N. Aziz

Nonlinear Modelling Application in Distillation Column

Distillation columns are widely used in chemical processes and exhibit nonlinear dynamic behavior. In order to gain optimum performance of the distillation column, an effective control strategy is needed. In recent years, model based control strategies such as internal model control (IMC) and model predictive control (MPC) have been revealed as better control systems compared to the conventional method. But one of the major challenges in developing this effective control strategy is to construct a model which is utilized to describe the process under consideration. The purpose of this paper is to provide a review of the models that have been implemented in continuous distillation columns. These models are categorized under three major groups: fundamental models, which are derived from mass, energy and momentum balances of the process, empirical models, which are derived from input-output data of the process, and hybrid models which combine both the fundamental and the empirical model. The advantages and limitations of each group are discussed and compared. The review reveals a remarkable prospect of developing a nonlinear model in this research area. It also shows the discovery of new advance methods in an attempt to gain a nonlinear model that is able to be used in industries. Neural network models have become the most popular framework in nonlinear model development over the last decade even though hybrid models are the most promising method to be applied for future nonlinear model development.

Comparison of various Wiener model identification approach in modelling nonlinear process

An accurate and simple model is essential to implement a model based controller. Wiener model is one of the simplest nonlinear models that can represent any nonlinear process. However, in Wiener Model development, there are several identification approaches available and need to be selected to produce the most accurate model. In this work, the nonlinear - linear approach, the linear - nonlinear approach, and the simultaneous approach are compared in identification of the Wiener model for nonlinear pH neutralization process. The parameters of linear block and the inverse of nonlinear block were obtained from several sets of data that are generated. These approaches are then compared in terms of model accuracy, calculation time, data requirement, and their flexibility.

Development of Process Inverse Neural Network Model to Determine the Required Alum Dosage at Segama Water Treatment Plant Sabah, Malaysia

Abstract The determination of the optimal coagulant dosage in the coagulation process of a water treatment plant (WTP) is very essential to produce satisfactory treated water quality and to maintain economic plant operation such as reducing manpower and expensive chemical costs. Failing to do this will also reduce the efficiency in sedimentation and filtration process in the treatment plant. Traditionally, jar test is used to determine the optimum coagulant dosage. However, this method is expensive, time-consuming and does not enable responses to changes in raw water quality in real time. Modeling such as neural network can be used to overcome these limitations. In this work, an inverse neural network model is developed to predict the optimum coagulant dosage in Segama WTP in Lahad Datu, Sabah, Malaysia. Real data from the WTP was obtained along with extensive data analysis and preparation, significant input-output selection and consideration of important raw and treated water lag parameters were carried out. The modeling results shown that the prediction capabilities are improving with the consideration of appropriate input parameters. Neural network models with different network architectures, including single and two hidden layers were developed and the optimum network architecture obtained was [11-27-9-1]. This model performed very well over the range of data used for training, with r-value of 0.95, mean square error (MSE) of 0.0019 and mean absolute error (MAE) of 0.024 mg/l when applied on the testing data set. Hence, the proposed techniques can significantly improve and have a great potential of replacing the conventional method of jar test due to its advantages; quick responsive tools, economical operating cost and its ability to be applied in real time process.

Nonlinear Model Predictive Control of a Distillation Column Using NARX Model

Abstract Distillation column is an important process unit in petroleum refining and chemical industries, and needs to be controlled close to optimum operating conditions because of economic incentives. Nonlinear model based control (NMPC) scheme is one of the best options to be explored for proper control of distillation columns. In this work, NMPC scheme using sigmoidnet based nonlinear autoregressive with exogenous inputs (NARX) model has been developed to control distillation column The Unscented Kalman Filter (UKF) was used to estimate the state variables in NMPC and the nonlinear programming (NLP) problem was solved using sequential quadratic programming (SQP) method. The closed loop control studies have indicated that the NARX NMPC performed well in disturbance rejection and set point tracking.

Simulation Studies and Sensitivity Analysis of Methyl Tert-butyl Ether Reactive Distillation

Abstract Methyl Tert-butyl Ether (MTBE) is an important chemical used as an octane booster in gasoline to replace tetra ethyl lead. Maximum production of the MTBE can be achieved using reactive distillation (RD) process that is operated at the optimum operating conditions and column configuration. However, optimizing the column configuration such as tray or catalyst location is experimentally expensive. Therefore, a reliable model of the MTBE reactive distillation is important to find the optimum conditions for MTBE production. In this work, continuous RD processes is simulated based on dynamic model using Aspen Dynamics. The model is then further used in Simulink for the singular value decomposition (SVD) analysis in order to select the best input–output pair for the control implementation. Finally, the step test is conducted in order to observe the sensitivity of the MTBE process toward changes of selected input variables. The results show that the model obtained from the Aspen produced a comparable result with the literature. The results also show that the tray temperature number 3 and 8 are the most sensitive output variables toward changes of reflux flowrate and reboiler duty.

Comparison of MIMO MPC and PI Decoupling in Controlling Methyl Tert-butyl Ether Process

Abstract The main control objective of Methyl Tert-butyl Ether (MTBE) reactive distillation is to maintain the MTBE product purity within the desired range at the highest isobutene conversion possible. However, MTBE purity and isobutene conversion have strong interaction, hence need a multiple input – multiple output (MIMO) based control system. In this work, MIMO model Predictive control (MPC) and Proportional Integral (PI) decoupling have been implemented to control both the MTBE purity and isobutene conversion by manipulating the reboiler duty and reflux flowrate. The purity of MTBE is at 99% and the isobutene conversion is at 99.83%. Performance of both controllers are then compared in term of integral absolute error (IAE), integral squared error (ISE), integral of the time-weighted absolute error (ITAE) and settling time. The results showed that the MIMO MPC is better than the PI decoupling.

Development of Sigmoidnet Based NARX Model for a Distillation Column

Distillation exhibits highly nonlinear dynamic behavior and the development of suitable nonlinear model to distillation pose a challenging problem to current process industry. In the absence of a reasonably accurate nonlinear model, distillation column is difficult to control using advanced model based control strategies. In this paper, a novel sigmoidnet based nonlinear auto-regressive with exogenous inputs (NARX) model is developed for high purity distillation column and verified using the experimentally validated first principle model. The model validation and regressor analysis proved that the developed NARX model was capable of capturing the nonlinear dynamics of a distillation column.

Kinetic Model with Effect of Water Content for Enzyme-Catalyzed Citronellyl Laurate Esterification Process

A kinetic model with effect of water content for enzyme-catalyzed citronellyl laurate was developed. These models incorporate the combined influences of established kinetics model with the function model on the effect of initial water content with kinetic parameters. The model development was carried out by performing a linear and nonlinear regression based on the behavior of the kinetic parameter profiles and validated with experimental data. Using the developed models, the influence of water content towards the enzyme-catalyzed initial rate of reaction was theoretically explained. It has been shown that the proposed model have good agreement between experimental data and intends to capture the effect of water content towards the conversion of ester. With this model, the optimal value of initial water content for this process could be estimated.

NARX-Model-Based Control (NARX-MBC) for Citronellyl Laurate Esterification Reactor

Abstract Model-based control nowadays appears to be a very promising control strategy for various processes. However, this type of control strategy requires an accurate, low complexity, easy identifiability, structural flexibility and if possible, invertibility type of models. In this paper, a non-linear autoregressive exogenous-model-based control (NARX-MBC) has been designed and implemented to a batch citronellyl laurate esterification reactor. Multi-input-single-output (MISO) model has been developed to representing the process using NARX modelling approach. The performance of the developed NARX-MBC is evaluated and compared with a conventional PID controller. Overall, it is observed that the former has outperformed the latter.

Implementation of neural network inverse-model-based control (NN-IMBC) strategy in batch reactors

Neural Network Inverse-Model-Based Control (NN-IMBC) strategy is used to track the optimal reactor temperature profiles and its performance is evaluated through a few robustness tests. A complex exothermic batch reaction scheme is used as a case study. The optimal reactor temperature profiles are obtained by solving optimal control problems off-line using Control Vector Parameterisation (CVP) and Successive Quadratic Programming (SQP) techniques. The NN-IMBC strategy is evaluated in tracking both the constant and dynamic optimal set points. Neural Network estimator is embedded to the strategy as the on-line estimator to estimate the amount of heat released by the chemical reaction. The NN-IMBC is found to be well performed in tracking both set points and accommodating changes within its range of training. It also promises robust controller if it is trained with a wide range of the reactor temperature covering all possible conditions of the process and is much easier to implement compared to other typical types of controllers because no tuned parameter is needed. Therefore, it can lead to efficient and profitable operation and provide a better business decision making in setting up new plants or improving the existing operations.