fermentation bioreactor using optimally designed artificial neural networks

    Chemical Engineering Journal

    (2007)
  • Z.K. Nagy et al.

    Evaluation study of an efficient output feedback nonlinear model predictive control for temperature tracking in an industrial batch reactor

    Control Engineering Practice

    (2007)
  • M. O'Brien et al.

    Model predictive control of an activated sludge process: A case study

    Control Engineering Practice

    (2011)
  • J. Ou et al.

    Grouped neural network model- predictive control

    Control Engineering Practice

    (2003)
  • G. Özkan et al.

    Generalized predictive control of optimal temperature profiles in a polystyrene polymerization reactor

    Chemical Engineering and Processing

    (1998)
  • G. Özkan et al.

    Non-linear generalised predictive control of a jacketed well mixed tank as applied to a batch process—A polymerisation reaction

    Applied Thermal Engineering

    (2006)
  • G. Özkan et al.

    Nonlinear control of polymerization reactor

    Computers & Chemical Engineering

    (2001)
  • L.F.T. Perea et al.

    Development of on-line optimization-based control strategies for a starved-feed semibatch copolymerization reactor

    Control Engineering Practice

    (2010)
  • Cited by (117)

    • Recurrent Neural Network-based Internal Model Control design for stable nonlinear systems

      2022, European Journal of Control
      Citation Excerpt :

      Owing to their modeling capabilities, NNs have been widely adopted in conjunction with model-based control strategies. For example, the use of RNNs as system models for Model Predictive Control (MPC) design has been explored both in the industry [17,21,32], and in the academia [8,26,31,33]. hile MPC generally attains remarkable performances, constraint satisfaction and – if suitably designed - closed-loop stability guarantees [31], it requires to solve an online nonlinear optimization problem at each control step, which might be computationally prohibitive in some applications.

    View all citing articles on Scopus
    View full text