Ilknur Atasoy

A software for parameter estimation in dynamic models

A common problem in dynamic systems is to determine parameters in an equation used to represent experimental data. The goal is to determine the values of model parameters that provide the best fit to measured data, generally based on some type of least squares or maximum likelihood criterion. In the most general case, this requires the solution of a nonlinear and frequently non-convex optimization problem. Some of the available software lack in generality, while others do not provide ease of use. A user-interactive parameter estimation software was needed for identifying kinetic parameters. In this work we developed an integration based optimization approach to provide a solution to such problems. For easy implementation of the technique, a parameter estimation software (PARES) has been developed in MATLAB environment. When tested with extensive example problems from literature, the suggested approach is proven to provide good agreement between predicted and observed data within relatively less computing time and iterations.

Comparison of Control Strategies for Dissolved Oxygen Control in Activated Sludge Wastewater Treatment Process

Abstract Six control strategies; PID control, Model Predictive Control (MPC) with linear model, MPC with non-linear model, Nonlinear Autoregressive-Moving Average (NARMA-L2) control, Neural Network Model Predictive Control (NN-MPC) and optimal control with sequential quadratic programming (SQP) algorithm were evaluated via simulation of activated sludge wastewater treatment process. Controller performance assessment was based on rise time, overshoot, Integral Absolute Error (IAE) and Integral Square Error (ISE) performance criteria. As dissolved oxygen level in the aeration tank plays an important role in obtaining the effluent water quality, and in operating cost, it was chosen as the controlled variable. It was concluded consequently that NARMA-L2 controller and optimal control with SQP would outperform the others in achieving the specified objective.

An integration based optimization approach for parameter estimation in dynamic models

Abstract A common problem in model verification is to determine the values of model parameters that provide the best fit to measured data, based on some type of least squares or maximum likelihood criterion. In the most general case, this requires the solution of a nonlinear and frequently nonconvex optimization problem. some of the available software lack in generality, while others do not provide ease of use. As the need for a user-interactive parameter estimation software, especially for identifying kinetic parameters, was needed; in this work we developed an integration based optimization approach to provide a solution to such problems. For easy implementation of the technique, a parameter estimation software (PARES) has been developed in MATLAB environment. When tested with extensive example problems from literature, the suggested approach is proven to provide good agreement between predicted and observed data within relatively less computing time and iterations.

Molecular weight control in acrylonitrile polymerization with neural network based controllers

Abstract Acrylic fiber is commercially produced by free radical polymerization, initiated by a redox system. Industrial production of polyacrylonitrile is a variant of aqueous dispersion polymerization, which takes place in homogenous phase under isothermal conditions with perfect mixing. The fact that the kinetics is a lot more complicated than that of ordinary polymerization systems makes the problem of controlling molecular weight a difficult one. On the other hand, abundant data is being gathered in industrial polymerization systems, and this information makes the neural network based controllers a good candidate for a difficult control problem. In this work, neural network based control of continuous acrylonitrile polymerization is studied, based on our previously developed new rigorous dynamic model for the polymerization of acrylonitrile. Two typical neural network controllers are investigated: model predictive control and NARMA-L2 control. These controllers are representative of the variety of common ways in which multilayer networks are used in control systems. As with most neural controllers, they are based on standard linear control architectures. The concentration of bisulfite fed to the reactor as the manipulated variable and weight average molecular weight of the polymer as an output function are used in control studies. The results present a comparison of two common neural network controllers, and indicate that the model predictive controller requires larger computational time. Furthermore, the model predictive controller involves difficulties in determining the values for the weighting factor and the prediction horizons. The NARMA-L2 controller requires minimal online computation.

MOLECULAR WEIGHT CONTROL IN ACRYLONITRILE POLYMERIZATION

ABSTRACT Acrylic fiber is commercially produced by free radical polymerization, initiated by a redox system. The fact that the kinetics is a great deal more complicated than that of ordinary polymerization systems makes the problem of controlling molecular weight a difficult one. In this study, dynamics and control of continuous acrylonitrile polymerization are studied based on a previously described kinetics by Peebles (Applied Polymer Science, 1973, 17, 113–128). As the conventional feedback controller was found to be unsuccessful, a model state feedback (MSFB) control strategy was implemented. The performances of linear and nonlinear controllers have been compared via simulation, and it was concluded that the nonlinear form would be effectively employed for set point tracking as well as disturbance rejection.

Optimization of Molasses and Air Feeding Profiles in Fed-Batch Baker's Yeast Fermentation

Abstract This work focuses on maximization of the amount of biomass in the production of bakers yeast in fed-batch fermenters while minimizing the undesirable alcohol formation, by regulating the molasses and air feed rates. An optimization mechanism coupled with a state estimation algorithm and an Artificial Neural Network model based on original industrial data has been designed. As substrate and biomass concentrations required within this structure can not be measured on-line, these variables were predicted by an artificial neural network model using other measurable variables. Nonmeasurable substrate concentrations were successfully estimated by Kalman filtering using industrial data and thus, obtained new data sets were used for training the neural network model. Subsequently, through an SQP based optimization algorithm feeding profiles yielding maximum biomass and minimum alcohol formation were obtained. When computed results were compared to the industrial data, it was seen that molasses feeding profiles were compatible whereas aeration profiles were considerably different. The reason of this discrepancy was due to the agitation of the industrial fermenter with air at high air flow rates in order to provide better mixing in the reactor. Since the aeration profile obtained is associated with only the reproduction of microorganisms, it is postulated that the suggested optimization strategy may be industrially applicable for the maximization of biomass where enough agitation is provided by other means.,

A Model for Molecular Weight Prediction in Acrylonitrile Polymerization

The number and weight-average molecular weights in acrylonitrile polymerization have been calculated previously by Peebles[ 2 ]. However, the foundations for the two critically important expressions leading to the calculation of molecular weights were not disclosed in detail, no dynamics were presented, and predictions were not in good agreement with the experimental data, particularly in terms of polydispersity index. The present work focuses on the same issue, and brings a new rigorous dynamic model, based on the kinetics given by Peebles[ 2 ]. The new, more detailed model defines the chain lengths in terms of the leading moments of active and dead polymer, provides the prediction of reactor dynamics in compliance with the practice in industry, and estimates the polydispersity index of the polymer with better agreement to the experimental data.

İZOTERMAL OLMAYAN SÜREKLİ KARIŞTIRMALI BİR TANK REAKTÖRDE YAPAY SİNİR AĞI İLE DERİŞİM KONTROLÜ

Bu calismada, surekli karistirmali bir tank reaktorde (CSTR) cikis derisiminin yapay sinir agi (YSA) temelli kontrolu incelenmistir. Izotermal olmayan, ekzotermik ve birinci mertebeden tersinmez bir tepkimenin gerceklestigi bu reaktorde cikis derisiminin kontrolu sogutma suyu akis hizi ayarlanarak saglanmistir. Sistemin dogrusal olmama yapisinin cok yuksek olmasindan dolayi kontrol stratejisi olarak yapay sinir agi kontrol yapilarindan YSA-Ongormeli kontrol ve NARMA-L2 (Nonlinear Auto Regressive Moving Average) kontrol stratejileri olusturulmus, klasik PID kontrol edici ile performans karsilastirmasi yapilmistir. Calismada kullanilan YSA kontrol edicilerin her ikisi de cok hizli ve kisa zamanda set noktasina ulasmayi basarmistir. YSA-Ongormeli kontrol edici ve NARMA-L2 kontrol edici yapilari, PID kontrol edici ile kiyaslandiginda daha iyi bir performans gostermislerdir. Sonuclar, literaturde yer alan ve ayni ornek problem icin kurulan YSA temelli DNNC (Dynamic Neural Network Control) ve NIMC’den (Nonlinear Internal Model Control) elde edilen sonuclardan daha iyi gorunmektedir.

A semi heuristic MINLP algorithm for production scheduling

Abstract A large number of process synthesis, design and control problems in chemical engineering can be formulated as a mixed integer nonlinear programming (MINLP) model, involving continuous variables and integer decisions. In this paper, we present an MINLP formulation for production scheduling of multi-product batch plants. The binary (0/1) integer varibles of the non-convex MINLP problem make it impossible to use the general purpose algorithms for solution. In order to overcome this difficulty, a semi-heuristic algorithm for production scheduling was developed. Using this approach, the non-convex MINLP problem is first considered as an MILP problem without dividing the orders. Thus, the order that causes prolonged delivery time can be identified. Constraints for this order are then relaxed and MILP problem is re-solved using the new constraints. Having reached the new schedule, the quantitative distribution of the specific order to different units can be determined by solving the LP problem that does not contain integer variables since allocation of the orders to the units and processing order are known. The results obtained with some example problems indicate improvements over previous schedules and therefore, give promise that the suggested strategy might be used in moderately-sized industrial applications.