Binary Distillation Column Research Articles

A model-free approach for auto-tuning of model predictive control

A two-layer approach for the auto-tuning of model predictive control (MPC) is proposed. The bottom layer computes the weighting matrices of the cost function from a desired closed-loop bandwidth while the top layer aims at finding the optimal bandwidth. This optimum corresponds to the optimal balance between the robustness and nominal performance of the closed-loop system. To find the optimal bandwidth, the extremum seeking (ES) algorithm, a form of non-model-based adaptive optimisation, is proposed. The auto-tuning approach is tested on a binary distillation column model. It is shown that the auto-tuning approach enables the MPC system to track its optimal closed-loop bandwidth and therefore obtain the minimum output variance.

A Tuning Approach for Offset-free MPC with Conditional Reference Adaptation

Model predictive control has become a widely accepted strategy in industrial applications in the recent years. Often mentioned reasons for the success are the optimization based on a system model, consideration of constraints and an intuitive tuning process. However, as soon as unknown disturbances or model plant mismatch have to be taken into account the tuning effort to achieve offset-free tracking increases. In this work a novel approach for offset-free MPC is presented, which divides the tuning in two steps, the setup of a nominal MPC loop and an external reference adaptation. The inner nominal loop addresses the performance targets in the nominal case, decouples the system and essentially leads to a first order response. The second outer loop enables offset-free tracking in case of unknown disturbances and consists of feedback controllers adapting the reference. Due to the mentioned properties these controllers can be tuned separate and by known guidelines. To address conditions with active input constraints, additionally a conditional reference adaptation scheme is introduced. The tuning strategy is evaluated on a simulated linear Wood-Berry binary distillation column example.

Incipient fault diagnosis using support vector machines based on monitoring continuous decision functions

Support Vector Machine (SVM) as an innovative machine learning tool, based on statistical learning theory, is recently used in process fault diagnosis tasks. In the application of SVM to a fault diagnosis problem, typically a discrete decision function with discrete output values is utilized in order to solely define the label of the fault. However, for incipient faults in which fault steadily progresses over time and there is a changeover from normal operation to faulty operation, using discrete decision function does not reveal any evidence about the progress and depth of the fault. Numerous process faults, such as the reactor fouling and degradation of catalyst, progress slowly and can be categorized as incipient faults. In this work a continuous decision function is anticipated. The decision function values not only define the fault label, but also give qualitative evidence about the depth of the fault. The suggested method is applied to incipient fault diagnosis of a continuous binary mixture distillation column and the result proves the practicability of the proposed approach. In incipient fault diagnosis tasks, the proposed approach outperformed some of the conventional techniques. Moreover, the performance of the proposed approach is better than typical discrete based classification techniques employing some monitoring indexes such as the false alarm rate, detection time and diagnosis time.

First Principle Modeling and Neural Network–Based Empirical Modeling with Experimental Validation of Binary Distillation Column

Abstract To get the better product quality and to decrease the energy consumption of the distillation column, an accurate and suitable nonlinear model is crucial important. In this work, two types of model have been developed for an existing experimental setup of continuous binary distillation column (BDC). First model is a theoretical tray-to-tray binary distillation model for describing the steady-state behavior of composition in response to changes in reflux flows and in reboiler duty. Another model is an artificial neural network (ANN)–based input/output data relationship model. In ANN-based model, temperature of first tray, feed flow rate, and column pressures have been taken in addition to reflux flow rate and reboiler heat duty as inputs to give the more accurate I/O relationship. The comparison of output of ANN model and the equation-based model with the real-time output of the experimental setup of BDC has been given for the validation of developed models.

Object-oriented approach applied to ANFIS modeling and control of a distillation column

Neurofuzzy networks are hybrid systems that combine neural networks with fuzzy systems, and the Adaptive Neuro-Fuzzy inference system (ANFIS) is a particular case in which a fuzzy system is implemented in the framework of an adaptive neural network. This neurofuzzy approach represents an effective structure to the modeling of plant dynamics, and the oriented-object programming environments offer an intuitive way to address this task. In this paper the MODELICA object-oriented environment has been applied to the ANFIS modeling and indirect control of the heavy and light product composition in a binary methanol-water distillation column by using the adaptive Levenberg–Marquardt approach. The results obtained demonstrate the potential of the adaptive ANFIS scheme under MODELICA for the dual control of composition both for changes in set points with null stationary error even when disturbances are present.

Branch and bound method for regression-based controlled variable selection

Self-optimizing control is a promising method for selection of controlled variables (CVs) from available measurements. Recently, Ye, Cao, Li, and Song (2012) have proposed a globally optimal method for selection of self-optimizing CVs by converting the CV selection problem into a regression problem. In this approach, the necessary conditions of optimality (NCO) are approximated by linear combinations of available measurements over the entire operation region. In practice, it is desired that a subset of available measurements be combined as CVs to obtain a good trade-off between the economic performance and the complexity of control system. The subset selection problem, however, is combinatorial in nature, which makes the application of the globally optimal CV selection method to large-scale processes difficult. In this work, an efficient branch and bound (BAB) algorithm is developed to handle the computational complexity associated with the selection of globally optimal CVs. The proposed BAB algorithm identifies the best measurement subset such that the regression error in approximating NCO is minimized and is also applicable to the general regression problem. Numerical tests using randomly generated matrices and a binary distillation column case study demonstrate the computational efficiency of the proposed BAB algorithm.

Effectiveness of Signal Excitation Design Methods for Identification of Ill-Conditioned and Highly Interactive Processes

A comparison of the effectiveness of five methods to generate excitation input signals for identification of ill-conditioned processes is presented. The FIT index of the identified models and the crest factor (CF) of the generated signals are employed to quantify the performance of each signal generation method. The scattering factor (SF), which measures the dispersion (nondirectionality) of the system output plane, is used as a third approach to evaluate the performance of the methods. All the methods are evaluated using a second order 2 × 2 MIMO model of a binary high-purity distillation column. The discussion of the results shows a consistent better performance of some of the studied methods under the conditions of the conducted experiments, as well as the influence of the plant dynamics and the model structure employed for the system identification.

Application of Feed Forward and Recurrent Neural Network Topologies for the Modeling and Identification of Binary Distillation Column

This paper presents identification of artificial neural network model of a Binary Distillation Column (BDC). In this paper, the two most common topologies of artificial neural networks in the area of control are introduced: Feed forward neural network and recurrent neural networks. The training of neural network has been performed by the data set acquired from real 9-tray continuous BDC setup available in laboratory. The network model is composed of two layers. A hyperbolic tangent sigmoid function and a pure linear function have been utilized as activation functions in the first and the second layers, respectively. The developed neural network model has been validated by an extensive data set of practical data received from real BDC setup.

Singular linear parameter‐varying observer for composition estimation in a binary distillation column

The aim of this study is to provide a full comprehensive study of the singular linear parameter-varying (LPV) systems and observer synthesis in a real process. The states and unknown inputs estimation problem for a binary distillation column is solved by using an LPV proportional-integral observer (PI-Observer). A singular LPV model of the process is developed. This model facilitates the design of the PI-Observer for singular LPV systems. The observer proposed is tested using experimental data of a distillation column separating an ethanol-water mixture. The results indicate that the observer is able to accurately reconstruct the product composition dynamics and the unknown inputs of the process.

Hierarchical control of large scale systems: A zone control approach

A methodology for hierarchical control of large scale systems is presented. This approach uses an integration of dynamic real time optimization (DRTO) and model predictive control (MPC) with zone control (coordinator). The dynamic real time optimization problem is solved in the upper layer and addresses the overall economic objective to provide the reference trajectories for the coordinator (middle layer) that calculates hierarchical input variables and output references that are always into the designed zone, this information is taken by the subsystems controllers (lower layer) to generate the input variables that is going to be applied, and that reach the reference that the coordinator generate. This methodology is applied to a complete chemical process composed by three chemical reactors type CSTR (Continuous Stirred Tank Reactor) and three non reactive binary distillation columns.

Sensor location for nonlinear dynamic systems via observability analysis and MAX-DET optimization

This paper presents an approach for determining the optimal placement of multiple sensors for processes described by a class of nonlinear dynamic systems. This approach is based upon maximizing a criterion, i.e., the determinant, applied to the empirical observability Gramian in order to optimize certain properties of the process state estimates. The determinant directly accounts for redundancy of information for placing multiple sensors via the covariance terms in the observability matrix. However, the resulting optimization problem is nontrivial to solve as it is a mixed integer nonlinear programming problem. In order to address this point, this paper also presents a decomposition of the optimization problem such that the formulated sensor placement problem can be solved quickly and accurately on a desktop PC. Properties of the presented technique are demonstrated and discussed in two case studies, one involving a binary distillation column and the other a packed bed reactor.

Firefly algorithm approach based on chaotic Tinkerbell map applied to multivariable PID controller tuning

Nowadays, a variety of controllers used in process industries are still of the proportional–integral–derivative (PID) types. PID controllers have the advantage of simple structure, good stability, and high reliability. A relevant issue for PID controllers design is the accurate and efficient tuning of parameters. In this context, several approaches have been reported in the literature for tuning the parameters of PID controllers using evolutionary algorithms, mainly for single-input single-output systems. The systematic design of multi-loop (or decentralized) PID control for multivariable processes to meet certain objectives simultaneously is still a challenging task. This paper proposes a new chaotic firefly algorithm approach based on Tinkerbell map (CFA) to tune multi-loop PID multivariable controllers. The firefly algorithm is a metaheuristic algorithm based on the idealized behavior of the flashing characteristics of fireflies. To validate the performance of the proposed PID control design, a multi-loop multivariable PID structure for a binary distillation column plant (Wood and Berry column model) and an industrial-scale polymerization reactor are taken. Simulation results indicate that a suitable set of PID parameters can be calculated by the proposed CFA. Besides, some comparison results of a genetic algorithm, a particle swarm optimization approach, traditional firefly algorithm, modified firefly algorithm, and the proposed CFA to tune multi-loop PID controllers are presented and discussed.

INCA MPC4Distillation: A Novel Industrial Solution for Cost-Efficient Non-Linear Distillation Column Modeling and Control

Distillation columns are one of the first unit processes to which MPC (Model-based Predictive Control) was successfully applied in an industrial context. The methodology that is used to execute such APC (Advanced Process Control) projects has not fundamentally changed in the decades that have passed since its inception. Linear black-box models are still the most commonly used type of models in such applications. However, obtaining black-box models by performing dedicated plant tests is often prohibitively time-consuming (and hence costly), which limits the cost-efficient application of APC. On the other hand, in several applications non-linear models offer significant benefits when combined with a high-performance non-linear APC controller.INCA MPC4Distillation employs a novel modeling approach, by combining physical knowledge and historical data in order to identify dynamic models that can be used for APC. This new technology results in accurate non-linear models and does not rely on dedicated step tests, which results in increased benefits to the customer and at a lower cost than existing methods. Experience shows that INCA MPC4Distillation models can typically be obtained in a matter of days, as opposed to several weeks when one would use the classical modeling approaches. The results are illustrated using results obtained on a high-purity binary distillation column.

Convex formulations for optimal selection of controlled variables and measurements using Mixed Integer Quadratic Programming

The appropriate selection of controlled variables is important for operating a process optimally in the presence of disturbances. Self-optimizing control provides a mathematical framework for selecting the controlled variables as combinations of measurements, c=Hy, with the aim to minimize the steady state loss from optimal operation. In this paper, we present (i) a convex formulation to find the optimal combination matrix H for a given measurement set and (ii) a Mixed-Integer Quadratic Programming (MIQP) methodology to select optimal measurement subsets that result in minimal loss. The methods presented in this paper are exact for quadratic problems with linear measurement relations. The MIQP methods can handle additional structural constraints compared to the branch and bound (BAB) methods reported in literature. The MIQP methods are evaluated on a toy test problem, an evaporator example, a binary distillation column example with 41 stages and a Kaibel column with 71 stages.

An artificial immune system for adaptive fault detection, diagnosis and recovery

The human immune system provides rich metaphors for adaptive pattern recognition. Fault detection and diagnosis in chemical processes is commonly formulated as a pattern recognition problem. However, conventional methods for fault diagnosis often do not have a mechanism to adapt and learn as the process changes over time. In this paper, we propose an Artificial Immune System (AIS) framework that endows learning to statistical process monitoring techniques such as Principal component analysis. The proposed AIS framework also provides a direct means to incorporate recovery actions after a failure has been detected and diagnosed. We demonstrate the efficacy of the proposed framework using a simulated binary distillation column case study.

A generalized design methodology for the output feedback regulation of a special type of systems with bounded inputs

SUMMARYIn this work, a generalized design scheme for the output feedback regulation of a special type of systems with bounded inputs is proposed. It gives rise to a simple dynamic controller that guarantees the regulation objective avoiding input saturation, for any initial condition within a specific set that may comprehend the whole state space and that does not require any additional system data (apart from the output variable). Several processes, like double‐pipe heat exchangers, bioreactors, and binary distillation columns, are shown to be part of the type of systems that may be regulated through the developed methodology. The efficiency of the proposed scheme is corroborated through experimental and simulation results. Copyright © 2011 John Wiley & Sons, Ltd.

Comparative performance analysis of various binary coded PSO algorithms in multivariable PID controller design

In this paper, comparative performance analysis of various binary coded PSO algorithms on optimal PI and PID controller design for multiple inputs multiple outputs (MIMO) process is stated. Four algorithms such as modified particle swarm optimization (MPSO), discrete binary PSO (DBPSO), modified discrete binary PSO (MBPSO) and probability based binary PSO (PBPSO) are independently realized using MATLAB. The MIMO process of binary distillation column plant, described by Wood and Berry, with and without a decoupler having two inputs and two outputs is considered. Simulations are carried out to minimize two objective functions, that is, time integral of absolute error (ITAE) and integral of absolute error (IAE) with single stopping criterion for each algorithm called maximum number of fitness evaluations. The simulation experiments are repeated 20 times with each algorithm in each case. The performance measures for comparison of various algorithms such as mean fitness, variance of fitness, and best fitness are computed. The transient performance indicators and computation time are also recorded. The inferences are made based on analysis of statistical data obtained from 20 trials of each algorithm and after having comparison with some recently reported results about same MIMO controller design employing real coded genetic algorithm (RGA) with SBX and multi-crossover approaches, covariance matrix adaptation evolution strategy (CMAES), differential evolution (DE), modified continuous PSO (MPSO) and biggest log modulus tuning (BLT). On the basis of simulation results PBPSO is identified as a comparatively better method in terms of its simplicity, consistency, search and computational efficiency.

Binary Distillation Column Startup Using Two Temperature Measurements

The problem of designing the startup (SU) control scheme for a binary distillation column with two temperature and two level measurements is addressed. The objective is the quick attainment, with minimum off-spec production and heat consumption, of closed-loop stable operation in continuous regime. The combination of advanced constructive control and conventional distillation dynamics and control tools leads to a variable control structure scheme with: (i) an event controller which, on the basis of the two temperature measurements, decides the duration of an initial open-loop batch operation period at total reflux, and (ii) a set of four (two level and two temperature) linear-decentralized controllers activated in a sequential manner, which steers the column toward its nominal steady-state in a continuous regime, by manipulating the heat duty, reflux, distillate, and bottom flow rates. The proposed SU control scheme is tested through simulations with three high-purity column examples with ideal and nonideal thermodynamics, yielding a behavior that matches or improves the ones obtained in previous studies with detailed model-based schemes.

Study of the separation of simple binary and ternary mixtures of aromatic compounds

AbstractConsider binary and ternary mixtures of aromatic compounds composed of benzene/toluene and benzene/toluene/p‐xylene, respectively. The present study shows how one can apply both mass and energy balance equations in order to understand the separation of such ideal mixtures by distillation. An adiabatic flash distillation problem is solved graphically for a mixture composed of benzene/toluene. Rigorous resolution of a steady‐state binary distillation problem, using Mathematica® and the same benzene/toluene mixture, shows perfect agreement with results obtained using HYSYS. Results of a dynamic simulation involving the solution of a relatively large system of differential algebraic equations are presented and discussed for the benzene/toluene mixture. SIMULINK© and Mathematica® are used to perform control of this binary distillation column. Finally, a steady‐state simulation of a simple multicomponent mixture, composed of benzene/toluene/p‐xylene, is studied and some qualitative results are drawn from both the temperature and composition profiles. © 2011 Wiley Periodicals, Inc. Comput Appl Eng Educ 22: 87–98, 2014

Identification of a Pilot Scale Distillation Column: A Kernel Based Approach

AbstractThis paper describes the identification of a binary distillation column with Least-Squares Support Vector Machines (LS-SVM). It is our intention to investigate whether a kernel based model, particularly an LS-SVM, can be used for the simulation of the top and bottom temperature of a binary distillation column. Furthermore, we compare the latter model with standard linear models by means of mean-squared error (MSE). It will be demonstrated that this nonlinear model class achieves higher performances in MSE than linear models in the presence of nonlinear distortions. When the system is close to linear, the performance of the LS-SVM is only slightly better than the linear models.