Design of Operable Reactive Distillation Columns Myrian Andrea Schenk A thesis submitted for the degree of Doctor of Philosophy London. University the of of Department of Chemical Engineering University College London, London WClE 7JE September, 1999 It. Abstract Reactive distillation is an integrated process which considers simultaneous physical and chemical transformations. It is increasingly receiving attention both from industry and academia. Significant advances have been made in the area of modelling and simulation as well as the implementation of such industrially. However, the area of control and optimisation of such units has been units not explored thoroughly. The thesis presents a general framework for simulation and design, which can handle reactive and non-reactive systems. Various different aspects of the distillation described have been in modelling and simulation of order to understand the behaviour of the reactive distillation columns. In the framework both simulation modes, steady-state and dynamic, are considered and the described by In is a equilibrium and non equilibrium-based models. process (or liquid between transfer and non equilibrium) model, mass rates rate-based based Maxwell-Stefan the equaon vapour phase are considered explicitly, tions. Equilibrium is attained at the phase interface in the non-equilibrium from it A to the to go one model switching policy makes possible model. free by following Gibbs knowledge based the the gained, energy as on other, has helped determining in Tray function the time. switch efficiency also of a between the non-equilibrium and equilibrium models, and has been studied for various systems. The existence of multiple steady state has been verified through simulation diagrams Bifurcation hybrid the the also confirmed existence of model. with in the simulations. output multiplicity obtained Analysis of process controllability at the design stage has been shown to proIn for thesis improving the operation. controllability process vide guidance first tofor the studied are presented as a step systems reactive measures wards control structure selection. design for the A method of reactive separation columns at miniobtaining (investment and and which will operating costs) total cost annualised mum is in the be able to maintain stable operation also presence of variability presented. Acknowledgements I would like to thank my supervisors David Bogle, Rafiqul Gani and Stratos Pistikopoulos. Their continuous support and guidance has led to the production of this work. It has been a pleasure to work next to them and I also thank them for the opportunity of this joint research. The many technical discussions at ICI, especially with Tom Malik, are also greatly appreciated. Financial support from the Centre for Process Systems Engineering, CAPECResearch Center and ICI is gratefully acknowledged. Special thanks go to Carlos, Nicolas, Sonia, Alejandro and Marta for introducing me to the challenge of research. It has been a pleasure to work amongst different groups, in Lyngby and here in London. The time would not have been memorable and enjoyable withlike have had I I to the thank particularly would out nice people around. to: Eduardo, with whom I have worked and discussed a lot about reactive distillation, and with whom I shared the difficult times far away from home; Cecilia, my office-mate from the old times, it have been a pleasure to work discussed Celeste, her; I I to piles of work and with whom with whom next keep my Spanish 'up-to-date'; my part-time flat-mates in the 'London time', the 'adventure guys', Soraya and Geoff; Vik, my desk neighbour at IRC, who day. has to the working start a smile ready always My dearest friends Marianne, Ari, Stefan and Angelika will always be refor in friendship for their their encouragement good and sincere membered bad times. and I would also like to thank my family for their love and everlasting confidence through the years. Last but not least, I would like to thank Jes, my lovely husband, without infallible in been. His have trust this my support and may never whom all these has years. over all me motivated work 2 Contents Introduction 1.1 Reactive Distillation 12 1.2 Motivation 14 1.2.1 1.3 2 3 12 ....................... Aim Work the of and ................ Problem Overview Definition .................... 15 .............................. Review Literature Background: 18 2.1 Reactive and Non-reactive Distillation 2.2 Modelling- Design- Synthesis Issues 3.1 3.2 Equilibrium 2.2.2 Non-equilibrium Equilibrium 21 .......... ................ Non-equilibrium Models ............. 3.2.1 Maxwell-Stefan Equations 3.4 Model Formulation 3.5 Equations 28 33 Models Model Equations 3.7 ................ .................... (rate-based) Models 18 21 Issues 3.3 3.6 in Process Engineering Models 2.2.1 Modelling 15 ................. ................ ..................... Generation of Models ............... 3.6.1 Tray Efficiency .............. Condensers and Reboilers ............ 3 ........ ........ ........ ........ ........ ........ ........ ........ ........ ........ 35 35 36 39 41 42 59 66 69 3.8 Computational and Numerical Aspects ............. 3.8.1 Specifications for Equilibrium and Non-equilibrium Sim- 71 ulations 3.8.2 .......................... Solution of Model Equations 3.8.3 Initialisation 3.8.4 3.9 71 ............... 74 ........................ Switching between Models and Monitoring of Phenomena 76 Conclusions 76 ............................ 4 Validation 4.1 of models Systems Studied .................... 4.1.1 MTBE: Methyl- Tert- B utyl- Ether 4.1.2 78 ...... ...... Ethyl Acetate ................. 4.1.3 Depropanizer Column ............ 4.2 Initialisation ...................... 4.3 Validation of Models ................. 4.3.1 MTBE ..................... 4.3.2 Ethyl Acetate ................. 4.3.3 Depropanizer Column ............ 4.4 Bifurcation Analysis 4.4.1 4.5 ................. Applied Bifurcation Analysis Conclusions ...... ...... ...... ...... ...... ...... ...... ...... ...... ......... ...... ...... ...................... 79 79 86 91 91 94 94 102 106 107 109 114 116 5.3 Models Hybrid with Monitoring of Phenomena through Gibbs Free Energy ..... Tray Efficiency Analysis ..................... Illustration of Hybrid Modelling Approach ........... 5.4 Conclusions 132 5 Analysis 5.1 5.2 6 71 Controllability 6.1 ............................ Issues and Dynamic Controllability 6.1.1 Issues Linear Model ....................... ....................... 4 Optimization 117 119 128 133 134 135 6.1.2 6.2 6.1.3 .................. Controllability Tools 6.1.4 Control Configurations 6.1.5 Examples of Application Dynamic Optimization 6.2.1 6.3 7 Scaling Optimization Conclusions ......... .......... ......... ......... ......... ........ ......... ............. problem ......... ......... ......... ................... General Conclusions ......... and Future Work 137 139 140 143 150 153 159 160 Nomenclature 163 References 167 A Properties 179 A. 1 Enthalpy 179 A. 2 Density .................... .......... .................... .......... A. 3 Viscosity .................... A. 4 Surface Tension ................ A. 5 Thermal Conductivity ............ A. 6 Binary Diffusion Coefficients ......... .......... .......... .......... .......... 180 181 181 182 184 B Results B. 1 Simulation Results for MTBE .................. B. 1.1 Controllability Analysis for the MTBE production ... B. 2 Esterification of Acetic Acid with Ethanol ........... 184 189 203 219 C Optimization CA 180 Objective Function and Disturbances 5 .............. 219 L0ist of Figures 1.1 Exemplary reactive distillation 2.1 Residue curve map for MTBE. (P=Iatm)(Ung 1995b) ............................... 2.2 3.1 4.1 column 16 ............. Doherty, and 25 McCabe-Thiele like diagram for MTBE. (Perez Cisneros et al., 1997a) ............................... 26 Schematic representation of an stage in the column (KrishnaTaylor, 1985) murty and ..................... 40 Multiple solutions in the MTBE column 96 ............ Temperature profiles for low and high conversionin the MTBE column ............................... 4.3 Equilibrium constant for MTBE synthesis ........... Colombo Hoffmann, Rehfinger Simulation 4.4 et and results with for MTBE DIPPR the column ............. al. and for ASPEN PLUS those Simulation 4.5 of results compared with the MTBE column ........................ 4.2 4.6 Simulation results with the equilibrium and non-equilibrium for MTBE the column ................... model for different thermodynamic Simulation models 4.7 results with the esterification column ..................... for data the Simulation 4.8 results compared with experimental esterification column ....................... 4.9 Mass Transfer rate for the esterification column ........ literature Comparison transfer results .... 4.10 rate with of mass 6 97 98 99 101 102 103 104 105 106 4.11 Simulation results with the non-equilibrium propanizer column ........................ for dethe model 108 4.12 Vapour compositions compared with ChemSep simulations 109 .. 4.13 Bifurcation diagram for the esterification column varying reflux flow rate ........................... 4.14 Bifurcation diagram for the MTBE column varying reflux flow 111 112 rate ................................ 4.15 Bifurcation diagram for the MTBE boil-up ............................... 5.1 column varying vapour 113 Excess Gibbs energy for the esterification column. Trays I 7................................ and 118 5.2 Excess Gibbs energy for the MTBE column. Trays 4 and 10. . 119 5.3 Vapour compositions for the esterification column ....... 5.4 Efficiency for the esterification calculated 5.5 Efficiency depropanizer for the calculated 5.6 Vapour composition 5.7 Efficiency calculated for the depropanizer for the depropanizer column column 120 121 ....... 122 ....... 123 column ....... from far column steady-state ............................ 5.8 Efficiency calculated for the depropanizer, compared with those Ethane ChemSep: of ....................... 5.9 Efficiency calculated for the depropanizer, compared with those Propane ChemSep: of ....................... Propane depropanizer for the Efficiency 5.10 Point column: .... fixed 5.11 Liquid composition profile with equilibrium model with model .............. efficiency and non-equilibrium fixed 5.12 Vapour composition profile with equilibrium model with model ............. . efficiency and non-equilibrium 5.13 Hybrid the to esterification model applied column ...... . 124 125 126 127 129 130 131 (10% linear from the model Response non-linear and 6.1 obtained 138 step change)............................ in increase 10% MTBE reboiler 6.2 High conversion of as result of 145 heat duty .............................. 7 6.3 Low conversion of MTBE as a result of 15% decreasein reboiler heat duty 146 ........................... 6.4 Lower purity of MTBE as a result of adding extra reactive stages147 6.5 Reactive distillation column for the design study 154 ........ 6.6 Top and bottom compositions of ethyl acetate. Results from the simulated nominal casewith disturbance 156 ........... 6.7 6.8 Ethyl acetate composition. Results from the dynamic optimization ............................... Controlled variable: Reboiler Heat Duty(* 10'(MJlh)) Re. from dynamic the sults optimization ............... 157 158 B. 1 Decreasing 10% methanol feed flow-rate ........ ..... B. 2 Increasing 10% methanol feed flow-rate ......... ..... B. 3 Increasing 10% isobutene feed composition ....... ..... B. 4 Decreasing 15% isobutene feed composition ...... ..... B. 5 Decreasing 10% isobutene feed composition ...... ..... B. 6 Poles and Zeroesfor Case I (MTBE) ......... ..... B. 7 RGA element (1,1) as a function of the frequency for Case I (MTBE) .............................. B. 8 Singular values for Case I(MTBE) ............... B. 9 Step responsefor Case I (MTBE). (10% step size) ....... B. 10 Poles and Zeroes for Case 2 (MTBE) .............. B. 11 RGA element (1,1) as a function of the frequency for Case 2 MTBE of ............................. 185 B. 12 Step response for Case 2 (MTBE). (10% step size) ....... B. 13 Poles and Zeroes for Case 3 (MTBE) .............. 198 B. 14 RGA element (1, I) as a function of the frequency for Case 3 (MTBE) .............................. B. 15 Singular values for Case 3 (MTBE) ............... (10% (MTBE). for Case 3 Step B. 16 step size) ....... response B. 17 Poles and Zeroes for Case 1 (Esterification) ........... 8 185 187 187 188 190 191 192 193 195 196 199 200 201 202 204 B. 18 RGA element (1, I) as a function of the frequency for Case I (Esterification) .... 205 ...................... B. 19 Step response for Case I (Esterification). (10% step size) ... 207 B. 20 Poles and zeroes for Case 2 (Esterification) ........... B. 21 RGA element (1, I) as a function of the frequency for Case 2 (Esterification) .... ...................... B. 22 Singular values for Case 2 (Esterification) ............ B. 23 Step responses for Case 2 (Esterification). (10% step size) . B. 24 Poles and zeros for Case 3 (Esterification) .......... B. 25 RGA element (1, I) as a function of the frequency for Case 3 (Esterification) ..................... .... B. 26 Singular values for Case 3 (Esterification) ........... B. 27 Step responses for Case 3 (Esterification). (10% step size) . 9 209 210 212 . . . . . 213 215 216 217 218 List 2.1 2.2 3.1 Tables of Thermodynamic equilibrium approaches for the simulation of distillation reactive columns ................... Design approaches for reactive distillation columns ...... Thermodynamic 3.2 3.3 List of equations per stage in the non-equilibrium 3.4 Various forms of non-equilibrium 3.5 List of variables per stage in the equilibrium 3.6 List of equations per stage in the equilibrium 3.7 Various forms of equilibrium 3.8 Computational equations 4.1 4.2 models model .... model ... model model 61 62 63 ............ models ............... times for solution of different 24 47 models ..................... List of variables per stage in the non-equilibrium 22 ...... ...... 63 64 65 sets of model 75 ............................. Range of Variables ........................ MTBE Column Design. Data taken from Jacobs and Krishna (1993) except for the tray design which has been calculated in this work .............................. 82 83 Modelling alternatives for MTBE system ............ 4.4 Ethyl Acetate Column Design. Data from Suzuki et al. (1971) in been has design this the tray work. calculated which except 86 Modelling alternatives for Esterification system ........ 4.6 Depropanizer Column Details. Data taken from Taylor and Krishna (1993) ........................... 90 4.3 4.5 10 88 92 4.7 Selected set of numerical statistics for the three distillation systems .............................. 6.1 Basic control configurations 6.2 Ratio control configurations 6.3 Measured variables for the controllability study on MTBE production ............................... Controllability indices for the MTBE production ........ Measured variables for the controllability study on ethyl acetate production ......................... 6.4 6.5 6.6 Controllability 6.7 Optimization 6.8 Optimization 94 141 ................... 142 ................... indices for the ethyl acetate production 148 148 149 149 .... details ....................... results: nominal and optimal values ...... 11 155 . 159 Chapter 1 Introduction 1.1 Reactive Distillation Reactive distillation is a hybrid unit which comprises distillation and reaction in a single unit operation. Recently reactive distillation has become a strong interest in chemical endistillation has been the gineering research, although concept of reactive known for a long time. In the 1920's the technique was applied to esterification processes using homogeneous catalysts (Backhaus, 1921). In 1971, Sennewald described a development employing solid heterogeneous catalysts. The most important benefit of reactive distillation lies in the economics: a in capital cost, energy saving, raw materials and solvent reducreduction tion. By carrying out distillation and chemical reaction in the same unit, is the associated pumps, piping and with along one process step eliminated, (the This instrumentation. area or gives safer environmental performance just hazards is one continuous significantly reduced with scale of risk and (the heat better of reaction energy management operating unit) as well as but increase better transfer, boil-up no phase vapour-liquid and more creates in temperature; therefore, no cooling is required). Particularly good candidates for reactive distillation are processesin which by: the chemical reactions are characterised 12 unfavourable reaction equffibrium: all chemical reactions have an equilibrium reaction. There are chemical reactions for which, at operating temperature, the mixture at chemical reaction equilibrium conditions still contains considerable concentration of reactants. Even if one of the reactants is present in a high concentration the reaction will not proceed. Such reactions are For these chemical reactions the connormally so-called equilibrium-limited. be increased by from the reacting version can continuous removal of products mixture. high heat of reaction: some chemical reactions have large heat of reaction (exothermic or endothermic). If these reactions occur in reactors they will In distilthe temperature the change and modify reaction progress. reactive lation the heat of reaction will not modify the temperature, the phases will boiling heat the the point, and of reaction will not affect the reacremain at tion equilibrium. In the case of exothermic reactions, the heat of reaction is directly used for the distillation process. large excess of reactants required: reactive distillation is also potentially attractive whenever a liquid phase reaction must be carried out with a large large In this situation, conventional processes need excess of one reactant. distillation however, (for the could reactive excess of reactant); rec y7cle cost be carried out closer to stoichiometric conditions minimising the recycle cost. Reactive distillation prevents side reactions and overcomes limitations due to chemical equilibrium by its natural separation. limitations imdistillation the can overcome azeotropic conditions: reactive by azeotropic mixtures: simultaneous chemical and phase equilibrium posed has the most beneficial effect of reacting away some of the azeotropes and behaviour. the thereby, simplifying phase is distillation to a applicable especially reactive catalysts: systems with solid The important that points catalysts. solid employ certain class of reactions disthe the catalyst at activity of that characterise the catalyst systems are the the reactants and products. tillation conditions and relative volatility of The balance between these two characteristics makes some chemical systems for technique. this perfect candidates highly is it is distillation that disadvantage system The main of reactive for be its to each process. separately assessed needs suitability and specific 13 Reactive distillation has a strong dependency on the properties of the chemical system that is dealt with. The poor knowledge of the chemical reactions features, (catalyst, kinetics, hold-ups) and distillation (vapour-liquid equilibria, thermodynamics, plate and/or packing behaviour) together with their in distillation difficult to simcombination a reactive unit, makes such a unit ulate and operate. Reactive distillation appears to be an interesting process unit. Models capable of dealing with different processes are still not available. Although fully limits for the equilibrium considerations may give reaction and separation, distillation (real reactive columns will not operate at equilibrium operations are not at equilibrium). 1.2 Motivation Aim Work the and of The motivation of this study has been the interactions between reaction and interest in for Increased the that units a reactive column offers. separation the application to fine chemicals and pharmaceuticals, as well as in the comhas been incentive industry this of work. an also modity chemical The main objective of the present work has been to generate appropriate tools to derive an optimal design of reactive distillation operations. The objective has been tackled by: distillation for developing reactive rigorous models e systems; best the to dynamic formulating choose optimisation problem able a e for task; (from the given a possible alternatives) all structure the openusing studied, system on analysis controllability 0 performing loop controllability indicators; and finally, (dynamic optimization) 0 applying optimal control distillation columns. active 14 for the design of re- Problem Definition Consider the reactive distillation column in Fig. 1.1 where a mixture is to be separated into two product streams. The mixture of A and B will react to C and D. Operational constraints, such as product specifications, be in must met spite of the presenceof variability in the feed streams over time (disturbances in composition of A or B). The is to design objective the reactive column at minimum total annual cost with the goal of ensuring feasible operation for all product specifications over time. In order to undertake such problem, a number issues be to of need addressed: Rigorous (equilibrium based) models for and non-equilibrium accounting highly non-ideal mixtures need to be implemented. Simulation and sensitivity analysis should be performed to study whether operability problems, such as multiplicity, can occur. Optimisation techniques to choose the best design for a given task should be considered. An objective function including economics of the process and factors, other such as operability indicators, should be included. 1.3 Overview This thesis is structured as follows, Chapter I gives the introduction, motivation and the aim of this work. Chapter 2 briefly shows the most relevant literature in the area of reactive distillation. The Chapter is divided into two Sections. The first one addressesthe literature in terms of equilibrium models presented. The second the area of non-equilibrium models. one addresses Chapter 3 gives a full description of the hybrid model presented and used in this work. A brief discussionhighlighting the equilibrium and non-equilibrium The is Maxwell-Stefan A detailed equations are also given. models presented. Section with the set of equations describing the hybrid model is presented. Generation of models from the hybrid framework as well as computational Chapter. the aspects are given at end of and numerical 15 aillate c B -->C+D Ittom D Figure 1.1: Exemplary reactive distillation 16 column Chapter 4 presents the validation of models. First the systems studied in this thesis are presented, then the model validation for each system is studied. Finally a continuation method integrated into the hybrid approach is also bifurcation the presented showing analysis. Chapter 5 gives the application of hybrid models. Whenever computational time is to be saved, then combination of non-equilibrium and equilibrium be from determine To it is to models should used. when possible switch one free Gibbs to tray tray energy over the another, efficiencies and excess model highlighted. be Hybrid time modelling results are should analysed. simulation Chapter 6 gives the controllability issues and dynamic optimization. theory description precedes the results. A brief Chapter 7, as last Chapter of this thesis, presents the conclusions and future directions from this work, as well as highlighting the main contributions of this thesis. 17 Chapter 2 Background: Literature Review "The idea of evolution is closely associated with an increase of organisation the to giving rise creation of more and more complex structures" (Glansdorf and Prigogine, 1971). 2.1 Reactive and Non-reactive Process Engineering Distillation in Different pieces of equipment configure a chemical plant. Some of these be distillation likely Distillation is to columns. one of the equipment are The 'stage', in concept of chemical engineering. which oldest unit operations describes distillation columns has been known for over a hundred years. The description of a distillation column continues by characterising defined flows flows heat These leaving they of mass or and consist stage. any entering and feed flows draw flow be the through the column, or any vapour/liquid can flows. Distillation displays different types of multiple phenomena such as heat flow, heat transfer, phase equilibrium, mixing, material and mass and 'equilibrium distillation is in kind the The column a of stage simplest etc. has the implies This the that of any component chemical potential stage'. liquid in the phase. vapour and same value big have that It is well-known advantage: a consisone equilibrium models big is in that the However their solution tent model. cases: most weakness is potentially very far from the real process. 18 This weakness has been overcome by chemical engineers with the help of the efficiency concept. The Murphree efficiency is commonly used for tray because its implementation in equilibrium models is straight columns mainly forward. When using this efficiency we assumethat the vapour composition obtained with the equilibrium model is not the real composition but that the fraction is tray real composition a of the tray composition. For binary sephave the two the same value, the concept works arations, where efficiencies but failing it quite well, starts when multicomponent mixtures are considered (Wesselingh, 1997). Efficiencies are often assumed'constant' throughout the Practical helps deciding 'value' in be column. experience which should used for a particular example. What is to be done if we do not want to use the efficiency concept? Recently, Wesselingh (1997) suggested the 'dream of a chemical engineer', saying: 'subdivide each piece of equipment into zillions of little elements dx. dy. dz, each of in Using the the then the phases equipment'. which will contain part of one of difference form of Navier-Stokes equations for the hydraulics, the MaxwellStefan equation to calculate the diffusion flows, and Fourier equations to describe the energy transfer, together with the corresponding balances and boundary conditions we should be able to solve the 'real problem'. However, be term it the to on a short solution obtain possible won't unfortunately basis: it is a dream. Between this 'dream of engineers' and the usual equilibrium approach, there is the non-equilibrium model, which in these days takes into account the it that the the important vapour assumes column: characteristics of more Equilibthermodynamic liquid in the equilibrium. column are not at and The interface. the is equations of equilibrium phase only assumed at rium Well for forces driving transfer. the the to mixed then mass obtain used are description the liquid and vapour phases are mainly considered and of the included. is between heat the transfer phases mass and in 1970s the been the have area of Significant advances early made since Simulations distillation state of steady columns. of modelling and simulation including shutdown and start-up operations state unsteady operations and operations are quite common nowadays. different for the the simulation problems At the end of models used century, including issues, the to simulation and modelling complex address attempt 19 those related to separation of azeotropic mixtures and simultaneous separation and reaction: reactive distillation. The main technical advantage of reactive distillation is the continuous removal of reaction products. However, as reactive distillation still involves longer lead time for research, which kind of models should be applied to the Can distillation be unit? a model extended to consider the simultaneous physical and chemical phenomena? Would one need a 'special' model to describe the proposed problem?. Reactive distillation be (simultaunits can modelled as an equilibrium system neous physical and chemical equilibrium); at physical equilibrium with kineticall y7controlled reaction; or as a non-equilibrium system (separation through kinetically transfer mass/heat plus controlled reaction); or even maybe as a combination of these models which could also be an interesting way of looking at the process. The selection of the appropriate model type, form and simulation mode (steady state or dynamic) for a particular simulation problem depends on the problem being solved. The separation task will provide details for the being the the separated, required simulation mode, existence of reacmixture tions and/or azeotropes and many more. The choice between non-equilibrium (or rate-based) and equilibrium models, although obvious, is not always followed. That is, since any distillation operation will not attain true equilibis the the most appropriate selection. always non-equilibrium model rium, The equilibrium-based model, however, predominates in terms of application A here is implement the it is to to arises again: solve. question and easier as the to all phenomena occurcapture non-equilibrium model always needed in the reactive column? ring Ideally, what is needed is a computer aided system with a general distillation be forms that from the generated so can necessary model which all model for If, be a particular simulation problem, more answered. our question can becomes the type than one computer aided system should necessary, of model hybrid through the the be able to generate use of a problem specific model hybrid The generating consists of approach modelling approach. modelling for defining their hybrid models and example, modelling of solution method, the trays distillation or use and non-reactive with reactive column single a trays model equilibrium-based and model on some non-equilibrium-based of 20 on the remaining trays. Also, for any design/ analysis/ control problem, the hybrid modelling approach will allow switching between steady-state and dynamic modes and different forms of models. This will make the transfer of data (information) from one simulation sub-problem to another transparent 'noise-free'. Smooth transfer of data also means better initialisation and and consequently, a more robust and efficient solution technique. In this thesis has been an attempt made to present such a model. 2.2 Modelling-Design- Synthesis Issues This section has been divided into two parts, one is related to equilibrium models presented in the open literature. The other one treats nonequilibrium-based models. 2.2.1 Equilibrium Models The literature review on distillation has been extensively addressed in the Several books past. offer the possibility of understanding, as well as showing how to model, design and control a distillation unit. Examples are, King (1980), Kister (1990,1992), Luyben (1974), Skogestad and Postleth(1996), few. For have focused to this here waite name a reason we on the description of the researchover the years of only reactive distillation columns. In recent years a large number of computational algorithms have been rebalance to the solve ported mass and energy equations together with the data describing distillation The the phase equilibrium reactive problem. unknown variables determined by solving these equations are mole fractions both flow-rates temperatures, rate of reaction and of phases, stage of each for The is the equilibrium model standard model simulating reactive phase. distillation. key The in is these that the assumption models or non-reactive leaving liquid in a non-reactive stage reactive or are equilibrium vapour and (liquid both temperatures the vapour) on stage are assumed equal. and and Table 2.1 summarises some of the simulation approaches using a thermody(many for distillation reactive more are presented namic equilibrium model in the literature, but all of them offer more or less the same characteristics). Most of the research has focused upon steady-state solution of the model The for earliest algorithms steady state simulation of staged reequations. 21 Authors Suzuki et al. (1971) Saito et al. (1971) Komatsu and Holland (1980) Perez Cisneros et al. (1996) Table 2.1: Thermodynamic distillation columns active Equilibrium- Simulation Features Steady-state model. Kinetically controlled reaction. Margules's modified equation for the thermodynamic model Experimental results from a pilot plant. Steady-state model. Instantaneous reaction. Model specific for the esterification example treated. Steady-state model. Kinetically controlled reactions. Homotopy- Continuation method of solution. Dynamic model. Kinetically controlled reactions. General model including nondistillation. reactive for the simulation of reequilibrium approaches (1971) (general Suzuki by distillation et al. columns were reported alactive (1971) Saito Newton methods); et al. gorithm employing quasi-linearisation (rather specialised, involves instantaneous reaction equilibrium, etc.); and Nelson (1971) (employing modified Newton-Raphson). During the years several other algorithms with different methods of solu(1980), Holland HolKomatsu for been have tion and example, studied, land (1981), including quasi-Newton methods combined with the 0-method (1977) (1976), Komatsu Hlavacek Jelinek used relaxation and of convergence. (1991) (1988) Bondy Seader Chang (i. false transient); and and e. methods homotopy-continuation The homotopy-continuation methods. worked on fail. Through Newton is the to methods problem when solve a way method into is transformed the one problem can solve a problem pararnetrisation in that by occur naturally or parameters parameters additional using either (parametric the model continuation). belong to tearing methods and take a two tier approach Inside-out algorithms 22 in which the physical properties are approximated by in the simple models loop outer while the model equations are solved at the inner loop (simultaby Newton's neously methods). These kinds of methods are the subject of industrial research, as is shown by Venkataraman et al. (1990). Dynamic equilibrium models for distillation have been presented by Gani (1986,1989), Cameron et al.(1986) and Ruiz et al.(1988). Their adet al. vanced equilibrium stage model including rigorous hydraulics has been applied not only to continuous columns but to shut-down/start-up for columns Recently, the model has been adapted to handle reactive distillation as well. (Pilavachi et al., 1997). To the best of our knowledge there are few papers in the literature, apart from our own work (Perez-Cisneros et al., 1996,1997a; and Pilavachi et al. 1997), that addressed dynamic simulation of reactive distillation columns: Ruiz et al. (1996) and Schrans et al. (1996), Scenna et al. (1998). From the 1980's the interest focused mainly on the design aspects of reactive distillation columns. Several research groups have been dedicated extensively for many years to this area and to the understanding of the interactions between separation and reactions. The design of reactive columns has followed different kinds of approaches: transformed variable approach; based balance approach; - element (including design through techniques synthesis). optimization Table 2.2 highlights some of the main features obtained for the design of distillation columns. reactive In the two first approaches, full equilibrium is considered. Doherty and coGani been the transformed have variable approach, while working on workers based Both balance focused have the approach. element on and co-workers less for in terms containing systems visualisation of excellent are methods than four components. for design been has the Optimization as well as synthesis of applied also kind This distillation a superstrucutilises of approach processes. reactive 23 Authors Design Features Barbosa and Doherty (1987,1988a, Developed tools for analysis of reb), Ung and Doherty (1995a,b,c) active mixtures with a single equilibrium reaction. Later extended to multiple equilibrium reactions with inclusion of inerts. Perez Cisneros et al.(1996,1997a,b) Development of tools based on the balance for element approach multiple equilibrium reactions. Ciric and Gu (1994) MINLP for the optimal reactive col(minimising umn annualised cost) for multicomponent mixtures and kinetically controlled reaction. (MINLP) Synthesis framework Ismail (1998) for multi-component and multi(kinetically controlled) reactions Hold-ups and catalyst loads are for. explicitly accounted Table 2.2: Design approaches for reactive distillation columns feed is to ture which streams, multisystems containing multiple applicable kinetic component mixtures and multiple reactions which are considered as Pistikopoulos instead based and of equilibrium reactions. expressions, rate for developing been have a superstructure approach working on co-workers design and synthesis of reactive and non-reactive distillation columns. Barbosa and Doherty (1987) introduced a new set of composition variables, idenis to the which a problem reaction considering problem which reduces dimensionality the the thus phase of tical to one without reaction reducing for systems. multicomponent problem and chemical equilibria diadeveloped Different tools they have are: computation of reactive phase for azeotropes, derivation reactive conditions general and of conditions grams, b; Doherty, 1987,1988a, (Barbosa and maps curve of residue computation the As the bc). 1995a, of Doherty, visualisation Ung and an example of for the in transformed system reactive the variables curves residue method, is inert presented component with n-butane as isobutene-methanol-MTBE in Fig. 2.1. 24 Image removed due to third party copyright Figure 2.1: Residue curve map for MTBE. 1995b) (P=latm)(Ung Doherty, and The element balance based approach was developed originally by Michelsen (1995) and extended by Perez-Cisneros et al.(1997b). Again, in this apin dimensionality is The the the reduction problem used. of proach, approach is based on the fact that the number of 'elements' needed to represent a reless is in than the number of components present the reactive active systems These 'chemical but they elements are not necessarily elements', mixture. be the chemical single element, a part of molecule or a or even comcan The for itself. this to the solution method approach reduces wellponent known physical equilibrium problem, which obviously gives a great advantage. Tools developed for this approach for the design of reactive distillation include: diagrams, for reactive phase general conditions element recolumns McCabe-Thiele-like (Perez-Cisneros diagrams 1996, et al., active azeotropes, 25 Image removed due to third party copyright Figure 2.2: McCabe-Thiele like diagram for MTBE. (Perez Cisneros et al., 1997a) 1997a, b, Pilavachi et al., 1997). As an example of the methodology and diagram like is Fig. McCabe-Thiele for in 2.2 the the shown visualisation, MTBE reactive system. A design methodology is also described by Pilavachi et al.(1997) which inthree steps: cludes (i) validation of models (including as well selection of thermodynamic models); (ii) identification of variables sensitive to design, and; (iii) analysis of these effects on the design and operation of reactive units. 26 Ciric and Gu (1994) developed a synthesis schemefor reactive distillation fairly processescombining a complex equilibrium model with rate-based kiThe netic expressions. resulting optimization model is a MINLP. The total is function. The solution of such problems the annualised cost objective yields the optimal number of trays, reflux ratio, heat exchanger duties and liquid hold-ups throughout the column. Recently, Ismail (1998) developed a mass/heat transfer framework within a superstructure environment which can handle reactive separation systems. The approach is based on fundamental mass and heat transfer principles. Units of process alternatives are not prepostulated, but instead these are generated according to the synthesis objectives. Summary Since the 1950s new algorithms are presented frequently for solving the equilibrium stage model equations for distillation columns. However this is not has been Still, done implication in that the everything area. many sepaan Some hard them to remain. of yet or are are not solved solve ration problems (Taylor and Lucia, 1995). The same problems apply to reactive distillation few due fact in the that to the research area started a years more strongly later (1970s) with equilibrium staged models. The equilibrium model has been used for reactive distillation without considlack knowledge because the the tray of about of efficiencies, mainly eration of data Experimental these are practically not available. efficiencies. values of For the design of reactive distillation columns, the variable transformation limited based balance to the simultaneous are approaches element as well as limits This, the the of of course gives chemical and physical equilibrium. Variable this but condition. units will rarely work at possible separation, From 0-1. the 'compositions' the leads to eletransformation range outside is to to there based compositions. component go no way compositions ment both For These element compositions are not always useful. approaches, four for difficult than become techniques commore of mixtures visualisation (or elements). ponents in Using the superstructure approach results a probably more realistic and is The distillation this design accuracy of price columns. of reactive accurate 27 for by computationally demanding models. At present, limited paid steady state equilibrium models or limited mass/heat transfer models have been used. This highlights the need for a complex model, able to capture all the phenomena present in the reactive distillation process for the superstructure formulation. 2.2.2 Non-equilibrium (rate-based) Models Rate-based models appear to be the most appropriate for reactive distillation due to the fact that the a priori computation of stage efsystems, primarily ficiency is avoided (Taylor and Krishna, 1993; Sivasubramanian and Boston, 1990). In general, a non-equilibrium model includes separate mass and energy balfor each phase, equilibrium relations and mass and energy transfer ances The for linked by models. conservation equations each phase are material balances around the interface, since there is no accumulation at the interface (the mass lost by the vapour phase is gained by the liquid phase). Equilibinterface. to the rium relations are used relate compositions at either side of Mass and energy are transferred across the interface at rates that depend on the extent to which the phases are not in equilibrium with each other. These from in transfer correlations mass multicomponent sysrates are calculated tems (Taylor and Krishna, 1993). Note that equilibrium stage models are a limiting case of the non-equilibrium model. If non-equilibrium simulations larger done interfacial 100 times than they areas about considering really are (Powers be the then matched et al., results of an equilibrium model can are, 1988). Physical reality is more accurately captured through the non-equilibrium between Interactions reaction and separation are considered explicmodel. itly through mass and heat balances around the vapour and liquid phases, in the case of reactive systems. respectively, The non-equilibrium models that have been developed for distillation prointo interesting insights the have effect of multicomponent mass given cesses transfer. The model developed by Krishnamurthy and Taylor (1985) has been the 28 'base' for any other further development of non-equilibrium models. The model, with or without modifications has been applied for many researchers to the study of non-reactive columns. Recently reactive columns have also been studied with the 'base-model'. Well-known applications of reactive distillation include heterogeneous reactive distillation processes in packed columns such as et herificat ions; processes involving highly non-ideal mixtures such as esterifications or etherifications; high in least and systems with a reaction equilibrium constant which at one of the component concentration is low such as esterification of acetic acid with (1994) Taylor ethanol. et al. pointed out that rate-based models are particfor modelling packed columns; strongly non-ideal systems; sysularly useful tems with trace components; columns that exhibit minima or change rapidly; A or processes where efficiencies are unknown. non-equilibrium model seems the way of modelling a reactive distillation column, instead of the well-know equilibrium model. Another key reason for using a non-equilibrium model instead of an equilibfor is that typically no good predicsystems reactive separation rium model tive methods exist nor is experimental data available for stage efficiencies. Kooijman and Taylor (1995) have developed a rate-based model for the sim'second The dynamic tray generation' of columns. operations of ulation of behaviour dynamic it is the of called, considers non-equilibrium model, as distillation drop the throughout the column as well as pressure column. Gorak and co-workers have done extensive research in the area of packed for their including with comparison obtaining experimental results columns, devices the Packed more comand contact columns are continuous models. differential is transfer-based technique model using mulmass used a mon Gorak, (Gorak, Kenig 1995). 1987, theory transfer and ticomponent mass for is from the developments this One of the latest rate-based model group for the batch experiments scale as pilot well as columns, packed reactive (Kreul 1998,1999). al., et methanol with acid of acetic esterification case of the described (1997) have Pelkonen of Also, experimental validation et al. distillation; for dynamic multicomponent of simulation approach rate-based imple(1999) in the Kenig simultaneous which a work presented al. et and, for reactive approach and non-equilibrium model equilibrium of mentation has been distillation simulation addressed. 29 Although rate-based models for distillation are now (see Seader, quite common 1989; and Taylor and Lucia, 1995, for an interesting review) such for models distillation in development the The first reactive in are still the stage. work field of rate-based model, to the best of our knowledge, Sawistowski was et (1979) al. who modelled a packed reactive distillation column for the esterification of acetic acid with methanol to methyl-acetate. An effective diffusivity type was used for the mass transfer model along with irreversible kinetics for the reaction which can lead to a high percentage of error (40%) in the prediction of the mass transfer rates. This may be due to the fact that there is lack a of theoretical and physical understanding with this effective diffusivity (Taylor Krishna, 1993). method and Zheng et al.(1992) and Sundmacher (1995) presented rate-based models for the production of MTBE on a heterogeneouscatalyst, which usesthe MaxwellStefan equations for the description of the mass transfer. The heterogeneous is being (mass transfer effects to reaction considered as pseudo-homogeneous from lumped the together into a catalyst efficiency term). and catalyst are Just recently more research in the area started to be published (Higler et b; Kreul 1997,1998,1999a, 1998,1999) al., et al., although of a commercial (ASPEN Guide, PLUS User 1997) of a generalised steady state rateversion based model (RATEFRAC) capable of handling reactive systems has been for in industry. The has been ten the almost years available model which described by Sivasubramanian and Boston (1990) is a steady-state model based on that of Krishnamurthy and Taylor (1985) for non-reactive columns. It considers resistance to mass transfer in both phases, but the examples literature in in to the are simplified resistance only one phase open shown (liquid or vapour). Recently, Higler et al.(1997,1998,1999 a,b) presented a non-equilibrium (applied distillation homogeneous for the to esterification of reactive model by transfer simultaneous accompanied acetic acid with ethanol), where mass Maxwell-Stefan described is the continuity equations. using chemical reaction In their model no expression for calculating the interfacial area is presented infor both These fixed is Lewis the phases. assumptions are number and (liquid Only first in by the the troduced phase) paper. one phase authors is to the the transfer case resistance considered when applying model mass both, liquid disregarded, is in Heat transfer that temperatures so and study. 30 vapour phases, are the same. Summary Reactive distillation involves multicomponent mixtures, but other distillation involve Systems binary multicomponent also mixtures. containing mixtures (which are practically not used) are more simple to model: efficiencies, for exfor both ample, are equal components. In multicomponent systems, however, diffusive interaction, reverse diffusion, osmotic diffusion and mass transfer barriers are of importance. It is clear that efficiencies can be different for (Taylor in Krishna, Simulation 1993). tray each each component and and importance then together with a good unand optimization are of particular derstanding of the process. It has never been presented when and for which (such the to the assumed simplifications model as equirange of operation librium, constant flows, etc. ) are valid. Lack of accurate simulators restricts for distillaimprovement direction, in this reactive specially even more any tion. Non-equilibrium models will continue to develop. Indeed they will be used for the case of multicomponent non-ideal mixtures, reactive sepamassively (Taylor Lucia, low-efficiency 1995). and processes ration and The main needs for using a non-equilibrium model at the moment are: kinetics; - reaction for transport better pure properties and physical of predicting methods dynamic (specially for for simulation); mixtures components as well as large for that a range over cover energy coefficients and mass correlations the operating conditions; the hydraulic units. of performance be it been though have Very sophisticated models poswill presented, even it is However, further. them such whether clear not to still extend sible in insights the be developments can problem. solving warranted or give more Most of the non-equilibrium models can, at this stage, accurately represent here that is this it data. However, to remarking without industrial say unfair data that find the hard data that is to of nature often and very experimental is unknown. There comes a point where the value of complex models must Can it be be compared with what can obtained using simplified models. 0.05% large to in tedious time get a models to complex solving expend pay 31 increase in accuracy? Probably the accuracy of the thermodynamic or transdata lower is than that. port property Model integration, where equilibrium and non-equilibrium models can be acfor distillation be the cessed same unit; where steady-state simulation can dynamic be to the equilibrium model can applied switched simulation; where for any stage of a column (if thermodynamic equilibrium is attained) while the rest of the unit will continue in the simulation as a non-equilibrium model. Model integration also where different types of models, simplified or rigorous directly. both be In types that of model, equilibrium and sense, accessed can Still have been integrated. time today computational never non-equilibrium, is an important issue, and the non-equilibrium model requires longer time to achieve the solution. A model combining both types of model is called for To be 'hybrid to a specific column configuration account able model'. a dynamic the of model equations optimization and control structure as well as be these tasks be in that implemented performed. can all a way such should In this thesis an attempt to develop a complete hybrid model is presented In both between together with a way of switching addition, steadymodels. be Dynamic dynamic optimization can also modes are considered. state and performed. 32 Chapter Modelling 3 Issues Within the chemical industries distillation is the most used unit. Its inherent (multicomponent dynamics) it difficult nature mixtures and complex make to model and to simulate. In this Chapter a detailed hybrid model (dynamic which can be also switched to steady-state) including both equilibrium and non-equilibrium approaches for continuous distillation and reactive distillation is presented. In the case Murphree the to the model, efficiency values are added equilibrium of equilibrium stage to account for deviation from ideality while for the rate-based (or non-equilibrium) model mass transfer is considered explicitly using the Maxwell-Stefan equations together with binary mass transfer correlations. The hybrid model approach is presented addressing important characteristics behaviour dynamic the of multicomprocesses such as steady-state and of implemented framework The using model equations are ponent mixtures. the ICAS-integrated simulator (Gani et al., 1997) and gPROMS(PSE Ltd., 1997). Validation of the model is highlighted by matching the results with experi(whenever literature in data the possible) or against open available mental for The several reactive and nonresults achieved other simulation results. is the that the to accurately propredict show model able reactive systems Through time. comparison of equilibrium and non-equilibrium cessesover for distillation is it to that the show reactive possible simulation results largest differences occur when the rate of reaction becomes important. The 33 in bulk liquid between the transfer the the the phases, reaction affects mass of thus moving the equilibrium. Rate-based models lack appropriate experimental correlations, especially for heat (as transfer; mass and equilibrium-based model are not accurate previKrishna, hence, Taylor both 1993), ously reported, and a model combining forward fast towards approaches gives a step and relatively accurate simulation. In Sections 3.1 and 3.2 some details of equilibrium and non-equilibrium moddescription Maxwell-Stefan together the with a of equations. els are given, From Section 3.3 the model developed in this work is described. Assumptions (with the corresponding justifications) are highlighted. 34 3.1 Equilibrium Models Since the 1950's modelling of distillation has been based on the equilibrium feature key is leaving The liquid that the stage vapour and streams stage. are at equilibrium with each other. Three different types of equilibria are presented in the equilibrium term: material equilibrium (there is no net change of material between the phases), (there thermal are no velocity gradients), and equimechanical equilibrium librium (there is no difference between the phase temperatures) (Bird et al., 1960). Since no distillation unit operates at equilibrium, efficiencies are used to account for departures from equilibrium. For binary systems, efficiencies between This bounded 0 1. for both they and components and are are equal does not hold for multicomponent mixtures, mainly because of the diffusional interactions, as defined by Toor(1964). The diffusional interactions include: its diffusing diffusion(when is own concentraagainst a component reverse has diffuses it if diffusion(when tion gradient), osmotic no even a component barrier(when transfer the a component mass concentration gradient), and does not diffuse despite its concentration gradient). 3.2 Non-equilibrium Models When deriving a non-equilibrium model, one of the following theories should be applied at the interface (if fluid-dynamic equations are not considered), be forty (there that theories than for example applied), can are more theory, two-film the model 9 boundary-layer theory, the * theory, the or penetration 9 theory. film penetration 9 because is (Whitman, 1923) experimostly used The two-film model theory is This theory is theory. to this data reported also adapted available mental factors such as chemical reactions causes to give accurate results, even when The (Taylor though, Krishna, 1993). theory, transfer in the and mass changes includes various simplifications. 35 3.2.1 Maxwell-Stefan Equations The Maxwell-Stefan equation for binary ideal mixtures is written as: 1 df, -= VPJ p XIX2(Vl - V2) D12 df be be driving force for diffusion I to the where 1 can considered of species in an ideal gas mixture is at constant temperature and pressure. The symbol D12 is the Maxwell-Stefan diffusivity. The Maxwell-Stefan coefficients do not depend on the compositions, and this gives the Maxwell-Stefan equations an (Taylor law for Fick's the multicomponent systems advantage over extended Krishna, 1993). and The theory for binary ideal systems can be extended to multicomponent ideal systems, so that, NC XiX3(Vi dfi = -1: j=l - Dij Vj). (3.2) If the approach is to be extended to non-ideal systems,then the driving force for diffusion in non-ideal systems is the gradient of the chemical potential for liquid for According to this, Krishna, (Taylor and 1993). mixtures or dense gasesexhibiting deviations from ideal gas behaviour, we can apply the condition, dfi Xi =- RT VTPPi ' (3-3) from departure the equilibrium the give gradients chemical potential where GibbsDue the for ideal to 11P (this driving force reduces to V pi gases). Hence independent. the forces driving NC Duhem restriction, only are -1 forces driving the vanishes: sum of NC Edfi i=l =0 Eq. 3.3 we can obtain, Working with 36 (3.4) NC-1 df i=E ]Pij 7 xj (3.5) j=j for liquids, where non-ideal dln-yi 6ij + Xi IT, Pxk, dxj k:A3,=l,.., n- 1 (3-6) The above formulation can also be used for non-ideal gasesby changing the for fugacity 0j, the activity coefficients -yj coefficients, 6ij dlnoi + Xi-IT, Pxk, dX3 kOj=l,.., n-1 (3.7) The driving force for the diffusive flux j of the component i within a mixture (Taylor df by Krishna, is 1993), NC and components, j, given of NC Z dfý=[F] jy - y.Jt jy y, --3;7vDrj - (3-8) Dý3y is the Maxwell-Stefan diffusivit coefficient, which for diluted gases is identical to the binary diffusion coefficient. The diffusion fluxes Jý are given by NC-1 iv []P][K]v N7yj + Jtvyi - pv (3-9) j=l by, the elements given with [K]V Jy; E" JtV = i=1 i and where, Ci, cij Yi, NC i - Di, NC yi, j NC Yi, k + k=l, k:oi =ýý (IDij 37 Di Di, NC (3-10) k (3.11) Because there are NC -I flux is calculated by independent equations, the remaining diffusional NC-1 iv (3.12) NC (Note that the equations presented here for the vapour phase, similar are equations are valid for the liquid phase). Using the two-film theory model requires a value for the film thickness for both phases. Different types of phenomena, such as entrainment, backmixing, turbulent eddies, and maldistribution, have been usually taken into account by means of mass transfer coefficients, instead of considering only diffusion (given by the Maxwell-Stefan equations). These mass transfer coefficients are purely empirical parameters and are a function of the diffusion coefficients, among others parameters. The linearised theory of Toor (1964) and of Stewart and Prober (1964) is a interesting for diffusion the very method solving multicomponent problems, but it is often limited to situations were the diffusion coefficient does not during diffusion the the change significantly as concentration changes process (Taylor and Krishna, 1993). The solution of the linearised model normally favourably fluxes but the compares with computed with exact solutions, com(Krisnhamurthy differing Taylor, For 1982). position profiles are often and the exact solution, the approach of Krishna and Standart (1976) may be used in conjunction. The method considers that the matrix of mass transfer coefficients can be calculated from Eq. 3.13 in situations where the film thickness is not known. This method is recommended by Taylor and Krishna (1993) because it is much easy to solve. [K]v = Ciks [Cks]-l NC Yi, NC ki ,NC 1=1,10i 38 Yi, l ki,l (3-13) (3.14) Cks ij kij ki, (3.15) NC The binary kij can be calculated as a function of the Maxwell-Stefan diffufrom from sion coefficients a correlation or a physical model. The approach Krishna Standart is of and used and recommended by Taylor and Krishna (1993) only because it is 'simple and easy to compute'. 3.3 Model Equations In this section a general distillation model is presented. The model allows the generation of problem specific simulation schemesbased on the use of a hybrid modelling approach, where equilibrium and non-equilibrium models be can combined. The main difference between the equilibrium model and the non-equilibrium is in the way which the conservation of mass and energy equations model For balance derived the the are used. non-equilibrium model, equations are for each phase and linked through mass and energy balance equations at the balance interface. For dethe the equilibrium model, phase equations are (distillation liquid tray) that the rived around an equilibrium stage assuming leaving Similarly, the thermodynamic stage are at equilibrium. and vapour leaving for the to stages relate compositions equilibrium relations are used the equilibrium model, K-values are calculated at the composition of two (usually both in temperature the the same stage well-defined streams and at For to the the equilibrium relations are used model non-equilibrium phases). K-values interface, the that are phase so relate compositions at each side of In interface this temperature. general modand composition evaluated at for framework the conservation of mass and energy equations are split elling it to eachphase, which makes easier consider equilibrium and non-equilibrium distillation For distillation tray, the for the neceach column. same models defining functions for the corresponding each phase and essary equations (mass transfer flux for the liquid and vapour phase) is the same throughout A interface. is it the the the tray, or, schematic representation same only at Fig. in 3.1. is in this general model given of an stage The general distillation model has been incorporated into two computer aided dynamic both, in mode, as well steady state and simulation providing systems different models, of complexity and non-equilibrium-based and equilibrium as 39 Image removed due to third party copyright Figure 3.1: Schematic representation of an stage in the column (KrishnaTaylor, 1985) murty and 40 form. Both, equilibrium and non-equilibrium models have reactive and nondifferent the forms reactive options and allow generation of model according to the specified simulation problem. This means that starting from the gendistillation eral model, many of the models reported earlier (for example, Cani et al., 1986, Krishnamurthy and Taylor, 1985, Rovaglio and Doherty, 1990 and Skogestad, 1997) are available for use with respect to a specified simulation problem. The presence of the various model options in the general distillation model development the gradual allows of problem specific models such that acceptance of every additional assumption makes the models more simple while Simulation the rejection of an assumption makes models more complex. refrom for initial the a simpler model serve as estimate sults a more complex hybrid The important modelling approach also allows monitoring of model. heat Gibbs free total transfer rates, excess mass and enphenomena such as during tray, and an efficiency-like parameter any simulation. ergy on each Also, during any simulation, a switch from one model form to another as be from to made. well as one simulation mode another can 3.4 Model Formulation In order to present the proposed model, certain simplifying assumptions have to be made. The simplifications are needed in order to simplify the calculain for this the tions and are not relevant work and seem reastudies made the Together the specific column simulated will assumptions, with sonable. following: listed in the have be assumed to the conditions and configurations Liquid and vapour phases are perfectly mixed for the equilibrium model, (liquid bulk for the vapour) and phases model non-equilibrium while are well mixed. leave liquid For each stage the phases vapour and model: equilibrium 9 For the (same temperature nonthermal pressure). and equilibrium at leave liquid the conditions stage at vapour and model: equilibrium determined by the mass and heat transfer rates. Murphree-like in for the equilibrium plate each efficiencies will apply model. 41 The feed(s) can enter any tray as liquid, vapour or a combination of both. Liquid and vapour products can be withdrawn from any tray. No heat is lost from any tray, but heat can be added or taken out at any stage in the column. Pressures * varies with time as well as from tray to tray. The reboiler heat input may be specified or can be used to control temperature or composition at the bottom of the column. The reflux (or distillate) flow may be specified or can be used to control temperature or composition at the top of the column. The column can handle sieve or bubble-cap trays. Plate hydraulics, flooding be determined. entrainment, weeping and can Physical properties can be determined by using different thermodynamic packages. Reaction liquid in for takes the the equilibrium model, and 9 place phase for the non-equilibrium model the reaction occurs in the liquid bulk. Reactions are always considered as kinetically controlled. The list of assumptions hold for the most complex model, however in order to obtain simplified models from the general model, other simplifications can be made, and they are stated in Section 3.6. The general hybrid model develin is Different this work presented as general as possible. oped combination lead literature. Different in to the previous models presented of equations for solving a specific problem are readily available within the assumptions framework. The equations and correlations used here have been used before for equilibrium and non-equilibrium models. The advantage of the general be is different it is the that generated and can choice many models model different to the correlations, mode of simulations, as well as select user of equilibrium or non-equilibrium model equations. 3.5 Equations (considering for distillation tray the The equations a general column model derived from: or a section of a packed column) are 42 Material and energy balance equations (ordinary differential equations for dynamic models and algebraic equations for steady-state models). Equilibrium 9 relations and physical properties. Kinetic models. Mass * and energy transfer models. 9 Hydraulics. * Defined functions. The above set of equations gives a system of differential and algebraic equations (DAEs) for a dynamic model or a system of algebraic equations (AEs) for a steady state model. Typically, the mass and energy balance equations (ODEs) differential AEs. are ordinary equations while all other equations are The defined functions are the equations that compliment the model, for exhold-ups the total ample mass and energy equations and efficiencies. The full set of equations (including differential and algebraic equations) is below. given 1 ODEs balances: Mass Vapour * phase: dmr21P y Fvpz, ,p = p+ dt p+i + EVp-lyi, + Vp + EVp)yi, -(PVp i=ly... Vp+lyi, p-l (3.16) + nr p 19P qp p=ly ... INP INCy Liquid * phase: dm-ý zip = dt FL, ý pz, +L pp p-l + Lp + ELp)xi, -(PLp i=lp -lxi, p=ly ... INP INC, 43 + ELp+lxi, p + Si, p+l p -nL i, p (3.17) liquid hold-up the where mýZip'mvlip are and vapour of a component Kmol/h, Vp Lp, in respectively; and i on plate p are the vapour liquid flow-rates leaving in Kmol/h; F is and stage p respectively the feed flow (the subscript V or L denotes liquid or vapour feed) in Kmol/h; x, y, z are the mole fractions of component Z in stage for liquid, liquid feed; EV EL vapour and p and are and vapour liquid PL, PV entrainments; are or vapour extracted or added to the column; Si, is the reaction rate in the liquid bulk in Kmol/h; p due i loss interface to the and, ny21p, net nýzipare or gain of species transport, mass transfer rates (note that transport from the liquid to the vapour is assumedto be negative) given by nvi,p Ný Pdap "P nLi,p Ný, Pdap p in twothe flux is Ni, the molar point a particular at 1 species of p dispersion. phase folit interface, the Since there is no mass accumulation at phase lows that, MT = nrlip - njL ýlp ,p balances: Energy Vapour phase: * dHýP dt FvpHvfp = -(PVp Hvp-i EVp-, Vp+lI-Ivp+l + + Qv Ov EVp)Hvp + Vp + + + PP P=11... pNP Liquid phase: 44 (3.20) dHIEP dt FL, + Lp-, pHlfp Hlp-l + Lp + ELp)Hlp -(PLp + ELp+, + Hlp+l + DSp QL p (3.21) -oLp p=1,..., NP HEHE hold-ups for liqthe the where V) II are enthalpic vapour and Hvfp, Hlf, Hvp, H1p uid phase respectively; and are the vapourfeed, liquid-feed, vapour and liquid enthalpy respectively; Qp is heat DSp heat is to the any extracted or added stage p; and of heat 0' OL the transfer rates. reaction; PPand are Since there is no accumulation at the interface hence: ET OV = pp OL = 0. 2 Algebraic Equations (AEs) Phase equilibrium: - For the equilibrium model, * yi, p - Ki, pxi, p (3.22) function is the equilibrium constant, of a as expressed p, temperature and pressure. compositions, Ki, NC (3.23) yi, p Interface model for the non-equilibrium model, (the non-equilibrium model assumesthat at interface physical is attained) equilibrium II Ki, yi, p = pxi, p NC Exii. NC yt, p i=l 45 (3.24) (3.25) ,p Physical properties: K-Values: Vapour-liquid i on tray p is given by, Ki, p or i=ll equilibrium constant of component (T) pat sp zip Oy P lip Ki, - lip P OV zip (3.26) (3.27) p=ly INC, ... yNP Equation 3.26 represents the 7-0 approach, where an activity liquid is the to coefficient model used calculate phase activities while an equation of state is used to calculate the vapour fugacities. (EOS) If is phase an equation of state chosen to (0 fugacities for both the the calculate approach), phases -0 by is Eq. 3.27. vapour-liquid equilibrium constant given Different types of thermodynamic models can be used, according to the mixture of the problem studied. For the systems been have in this thesis, used and are studied several models listed below. For the where -y -0 approach was needed: systems The following models for activity coefficients were used, UNIFAC-vle (3 parameters) UNIQUAC-modified (Suzuki 1970). Margules al., et equation modified The vapour phasewas considered as ideal vapour, so that, OV = 1. 0-0 For the needed: was approach systems where Two different equation of state were used: Soave-Redlich-Kwong and, state; of equation Peng-Robinson equation of state The general model of this work is not limited to the thermodynamic models listed above. Table 3.1 give a more complete 46 list of models available within the framework of ICAS (Gani 1997). et al., Activity Coefficients Models Equations of State Ideal Gas SRK-eos PR-eos ALS-eos SWL-eos Virial UNIFAC-vle; Ile UNIQUAC-original UNIQUAC-modified NRTL Wilson Margules equation Van Laar Table 3.1: Thermodynamic models Enthalpies: * Enthalpies are calculated as a function of temperature, composition and pressure. Hvp =f Hlp =f (yi, Pp Tp) pI 7 (xi, Pp, Tp) pi (3.28) (3.29) See Appendix A for details. (Antoine Equation): Vapour * pressure at = p is, p Bi exp[Ai _1 ci + TP (3.30) Density: * Densities for the liquid phase are calculated through a suitable (Daubert Danner, DIPPR and correlation as such correlation Ideal by the for the vapour phase are calculated 1986) and Gas Law or an equation of state (EOS). TLP =f (Xi,PlTP,PP) (Yi, TPIPP) TV P7 p=f See Appendix A for details. 47 (3.31) (3.32) Molecular * weight: NC ýWL p ML i, pXi,p (3-33) mv i, pyi,p (3-34) NC vV p Viscosity: * ViscositY for liquid and vapour phase are calculated through (Daubert DIPPR Danner, a suitable correlation such as and 1986) (Xi, PL pf p iTp) (yi, TP) p, pvpf (3.35) (3-36) See Appendix A for details. Surface tension: * Surface tension is calculated through a suitable correlation (Daubert DIPPR Danner, 1986) and such as UP (xi, =f p) (3-37) See Appendix A for details. Thermal * conductivity: Thermal conductivity for liquid and vapour phase are calculated through a suitable correlation such as DIPPR (Daubert Danner, 1986) and ALp f (Xi, piTp) -xv pf (yi, TP) pI (3.38) (3-39) See Appendix A for details. diffusion Binary * coefficients: 'The coefficient of Mterdiffusion of two liquids must be considdepending on all the physical properties of the mixture ered as 48 laws be to according which must ascertained only by experiC. Maxwell in Encyclopedia the Brittanica, ment'-J. writing 1952. Dý3 f (T, M V) ij 1 Dý3 (xi, f T, M, V) 13 (3.40) (3.41) See Appendix A for details. Kinetic models: Reaction kinetic is assumedto follow, for example, an 'Arrheniustype' equation in the form, Nr Np ]I Il (T) k+ (T) A' kAO ri, p = Ea k+(k-) k,, with exp(- RT = or i=l) where K,, q k- -1 K, (3.42) (3.43) (3.44) q p=lj INC, ... )NP IT A2 f (T) = K(T,, )exp(A, ((I - -) + In(-) T T,, TO 2-T,, 3 2) + A5(T _T3) 0 +A3(T - T,,) + A4(T 4 (3.45) +A6(T _T4) 0 A be where can compositions, concentrations or activity coeffiNp Nr the reactants, represents the products; cients, represent NR is the number of reactions occurring in the system. Total reaction rate (sum over all reactions occurring in the stage) for component Z in the tray p is given by SP (term added to the zip balance in tray): component mass each SP lip i=l) NR E(rk, = pVip) k=l p=ly yNC, ... yNP 49 (ç) (3.46) In this work, we have assumed the reaction to occur in the bulk liquid only and not in the mass transfer film, although this aslimits for the the general model. The sumption application range thickness of the liquid film (as well as the vapour film) is very difficult to predict (or estimate) and it is normally considered to be relatively small, also with respect to the vapour film. So that, the interactions between diffusion and chemical reaction in the liquid film have not been taken into account due to the 'predicted' small thickness of the liquid film. Moreover, in a recent paper, Higler et (1997) al. pointed out that the reaction volume for the mass transfer film is negligible under certain conditions, with respect to the bulk liquid volume. The main limitation of the model under this be for fast assumption will reactions. Rate equations - Mass Transfer * One of the first assumptions here is that the bulk phases are has The for been the transfer completely mixed. model mass based on the model of Krishnamurthy and Taylor (1985) for distillation steady state simulation of columns. The molar fluxes (mass transfer) in each phase are given by, Ji' + Nýyj, Ni' p 21P ttp NT' Nil Jil + xi, p zip zip (3.47) (3.48) i=l,..., NC-1, p=l,..., NP The relationship between the molar fluxes and the mass transfer rates is given by Eq. 3.18 for the vapour phase and Eq. 3.19 for the liquid phase. Since many multicomponent distillation systems and specially behaviour by of reactive systems are characterised non-ideal the mixtures, the influence of the non-idealities on the mass transfer fluxes should be included when defining the fluxes. The good representation of the mass transfer fluxes can be compared with a good representation of the phase equilibria diffusion fluxes, Hence, the when equilibrium models are used. 50 J, in our model are given by, Jiv = ;5v [K'] (yi, yip) p tfp p [IN][Kl] (xi,, xi, i= ; 5L i"P p 21P p) i=ll (3.49) (3.50) NC-1, p=l,..., NP The mass transfer rates (nY ný, are obtained by combining P) 'P, Eqs. 3.47-3.48and 3.49-3.50and multiplying by the interfacial for (Krishnamurthy transfer Taylor, area available mass and 1985). nvi, = pva[K'](yi, p n',ipi, = TLa[]F'][K'](x, i=ll + y, n'yi, i, p) T p p + xi, n' TXi, P p) p (3-51) (3-52) NC-1, p=l,..., NP High flux corrections have not been used because they are (normally Nt for distillation the unity when case =0 units). It can be seen from Eqs. 3.51 and 3.52 that the mass transfer functions rates are of multicomponent mass transfer coeffiQKv], [K]), (a), interfacial density cients area of the mixture (p), compositions in the bulk and at the interface as well as the total mass transfer rate of the components (nvTT n' ), present , in the mixture. A further is depending the optional assumption, which on be is is that to the to assume going problem solved, nature of This is transfer, nvTT =0 and n' =0. equimolar counter for for distillation distillation in general and reactive valid in during is total the there number of moles when no changes diffusion with chemical reaction. Following this, Eqs. 3.51 be 3.52 and can reduced to: v (yi, ni, p = pvpa[Kv] yi, p p) 1 (xjI, j5LPa[rli[Kl] ni, p = xi, p) tip i=lp (3-53) (3-54) NC-1, p=l,..., NP from the genwhich makes another possible simplified model hybrid is dealing the eral model, and usual approach when 51 distillation. This is with assumption equivalent to consider that the heat of vaporisation of the components are equal to each other. When the assumption is not taken into account, an iterative is method needed to calculate nTand n'il niv. Repeated subfluxes (for from the stitution of starting an estimate example, few 0) iterations If take to nt = will converge. nTis assumed fluxes be to then the equal zero, can calculated without any iterations. The interfacial area is calculated through: for tray columns the net interfacial area a is (taken from Taylor and Krishna, 1993): (3.55) a'hfAb froth; is interfacial the where a' area per unit volume of hf is the froth height; and, Ab is the bubbling area. The interfacial area per unit volume a' is calculated through the following expression: (Bravo and Fair's correlation) e. a 19.78Ap(Rc'Ca 5 1)0.392 HOA (3.56) Reynolds Rev is the the vapour phase,and of number where, Cal is the liquid capillary number given by, v Re Cr PVPU (3.57) pvAp ulill al = i5: (3.58) The matrix of multicomponent mass transfer coefficients for [K']), binary QK"], from transmass are calculated mixtures, fer coefficients, following Krishna and Standart (1976) method. This method is applicable when the film thickness is not known, in is is this the the approach. used method reason why and The film thickness is (unfortunately) almost never known. 52 for (Krishna the Standart, 1976): vapour phase and [R]-' (dimension (NC-l)x(NC-1)) Rý,, R, by following ý, the with and given expressions: j j NC Yi,NC. Ri. t't - ki NC , k=l, k:oi ( -yi, j ikiij j R' Yi,k ki, ki (3-59) k (3-60) NC i=l,..., NC-1, p=l,..., NP for (Krishna liquid Standart, the 1976): phase and [Rll-' (dimension (NC-l)x(NC-1)) ý by following R, R(, the with and ,j given expressions: ,j NC Xi, NC w. 2,2 Rý3-= i=lj... ki,Nc + Xi, k Y-d ki, k k=l, k:Ai I1 -Xi, -'-( ki, j ki, (3.61) (3.62) NC p=lj jNC-1j ... INP The binary mass transfer coefficients ki,k are obtained from a suitable correlation, such as: (AIChE, AlChE Method 1958) the for the vapour phase, kv = (0.776 + 4.57h,, - 0.238F, + 104.89-1) w SCO. 5atg (3.63) Eq. 3.63 is derived for the case where mass transfer rein ideal law is the gas phase, and sistance exists only gas h,, (m), is length F, to the also assumed apply. exit weir is the is the superficial factor (defined by, F, = u,;5vo-5), Q1 is the volumetric liquid flow-rate (m'/sec), W is the 53 (m), length Sc is the Schmidt number for the vapour weir (defined by, Se = TvDv "" ), a is the interfacial area phase is t9 the gas contact time. and for liquid the phase, (1970OD' 0-(0.4F, + 0.17)tl) kl = (3-64) atl D' is Wlsec), Fick liquid diffusivity the the where of system is liquid-phase (defined tj the by, time tj and residence z U+I The AlChE correlations are applicable for bubble cap trays or sieve trays. It is important to note that these binary mass transfer coefficients are functions of tray delayout (or in type sign and packing and size, a packed This design be that the column). means column must known in advance in order to solve the equations for the be However, these rate-based model. parameters can also determined by carrying out design calculations simulta(Taylor the the model equations neously with solution of 1992). et al., The correlation for calculating the binary mass transfer fluxes is derived under the assumption of vapour-phase (as that this correlaso stated earlier), resistance only tions should be used with models which considers only rein the vapour phase, as a matter of consistency. sistance Furthermore, the correlation is derived from experiments in steady-state processes. The validity of them for dyliterature, in been have the reported not namic operations (1998) Kreul this the where et al. work of with exception of dynamic for has been tested packed columns correlation given acceptable results. In distillation the transfer resistance is almost entirely located on the vapour phase, according to experience and (Vogelphol, literature in the experimental work presented 1979; Kayhand et al., 1977, Dribika and Sandall, 1979). However, there is no evidence that this can hold for redistillation. Furthermore, active some experimental work 54 liquid in distillation the that resistance a shows column be 20% (Chen the can as much as of entire resistance and Chuang, 1995). New correlations are neededwhen both phasesresistances believed It is that new correlations are are considered. for dynamic also needed system and specially for reactive Nevertheless, in this work mass transfer correlasystems. tions as described above are used for dynamic models and for reactive systems, in an attempt to verify if the correlations lead to acceptable results. Pelkonen et al. (1997) be that transfer proved steady-state mass correlations can for dynamic distilapplied simulation of multicomponent lation. If the stagesof the model are consideredvery closeto each other then we could used the model to simulate packed The columns as well. sameequations are used with exception of the binary mass transfer coefficients and interfacial area. So that, for packed columns the interfacial area is calculated from, for packed columns the net interfacial area a is (taken from Taylor and Krishna, 1993): a= ahA, (3.65) h is the interfacial is the volume; unit area per where a' height of the packing section; and, A, is the cross-sectional the column. of area The binary mass transfer coefficients can be calculated through different expressions,for example, for (Onda 1968) Onda's random packal., et correlation ing in packed columns, for the vapour phase: - 55 The binary kij are calculated from the Sherwood number as: k.2937S 0.33 (ApDp )-2 (3.66) Sh A,,Re'o- Cv DjjAp Dp is Ap is the the specific where nominal packing size and 2/M3) (m A, is the that surface area of packing a constant . takes value of 2 if the nominal size of packing is less than 0.012m or 5.23 if Dp is higher or equal to 0.012m. Re' is the Reynolds number of the vapour phase (Eq. 3.57). The vapour phase Schmidt number is, scv it - (3.67) pv Dv for the liquid phase: The binary liquid phase mass transfer coefficient is obtained from the following equation, /ig)0.333 (; 5L kl )0.667SCI --- 0.0051(Re" -0-5 (apdp )0.4 (3.68) liquid for Schmidt Scl is the the phase number where Scl = pll(; 5LD') (3.69) interfacial based the Reynolds Re" is the on number and area Re" = pLul/(ttla') (3.70) Using this correlation with the expression from Bravo and k, different lead to Fair for the interfacial area will value of a is for interfacial Onda's the if area used. than the correlation However, there are no explanations about which one of the Many knowledge. best the is to the of our one correct values 'combined' literature this in taken the are works presented (Taylor is Accuracy or consistency not reported equations. Krishna, 1993). and Each stage model represents a section of packing and the height of this packing section should be specified. Even small 56 heights therefore, stages, more calculameans more section tions are as well needed. This is directly reflected in the increasing of computational time. As more stages are then included, the model accuracy increases,but how many stages be Experience imshould used still remains vague. plays an determining 'number the portant role when of stages' repreheight senting a of packing. Heat transfer equations: * The energy flux for each phase is defined by, NC Ov = a'(Tv - TPI)+ pppi, (3.71) nyPH" p NP p=1,..., NC (TI-Tl)+Enli, a' ppppp Hi, i=l (3.72) P P:--:: ý'll INP Wv, H1 heat the transfer the are partial and coefficients, a are for in the vapour and p, stage 1 molar enthalpy of component liquid phase respectively. Heat transfer coefficients are calculated following the wellknown Chilton-Colburn analogy. The energy transfer rates for the vapour and liquid phases are obtained after multiplying Eqs. 3.71 and 3.72 by the interfacial area and adding the high flux correction factor expe'vC,-1 as follows NC OP v = avP a expev --v H (T' - TI) + Ppt, nyp i, P (3.73) ni,pHi,p (3.74) Pýll ... INP NC pP a' a expev NP p=1,..., 57 (TI - Tpl)+ P A simplification of these equations (and the model in general) be by can achieved considering that there is no differences in temperature between the phases. So that, Eq. 3.73 and 3.74 be can reduced to: NC ov = Env i,P i,p p i=l (3.75) NC -1 1: OPI nýzip (3.76) These are the equations (3.75-3.76) that are used for the simin this work. ulations Hydraulic equations The pressure drop between trays is calculated from the pressures of the adjacent trays: App= pp-1- pp (3.77) Flow-rates of liquid, vapour and entrainments rates (liquid and functions hold-up tray geometry, of and pressure vapour) are all drop between trays (Gani et al., 1986) (AP, holdup, geometry of the tray Lp =f (holdup, geometry of the tray) ELp =f (AP, holdup, geometry of the tray) (3.78) EVp =f (AP, holdup, geometry of the tray) (3-81) V=f P (3-79) (3-80) functions: Defined Hold-ups: NC mv = p mvi,p (3.82) Mý27P (3-83) NC Mi = pE 58 P=11... yNP Hm' HE vp E%=l INCmi,vp HE lp p Brm J: ýVc Mý zip 1=1 p (3.84) (3.85) NP p=1,..., Efficiency parameter: pMp J-Ii, Yi, p+l - Yi, p (3.86) yi,p+l - Yý lip i=ll 3.6 Generation NC-1, p=l,..., NP Models of Hybrid models indicate the use of more than one type of model for the simdistillation For hybrid ulation/design of a single column. example, models having trays that are at equilibrium and trays that may represent a column are not at equilibrium or a column with reactive and non-reactive trays, or designs. different tray a column with Also, for a specific simulation of distillation operation, use of different model forms may require the use of mixed simulation modes (steady state and/or dynamic simulation modes). Hybrid models are useful for synthesis/ design how because indicate they a specified separation can of separation processes be achieved. Note that the same hybrid model is applicable even though different distillation operations may be necessary to achieve the specified separation. Use of the hybrid modelling approach may result in improved computational better because it inito the can provide single model efficiency compared For flexibility in terms tialisation and example, since of modelling options. have the rate-based models usually more equations than their corresponding equilibrium-based form, if any tray reachesequilibrium, the equilibriumbased form for the remainder of the dynamic simulation run may model it. Also, most reactive distillation columns usually have non-reactive trays. In 59 be such cases,simulations may started with a totally reactive or non-reactive column and gradually changed to the required hybrid form. In addition, the have trays reactive and non-reactive may rate-based or equilibrium-based models. The non-equilibrium model is obtained by including massand energy transfer models and specifying that equilibrium is attained only at the interface of the liquid and vapour phases. The rate-based model consists of Eqs. 3.16-3.17, 3.20-3.21,3.24-3.25,3.26 or 3.27,3.51-3-52,3.73-3-81, plus defined relations for If be is to and a selected set of equations properties. reaction considered, Eq. 3.46 is added to the system. Models of various degreesof complexity are by different forms For the using obtained of algebraic equations. example, for 3.26(or 3.27) 3.51-3.52 expressions equations simple or or 3.73-3.74 will lead to simpler versions of the rate based model. The number of variables (rate-based for in Tables tray 3.2 a single and equations given model) are and 3.3 respectively. Note that the procedure equations (for example, representation of property models as library routines where the calculated property for specified values of temperature, pressure and/or comvalues are obtained listed in In these tables. they and variables are not principle, can position) (temperaknown be intensive the values of variables calculated with always ture, pressure and phase compositions). For all the types of non-equilibrium by ODE form derived dynamic the the of mass models are using models, derived by balance setequations while steady state models are and energy ting the left hand side (LHSs) of the same set of mass and energy balance belonging dynamic Note that to models all non-equilibrium zero. equations to the general distillation model are represented by a set of differential and (DAEs) index 1. of algebraic equations From the general non-equilibrium model, various model forms can be genif For there by are no example, making simplifying assumptions. erated Similarly, drop ((NC terms 1)NR) + out. variables and reaction reactions, if the pressure profile is specified, one variable (tray pressure) and one equaforms different Three drop (tray hydraulics) of ratetion examples of out. based models (reactive and non-reactive) are listed in Table 3.4. Similar Taylor by Kooijman been have and earlier reported models non-equilibrium (steady-state Taylor Krishnamurty 1995) (dynamic model, model, and and 1985) for example. is The equilibrium-based model a special case of the rate-based model, and 60 [Variable I Number Vapour and Liquid flow-rate Vapour and Liquid compositions Vapour and Liquid temperature Vapour and Liquid interface compositions Interface temperature Mass transfer rates Energy transfer rates Stage pressure Rate of reaction Heat of Reaction Total 2 2NC 2 2NC 2NC 2 I NC*NR NR 6NC+8+(NC+I)*NR Table 3.2: List of variables per stage in the non-equilibrium model therefore, is a simplification of the rate-based model. Instead of considering is it is interface, that the equilibrium attained assumed equilibrium only at Therefore, it is in liquid the come contact. vapour and at all points where in in the the and entirely column. stage points assumed equilibrium at all This means that Eqs. 3.16-3.17 Eqs. 3.20-3.21 are combined as Eqs. 3.88 3.89 respectively. and dmT ! I'-P= dt Fpzi, p -(PVp + Vp+lyi, p+l + Lp-lxi, + Vp + EVp)yi, 11 NC, p ... I p- p-l (PLp + EVp-lyi, p-l + Lp + ELp)xi, + ELp+lxi, p + Si, p p+l (3.87) NP +, rny mil, mT where lip = lipp from drops term the hold-up is If the vapour out vapour assumed negligible, (as for dynamic the thereby, example, model the mT generating a simpler P) z? for the (1986)). The Gani by at equilibrium condition et al. one presented heat transfer (Eq. 3.25) 3.24the related equations interface mass and and but the 3.51-3.52,3.73-3.74) (Eqs. equilibrium condition are not needed, (Eqs. 3.22 and 3.23) are added. is for balance oban equilibrium-based model Again, the energy equation 61 M M E E R R R S H Q Equations Number Material balances for liquid and vapour Material balances at the interface Heat of reaction Energy balance equations Rate of reaction Transfer rate equations Energy transfer rate equations Summation equations Hydraulic equations Interface equilibrium calculations Total 2NC+l NC NR 3 NC*NR 2NC-2 2 2 2 NC 6NC+8+(NC+I)*NR Table 3.3: List of equations per stage in the non-equilibrium model tained as a special case of the rate-based model (employing the same asbalance for equations). mass sumptions as dHT' p + Vp+, Hvp+l FpHfp dt + Vp + EVp)Hvp -(PVp Hlp-l + Lp-, - (PLp i= where ... I NC, Hvp-l + ELp+, Hlp+l + Lp + ELp)Hlp (3-88) +Qp + DSp 11 + EVp-, p NP E HTE p= Hip + HýP. further is Eq. 3.89 is hold-up If the energy vapour simassumed negligible, EE Hip. HT Hence, drops term the out. energy vapour as plified P= If reactive distillation is considered, Eq. 3.46 is added to the equation set. has term the This form of the equilibrium-based model, without reaction (1997), for (1994) Skogestad Sorensen by example. been earlier reported and ODE form by derived the is the using The dynamic version of model again the balance of version state while steady equations the energy and mass of by setting the LHSs of the same set of mass and energy derived is model for The the to equations of variables and number zero. balance equations 62 Model Constant NEQ-1 NEQ-2 NEQ-3 Equal pressure Mass resistance in one phase temin perature both phases Number of equations to solve per stage yes no no 6NC+7 yes reactive 6NC+7+(NC+1)*NR for reactive 4NC+5 for non- yes reactive 4NC+5+(NC+I)*NR for reactive 4NC+4 for non- no yes yes yes for non- reactive 4NC+4+(NC+1)*NR for reactive Table 3.4: Various forms of non-equilibrium models listed in Tables 3.5 3.6 general equilibrium-based model are and respectively. Like the non-equilibrium dynamic models, the equilibrium dynamic models by DAEs index 1. are also represented a set of of Variable Vapour and Liquid flow-rate Vapour and Liquid compositions Temperature Stage pressure Rate of reaction Heat of Reaction Total Number 2 2NC NC*NR NR 2NC+4+(NC+I)*NR Table 3.5: List of variables per stage in the equilibrium model Various forms of the equilibrium-based models can be generated from the For Eqs. 3.88 3.89 form the of equilibrium model. example, and general by hold-up is that the be assuming molar vapour negligible simplified can liquid hold-up. form Using the the the to molar general of physical compared liquid in the two case of phases equilibrium with one equilibrium condition, (Bossen (1993), (1990)). is Rovaglio Doherty obtained et al. and phase vapour be by Further simplification can made assuming that the rate of change of 63 M E E R S H Q Number Equations Material balances Heat of reaction Energy balance equation Reaction Rate Summation equations Hydraulic equations Equilibrium calculations Total NC+1 NR I NC*NR 1 I NC 2NC+4+(NC+1)*NR Table 3.6: List of equations per stage in the equilibrium model the energy accumulation term (LHS of Eq. 3.89) is negligible. This assumption will convert Eq. 3.89 to an algebraic equation. Also, this assumption longer independent is tray the that variable. pressure no an usually means Therefore, the tray pressure is removed from the list of variables and the hydraulic equations are reduced by one (usually, the relationship of pressure drop, liquid hold-up on tray and vapour flow-rate is not used). Each verfor further in be terms these of models used simplified models can of sion lists Table 3.7 the hydraulics models tray some of and physical properties. Note that from be the that can model. general equilibrium-based generated (for been have these generated models are not new and reported previously Doherty, (1995), Rovaglio (1986), Gokhale Gani by and et al. et al. example, (1990)). Note also that the number of equations in models EQ-2, EQ-3 and different AEs is ODEs between distribution but the EQ-4 do not change and has EQ-2 ODEs the largest has EQ-4 the for each model. while number of lowest number of ODEs (not counting EQ-1). it distillation different for Hybrid models are useful operations where may in improved it be necessaryto achieve a specified separation, and will result it because the to procan model single compared computational efficiency hybrid decision The flexibility. model better initialisation a of using and vide distillation for for the with be reactive of case example made a prZorl, can for be it trays, example, posteriori, a made can or reactive and non-reactive (the between simulation models equilibrium and non-equilibrium switching have determine trays to have equilibrium). reached which started should distillation the monitoring of a selected set of variThe general model allows For the function tray. time (or example, and/or of as a phenomena) ables 64 Model EQ-1 EQ-2 EQ-3 EQ-4 Assumption Number of equations to solve per stage Rate of change of en- NC+I for nonergy neglected. Con- reactive stant molar overflow. NC+I+(NC+1)*NR for reactive Rate of change of en- 2NC+3 for nonergy neglected reactive 2NC+3+(NC+I)*NR for reactive Liquid hold-up and 2NC+3 for nonhold-up energy reactive considered 2NC+3+(NC+1)*NR for reactive Liquid and vapour 2NC+3 for nonhold-up considered reactive 2NC+3+(NC+I)*NR for reactive Table 3-7: Various forms of equilibrium models V An, Ani general model computes the values of n' ntv+l nt = nl+, = tt , Gibbs G' free EM funcexcess energy, and an efficiency-like parameter, as a tion of time and tray. For the rate-based model, the excessGibbs energy for the liquid phase can be estimated without significant additional computation because the liquid (see known Eq. 3.89). phase activities are already GE p RTp i=ll.. i In xi, p 7i, p (3-89) p=lj ... INP )NC, Note that these estimated values of tray excess Gibbs energy change with time for the dynamic model and that the plots of excessGibbs energy as a function of time are not necessarily the same as plots of excess Gibbs enfunction Therefore, it is to the of composition. possible monitor ergy as a Gibbs for Those trays. that approach an excess energy transient path of all 65 asymptotic minimum value may be considered to be at a pseudo-equilibrium state. For the non-equilibrium model, the difference of the values of n', nv (repreby Anj) Anv be function sented and can also monitored as a of time and tray Generally, difference time. the without significant additional computation of the values of n', nv in a time step before and the present time should be ex(approaching decrease Gibbs to the pected zero) as excess energy approaches a minimum. The values of AnI, An, and GE can then be used as a criterion by which the be its to switched rate-based model may corresponding equilibrium-based form. At equilibrium, the Gibbs energy must be at a minimum. Also, at Ani An, be and must approximately zero. equilibrium, Efficiencies are also a way to look for equilibrium. type of efficiency have been investigated. 3.6.1 Tray In this work, two different Efficiency Two types of efficiencies have been used in this work. 'Standard' Murphree deBoth Murphree types are vapour phase point efficiency. efficiency and below. scribed Murphree 'Standard' efficiency o Taken the Murphree tray efficiency (King, 1980) as a basic, an efficiencylike parameter that is a function of only the simulated bulk and interface compositions from the non-equilibrium model, is proposed, Em Yi,p+i - Yi,p i',p = yi, Yi, p+l lip (3.90) i=l,..., NC-1, p=l,..., NP for bulk the In the above equation, Yki represents vapour composition k in the tray composition while equilibrium y*j represents Z component k 66 for component Z in tray k (at the interface). With the calculated comfrom hybrid the positions available model, the component efficiencies for all components on each tray can be monitored as a function of time. Note that the use of measured bulk compositions in Eq. 3.90 instead bulk the different (see of simulated compositions may also give values Chapter 5). Murphree vapour phase point efficiency Point efficiencies for multicomponent systems are given by (Toor, 1964), (AY)= [Q](Ay') (Ay) (y* (Ayl) where yp), = pp [Nov] by with given (y* ), [Q] [Nov]], = = exp[- yp+, and, 11+ [Nov] (3-91) M(VIL) [Nv] (3.92) [NLI [Nv], [NL] are the numbers of transfer units for the vapour and liquid The be by, the phases respectively. elements of matrix can calculated Nvi = k'ahf /u, NLi = klahfZI(QIIW) (3-93) (3.94) hf froth height, is is interfacial Z the the where a area per unit volume, is the liquid flow path length, W is the weir length, Q1is the volumetric liquid flow-rate and u, is the superficial vapour velocity (based on the bubbling area of the tray). Other models of flow and mass transfer on [Q], form but different lead distillation to tray of calculating may a a the form of Eq. 3.91 can be retained. The relationship between the elements of [Q] and the component Murfollows is for a quaternary system: phree efficiencies as There are only three independent efficiencies given by Elov Qll Q12 AAY2E AYlE 67 Q13 AY3E AYlE (3.95) Eov 2 =IEov =I3 Q22 Q33 - Q21 AYlE AY2E AYlE -Q 'JJLAY3E Q23 AY3E AY2E AY2E 32AY3E Q (3-96) (3.97) the fourth dependent efficiency can be calculated with the help of the (3-95-3.97) three other equations given, V E40 + AY2 + AY3 141 +, + ý42E 1ý41E 4ý43E (3-98) If the vapour leaving the stage is assumed to be at equilibrium with the liquid leaving the stage, then the matrix [Q] is null, and therefore (as be from Eqs. the 3.95all component efficiencies are unity can seen 3.98) [Q] will be different from the 0 matrix whenever the stage is not at be There equilibrium. could a case where the approach to equilibrium is be by That the the the of all components same. will represented [Q] matrix being diagonal with all the elements in the diagonal equal [Q] ie. is This is to the a constant. one another, = q[I], where q have that the the same efficiency: all components same as considering According to Agarwal and Taylor (1994) this situation EO' =1-q. by happen the mixtures are constituted similar components when will in which the binary pair mass transfer coefficients are equal and the liquid is in the transfer phase entirely negligible. resistance mass [Q] diagonal with unequal diagonal values is the situation most often data in to some components match experimental encountered practice: [Q] be diagonal different when will also values. are given efficiencies of the effective diffusivity model of mass transfer is used. If we assume that [Q] is a function of [NOV] and, since the elements of the matrix [NOV] are a function of the mass transfer coefficients of both vapour diagonal [Q] be longer in liquid then general. will not any phases, and Thus, [Q] will have non-zero off diagonal elements together with unbe diagonal These arbitrarily specified elements. elements cannot equal (mass balances and non negativwithout violating physical constraints ity of mole fractions) (Agarwal and Taylor, 1994). 68 Efficiencies computed with the corresponding mass transfer rates Eq. 3.91 may be quite different than those calculated with Eq. 3.90. 3.7 Condensers Reboilers and Condensers and reboilers for distillation or reactive distillation are modelled differ from These the other equilibrium stagesin stages as equilibrium stages. the column becausethey are a heat source or a sink of the column. They can be total or partial. Efficiencies of reboiler and condensersare usually taken to be unity (so that the efficiency-like parameter is not applied to these is liquid the the the vapour entering stages): condenser at equilibrium with liquid is the the entering reboiler at equilibrium condensate and, similarly leaving Different the the vapour with stage. specifications are usually made for these stages (two are needed for the simulation of the full distillation for for the condenser): the reboiler and one column, one distillate flow-rate product stream of * bottoms flow-rate product stream of 9 flow-rate e reflux ratio or reflux boil-up boil-up ratio or e vapour dutY heat of condenser or reboiler * distillate in the either some specific composition of a given component bottoms product or The independent variables in both casesare the compositions, the stage temfor both The flow-rates. the condenser and reboiler equations perature and below. listed are Condenser from in is The condenser which vapour a vapour-liquid equilibrium system All heat flows the through tray exchange coil. a the top condenses as coolant in is total fraction it the or a condenser of a case condensed of a or vapour respectively. condenser, partial Equations: * 69 9 Mass balance: dT mj,C VNPYi, -'"::: dt NP - Dxi, rxi, D D (3-99) Energy balance: 9 dHT' C= VNpHVNP- DHID - QC rHID- dt (3-100) Physical Equilibrium: 9 Yi, D =Ki, (3.101) DXi, D In a total condenser, the vapour compositions (used in the equilibrium relations) are those that would be in equilibrium with the liquid stream that actually exist. Reboiler The reboiler receives the liquid from the bottom tray of the reactive distillation column and partially (or totally) vaporises the liquid using the heat bottom is The the to the tray remaining returned while vapour provided. liquid is the bottom product of the reactive distillation column. Equations: * 9 Mass balance: T dm i, B dt Ljxj, VByi, + = B j - BXi, B (3-102) balance: Energy o dHTE B= dt LjHIj - BHIB- VBHVB + QB (3-103) Equilibrium: Physical o Yi, B =A, 70 i, BXi, B (3.104) 3.8 3.8.1 Computational Specifications Simulations Numerical and for Equilibrium Aspects Non-equilibrium and Both models require similar specifications in terms of distillation decolumn Feed flow-rate, feed composition as well as the state of the sign variables. feed need to be specified for a typical simulation problem. The decolumn tails include feed location, number of trays, product flow-rates, and product Additional specifications. specifications such as reflux rate and vapour boilfor both if has the up are needed models column condenser and reboiler, A respectively. rate-based model also needs the specification of model pafor the transport properties models needed for calculation of mass rameters heat transfer coefficients. Finally, variables related to column geometry, and diameter, tray geometry, etc., need to be specified for dysuch as column for (steady the namic equilibrium models and non-equilibrium model state dynamic In or simulation modes). general, the rate-based model needsmore properties than the corresponding equilibrium-based model and for some of these properties, if suitable prediction methods are not available, correlations be Obviously, limit to the ability to predict the need used. use of correlations distillation column behaviour. The correlations used in this work, have been because best but because known they the they chosen not are are well and be introduced Most to the can easily model. of the pure component property (Daubert from databank DIPPR Danner, 1986). taken and correlations are 3.8.2 Solution Model of Equations The model equations listed in Tables 3.4 and 3.7 are solved by a collection of DAE and AE solvers. Details on the various solution approacheshave been (see for Cameron 1986). et al., example, given earlier In general, for the dynamic simulation mode, initial conditions for the ODEs for be Note AEs to the to the need supplied. variables related and estimates (Gani be initial initial to the that the condition and consistent estimates need 1989). Cameron, and The general model and each of the models which can be generated from it, defines a set of ODEs and a set of procedures to represent the process: 71 o ODEs: mass balances Procedures: thermodynamics, reaction term, mass and energy transfer rates, etc. The model is set up in a specific manner so that it can easily be switched between solving a steady state or dynamic problem. This allows using the dynamic model as a relaxation method for solving steady state problems (Gani and Cameron, 1989). A problem is started with a steady-state simulation to initialise all the variables. After that, starting from the initial state (all the differential variables are known) the solution of the set of ODEs and for be the equations representing algebraic model a particular problem will following: as determining the algebraic variables using mass and energy transfer ahydraulic equations, reaction equations rates, equilibrium equations, defined determine hand then, the equations and side of all and right the ODEs. b- determining values of the differential variables at time tj = tj-l + At, for from end. aj and repeating At each time tj, a set of ODEs and algebraic equations are solved simultain to an equation-oriented package. manner neously a similar (Note that here the procedure is described when solving the model in ODEs form into the WAS simulator. WAS simulator also allows to solve the equations as a DAE system. The model is solved as a DAE system with analytical Jacobian when gPROMS is used). The algebraic equation set which includes all the algebraic equations and ODEs, the determine the to algebraic variables of all correlations required been has defined for example, prediction of physical properties, equations, divided into subsets for easy numerical solution (Gani et al., 1986). (most below is the to The detailed procedure given most general model solve ICAS inside the simulator. complex non-equilibrium model) Proc-edure: 72 Given: differential initial the the 9 state; variables at design the o variables; and, the 'input-output' variables (for condenser and reboiler): one product heat duty, rate and one definition (the liquid the hold-ups) total using equations molar and vapour the bulk-liquid and bulk-vapour compositions can be determined. Then, 1. Assume the plate pressure. 2. Given the plate pressure, the compositions of liquid and vapour phase and the enthalpy of the plate, the temperature of the phases in the be determined. plate can 3. Given, pressure, temperature, compositions and an estimate of the liqinterinterface, the the the vapour composition at uid composition at face can be calculated through calculation of the physical equilibrium (using by iterative thermodynamic package an constant means of any If 'flash the the to this sum of vapour comproblem'). method, solve from different is interface the unity, continue, otherwise go positions at to step(5). 4. Assume a new value for the plate pressure (this can be done by using the methods of successive substitution, secant method or newton-Raphson method) go to step(3). 5. Given the compositions in the bulk and at the interface, the mass transfer rate can be calculated and new values for the bulk liquid compositions can be obtained. If the new interface liquid compositions minus the estimated interface liquid compositions are greater than a specified from (I the 04) then now changing estimated step(3) repeat error -Oeliquid for interface the the to composition. calculated values values 6. Use the hydraulic model to determine the liquid and vapour flow-rates, (if any). entrainments rates liquid hold-ups the the Calculate 7. reaction rates, given and the compobulk in the phase. sition 73 8. Determine reboiler section variables. 9. Determine condenser section variables. In the caseof the steady state simulation mode, initial estimates for the set of be It is important difference to to the unknown variables need supplied. note between the initial estimates for the set of AEs and the initial condition for the set of ODEs. Naturally, if a reference steady state condition is known, the initial condition (for the ODEs) will be the same as the initial estimate balance RHSs In the the of equations will approach zero. this thesis, only and the use of the hybrid modelling approach with respect to solution efficiency been has investigated. initialisation and Since the dynamic models are represented by sets of DAEs of index 1, except for initialisation, no attempt has been made to improve the method of solution. The dynamic model employs BDF (backward-differentiation-formulae) RHS is Jacobian the the the of all equations numerically comof solver and Jacobian Jacobian in The the to order save method retains a copy of puted. fast. The the the model problem rather solution of evaluations, which makes is not yet optimised for speed or memory requirements. 3.8.3 Initialisation Initialisation has an important role in the solution of DAE or AE systems. The hybrid modelling approach provides a gradual, step by step procedure for (an first for That is, initialisation. a simple model any simulation problem, balance equations and simple propequilibrium-based model with only mass In the is model complex a more step, next solved. and generated erty models) is generated and solved using the results from the previous simulation as the initial estimates and providing additional initial values for only the new variThis by introduced the (additional model). complex more are variables ables The this is desired the is of advantage reached. model repeated until step for be is desired to by is the time the that simulation, used model procedure behaviour the the information to large of with respect of valuable amount a Consequently, from is the distillation column a maearlier steps. generated distillation design for data the column jor portion of and analysis of needed final before the becomes already available even model equations are solved. This is becomes to Therefore, each new simulation problem solve. easier for DAE systems, where, the simulation problem is solved true particularly (the AEs ODE-mode are solved separately as procedure equations) in the 74 before switching to the DAE-mode. In this way, the necessaryinitialisation is obtained for the DAE-mode. The non-equilibrium model is somewhat more demanding with respect to computer time than the corresponding equilibrium model. Table 3.8 gives idea for (the times total computaan of computational required each model tional time is divided, for purposes of illustration, into 10 CPU time units). To compare the simulation times, for the same distillation system, one form of non-equilibrium model and the corresponding equilibrium model with the distillation design have been thermodynamic same column and models con(the depropanizer is sidered column taken as an example). Taylor et al.(1994) have also reported similar comparison of computational times between equilibrium and non-equilibrium models. Model Non-equilibrium Equilibrium K-values 0.1 0.1 Mass transfer Solving equations (RHS calculation) 2.0 0.0 7.9 5.0 T-5.1 Total (CPU-units) Table 3.8: Computational tions -- TI-O -0 times for solution of different sets of model equa- The rate-based models need to solve twice as many mass and energy balMass transfer the rates are only model. equilibrium-based ance equations as Since the total the number of equations rate-based model. calculated with for the rate-based models is about three times the number of equations of the equilibrium based, the solution of the non-equilibrium model equations takes longer than the corresponding equilibrium model equations. From the time requirements presented in Table 3.8 it can be noted that the equilibrium by the 50% time the non-equilibrium needed of model needs approximately incentive Therefore, to for the reduce an same simulation problem. model (or for the times the the computational nonmodel using non-equilibrium for dynamic is equilibrium model only when necessary) worth considering hybrid By introduction the approach where equilibrium-based of simulation. be it the to simulapossible reduce and rate-based models are combined, may For in 20 if trays time with we consider tion column a example, significantly. hybrid in the the them 10 are equilibrium, modelling that simulation with of % for 75 take the the time approximately of corresponding nonapproach will 75 equilibrium model without any loss of accuracy. This simple analysis serves indicator for the as an reduction of computational time. It of opportunities be however, that other means of reduction of computational must clarified, time, not considered in this thesis, are also possible. 3.8.4 Switching between Models and Monitoring Pheof nomena Excess Gibbs energy for a single phase is a function of the chemical potential: G= RT Eni In tlE (3-105) ii Since during a dynamic simulation (rate-based model), the component comknown in positions and chemical potentials are each phase, evaluation of the (independent Gibbs does the tray time at any excess energy on variable) Analysis time. not require additional computational of the transient path of the Gibbs energy for each tray (with the rate-based model) then indicates the approach to equilibrium for each tray. If the Gibbs energy is found to be approximately at a minimum (or asymptotically approaching a minimum be While there to the model can made. equilibrium-based value), a switch is no gain in switching from the rate-based model with respect to simulation behaviour, dynamic there are obvious computational results or prediction of for is For the tray the reduced number of equations example, advantages. by nearly 50 %. Also, from a design and analysis point of view, information has distillation by the trays the to column of equilibrium on the approach importance - the upper limit of conversion will be attained if equilibrium is Gibbs is the to To that the a minimum, energy close ensure approached. (see Chapter Bifurcation An, Anj analysis are also monitored. and values of 4) provides a further verification alternative. If the Gibbs energy is not at a bifurcation lower (that it the is, if to analysis value), a can move minimum will point to multiple steady state solutions. 3.9 Conclusions A general hybrid model has been developed for simulation of distillation different between The the of complexswitch models model allows columns. (steady dynamic). between ity as well as state and simulation modes 76 This general framework implemented into WAS (Gani et al., 1997) permits the simulation of reactive as well as non-reactive systems. The framework Ltd., 1997) allows the use of optimization techniques within gPROMS(PSE for the optimal control objective of this thesis as well as simulation. Both have been implemented it models almost simultaneously and was not the idea of this work to compare the results from the two systems. The simulation results obtained from the WAS simulator have been used as the initial for However, has be the to gPROMS simulation. estimate gPROMS showed first has been tool the once a very reliable convergence achieved. In the following Chapters the validity of the general model is presented tohybrid Furthermore, gether with some applications of modelling. controllability analysis, based on a linearised version of the general model is assessed. Finally the dynamic optimization results, from the gPROMS framework, for the design of reactive distillation columns is presented. 77 Chapter Validation 4 of models In this Chapter the validity of the hybrid model presented previously (Chapter 3) is shown for the systems studied during this work. First the systems are presented, given some insights into the phase equilibkinetics distilthe the the mixtures, reaction rium of and characteristics of lation column where the separation/ reaction is taking place. These systems have been chosen in order to cover a wide range of examples. Different characteristics of the systems make them interesting to analyse, for MTBE the production, availability of exexample, reported multiplicity of data for the esterification column. perimental Before going to the simulation results, an initialisation procedure is prefaster This to the more complex modconvergence allows procedure sented. finding 'slow', helps it look 'tedious' it the soand may els, and even when lution (convergence) particularly for non-equilibrium models. Without this in the it to times more convergence attain possible was not procedure, many hybrid from be the that general model. generated can complex models Then, simulation results with different models, generated from the general (whenever data hybrid model, are validated against experimental possible) literature results. or against other open A detailed study of a vast number of parameters together with simulations for is the different which model given shows more appropriate a models with 78 system or which set of parameters should be used when simulating the column. A brief study on bifurcation behaviour of the models is also presented in this Chapter. Bifurcation analysis for the reactive systems is investigated. Bifurcation theory is used to explore all the possible steady states (stable for and/or unstable) reactive systems. The general hybrid model has shown to be a reliable simulator for distillation It has been find to columns. always possible a model which can match (or literature experimental results results). 4.1 Systems Studied Methyl-tert-butyl * Ether (MTBE) production Esterification Acetic 9 of acid with ethanol Depropanizer e column In the following sections the systems are described. 4.1.1 MTBE: Met hyl-Tert-B her utyl-Et What is MTBE? MTBE, or Methyl Tertiary Butyl Ether, is an ether compound in the same boiling range as gasoline. It is soluble in water, alcohol and other ethers by MTBE is isobutylene has combining made a characteristic smell. and (isobutene) and methanol in presenceof a suitable catalyst, typically a cation for OrigiMTBE is gasolines. mostly used as an additive exchange resin. it function instead its to the was raise octane of gasolinesand was used nally It in the the lead also gasolines and raises oxygen content components. of from by hydrocarbon and carbon monoxide emissions motor vehicles reduces MTBE for is the complete combustion. preferred additive promoting more in USA. has However, MTBE the gasoline reformulated received producing health tests publicity as a result of although some consider some negative 79 the tests to be inconclusive. The National Academy of Science(N. A. S.) con'Based does it the cluded: on available analyses, not appear that MTBE from the use of oxygenatedfuels are likely to pose a subexposuresresulting human health stantial risk". However, MTBE has now been banned in the state of California (25th March, 1999). They acted out of concern for water quality. MTBE has been found to enter Californian aquifers and show up in drinking water. As this state goes on environmental regulations so tends the rest of the United States. Strengths Weaknesses High octane Availability of economical isobutylene feed stocks is limited Low volatility Blending gasoline characteristics similar Widely accepted in market place by consumers and refinery Possible methanol supply constraints Environmental totally issues not cleared Reduces carbon monoxide and exhaust hydrocarbon emissions Economics not dependent on subsidies MTBE Production Methanol and isobutene, under appropriate process conditions, react in the in is isobutene Typically, form MTBE. the to the a catalyst of presence isobutene, 50% from 10% to butane that about about contains stream mixed isobuThe depending on the source. other major components are n-butane, inert the These process under tane and n-butenes. other components are interest In temperature the the of range catalyst. chosen and conditions With high. MTBE is (31OK-370K), the equilibrium conversion of an excess isobutene is it to theoretically of a conversion achieve possible of methanol % two 90-97 in the range of with a single-stage conventional process, using by followed of of excess recycle and external separation reactors series-flow methanol. 80 Separation of MTBE from the inerts and the excessof methanol is very difficult through distillation. The methanol excessin the distillate is recovered in another separation step and recycled. MTBE (high purity) is obtained as bottom The isobutene from the normal a separation of unreacted product. butenes by distillation is expensive due to the low relative volatility between them. Therefore, the unreacted isobutene leaveswith the inerts, thus, a certain amount of isobutene is not converted to MTBE. On the contrary, using distillation instead technology, is the reactive of conventional units, possible to achieve up to 99 % of isobutene conversion. The MTBE reaction is not severely equilibrium limited. The net reaction fairly high isobutene is until a rate remains substantial amount of converted. Only when the liquid composition (where the reaction takes place) is close to the equilibrium compositions the net reaction rate decreaserapidly. Conventional processes adopted by industries are, in general, made up of: two liquid-phase catalytic reactors placed in series (the isobutene con% is is 90-95 used) around and excess of methanol version the reactors are followed by one or two distillation high purity MTBE fractionation methanol extraction and for recycling. pure methanol The MTBE process using reactive distillation column. columns to recover to recover columns are required technology consists of a single Co., (Koch Engineering distillation industrial A particular column reactive Inc. ) has a diameter of 203cm with a height of 5.2m. It is insulated and The for heat concolumn operation. adiabatic compensation with equipped Flexipac; (top bottom: beds four mid-top and midtains and of packing bed. distributors liquid Katamax), bottom: above each with vapour and The Katamax packing contains Amberlyst 15 catalyst. The column was opfor demonstrations the for variables of ranges runs, and months six erated ). Co., Inc. Table Engineering in 4.1(Koch studied are presented 81 Variable Total Feed Rate Feed Isobutene Concentration Feed Methanol- Isobutene Molar Ratio Average Katamax Packing Temp. Reflux Ratio Value 20-30 bbl. /stream day 11-13 wt % 1.0:1 to 2.6:1 333K-353K 1.0:1 to 6.0:1 Table 4.1: Range of Variables Why Use Reactive Distillation for MTBE Production? Reactive distillation achieveshigher conversionlevels than conventional for processes this selected reaction. (in the The the temperature which catalyst catalyst of range preferred 9 for distillation, be the that a characused more efficiently) match can distillation. for is this that that teristic reactive process excellent ensure less due Lower to equipment. operating cost 9 (higher higher due Lower to rate of conversion per pass operating cost 9 production). However, one main disadvantage of reactive distillation for MTBE production is the poor knowledge of the reaction kinetics as well as the poor understanding of the plate or packing behaviour. Case Study discussed industrial from differs the in The MTBE column studied this work the is to The reactive simulate used program general simulation above. flow bottom The Table 4.2. in rate was specified distillation column specified industrially MTBE the to required. to 197 mol/sec amount of obtain Reaction Kinetics isobutene between The reaction on sulfonic catalytic resins methanol and (1978) found Anccilotti by et al. has been studied several research groups. first in the in the olefin. order alcohol and that initial rates were zero order 82 [_Variable Feed Rates: Methanol Butenes Feed Stages: Methanol Butenes Number of Trays Type of tray Tray design Value 215.5 mol/s 535 mol/s 4 11 17 Sieve tray Diameter = 2.5m TraySpacing ActiveArea Weirlength WeirheZght Reactive Section Catalyst Load Pressure Reflux Ratio Bottom Flow-rate 0.311m = 2 5.3m 3.66m 0-38m from tray 4 to tray 11 30OKg/stage II atm. 7.0 197 mol/s Table 4.2: MTBE Column Design. Data taken from Jacobs and Krishna (1993) except for the tray design which has been calculated in this work The MTBE synthesis reaction catalysed by Amberlyst 15 macroreticular sulfonic resin was also studied by Gicquel and Tork (1983). This catalyst acts beads The bonded its to the through porous spherical resin. sulfonic groups 4.9 equivalent protons/kg. contain The reaction mechanism and the rate expression are dependent on the relative concentrations of isobutene and methanol which affect the activity of the less is isobutene than When to the methanol ratio of acidic catalyst particles. hydrogen (due highly its the is to 0.7, methanol adsorbed polar nature) and dissociate. (protons link the bonds that or catalytic agents) sulfonic groups Then methanol becomeshydrogen-bonded to the sulfonic group to react with it is Therefore, in the the isobutene solution within asgel phase. pores and The by Eley-Rideal MTBE the that mechanism. synthesis proceeds sumed (Al-Jarallah by is 1988): initial surface reaction rate given et al., 83 CIOB5CMe k., KMeOH _0I, KMeOHCMeOH + - 15 CMTBEIK +KMTBECMTBE)l The rate constant k, is given by: 1013 ks = 1.2x (-87900/RT) exp (4.2) The liquid adsorption constant for methanol and MTBE are: KMeOH= KMTBE: IXIO-13 5. ---':1.6xl 0-16 (97500/RT) exp (4-3) exp(1190001RT) (4.4) The chemical equilibrium constant is given by an expression like: In[ K(T,, ) 1-I)+ A, ( T T, +A4(T 2- T)+ A21n( A3(T - T,,) TO 3_ 4 T02)+ A5(T Toý')+ A6(T _T4) (4.5) 0 When the concentration of isobutene begins to be significant, the LangmuirHinshelwood mechanism, where both methanol and isobutene absorbed react to give MTBE, starts to be operative. The sorption mechanism for hetero(Rehfinger Hoffmann, is 1990): geneouscatalysis and [aMeOH aIB aMTBEj r. = qkr 2 K aMeOH (4.6) is is the the q amount of acid catalyst mass, r, rate of reaction per unit is the the the of component. activity groups on resin per unit mass and a The rate constant kr is presented as: 012e -11115) kr = 3.67x1 (4.7) The chemical equilibrium constant is given as above (Eq. 4.5). When the is in initial is isobutene 10 to the methanol around a maximum rate ratio of decreases Thereafter, the the when methanol concentration more, observed. dependence first dependence in isobutene order and a rate shows a zero order diffusion Here, begins transport to control methanol transfer in methanol. from the bulk liquid phase. 84 The reaction rate is influenced neither by the diffusion inside the micro-sphere by the internal surface area. nor In order to preserve selectivity the presence of water should be avoided. Since water is more polar than the alcohol (methanol) it will be preferentially adsorbed on the sulfonic groups. The adsorbed water can react with the isobutene to give tertiary butyl alcohol. On the basis of the experimental results from the different authors (as disfollowing for the the reaction system: cussedabove), assumptions were made the amount of methanol feed to the column is close to stoichiometric isobutene is the to conditions, so ratio of methanol smaller than unity the column operating conditions are chosen so that the amount of in bottom is This the methanol very small. guarantees that the ratio isobutene is in to than the reaction section of methanol smaller unity in as well as the stripping section (dimerization formation isobutene, 9 side reactions are negligible of of tert-butyl alcohol from isobutene and water) With these assumptions the rate of reaction can be accurately described by the expression proposed by Rehfinger and Hoffmann(1990). Model Formulation The following assumptions were made when formulating the model: (equilibliquid 1. the vapour and on each stage are at equilibrium or not rium model or non-equilibrium model) 2. Since this section is actually a packed section when formulating the is through this a section packed considered as non-equilibrium model trays. number of specific I (not liquid liquid in in the the the chemical reaction occurs only phase film) follows the correspondly kinetic expression: rate4. chemical reaction based 85 the pressure drop through the column is negligible 6. there are no external heaters or coolers apart from reboiler and condenser. Phase Equilibrium The various physical equilibrium modelling alternatives considered in this in Table in liquid 4.3, terms work are given of and vapour phase models together with the sourcesfor their model parameters. Liq. Phase Model Liq. Phase Model Parameters Vap. Phase Model UNIFAC UNIFAC Fredenslund et al. (1977) Fredenslund et al. (1977) Rehfinger and Hoffmann (1990) Ideal SRK UNIQUAC SRK Vap. Phase Model Parameters Binary interaction coefficients set to zero Binary interaction coefficients set to zero Table 4.3: Modelling alternatives for MTBE system 4.1.2 Ethyl Acetate Esterification of acetic acid with ethanol to water and ethyl acetate is the distillation in frequently column. considered reactive system a reactive most In 1921 a patent was published by Backhaus. In the early 70s reactive distillation was described through computer simulations using rigorous equilibrium different From (Suzuki to the 1971). there algopresent on et al., models this to system were presented. solve rithms and models Especially compared with the esterification of acetic acid with methanol, this be to good very not appears acetate ethyl producing process esterification be it (Bock theo1997a) distillation for a reactive and should et al., column different high to processes. with ethyl acetate purity obtain possible retically Bock et al. also show that the equilibrium models, which are normally apdistillation inadequate this to model to are column simulations, reactive plied better. do task the that model will a non-equilibrium suggesting system, 86 Chang and Seader (1988) reported that the main problems in achieving high in purity ethyl acetate a reactive column are: * unfavourable reaction conversion K-values o similar of ethanol, ethyl acetate and water, and in temperature the column. 9 profile In this work ethyl acetate production is considered despite the disadvantages offered for the system when using reactive distillation. There is one big advantage of this system: experimental data is openly available in the literature. This was a major consideration when selecting reactive systems. Experimental data is hard to find. Moreover, despite the fact that high purity is not obtained with the reactive different interesting in this syscolumn, many as well as phenomena occur tem: azeotropes, similar volatilities, low and fast reaction in different trays. The system can definitely test the general model to its limits, pushing the its to ranges of prediction and accuracy. model The general simulation program was used to simulate the reactive distillation in Table 4.4. column specified Reaction Kinetics The equilibrium reaction of the esterification is described as follows: + + ý-+ ethyl acetate water ethanol acetic acid The rate of reaction is given by: r= ki exp(- Ei RT ) rj E2 CR - k2exp(- RT ) r, Cp (4-8) Both the forward and the backward reaction are second order (the forward first in in backfirst the is order acetic acid and order ethanol, while reaction first first is in in order water and order ethyl acetate, given ward reaction 87 [.Variable Value Feed Rate: Composition Acetic acid Ethanol Water Feed Stage: Number of Trays Type of tray Tray design Section I to 12 4885.0 mol/h Tray 13 Diameter =6 25m. . TraySpacing = 0.911m 2 AchveArea = 54.684m WeZrIength = 11.684m Weirheight = 1.193m Reactive Section Pressure Reflux Ratio from tray 1 to tray 13 1 atm. 10 0.4963 0.4808 0.0229 6 13 Sieve tray Diameter = 5.60m. TraySpacing = 0.911m 2 AchveArea = 50.684m Weirlength = 11.684m Weirheight = 0.18m Table 4.4: Ethyl Acetate Column Design. Data from Suzuki et al. (1971) has been in design this the tray work. which calculated except is for The the the same components all reaction rate secondorder reactions). because the stoichiometric coefficients of all components are equal. The rate of reaction used in this work is as follows: E r= ki exp(RT )CAACEt - k2eXP(- where: 483.33, ki k2 123-00, 88 E RT )CWCEtAc (4.9) E= 54970-51 R is the gas constant, and CAA, CEt, Cw, CEtA, (AA), the represent concentration of acetic acid ethanol (Et), water (W) and ethyl acetate (EtAC) respectively. The rate of reaction is given in Kmol/h. Model Formulation The following assumptions were made when formulating the model: the vapour and liquid on each stage are at equilibrium or not (equilibrium model or non-equilibrium model) 2. the reaction zone (all the column, in this case) is modelled as a series of trays 3. the chemical reaction occurs only in the liquid phase (not in the liquid film) 4. the chemical reaction follows the correspond kinetic expression: ratebased 5. the pressure drop through the column is negligible 6. the last tray in the column is bigger than the rest (this facilitates the distillation bottom the the column) of reactive reaction at 7. there are no external heaters or coolers apart from reboiler and condenser. Phase Equilibrium The system is strongly non-ideal. The following azeotropes are possible: - ethanol-water acid water-acetic ethanol acetateethyl - 89 - ethyl acetate-water - ethanol-ethyl acetate-water. Suzuki et al. (1971) determined the phase equilibrium for the system taking the reaction into account (they fitted coefficients to use in a modified MarSawistowski (1982) Pilavachi gules equation). and mention that the phase do data. the equilibrium not correspond with experimental Bock et al. (1997a) have determined experimentally NRTL parameters withhave the they reaction and out considering reproduced the experimental mole fraction profile quite well. In our work (Perez-Cisneros et al., 1996,1997; Pilavachi et al., 1997) different thermodynamic models were employed in order to match the experimental data. Table 4.5 surnmarises the different models used. Liquid Phase Model Liquid Phase Model Parameters Vapour Phase Model UNIFAC Fredenslund et al. (1977) Fredenslund et al. (1977) Ideal Fredenslund et al. (1977) Kang et al. (1992) Kang et al. (1992) Virial Eq. Suzuki et al. (1970) Suzuki et al. (1970) Ideal UNIFAC UNIFAC UNIQUAC UNIQUAC Margules Modif. Wilson SRK Vapour Phase Model Parameters Binary interaction coefficients set to zero Hayden and O'Connell(1975) Ideal SRK Binary interaction coefficients set to zero SRK Binary interaction coefficients set to zero Table 4.5: Modelling alternatives for Esterification system 90 4.1.3 Depropanizer Column Taylor and Krishna (1993) have also studied this depropanizer column. The distillation column details are given in Table 4.6. This column has been simulated with the equilibrium model as well as the non-equilibrium model. Models of types RB-3 and EQ-3 have been used in all simulations. For initialisation purposes, simpler model forms have been used. The depropanizer column is selected as a system to study in this work, mainly due to the availability for data for the column design as well as the property for the mixture. model Model Formulation The following assumptions were made when formulating the models: the vapour and liquid on each stage are at equilibrium or not (equilibrium or non-equilibrium model) 2. the pressure drop through the column is negligible 3. there are no external heaters or coolers apart from reboiler and condenser. Phase Equilibrium Since the system is a hydrocarbon mixture an equation of state for both behaviour is Ideal the not considmixture quite well. phases will represent (15atm. ). Peng-Robinson The high in due the to the column pressure ered both for in is the thermodynamic taken phases, model as equation of state (1993). SRKby Krishna Taylor to and compare results presented order EOS is also used as a thermodynamic model in order to investigate sensitive to simulation. parameters 4.2 Initialisation have been In two the tried. For the simulation of starting points systems, known is the first not condition although a reference simulation the case, 91 Variable Feed Rate Feed composition: Ethane Propane Butane Pentane Feed Stage Number of Trays Type of tray Tray design Section I to 15 -Value 1000.0 mol/S 0.10 0.30 0.50 0.10 16 35 Sieve tray Diameter =4 TraySpacing Act2veArea Weirlength WeirheZght 82m. . 0.5m = 2 14.96m 17.6m 0.05m Section 16 to 35 Diameter 6 17m. . TraySpacing = 0.5m 2 ActMeArea 24.516m Weirlength 0.037m 0.038m Weirheight Reactive section Pressure Reflux Ratio Bottom Flow-rate none 15 atm. 2.5 600 mol/s Krfrom Taylor Data Column Details. taken and Table 4.6: Depropanizer ishna (1993). 92 problem needsto predict the transient behaviour due to disturbances around the reference condition -this situation is often encountered in process design where the reference condition or designed condition of operation needs to be determined. This means that first an appropriate reference steady state be determined before the effect of disturbances can be studied. Since must the starting point for dynamic simulation mode is far from the referencecondition, simulation is started with the simplest model and after short periods of simulation, a switch to a more complex model is made. This is continued desirable has been the In the second case, a reference until model reached. is known from this condition to study condition and simulations are made disturbances in design variables. For these simulations, effect of changesor the reference condition provides a very good initial estimate and thus the initialisation procedure is not needed unless a switch from one model type be is to or mode made. For each distillation system, simulations were started with the equilibrium EQ-1 type model of when the corresponding reference conditions were not known. After an arbitrary number of integrations, for example 5 integration steps, a switch to equilibrium model of type EQ-2 was made and after further integration 5 EQ-3 After to a steps, a switch model was made. a further integration (t=1 hr), a switch to the steady state mode was made (this step was only taken if a steady state with the equilibrium model was for At design the purposes or of needed validation). computed equilibrium (type RB-3) to the non-equilibrium model model steady state, a switch was for formuin If is included the simulation problem a check stability made. lation, dynamic simulation for a further period of time (about 100 hrs) was (MTBE distillation involving For reaction, production and systems made. in increintroduced terms of small ethyl acetate production) reactions were final (Ar form 0.001) the to the switch model after ments of reaction rate = (equilibrium or non-equilibrium). The integration statistics (NFE: number NSTEP: Jacobian NJE: function evaluation; numnumber of evaluation; of ber of integration steps; Time: time of simulation; and, Residuals: residual for differential three time) the the the simulation equations after values of first Although in Table 4.7. to this may sight, seem at problems are shown be a lot of extra computation, use of the initialisation procedure did result in faster convergenceto the desired steady state solution when compared with (see Table 3.8). initialisation the also procedure simulations without use of In most cases,without using the initialisation procedure, it was not possible to obtain the steady state solution with the non-equilibrium model. 93 Model MTBE col.: -equilibrium -non-equilibrium Esterification col.: -equilibrium -non-equilibrium ODEs I NFE I NJE _ FNSTEP Time Residuals 68 68 136 136 518 731 2251 2807 7 7 18 19 26 34 198 101 ih. 100h. ih. 100h. 0.3809e-14 0.5658e-19 0.1824e-08 0.1257e-12 52 52 104 104 393 873 1069 1457 7 8 12 13 26 20 108 103 ih. 100h. 1h. 100h. 0.3753e-13 0.8962e-21 0.5283e-08 0.4016e-13 140 140 280 280 494 678 2938 3558 7 8 22 23 56 48 158 147 ih. 100h. Ih , 100h. 0.8872e-14 0.8978e-21 I 0.4027e-08 0.3075e-14 Depropanizer col.: -equilibrium -non-equilibrium Table 4-7: Selectedset of numerical statistics for the three distillation systems 4.3 4.3.1 Validation Models of MTBE In the open literature many authors have presented the MTBE profile in Sundmacher 1995, Hauan (Jacobs Krishna, 1993; the column and et al., and Hoffmann, 1996, Ung and Doherty, 1995a,b). In this thesis different models from the general hybrid model have been used to simulate the process. A set debeen in (Table has to 4.3) different thermodynamic order used models of be the to that the simulation results. termine sensitive can set of properties be found been have to The activity coefficient model parameters sensitive have been There (see 1996). Perez-Cisneros in et al., some special cases only hardly any differences between the simulated results when using UNIFAC, UNIQUAC or Wilson equation as thermodynamic models. liquid for UNIFAC the the With the combination of phase activity model for SRK the ideal the phase, vapour gas of state equation or coefficients, dependent (1990) for Hoffmann the temperature Rehfinger and expression dynamic it term, through and was steady state possible constant equilibrium (including in the the to reported results multiple solutions) match simulation 94 literature. Fig. 4.1 shows the variation of fractional conversion of isobutene feed function location, for dynamic the as a methanol of steady state and Temperature for low high simulation, respectively. profiles and conversions (for the multiple solutions) are shown in Fig. 4.2. In order to corroborate the multiplicity presented by this system, differdependence the temperature ent ways of calculating of the equilibrium conhave been (1990), Kq, Rehfinger Hoffmann Colombo stant, studied: and et (1983); from based Gibbs free data the the and al. expression on energy using from the DIPPR Data Bank (Daubert and Danner, 1986). K, q for the three expressionsis shown in Fig. 4.3. Simulation results using the three expressions for Kq, are given in terms of fractional conversion of isobutene with the methanol feed location in Fig. 4.4. From this figure it can be noted that while the Rehfinger and Hoffmann the two other expressions multiple yields multiple solutions, with expression be found. solutions cannot From Fig. 4.3, it can be noted that the differences between the three exfrom Fig. increasing for Kq increases 4.4 temperature while with pressions it can be noted that the differences between the different solutions (considFrom Fig. in high 10-14. the stages conversion solution)lies ering only one 4.3, however, it can be noted that these stages correspond to the 'higher' for between differences the temperthe the temperatures where expressions increases the temperature dependence the as equilibrium constant of ature increases. Somepilot plant experiments (Sundmacher, 1995) tend to confirm these mulHowever, for that MTBE the tiple solutions showed our analysis column. The for the there are specific cases multiple solutions can occur. which indein the these systems, remain specific conditions, under multiplicities, been fully have Three types the used: models of model used. of pendently fikinetically reaction and controlled equilibrium with physical equilibrium, nally non-equilibrium. been has the Multiple solutions with equilibrium model validated against Fig. The if is 4.5. as shown equilibrium results model comsimulated other 95 1.0 0.9 0.8 0.7 I Rehfinger and Hoffmann (1990) 0.6 0.5 0.4 0.3 0.2 0.1 13 From 2 to 14 & From 14 to 2 0.0 234567891011121314 Methanol feed stage Figure 4.1: Multiple solutions in the MTBE column 96 440 High conversion Low conversion 420 ii'. .. 1 a.. .. M. '--' 400 U 380 U F-4 360 / 340 10 3456789 11 12 13 14 15 16 17 Stage Figure 4.2: Temperature profiles for low and high conversion in the MTBE column 97 -6 -5 ,;::ý -4 0 -2 -1 0 2.4e-3 2.7e-3 3.0e-3 3.3e-3 l/T (K -1) Figure 4.3: Equilibrium constant for MTBE synthesis 98 3.6e-3 1.0 0.9 pp Q 0.8 0.7 Z CD 0.6 0--ý Gn 0.5 Z 0.4 0.3 Z c) 0.2 o. 1 0.0 10 11 12 13 14 METHANOL FEED STAGE Figure 4.4: Simulation results with Rehfinger and Hoffmann, Colombo et al. MTBE for the DIPPR column and 99 full (meaning ASPEN PLUS pared against results where equilibrium physical is and chemical equilibrium) assumed. Multiple solutions still remain when using the non-equilibrium model. This (1999) in Higler results are agreement with a recent paper of et al. where they also present their non-equilibrium model reproducing the same multiplicities presented earlier with the equilibrium model. Composition in the column with equilibrium and non-equilibrium profile It in Fig. is interesting 4.6. deviations beto that models are shown note tween the two models are largest on the reactive stages, but the differences bottom indistinguishable. the top the column are almost of at and The non-equilibrium model used to simulate the MTBE column assumesconFurthermore, the the packed section of throughout column. stant pressure the column has been simulated considering packed column expressionsfor the binary mass transfer coefficients (Onda et al., 1968) and interfacial area. Mass transfer resistance has been consider to be predominant in the vapour has been The liquid the that assumpneglected. phase resistance phase, so tions, however are not limiting the ability to provide a good representation the of system. The aim of this work has been to determine which model represents acculack The little time the as possible. computational rately system using as knowledge the number regarding of accurate correlations as well as a poor height the the to of packing section compares with represent of stages able the assumptions made. The complex non-equilibrium model, which considers both resistances and variable pressure takes around 40 % extra time during better it believed is it that percannot achieve significantly calculations and formance. due data be to the Unfortunately results cannot compared with experimental (very from the information obtained small) set of results presented restricted in the open literature. For this reason, high production of MTBE and the (in investigate to what actually will order existence of a multiplicity region happen if for any reason the systems gets into that operating conditions) Considering differences between the the this the work. models of aim were data in to this lack the principle point any and up experimental of and be MTBE the (equilibrium to could or non-equilibrium) used simulate model 100 1.0 0.9 0.8 iz 0 14-4 0 0.7 0.6 0.5 0.4 0.3 w ý4 0.2 0.1 0.0 68 10 Methanol feed stage 12 14 Figure 4-5: Simulation results compared with those of ASPEN PLUS for the MTBE column 101 1.0 0 0.9 0.8 (D o E 0.7 C 0 0.6 V) 0 CL 0.5 E 0 0.4 0.3 cr w 0.2 0.1 0.0 10 12 14 16 Stage Number Figure 4.6: Simulation results with the equilibrium for MTBE the column model and non-equilibrium (Chapter how be it later in 5) However, to this will show work column. determine which one is the best and which one should be used. 4.3.2 Ethyl Acetate Several thermodynamic models (for activity coefficients) were used in order been has by Suzuki It 1971. to reproduce experimental results given et al., for Margules the that activity coefficients modified equation unless proved (1970) dimerization is Suzuki the by of used or association and et al. given (acetic is into it is taken to acid) account, not possible the vapour phase Fig. 4.7 the the equilibrium model. shows experimental results with match different ethyl acetate composition profiles using the thermodynamic models Activity from 4.5. Table in listed coefficient model parameters estimated 102 0.6 0 0.5 4.0 0.4 0.3 0.2 0.1 0.0 10 12 14 Stage number Figure 4.7: Simulation results with different thermodynamic models for the esterification column data, in when used simulations of reacting systems vapour-liquid equilibrium and physical equilibrium may not give acceptable results. In Fig. 4.8, the experimental data given by Suzuki et al. (1971) are compared (using UNIFAC from thermodythe as equilibrium simulated values with final liquid Only the steady state namic model) and non-equilibrium models. Fig. in 4.8. the tray plots of are shown compositions of ethyl acetate on each Note that no tuning of model parameters has been performed at this stage. The non-equilibrium model performs better than the equilibrium model, a has been for the made with non-equilibrium model surprisingly good match is the the bottom the column, and neither of models able to match section of for the top section of the column when the reported experimental values UNIFAC is used as the activity coefficient model. Using the Margules model 103 0.6 0 --0+ Suzuki et al. (calculated values) Experimental values -40--0-9- Non-equilibrium model wiffi UNIFAC Equilibrium model with UNIFAC Non-equilibrium model wiffi Margules 0.5 0.4 0.3 Ow 0.2 4..1 w 0.1 0.0 10 12 14 Stage number Figure 4.8: Simulation results compared with experimental data for the esterification column fit better (1971) however, Suzuki gives a significantly parameters, et al. with for the entire column with the non-equilibrium model (shown in the plot) as (not figure). in for the the shown equilibrium model well as Figure 4.9 shows the simulated total mass transfer rates for the four compois liquid from Mass to the the transfer in this vapour phase system. nents is be As be the to more can expected, since ethyl acetate negative. considered behave differently (but multicomponent mixtures often volatile component from binary systems), it transfers from the liquid to the vapour phase in all the trays. The non-equilibrium model, therefore, qualitatively and quantitabehaviour. transfer the tively, gives correct mass The mass transfer values obtained with the non-equilibrium model, are com104 12( 10( 8( 0 E 4( cc 21 LCD cn c -21 co -41 -61 -81 -100+ 0 440 IV Stages Figure 4.9: Mass Transfer rate for the esterification column 105 100 Ethýl Acetate 90 Rate-based Model 80 70 i 2 60 ,v CC 5C cc 2 4C 3C 2C Higler et al. 10, 1 456789 10 11 12 Stage number Figure 4.10: Comparison of mass transfer rate with literature results (1998). The is Higler trend those comparison of et al. rather good pared with (ethyl for be from Fig. 4.10 the components one of acetate). as can seen 4.3.3 Depropanizer Column Different thermodynamic models have been also used for this system, in order (as the to identify to simulation results possible set of sensitive parameters There are practically well as good agreement with previous reported results). EOS Peng-Robinson between the the and results obtained with no variations the SRK-EOS. Since both equations of state lead to similar results, the rest has Peng-Robinson for done the this the equation of used system work of literature keep in to results. consistency with order state In Fig. 4.11, the simulated steady state liquid composition profiles with the for depropanizer The the comcolumn are shown. non-equilibrium model fairly by Taylor those well with reported puted composition profiles compare (1993). The Taylor Krishna in Krishna results of and are not shown and Fig. 4.11. These results validate, at least qualitatively, the non-equilibrium Taylor Krishna the this and and work model since approximately of model 106 behaviour is similar simulated predicted. Simulation with the equilibrium model has also been performed. The results differ somehow from those predicted from the non-equilibrium model, but the results from the equilibrium model are not unacceptable. The ChemSep package (Kooijmann and Taylor, 1996) has been used to comSimulations the the results obtained with general model. pare with the ChemSep package are compared with our non-equilibrium model. Fig. 4.12 four for The the the the profiles of vapour composition components. shows differences are acceptable. 4.4 Bifurcation Analysis In order to explore the dynamic behaviour of the reactive systems, a continubeen has implemented the the to general equations of use with ation method model. Due to the non-linearity of the general form of the model equations, it is behaviour have distillation that or unstable columns may reactive possible for been has the have case shown already multiple steady states, which may MTBE the production. of Bifurcation diagrams are believed to give insight, especially in the caseswhere have for but caseswhere multiple solutions also multiple solutions are present determine furthermore, the found. Bifurcation been exisanalysis will, not tence or not of multiple solutions. by final the The simulation are reactive systems obtained steady states of it is To bifurcation through poscontinuation, perform analysis. also verified (Petersen, PATH for to example, use some software already available, sible (1995); Marek, Schreiber 1991 (Kubicek V Marek, 1983; 1989); CONT and and bifurcation, into As introduction theory the 1976). Kubi'c"ek, several of an Thompson (Mess, Solari 1996; Hilborn, 1994; 1981; books are available et al., 1986). Stewart, and In this work, all continuations are performed with the software program 107 1.0 0.8 Ethane E Propane 0.6 Butane Pentane c 0 a 0.4 E 0 C) '0 .B 0.2 AT -i 0.0 5 10 20 15 25 30 35 Stage number Figure 4.11: Simulation results with the non-equilibrium propanizer column 108 defor the model 1 0.9 + 0.8 0.7 -0- PropaneC-hewn-Si) Propane (RB) Butane ChemSep + Butane (RBI E 0.5 0.4 Co +' 0.3 +. Ok 0.2 0.1 0 02468 10 12 14 16 18 20 Stage number 22 24 26 28 30 32 34 Figure 4.12: Vapour compositions compared with ChemSep simulations CONT (Kub'Cek and Marek). The software has been added to the set of ODEs, for AEs DAEs. the solvers and The continuation method-based solver makes it possible to calculate all the design to two steady-state solutions, stable and/or unstable, with respect (bifurcation for bifurTherefore, two any preselected parameters). variables bifurcation diagrams is it to and analyse possible generate cation parameters, the steady state behaviour of the distillation operation. After defining the feed stream (compositions, temperature, pressure) in the degrees freedom. These two there are remaining are chosen as of column bifurcation parameters. 4.4.1 Applied Bifurcation Analysis Bifurcation diagrams have been produced for the two reactive systems, ethyl MTBE production. acetate and 109 For the bifurcation analysis, two parameters (design variables) were chosen. The results show the analysis for a fixed value of one parameter. For the esterification column, the bifurcation analysis results are shown in Fig. 4.13. Ethyl acetate composition at the top of the column (product) as function for fixed heat duty is shown the of reflux rate a a value of reboiler in the Figure. As can be seen, only one steady state (stable) is found for all the range of reflux rate studied. The operating point of the nominal design is also marked in the Figure, showing that the design can be optimized to higher achieve slightly product purity. The analysis has also been made varying reboiler heat duty for a fixed value flow (stable) has been found in the reflux rate of and again only one solution the the of all range variable studied. For the MTBE column, the bifurcation analysis results are shown in terms of MTBE composition in the bottom product as a function of reflux rate for a fixed value of vapour boil-up in Fig. 4.14 and as a function of vapour boil-up for a fixed value of reflux rate in Fig. 4.15. It should be noted that for this column, the results of Figs. 4.14 and 4.15 cordistillation fixed the the to column where reactive configuration of respond the multiplicities have been already reported (the methanol feed is located in stage 10 and the butenes are fed at stage 11). It can be noted that while for the esterification case no multiple solutions heat for for be found the the range of reboiler range of reflux ratio or could duty investigated, for the MTBE column, multiple steady states were found for a range of reflux ratios. The region of operation for this system has been flowrates from flowrates in the the ±20% column, starting of calculated as a Fig. According 4.14 design the to from the the column. of nominal calculated have this some serious problems, probably changing column will operation of from a high purity of the product to almost no product recovery. 110 0.6 0 *p 0 0.5 co 75 Operating point ,E0.4 CL 0 4-0 cc (D 4-0 co cz 0.3 0.2 w 0.1 500 1000 1500 2000 Reflux flow-rate (Kmol/h) Figure 4.13: Bifurcation diagram for the esterification column varying reflux flow rate ill 1.0 c Co 2 0.0 0 2000 4000 6000 8000 10000 12000 Reflux Flow-rate (kmol/h) Figure 4.14: Bifurcation diagram for the MTBE column varying reflux flow rate 112 1.005 1.000 0 t5 U.Vvo 0.990 E U) 0.985 E 0 0 0.980 0 w 0.975 CO Operating point * * 0.970 Stable steady state Unstable steady state 0.965 1111111111-1 10800 11000 11200 11400 11600 11800 12000 12200 12400 12600 12800 Vapour boilup boilfor MTBE diagram Bifurcation the 4.15: Figure column varying vapour up 113 4.5 Conclusions The systems selectedin this work showeddifferent types of phenomena which helped to demonstrate the strength and the versatility of the general model its as well as applicability. The non-idealities of the systems have been captured through the general The be the to model. analysis of mixture separated or reacted away together first with simulation results gives an identification of sensitive parameters. The set of sensitive parameters that has been identified for each system helps in the design of a distillation column. Systems presenting multiple solutions, highly non-ideal mixtures, as well as 4simple' hydrocarbons for distillation 'normal' have a mixture of a column been selected in order to validate the general (hybrid) model as well as its handling capability of various problems. Two other systems have been tested with the general model although not in The this thesis. presented systems are, one reactive system: methyl ac(Papaeconomou, 1999) etate production, and one non-reactive: methanol(Pilavachi 1999), the capability of the genet al., enhancing ethanol-acetone eral model. The aim of this work was to cover a wide range of systems. Being able to definitely help different different when a new systems will aspects of cover be has to simulated or analysed. system The bifurcation analysis also contributes to the development of a computer different facilitates the simulation probsolution of aided environment which lems. The hybrid modelling approach helps to avoid problems of mismatch between (for between forms two of models any model and simulation results as well as be forms two approximately of models, a sub-set of simulation results should the same). The results presented in this Chapter have mainly focused on highlighting features In following the the the of of general some model. validating and 114 Chapter, the application of hybrid models, i. e, the solution of a combined is Through detailed equilibrium and non-equilibrium model, presented. a in terms of tray efficiency, mass transfer and reaction rate, and analysis Gibbs free energy as a function of time a switch between non-equilibrium be and equilibrium model can made. 115 Chapter Analysis 5 with Hybrid Models Hybrid models are useful in many distillation operations when a product be Hybrid in improved should met. specification models result computational efficiency compared to the single model because they provide better initialisation and flexibility. As stated before (Chapter 3) the decision to use a hybrid model can be made focus hybrid Chapter In this on using we will models a priori, or a posteriori. between has the the non-equilibrium started, i. e, switching simulation after to the equilibrium model. The general distillation model presented and validated previously, allows the function tray. time of and/or monitoring of a selected set of phenomena as a It computes the values of: o GE, Gibbs free and, energy, excess EM. 9 efficiency parameter, (most time be Both can estimated without significant additional computation durknown or calculated of the needed variables and parameters are already ing the simulation). The transient path of excessGibbs energy for all trays is monitored over the be Those that time. approach an asymptotic minimum value may simulation The difference be the to a pseudo-equilibrium state. of values at considered ) by An, An, (represented the and are also monitored at present nv n', of 116 time and are expected to approach zero when the excessGibbs energy ap(or Tray in the column is calculated proaches a minimum. point) efficiency during the simulation time, or after the simulation has finished. Efficiency (or to values close unity represent an equilibrium close to equilibrium) stage. These results can then be used as a criterion by switching between the nonto model equilibrium an equilibrium model. The main idea of this approach is to obtain useful insights to the reactive distillation operation as well as to save computational time while capturing believed The is the the non-idealities of all system. non-equilibrium model to reproduce accurately the behaviour of the mixture in the column but it is time consuming. Whenever the non-equilibrium model is approaching to (the limit is of any non-equilibrium model simulate an equilibrium model the equilibrium), all the extra equations are no longer needed because an be instead. used equilibrium model can 5.1 Phenomena Monitoring of Free Energy through Gibbs Since, during a dynamic simulation with the non-equilibrium model, the known in each phase, component compositions and chemical potentials are This is Gibbs time the tray the performed. at any energy on evaluation of indicate for Gibbs tray the transient will each energy path of analysis of the the approach (or not) to equilibrium for each tray. Studies of tray Gibbs energy as a function of time have been made for the funcGibbs in Fig. Plots 5.1 the tray two reactive systems. energy as a show be It (for 7). I for trays the noted can tion of time and esterification system that while the Gibbs energy value on tray 1 is asymptotically approaching Gibbs for Note is 7. that tray true the of plots while not same a minimum, (possibility be function smooth of composition may not energy surfaces as a be, function Gibbs to time local appeared of energy plots as a solutions), of first in all casesstudied, order. The been found behave I. have tray 3 to the 2 Trays and also same way as The is to free Gibbs rest of a minimum. energy asymptotically approaching 117 298 297 0 E -) LLJ (D 278 270 0.0 0.1 0.2 0.3 0.4 0.5 Time (h) Figure 5.1: Excess Gibbs energy for the esterification column. Trays 1 and 7. keeps free Gibbs The behaviour follows 7. tray the the column energy on of decreasing as a function of time. The final steady state is stable and the mass transfer rate differences (Anj likely, highly It is ) these therefore that An, are approximately zero. and for 1-3. to trays indicate a state close equilibrium results For the MTBE column, the tray Gibbs energies as a function of time (see Fig. 5.2) show a decreasingtrend, thereby indicating a state still further away from equilibrium. In this case,mixed model simulation is not necessary.Note Gibbs for does the energy a non-equilibrium process that at steady state, not have to be at a stationary value. This also means that the non-equilibrium be At that this the is the should used when simulating column. model model 118 292 285 0 E -) w 0 250 245 0.0 0.1 0.2 0.3 0.4 Time (h) Figure 5.2: Excess Gibbs energy for the MTBE column. Trays 4 and 10. be By Chapter it 4 analysis used. was not clear which model should end of dynamics All becomes issue the this the as well complex clear. system, of better through the different non-equilibrium captured non-idealities are as model. 5.2 Tray Efficiency Analysis for for been has the Efficiency analysis esterification case as well as studied the depropanizer column. Figs. 5.3, the the For the esterification case, using composition profiles of (shown have been in Fig. 5.4) parameters estiefficiency-like component 119 0.6 -0A 0.5 --0-0- Water Acetic acid Ethanol Ethyl acetate 14. 0.4 0.3 0.2 0.1 0.0 10 12 14 Stage number Figure 5.3: Vapour compositions for the esterification column Figure 5.3 the mated. shows corresponding simulated vapour compositions Without introduction the with non-equilibrium model. any of noise and with bulk in phases, the component efficiency-like assumption of perfect mixing lie between 0 1. and parameters As previously predicted through the analysis of the excess Gibbs free energy, for 3 1 to trays to are very close unity, giving an extra component efficiency believe that these to stages are actually at equilibrium. reason For the rest of the column (trays 4-13) efficiency values vary between 1 and 0.6. A tray efficiency of 0.7 should predict accurate results for the system. Tray efficiency analysis is also highlighted for the depropanizer column in Fig. 120 0.8 0.6 0.0 268 10 12 Stage number Figure 5.4: Efficiency calculated for the esterification column 121 1.00 0.95 .c 0.90 Ethane Propane Butane Pentane 0.85 0.80 3579 11 13 15 17 19 21 23 25 27 29 31 33 35 Stage number Figure 5.5: Efficiency calculated for the depropanizer column 5.5 where the calculated component efficiency-like parameters are shown. An important point to note here is that all the component efficiency-like paramlie between between largest 0 0.8 1, 1, the with eters again and varying and deviations from unity observed at the top of the column. The simulations for this system again showed the same phethe with non-equilibrium model for The the calculated vapour compositions esterification case. nomena as in Fig. in the the efficiency-like parameters are shown calculations of used 5.6. This result is not surprising since perfect mixing in the bulk phases have been assumed. If measured composition values are used, perfect mixing probably will not be valid, resulting in erratic efficiency values. This (for be by simulated adding noise example, randomly change condition can the simulated compositions by a factor of (le - 03) to the simulated steady (the ). between Figure 5.7 efficiencies vary state compositions -oo and oo 'noise' to the the erratic component efficiencies with added calculated shows for the non-equilibrium model. calculated compositions If we compare our results with those of Taylor and Krishna (1993) we can disagreement. From large it this practical experience of column observe a 122 1.0 0.9 0.8 0.7 0.6 0.5 E 8 0.4 0.3 0 CL 0.2 cd 0.1 0.0 10 20 15 25 30 35 Stage number Figure 5.6: Vapour composition for the depropanizer column 123 1.0 0.9 --qr--v 0.8 0.7 Q 0.6 0. ;Oil c (D 2 4-w Ethane Propane Butane Pentane 0.5 v -V"V. 0 0.4 0.3 0 0.2 0.1 0 0.0 02468 10 12 14 16 18 20 22 24 26 28 30 32 34 Stage number Figure 5.7: Efficiency calculated for the depropanizer column far from steadystate 124 0.8 Ethane 0.6 0.4 0.2 0 -0.2 o Non-equilibdum model ChemSep model -0.4 -0.6 -0.8 -1 05 10 15 20 Stage number 25 30 35 Figure 5.8: Efficiency calculated for the depropanizer, compared with those Ethane ChemSep: of is known that the global efficiency of this column is around 0.7, which are by in less this those the than predicted work. non-equilibrium model actually Looking back at the assumptions made when formulating the problem, we decide to include resistance in the liquid phase, which has been considered both Re-running the resistances,we simulations, now considering negligible. Taylor KrResults 0.7. to and are close obtain a global efficiency of around ishna (1993) but still we get the efficiencies in the bandwidth 0-1. This can be seen from Figs. 5.8 and 5.9 where the efficiency for ethane and propane the the against effinon-equilibrium model and plotted are calculated with (Kooimann Taylor, ChemSep 1996). the and program ciencYcalculated using The main point of the difference, is the way yi* is calculated. In this work, ie. interface, the the is yi* = xiK(Z) composition at yi* calculated using bulk for in Krishna, the the Taylor example, utilise composition and while This lead if ie. to the liquid, will not same value mass yi* = xjK(i). of the liquid has been in liquid film bulk is the the considered or transfer resistance fact The in that tray the real efficiencies should remain not perfectly mixed. 125 Propane 0.9- 0.8- 0.7- 0.613 50.5- 0.4Non-equilibrium model ChemSep 0.3- 0.2- 0.11 0 5 10 15 20 Stage number 25 30 35 Figure 5.9: Efficiency calculated for the depropanizer, compared with those ChemSep: Propane of limit 0-1 is not claimed here. However, efficiencies are very much composition dependent, so that, depending which compositions have been considered, different results are going to be obtained. With the definition presented in this limit. Consider0-1 the the efficiency-like parameters will stay within work, ing other definitions of efficiency, it is clear that they can go beyond the 0-1 limit, however it is still uncertain if the definition is consistent as well as if the for is bulk the correct. calculating equilibrium composition composition use of Taylor and Krishna (1993) in the concluding remarks of Chapter 13 (MultiEfficiency Models) Distillation: that the compopointed out also component to the computed equilibrium composition, nent efficiencies are very sensitive different it is that to They efficienpossible obtain completely also claim y*. is for different by thermodynamic computing y*, which model using a cies different be However, it to could even possible get efficiencies not surprising. but interaction thermodynamic the the model only changing same utilising here both models or parameter sets give a good parameters, and surprisingly data. the vapour-liquid equilibrium representation of 126 0.1 0.1 i o.«, w 0.0.ý 0.: 0.11 05 111111 10 15 20 Stage number 25 30 35 Figure 5.10: Point Efficiency for the depropanizer column: Propane This is exactly what it is possible to observe in this work. Even when vapour liquid composition are quite well matched, efficiencies are not. and In order to compare the well-know Murphree efficiency and efficiencies calhave knowledge transfer, the through also point efficiencies mass of culated been calculated for the depropanizer column. Figure 5.10 shows the results. Completely different values are obtain. More experience with this kind of 'right' is help in the one. understanding which one efficiency could Efficiency remarks definition be in few that In general, there are a our marked should points following The conditions, which of efficiency as well as model assumptions. follows from the assumption of perfect mixing in the bulk phase, must apply: 0< therefore then if > yi*,p yi, p+,, yi, p > yi, p+,, a). Em < 1. 0< therefore b). if yi, < yi, p+,, then yi* yi, p+,, p ,p< Em < 1. 127 The conditions, a.) and b. ), are fixed by the thermodynamics the of mixture. The equilibrium composition for a light component is always higher than its corresponding non-equilibrium value. Similarly, the equilibrium composition heavy less is than the corresponding non-equilibrium value. of a component These conditions are always satisfied in the simulation results when the conis vergence tight enough. When composition changes between plates are small, quite accurate simulation results are neededbecauseseveral places after the decimal point are significant. With all this we can conclude that while the non-equilibrium models give more accurate composition and temperature profiles than the equilibrium models, and provide more useful information, they are not able to match experimental tray efficiency values becausecom) ) b. This does not mean that experimental positions satisfy conditions a. or it values of efficiencies are wrong, just means that the non-equilibrium moddeveloped, including the one presented in this work, still are theoretically els limited by the assumption of perfect mixing of the bulk phase, given by ) ). b. conditions a. and 5.3 Illustration Hybrid of Modelling Approach With the results obtained in the previous Sections, the solution of hybrid be highlighted. depropanizer The models can now column and the esterifihybrid The the cation column are used as examples of modelling approach. for from the the the model results combined esterification, which reproduces less data in 20% compuexperimental quite accurately using approximately tational time, are shown. Using the tray efficiency parameters from the non-equilibrium model in the been for depropanizer have the performed equilibrium model, simulations for is 0.85 An the equilibrium model of used. specified efficiency column. A hybrid model switching between non-equilibrium and equilibrium model has also been performed. The top section has been simulated with the nonbottom the the section with equilibrium model, as equilibrium model and by the efficiency analysis. suggested The results in Figs. 5.11 and 5.12 show the simulated liquid and vapour comfor fixed HM the caseof equilibrium model with efficiency. position profiles in these Figures refers to simulations with fixed column efficiency, while the 128 1.0 0 0.8 as 75 E c 0 0 aE 0 cr --0- 0.6 --00.4 --0- Ethane Propane Butane Pentane Ethane(HM) Propane(HM) Butane(HM) Pentane(HM) 0.2 0.0 5 10 15 20 25 30 35 Stage number Figure 5.11: Liquid composition profile with equilibrium efficiency and non-equilibrium model fixed model with from A the the other results are ones obtained non-equilibrium model. simfrom the non-equilibrium model to the equilibrium model at the ple switch converged steady state with the non-equilibrium model performs this simulation (time used to convergethe non-equilibrium model: 2.53 min., time for the hybrid model: 1.25 min. in a SUN-Ultra I Workstation). With a fixed column efficiency value (for equilibrium model), small deviations can be noted. However, for the hybrid model, not surprisingly, hardly be noted. any variation can The same procedure has been repeated for the esterification case. In this 129 1.0 0.9 0.8 co 0.7 0 E 0.6 .20.5 0 0.4 0E 0 0.3 0 CL 0.2 co 0.1 0.0 5 10 20 15 25 30 35 Stage number Figure 5.12: Vapour composition profile with equilibrium efficiency and non-equilibrium model 130 fixed model with 0.6 Full non-equilibrium model Hybrid model 0.5 0.4 0 0.3 0.2 0.1 0.0 10 2468 12 14 Stagenumber Figure 5.13: Hybrid model applied to the esterification column Gibbs free the to the energy analysis, a excess efficiency and case, according hybeen The has 0.6 for the used. equilibrium model of specified efficiency brid model, comprising stages 1 to 3 simulated with equilibrium model and been has 4-13 performed. also simulated with non-equilibrium model, stages The results obtained from the hybrid model are presented in Fig. 5.13. Small deviations in the composition values (bottom of the column) in the hybrid model of combined non-equilibrium and equilibrium models, compared with (time the to full the non-equilibrium used converge non-equilibrium model SUN-Ultra for hybrid 1 in 1.46 time the 2.13 a model: min. min., model: Workstation). 131 5.4 Conclusions The general(hybrid) model contributes to the development of a computer distilfacilitates the the aided environment which understanding of reactive lation operation and the solution of different simulation problems. The new features have been model validated through reactive and non-reactive systems. The hybrid modelling approach not only allows different model types in the modelling of a single distillation column, but also the use of different during the solution of a specific simulation problem. simulation modes The hybrid modelling approach provides a powerful analysis tool through be different importance in the the terms models can established. of which In principle, the non-equilibrium model should provide a more correct and distillation behaviour the the reactive system. of consistent prediction of The non-equilibrium model, however, is more time consuming and requires information on a greater number of model parameters. Use of a suitable hybrid model, whenever applicable, makes the solution more efficient without loss of model accuracy. appreciable In addition to the analysis presented in this Chapter, the hybrid models, WAS or gPROMS can also be used for other types of analysis, such as bifurcation (which has already been shown in Chapter 4), controllability and dynamic optimization. Analysis with the hybrid model in terms of controllability timization is presented in the following Chapter. 132 dynamic opand Chapter 6 Controllability Issues and Dynamic Optimization Reactive distillation systems require tight process control to ensure that defrom 'optimum' be handled The development the viations can adequately. identification the of a good process control scheme requires of the process objectives, translation of these objectives to possible or achievable control integration the objectives and with the control strategy. Reactive distillation differs from conventional distillation due to the presence has has It been influthat this of reactions. proven a significant phenomena (Sneesby 1997). the the al., performance of et ence on control system Obviously, the main objective of any process is to maximise profitability. The objective for the reaction part of the column should be to maximise the be for to the the should separation part objective reactant conversion, while fulfil full Hence, the the the column should purity of products. maximise the overall objective (maximise profitability) by achieving in each process (reaction and separation) the particular objective. Other processobjectives such as contamination problems when working with dangerous products or to maximise the catalyst life can also be included. Before starting to implement any controller design some ideas about how distillation What to is to a measure? reactive unit are needed. control easy What to manipulate? What are the best pairings? 133 Newells and Lee et al.(1987) presented somequalitative rules answering these questions: * control outputs that are not self-regulating have favourable dynamics characteristics that e control outputs inputs have large that * select effects on the outputs inputs that rapidly affect the controlled variables 9 select These questions are characteristic of the control of the process itself: controllability. Thus, controllability is independent of the controllers and can only be changed when the design of the unit is changed (Skogestad et al., 1990). This is then the main reason for controllability analysis made at the design stage. For control system design, many mathematical methods are used, but for broad indication. A 'tethe only controllability results provide a number of dious' simulations normally replaces this analysis. Important features on controllability of reactive distillation are described in this Chapter. A linear model is shown. Configuration is defined as well as disturbances. measurements,manipulated variables and In the second section of this Chapter dynamic optimization is addressed. The optimal control results show that integrated design-control techniques distillation be to columns. reactive applied should 6.1 Controllability Issues There has been very little research in the area of controllability of reactive has been for batch disHowever some research presented reactive columns. tillation (Sorensen, 1994). In engineering practice a plant is called ccontrollable' if the plant is able to (Rosenbrock, 1970). the control objectives specified achieve 134 All controllability assessmentsstudies have been proposed with the main idea of perfect control. This idea is limited by time delays, right-half plane zeroes, model uncertainty and manipulated variables constraints. Controllability analysis is an extensive area of research but only some characteristics here. are presented Most of the controllability analysis found in the literature is presented in terms of linear measures. Although it may seem restrictive usually it is not, important the most since non-linearity, given by the constraints in the inbe handled linear (Skogestad However, 1990). puts, can with analysis et al., non-linear simulations are always required to confirm the results obtained from the linear analysis. The procedure for controllability analysis in this work is described by the following main steps (Wolff et al., 1992) starting with the linearised model, o scale the plant; * compute controllability measures; and, 9 analyse controllability. To analyse the controllability of reactive columns a control configuration be There is be the that probably no single configuration must assumed. can best for all the columns (Skogestad et al., 1990). Because of the large number of possible control configurations there is a clear need for tools that can best in the the selection of possible configuration. assist In the following Sections a linear model, some issues about scaling, controllability tools and possible control configurations are presented. Linear Model Applying linear control theory to nonlinear systems is an approach adopted by several authors who have demonstrated its success on a number of realistic by Morari Perkins The Barton works reported and et al(1985), examples. (1986), Perkins and Wong (1985) among others, confirmed that the closed loop performance of the control system was consistent with the predictions theory. the of 135 Almost all the tools developed for controllability analysis are in terms of linRGA, ). ear models: poles, zeroes, condition number, etc. So that, in order to analyse the controllability of reactive columns a linear model is needed. Based on the non-linear model of Chapter 3, a linear model has been deby linearising This the equations veloped at an operating point. operating is linearisation done the is the condition steady-state and since point around this point, the linearised model will only be valid for small deviations from the steady-state and the validity of the model should be probed. A linear model of any process can be described by the following equations, i= Ax + Bu + Ed Cx + Du + Fd (6.1) (6.2) where, y= x= [measurements d= [allstatesIT (6-3) variableS]T (6.4) T [control u= variables] [disturbance variableS]T Applying Laplace transformation (6.5) (6-6) to the linear model, we get, Gd(s)d(s) G(s)u(s) + y(s) = (6.7) describes functions the Gd(s) G(s) transfer efmodels which are and where fect on the output of the input and disturbances. The objective is to keep the outputs close to their set-points (r), i. e, e=y-r, by inverting The ideal disturbances. this control will accomplish and to reject become, inputs that the the process such manipulated G-'r - GW'd d (6-8) (control) linear the deriving When varimodel, measurements, manipulated have been defined disturbances as, and ables 136 The measured variables are temperatures and/or, composition of the (it is assumed that these variables can be practically meaproducts ). Interest in from these controlling sured! variables comes previous in distillation batch discontrolling results columns as well as reactive tillation (Skogestad et al., 1990; Sorensen,1994). The distillate flow the boil-up the manipulated variables are 9 and vapour heat the and/or of the reboiler. Disturbances are in feed flow-rate and in feed composition. The selection of disturbances in the feed composition will offer also an indirect looking disturbances Disturthe way of at on reaction parameters. bances in the feed composition affects the reaction stoichiometry. The linear and non-linear models are compared by looking at Fig. 6.1. A 10% step in distillate flow rate are introduced to both models. The linear (shown in describes in the top the graph) model responses the same way as the non-linear model (bottom graph) (the axes for the top graph are the same bottom Ethyl for the top the the graph: acetate composition at of column as linear The for be time). then model can controllability and control used vs. far from for linearised disturbances i. too the small point, e. only studies not (Skogestad et al., 1990). 6.1.2 Scaling Why is it important to scale the plant? Most of the controllability measures are scale dependent. It is crucial that following described in is One the way of scaling variables are scaled properly. (Skogestad and Postlethwaite, 1996), its (u): to Inputs allowed range. normalise uj with respect 9 its (y): to Outputs allowed range. normalise ej with respect * its (d): dkwith to Disturbances allowed range. respect normalise * Another way, which also achievesthe same objectives, is to scale the transfer d, Gd, G the that the signals e of u, assuming magnitude allowed and matrices (Havre Skogestad, The later frequency 1992). the do with and vary not r and in this work. is the scaling adopted 137 Step Response x 10, 3. 2.5 2 1.5 1 0.5 0 0.525 IE 0.52 0 0 a E 0 0.515 Co w 0.51 0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 Time (h) (10% linear from the Response Figure 6.1: and non-linear model obtained step change)- 138 6.1.3 Controllability Tools Which tools for controllability analysis are available? Functional e controllability The first thing that should be checkedis that the plant is functionally (Skogestad Postlethwaite, 1996). controllable and Essentially, a plant is not functionally controllable if the rank of G(S) is for all s less than the number of outputs that we want to control. This is extended for square plants to the requirement that det(G(s)) should be different from 0. For unstable plants it should be checkedthat the unstable states are state controllable and state observable. Nevertheless, this is not important becausethe states we really care about are in the output vector y. half Right delays time e plane zeroes and Right-half plane zeroes (RHP-zero): a zero is defined as the value for G(s) (for for loses be this s which rank square matrices may computed det(G(s)). A G(s) limits RHP-zero the the achievsolution of as of bandwidth holds the this able of plant and regardless of the type of (Holt RHP-poles limitaMorari, 1985). also put controller used and tions on the control system through stability considerations. If there direction, is important RHP-zeros in it that the the and poles are same RHP-pole at p is located at a higher frequency than the RHP-zero at i. z, e, p> z- Time delays have essentially the same effect as RHP-zeros. decomposition Singular value e Singular value decomposition (SVD) of any matrix G is: UH. (6.9) There will be rank(G) singular values. The SVD is useful to examine which manipulated input combinations have the largest effect and largest disturbances the give output variations. which 139 Condition * number The ratio between the maximum singular value and the minimum is the condition number. Plants with large condition singular value 'ill-conditioned', different input magnumbers are and require widely depending direction desired the (Skogestad the nitudes on of output Postlethwaite, The 1996). is and process sensitive to plant/model mismatch when the condition number is high. Relative o gain array The relative gain array (RGA) defined by Bristol (1966) for a square defined is 'open-loop' 'closed-loop' the bethe plant as ratio of and gains tween input 1 and output Z. Plants with large RGA are 'ill-conditioned'. The RGA is scaling independent and ideally should be a diagonally dominant matrix. In this case the control loops are largely decoupled. The general rule about pairing is to pair the controlled outputs yj with the manipulated variables uj in such way that the RGA values are positive and as close as possible to unity. RGA can also be calculated as a function of frequency. If the large values of RGA are at high frequencies One 'never' difficult to should control. use a controller with plants are large RGA values (Skogestad, 1992). One big disadvantage of RGA is that it contains no information about disturbances and dynamic be- haviour of the plant. 6.1.4 Control Configurations The control configurations of a distillation column can be divided into two categories: Configurations Basic e Schemes Control Ratio e basic the 6.1 Table configurations: summarises feasible but lead bad level VB to LD Both configurations are control and (via balance) the control overall mass and reduces composition affects which control response. 140 II Configuration LV LB LD DV DB VB Primary Variable Manipulated duty reflux rate or reboiler bottoms rate distillate rate distillate rate distillate rate bottoms rate Secondary Variable Manipulate I duty reboiler or reflux rate reflux rate reflux rate duty reboiler bottoms rate duty reboiler Table 6.1: Basic control configurations The LV configuration is the most common used in the industry. The main diis this that the manipulated variables affect configuration advantage of to compositions. redly The IV configuration is sensitive to disturbances when no control is used but RGA is In the insensitive this configuration, with one-point control. rather large large for high be to reflux. purity columns with expected The DB configuration it is believed violates the mass balance and is normally demonstrated has been its experimeneffectiveness not considered although tally (Skogestad et al., 1990). This configuration was considered as a non does basis it however basis, long term term on a short a on viable alternative long for time then the B If D behave a and are adjusted unreasonably. not is This then filled being by configuration up or emptied. column will end feed-back control. actually rather good when using composition For the DV configuration the RGA elements are always small contrary to the (thus, infinite RGA DB configuration where the at steady state elements are for this is RGA configuration). the measure not a reliable but the more For the ratio control schememany combinations are possible, by been has (L/D)(V/B) Table in 6.2. recommended presented are common The double known 'the Its Morari, 1987. Skogestad and ratio' scheme. as but for double is not control composition advantageous scheme ratio control (as decoupling implicit This for single composition control. scheme creates the schemes). ratio all well as 141 Configuration I Primary Manipulated Variable (L/D)V duty reboiler (L/D)(V/B) boil-up ratio reflux ratio or (V/B)L reflux ratio Secondary Variable Manipulated reflux ratio boil-up ratio or reflux ratio boil-up ratio Table 6.2: Ratio control configurations Important differences between configurations are related to interaction when (Shinskey, input 1984) to and sensitive using single-loop controllers gain unfor both large decoupler In two-point control. casesa certainty when using frequency in RGA the the range of closed-loop results at control value of problems. Recent work has been presented in the literature for an ETBE (ethyl-tertbutyl-ether) production (Sneesby et al., 1997). They recommended two configurations: LV and LB, both for a single composition control. They found that the location of the temperature control is critical and can dramatically due is interactions This to the the the control scheme. stability of affect between the chemical and phase equilibria. Dynamic simulation for studying the responsesof the system was recommended. A model for reactive distillation columns including control system and the (1997b). by derived Bock et al. possibility of coupling additional columns was in disturbances to They analysed sensitivities a manipulated variables and for the distillation esterificolumn column and a coupled recovery reactive (to it SISO Based the this make simple) control on cation of mystiric acid. structure was carried out. Bock et al.(1997b) also found that the influence of the reaction temperature is be known, has temperature the to the when reactive column profile within They for the the using system controlled purity. variable reference used as a to loops, SISO-control ensure conversion. not sufficient two as only one was by the top the manipuThey kept the temperature at column constant of fast leads to the the top heat a very lating the column, which of supply at its proximity; and the temperature in the middle of the due to loop control In heat by the the kept is most of reboiler. manipulating constant column A long time disturbances the was quite well. were eliminated the cases of influence the the the could not to of reaction when set-point achieve required 142 be compensated by a manipulated variable. For this work the DV configuration has been chosen. 6.1.5 Examples Application of It is known that the interactions between the top and bottom of distillation large. Changing lead the top to changes columns are very conditions at will in the bottom, and vice versa. Becauseof this distillation columns are known difficult impossible to control. as very or Using two single control loops it may be possible to control both the top and the bottom of the column. In the caseof reactive distillation it is desirable to control not only the combut loss be the the that position of product, also of reactants may at the be the that column so other end of only one point control will not able to achieve this. The temperature of the reactive section may be another variable desirable to however, it is in For this the work not examples studied considered. control, it seemsto be sufficient to control compositions or top and bottom temperatures. Controlling these variables the temperature in the reactive section (or in the column itself, in the case of full reactive column) suffices to achieve desirable temperature profiles. The DV-configuration was adopted for this controllability analysis: DV-configuration: controlling the distillate flow and the reboiler heat duty (or vapour boil-up) to control the distillate composition (or temperature) (or bottom temperature): the composition and Controlled variable: Manipulated variable XD XB TD) or (or, TBor D XD or XB (or, TBor TD) V Procedure 143 Assuming that we have made a decision on the plants inputs and outputs, G investigate to to the we want analyse model what control performance can be expected. The procedure is described from Skogestad and Postlethwaite (1996) and is only an extension from the procedure presented above. The described below. indices controllability are Linearise the model. * Scale all the variables (inputs, outputs, disturbances, references) to obtain a scaled model. Check functional * controllability. Compute the poles. 9 Compute the zeroes. e Obtain G(jw) frequency RGA the the 9 response and compute matrix. Compute the singular values of G(jw) and plot them as function of frequency. Compute condition number: the ratio between the minimum and maximum singular singular value. Analyse. de For the results shown in this thesis, the hybrid model included into the WAS simulator has been linearised, however, gPROMS also allows the linhave been indices All the obtained using controllability earisation of models. MATLAB. MTBE Production The MTBE production column has been first analysed by a series of simfor for The taken which multiple solutions exist was steady state ulations. butenes fed fed is in 10 11. Methanol tray the while remain at stage study. Simulations have been performed with the non-equilibrium model and it was hour disturbance for feed flows in have I to changes and composiassumed distillate flow-rate. duty found heat Full be in and tion, reboiler results can 144 -El 0 Le E E 0 0 1 CL 9 w Co 5 10 15 Time (h) Figure 6.2: High conversion of MTBE as result of 10% increase in reboiler heat duty. Appendix B. The reboiler heat duty was found to be a very sensitive variBelow high is There the conversion solution. a range of values giving able. low drastically to the the that conversion changes system range or above by (see This Figs. the 6.2,6.3). the conobtained results confirms solution tinuation method (bifurcation analysis in Chapter 5). Through simulation it has been shown that adding or reducing the number of it in However improve does the many cases column performance. not stages from Fig 5 be 6.4 where reactive stages seen may worsen performance, as can have been added to the system. The purity of MTBE at the bottom of the design. by half the the is to nominal achieved of purity almost reduced column between desirable is the in then the If any changes ratio number of stages between the the the size of reactive section, as well as stripping section and (see be the changed. reactive column should the rectification section and detail for B Appendix explanations). more described following the above Secondly the controllability analysis procedure 145 0.9 0.8 0.7 E E 0.6( 0 1 CD fm 0.5 0 1 0.4 CL 0.3 - Co 5 0.2 0.1 0. 0 0.5 1 1.5 2 2.5 Time (h) Figure 6.3: Low conversion of MTBE as a result of 15% decreasein reboiler heat duty. Once in this analysis, the steady state with multiple sowas studied. again lutions is analysed (methanol is fed in tray 10 while the butenes remain fed for DV-configuration be Three 11). the as can cases were analysed at stage Some indices from Table in 6.3. obtained are surnmarised controllability seen Table 6.4. Accounting for the process dynamics some controllability indices, frequencies. RGA and singular values were calculated over a range of such as The results from the frequency-dependent analysis are the same as obtained for the steady-state analysis. Detailed calculations are in Appendix B. From the analysis of all the indicators (steady-state and frequency dependent) it can be seen that: have indicating the the systems negative real parts, stability poles of 9 all the open-loop system, of limiting RHP-zeroes, the the present achievablecontrol studied cases " all performance, to to the plant/model very sensitive are mismatch, according processes " the high condition number, 146 1.0 --oA 0.9 Nominal design Simulated design with 5 extra stages 0.8 0.7 It: 0.6 0.5 0.4 0.3 0.2 0.1 0.0 05 15 10 20 25 Stage MTBE Lower 6.4: Figure as a result of adding extra reactive stages purity of 147 -- FMeasured Variables Cases Case I XMTBE, D XMTBE, B Case 2 XMTBE, TB D Case 3 TD I Disturbances Xisobutene, F TF Xisobutene, F TF Xisobutene, F TF TB Table 6-3: Measured variables for the controllability study on MTBE production Case RGA 1 1.063 2 -0.06 1.047 Singular values -0.06 1.063 -0.047 1.047 -0.047 0.71 0.29 0.29 0.71 3 mZn max mM max mZn max Table 6.4: Controllability 0.0223 2.8356 0.00387 0.12200 0.000316 0.320000 ondition Number RHP-zeroes 127.16 yes 31-52 yes 1012.7 yes indices for the MTBE production following for from RGA the the three the cases, control pairings values 9 be suggested, can Manipulated variable: Controlled variable: XMTBE, D D V TB or XMTBE, B Temperatures are easy to measure in a distillation column, and they have a linear dependence with the control measure. The pairing of the bottom in done be the temperature the of with vapour-boil-up should the column probably in spite of the very small minimum singular value (compared with that of Case 1). in is the the all cases, suggesting very small minimum singular value 9 desired be inputs the to large that outobtain may needed magnitude puts. Ethyl Acetate Production described following the procedure above was also The controllability analysis The Chapter in 4 for the column column. studied using esterification studied 148 the equilibrium model with tray efficiency is taken for the linearisation and Three for DV-configuration the controllability analysis. caseswere analysed Some in Table 6.5. as presented results showing the controllability indices (full in Table 6.6 Appendix in B). are given results are shown Cases Case 1 Measured Variables Disturbances TD Xethanol, F TF TB Case 2 XEtAc, D Xethanol, F TF XEtAc, B Case 3 XEtAc, D Xisobutene, F FF XEtAc, B Table 6-5: Measured variables for the controllability production Case RGA 1 1.17 -0.17 0.17 1.17 1.05 -0.05 1.05 -0.05 1.02 -0.02 1.02 -0.02 2 3 Singular values min max min max mZn max Table 6-6: Controllability 0.00106 8.08000 0.00018 0.24400 0.0008 0.1317 study on ethyl acetate Condition Number 7622.64 yes 1355.56 yes 164.62 no RHP-zeroes indices for the ethyl acetate production From the analysis of all the indicators (steady-state and frequency dependent) it can be seen that: indicating have the the stability negative real parts, systems poles of * all of the open-loop system, limiting but RHP-zeroes, the the one present casesstudied almost all Case for it 3, is RHP-zeroes there not achievable control performance, disturbances be that the temperature are measurements or noted may RHP-zeroes in the the system, generating the processes are very sensitive to plant/model (very high), the condition number 149 mismatch, according to from the RGA values for the three cases,the following control pairings be can suggested, Manipulated variable: Controlled variable: D XEtAc, D v XEtAc, B the minimum singular value is very small in all the cases, suggesting that large magnitude inputs may be needed to obtain the desired outputs. 6.2 Dynamic Optimization If we wish to determine, for example, the optimal diameter of the reactive in column addition to the optimal way of operating the column over the time then we need to use dynamic optimization. Since operation and design problems are of transient nature then both are dynamic known is applications of also optimization, which as optimal control. The dynamic optimization in this work is carried out by using gPROMS and deis Ltd., In 1997). to general, gPROMS able solve problems gOPT(PSE by following items, the scribed Mathematical Problem We consider a process described by: i(t), y(t), U(t), V) = g(x(t), i(t) differential the the are variables, and algebraic where, x(t) and y(t) are time derivatives, u(t) the control variables and v the time invariant parameters to be determined by the optimization. Initialisation by: described is initial the the that system condition of gPROMS assumes i(o), Y(O), U(O), V) =0 g(x(o), 150 By performing dynamic simulation we will completely determine the transient response of the system, once the time variation of the control is fixed. Objective Function determine: to gPROMSIgOPT seeks horizon, tf, the time * invariant the the time * v, value of parameters, horizon the the time the time variation of 9 control variables over entire final to the as well as minimise or maximise value of a single variable, z, mintf, u(t), V,tE[t, tf]Z(tf ) (6.12) differential is the variables x or the algebraic variables y: z either Z(tf) = 10tf Bounds of the optimal D(x (t) Iý (t) (t), u (t), v) dt Iy decision variables tmin f-f min < U v min < tf < < U(t) <V< tmax (6-14) Umax (6.15) Vmax (6.16) Vt E [07tf] End point constraints These are conditions that must be satisfied at the end of the operation. W* equality =#,. w(tf) = inequality =* W, in < Path constraints 151 W(tf) < Wmax (6.17) (6.18) These are conditions that must be satisfied at all the time during the operation. mtn w < W(tf) Vt E [01tf :5 Wmax (6-19) Classes of control variable profile but this is the one utilised within this gPROMS offers several possibilities, work, Piecewise 9 constant control: the controls remain constant at a certain value over a certain part of the time horizon before jumping discretely to a different value over the next time interval. Solution The solution comprises the following: horizon, the the time tf, 9 value of 9 the values of the time invariant parameters, 9 the variation of the control variables over the time horizon. The dynamic optimization is problem solved as: e parametrisation of the control variables fix the parameters values o informaDA-SOLV integrate the the to system and get sensitivity use tion NLP SRQPD to the to get new set of values. solve 9 use The control vector parametrisation approach (Vassiliades et al., 1994a,b) is implemented within the gPROMS process modelling (PSE Ltd., 1997). The dynamic the optimization problem problem of solving a approach converts (DAE, initial estimates, junction conditions and point equality and inequality finite (NLP). dimensional The to nonlinear programming a soluconstraints) DAE NLP the the tion of solution of a multistage are calculated via system 152 in the variable sensitivities. The control parametrisation approach requires that the DAE system of the problem being solved must have a solution for decision the variables may take. any set of values of Other method of solving dynamic optimization has been developedby Biegler (Cuthrell Biegler, In 1988). differentheir the and approach and co-workers tial equations are discretised by using Lagrange form polynomials and orthogThe onal collocation. resulting algebraic approximations are then written as in NLP. The large NLP's is for an solution constraints of still under study, that, the method works quite well for small systems but starts to fail when large number of equations are neededto represent the physical problem, since discretised. the equations are all 6.2.1 Optimization problem This study addresses some aspects of optimal control of a reactive distillation is The this the example since esterification column was used as an column. system that, using equilibrium models with tray efficiency, can accurately data. represent experimental Problem statement: The system under consideration is shown in Fig. 6.5. Table 6.7 gives the structure of the optimization problem. The overall objective is to design the column (diameter, areasof reboiler and (Jo horizon interest finite time feasible to of operation over a condenser) give hours) while maximising profitability, i. e. minimising the total annualised cost. minimise Cost = minimise (Cshell + Ctray + Chx + Cop) (6.20) Chx shell)Ctray is the tray, the is the the cost of column cost of whereCshell C,, is (for the the heat and is the reboiler and condenser), exchangerscost p operating cost. bottom loss in the the as well of product maximum There are constraints on 153 ]late > 0.50 Ac A EtAc 0.30 Figure 6.5: Reactive distillation column for the design study 154 Objective function: Fixed: Optimized-variables Specifications: Disturbance: Control variables: total annualised cost (13) trays number of feed location (tray 6) diameter of the column (3m < D, < 6m) (4m !ý D2< 8m) areas of condenser and reboiler ethyl acetate top composition (xtop > 0.525) bottom ethyl acetate composition (Xbot 0.3) ý, :! feed composition (sinusoidal, 1h period) distillate flow-rate (1000 < Dzst < 3800(mol/h)) heat duty reboiler 1 (0.25 x 107 <Q Table 6.7: Optimization :50.9X106(M details is The distillate. in the the subject system of product purity minimum as on Manipulated feed in disturbance the variables to a sinusoidal composition. flow-rate distillate the heat the for control are the according reboiler and of to the suggestions from the previous controllability studies. function together The 2609 with The problem comprises objective variables. C. Appendix in disturbances and constraints are given Results (through tested simulation) The nominal design for the reactive column was feed disturbance The comwas a sinusoidal in the presenceof a disturbance. large violations. there As constraint of number a were expected position. be top the seen composition, as can The most important was the violation of limits beyond the from Fig. 6.6, where the composition of ethyl acetate goes for highlights This the be 0.525). need (minimum purity at the top should the system. controlling 155 0.6 0.5 r. 0.4 0 -,4J4 -A th 0 a 0 u 0 , to 4J 4J pq 0.3 0.2 0.1 0.0 10 Time Figure 6.6: Top and bottom compositions of ethyl acetate. Results from the disturbance. simulated nominal case with 156 0. E 0.5 0.4 0.3 co -i 0.2 0.1 0.0 56789 10 Time Figure 6.7: Ethyl acetate composition. Results from the dynamic optimization. Dynamic optimization was carried out, controlling the reboiler heat duty and the distillate flow-rate. The results are shown in Fig. 6.7. Table 6.8 comFor the the the and optimal values obtained. optimal solution nominal pares there are no constrains violations. The system is rejecting the disturbance by is increasing heat the top the product specification maintained, of while (explaining implies is increased that the this the reboiler, reaction rate why the bottom purity increasesalthough remaining below the upper bound). It because is increased that the the be while operating well, cost of as noted can duty increase, in is heat the a notable reduction capital cost made reboiler by a reduction in the column diameter. 157 0.9 0.8 0.7 CY 0.6 0 U 0.5 0.4 0.3 0.2* 0 56789 10 Time Figure 6.8: Controlled variable: Reboiler Heat Duty(* 106(MJlh)). from the dynamic optimization. 158 Results Variable Nominal Optimal Diameter Reboiler heat duty Distillate flow-rate 5.6/6.25m 106 MJ1h 0.58 x 1150Kmol/h 4.5/5m 106 X 0.85 MJ1h 737.87Kmol/h Table 6.8: Optimization 6.3 results: nominal and optimal values Conclusions Controllability studies have been performed for two reactive systems. The for from have been the this the applied solution of analysis optimal results control problem. Following the controllability studies an optimal control problem has been distillation design From these the columns of reactive optimal results solved. has been determined. The optimal solution found is limited to the objective function, proposed disdisturbances Certainly, defined. types the to turbance, and of other model as however in be included the optimization problem, and uncertainties should The data (or the to results are not available. corroborate plant) experimental for is for be this other generally applicable analysis and used approach can similar systems. large Nowadays, for broad The results presented open up a research. area trouble, be simultaneous and much without solved can problems optimization design of reactive distillation columns and their associated control structure, for be the disturbances optimal solved to can uncertainties as well as subject solution. 159 Chapter 7 General Conclusions Future Work and The main contributions of this thesis have been: hybrid for distil9A model continuous staged reactive and non-reactive lation columns. The hybrid modelling approach allows different model types in the moddistillation It different the elling of a single column. also allows use of during the solution of a specific simulation problem. simulation modes The framework developed helps to avoid problems of mismatch between model and simulation results as well as between the two forms of models. New tools for the analysis of distillation columns. The importance of terms in the different models can be established. The analysis of the Gibbs free energy as a function of time has been distance from Bifurcation the analysis equilibrium. used as a measure of for reactive columns gives the existence or not of multiple solutions. Controllability analysis. 9 For the first time controllability analysis has been addressed for rehelped The have indices distillation to columns. controllability active determine the best control structure for the problem studied. Dynamic optimization As well as controllability issues, for the first time, to the best of our knowledge, dynamic optimization has been performed for continuous The here, distillation limited work presented systems. although reactive 160 in scope, opens a new area of study for the interaction between design and control of reactive distillation columns. The simulation and design of hybrid (reactive and non-reactive) distillation different types of models and combination of them is made columns with framework developed in this thesis. Different analythrough the possible sis tools, such as tray efficiency values, monitoring of the Gibbs free energy, bifurcation analysis, are available within the framework. Moreover, controllability indices can be determined as a first step towards developing the control finally design distillation the structure and optimal of reactive columns can be found through dynamic optimization. The motivation for this study has been the interactions between reaction and for Increased interest in that the a reactive column offers. separation units fine in chemicals and pharmaceuticals manufacturing, as well as applications in the commodity industry chemical industry, has also been an incentive of this work. The work has been mainly focused in developing models (dynamic and steady behaviour illustrate the to of complex state) which are sufficiently rigorous different The distillation thesis addresses columns. continuous staged reactive distillation design columns. of reactive aspects of operation, simulation and A novel integrated model has been presented, given the hybrid modelling different The integration types of models as well as modes of of approach. design for the tool of reactive columns. simulation gives a new Controllability issueshave also been addressedfor reactive distillation columns. it for been investigated has However, this analysis specific casesand only in Assessment depth. in be process controllability of greater studied should design should include many other aspects such as the ability of the system to handle process uncertainties. in has been dynamic this Optimal control or work. addressed optimization future for benefit that work This constitutes a major area of work would in this from the leading work. obtained results 161 The following future work is suggested: First of all the need for experimental data must be highlighted, since lacking literature. in data is the open seriously real be The both throughly tested 9 non-equilibrium model should more when in liquid film included. the transfer and vapour mass resistances are Models that consider reaction in the liquid film may be needed in cases (albe in framework fast included the general of reactions, and should though at the present such models have not given much advantage over the models which does not consider reaction in the liquid film). be investigated. in The the models should effect of changing correlations * Controllability e be in terms of non-linear models. may carried out Dynamic optimization should be carried out in order to determine the design the together of with optimal control structure and parameters the reactive column at minimum annualised cost. It has been shown that the simultaneous design of the process and the control structure (Mohideen design 1996; et al., offers great advantages over sequential Bansal et al., 1999). Important savings can be made by utilising the detecting the control possibility of simultaneous approach as well as design is be detected the that already made. when might not problems 162 Nomenclature a a Activity coefficient [-] [M2/M3] Interfacial area a' A Interfacial area per unit volume of froth [M2/rn3j Area [m 2] Ab [M21 Bubbling area [M2/Ml] Ap Specific surface area of the packing A, B, C,.. Coefficients in correlations C Concentration [kmol/hj Cal Liquid capillary number df Driving force 21S] Df Fick's coefficient [m [M2/S] Dij Maxwell-Stefan diffusivity (Inverse friction coefficient) Diffusion low D?. i in infinite in j coefficient of species concentration species Z3 Dp Packing size [m] DSp Heat of reaction [J/Kmol] [eq] Amount the of acid groups on resin per unit mass eq Activation energy [J/Kmol] Ea EM Efficiency parameter [-] EM Murphree tray efficiency P Liquid entrainment [Kmol/hl EL EV Vapour entrainment [Kmol/h] fij Friction factor [-] Feed flow [Kmol/h] F Superficial factor [-] F, Excess Gibbs free energy [J/Kmol] GE Froth height [m] hf Exit weir height [rn] h,, [m] Height the H packing section of [J/Kmol] Enthalpy H [J/Kmol] hold Enthalpic HE up 163 [CM 2/sl HI HIf Hvf Hv i K K K, q L M M, M ni N Ni, p NC Np Nr NR P sat P PL PV Q R Re Si, P Sc, S, A t t9 ti T us V V V V X XjXj Y Liquid enthalpy [J/Kmol] Enthalpy of a liquid feed [J/Kmol] Enthalpy of a vapour feed [J/Kmol] Vapour enthalpy [J/Kmol] [MOI/M2, S] Molar diffusion flux Matrix of volumetric mass transfer coefficients [s-1] Equilibrium constant [-] Chemical equilibrium constant Liquid flow rate [Kmol/h] Hold-up [Kmol/h] Molecular weight [Kg/Kmol] Mass transfer rate [Kmol/h] Number of transfer unit Molar flux of speciesi at a particular point in the two phase dispersion [Mol/,rn2 S] Number of components Products Reactants Number of reactions Pressure [atm] Vapour pressure [atm] Liquid extracted from (or added to) any stage [Kmol/h] Vapour extracted from (or added to) any stage [Kmol/h] Volumetric flow rate [m'/s] Gas constant [J/Kmol K] Reynolds number [-] Reaction rate in the liquid bulk [Kmol/hj Schmidt number [-] Sherwood number Time [h] Gas contact time [h] Liquid-phase residence time [h] Temperature [K] 21S] Superficial vapour velocity [m 21S] Velocity [m Molecular diffusion volumes 3/M01] Molar volume at the normal boiling point of the solute [cm Vapour flow rate [Kmol/h] [-] Liquid composition Factor of concentration weight [-] Vapour composition 164 Liquid flow path length [m] z Greek letters: 2 [W/m K] Heat transfer a coefficient [-] LY Point efficiency , r Factor accountingfor thermodynamic non idealities [-] Activity 7 coefficient AH vap V) 0 OP P A A Y Pi V a Enthalpy of vaporisation [J/Kmol] [W/M2] Energy flux Fugacity coefficient Heat transfer rate [W] Density of mixture [Kmol/m'] Density of pure components [Kmol/m'] Thermal conductivity [W/(mK)l Thermodynamic factor Viscosity [CP] Chemical potential [J/Kmol] Stoichiometric coefficient Surface tension [N/m] Superscript Liquid L Vapour v Subscript CP i1i P r AH Heat capacity Components Tray Reduced Enthalpy Mathematical Symbols: Gradient '7 Matrix Inverse of a matrix Abbreviations: Acetic acid AA Algebraic equation AE formulae differentiation Backwards BDF Differential algebraic equation DAE 165 EOS Et EtAC MTBE ODE RHS SRK w Equation of state Ethyl Ethyl acetate Methyl-tert-butyl ether Ordinary differential equation Right hand side Soave-Redlich-Kwong Water 166 References AIChE, 1958, Bubble Tray Design Manual, AIChE, New York. Agarwal, S., and Taylor, R., 1994, Distillation Column Design Calculations Using a Non-equilibrium Model, Ind. Eng. Chern. Res., 33,2631. Al-Jarallah, A. M., Siddiqui, M. A. B., and Lee, A. K. K., 1988, Kinetics of Methyl Tertiary Butyl Ether Synthesis Catalized by Sulfuric Acid, Chem. Eng. J., 39,169. Ancillotti, F., Massi Mauri, M., Pescarollo, E., and Romagnoni, L., 1978, Mechanisms in the Reaction between Olefins and Alcohols Catalysed by Ion Exchange Resin, J. Molec. Catal., 46,49. ASPEN PLUS User Guide, 1997, Aspen Technology. Backhaus, A. A., 1921, U. S. Patent 1400849. InterThe N., E. 1999, Pistikopoulos, J. D., Perkins, R., Bansal, V., Ross, and Control, J. Proc. Distillation, Effect Double Control: Design and actions of accepted. Azeotropic Diagrams Phase Theory F., M. 1987, Doherty, and of Barbosa, D., and A14, Soc. London, R. Systems, Proc. Reactive Two-Phase for Conditions 443. ChemThe Equilibrium Influence F., M. 1988a, Doherty, of Barbosal D., and Sci., Chem. Eng. Diagrams, 43,529. Phase Liquid Vapor ical Reactions on 167 Barbosa, D., and Doherty, M. F., 1988b, The Simple Distillation of HomogeReactive Chem. Mixtures, Eng. Sci., 43,541. neous Barton, G.W., and Perkins, J. D., 1986, A Case Study of Dynamic Simulation in Control System Design, presented at Technology & Industry, Sydney. Bird, R. B., Stewart, W. E., and Lightfoot, E. N., 1960, Transport Phenomena7 John Wiley & Sons Ltd., New York. Bock, H., Jimoh, M., and Wozni, G., 1997a, Analysis of Reaction Distillation Using the Esterification of Acetic Acid as an Example, Chem. Eng. Technol., 20,182. Bock, H., Wozni, G., and Gutsche, B., 1997b, Design and Control of a ReacColumn Including the Recovery System, Chem. Eng. Froc., tion Distillation 367 101. Bondy, R. W., 1991, Physical Continuation Approaches to Solving Reactive Distillation Problems, AIChE Meeting, Los Angeles. Bossen, B. S., Jorgensen, S.T., and Gani, R., 1993, Simulation, Design and Analysis of Azeotropic Distillation Operations, Ind. Eng. Chem. Res., 32, 620. Bravo, J. L., and Fair, J. R., 1982, Generalised Correlation for Mass Transfer in Packed Distillation Columns, Ind. Eng. Chem. Proc. Des. Dev., 21,170. Bristol, E. H., 1966, On a New measure of Interactions for Multivariable ProAC-11,133. Automatic Control, Trans. IEEE Control, cess Model for C. Gani, A Generalized DisA., R., 1986, Ruiz, T., I. Cameron, and Comp. Chem. Computational Aspects, Numerical Columns-11, and tillation Eng., 10,199. Synthesis Nonequilibrium Reactive Gu, D., Distil1994, R., A. Ciric, of and 168 lation Processesby MINLP Optimization, AIChE J., 40,1479. Colombo, F., Cori, L., Dalloro, L., Delogu, P., 1983, Equilibrium Constant for the Methyl Tertiary Butyl Ether Liquid Phase Synthesis by Using UNIFAC, Ind. Eng. Chem. Fundam., 22,219. Chang, Y. A. and Seader, J. D., 1988, Simulation of Continuous Reactive Distillation by a Hornotopy- Continuation Method, Comp. Chem. Eng., 12, 1243. Chen, GX, and Chuang, K. T., 1995, Liquid-Phase Resistance to Mass Transfer on Distillation Trays, Ind. Eng. Chem. Res., 34,3078. Cuthrell, J. E., and Biegler, L. T., 1988, On the Optimization of DifferentialAlgebraic Process Systems, AIChE J., 337 1257. Daubert, T. E., and Danner, R. P., 1986, DIPPR Data Compilation, AIChE: New York. Dribika, M. M., and Sandall, O. C., 1979, Simultaneous Heat and Mass Transfer for Multicomponent Distillation in a Wetted-wall Column, Chem. Eng. Sci. 34,733. Fuller, E. N., Ensley, K., and Giddings, J. C., 1969, Diffusion of Halogenated Hydrocarbons in Helium. The Effect of Structure on Collision Cross Section, J. Phys. Chem., 73,3679. Fuller, E. N., Schettler, P.D., and Giddings, J. C., 1966, A New Method for Prediction of Binary Gas-PhaseDiffusion Coefficients, Ind. Eng. Chem., 58, 19. Fredenslund, Aa., Gmehling, J., and Rasmussen, P., 1977, Vapour-Liquid Equilibria Using UNIFAC, Elsevier. Gani, R. and Cameron, I. T., 1989, Extension of Dynamic Models of DistilComp. Simulation, Chem. Eng., State Steady 13,271. to Columns lation 169 Gani, R., Hytoff, G., Jacksland, C., and Jensen, A. K., 1997, An Integrated Computer Aided System for Integrated Design of Chemical Processes,Comp. Chem. Eng., 21,1135. Gani, R., Ruiz, C. A., and Cameron, I. T., 1986, A Generalized Model for Distillation Columns-I, Comp. Chem. Eng., 10,181. Gicquel, A., and Tork, B., 1983, Synthesis of Methyl Tertiary Butyl Ether Catalized by Ion-Exchange Resin- Influence of Methanol Concentration and Temperature, J. of Catalysis, 83,9. Glansdorf, P., and Prigogine, 1,1971, Thermodynamic Theory of Structure, Stability and Fluctuations, John Wiley & Sons Ltd., NewYork. Gokhale, V., Hurowitz, S., and Riggs, J. B., 1995, A Comparison of Advanced Distillation Control Techniques for a Propylene/ Propane Splitter, Ind. Eng. Chem. Res., 34,4413. Gorak, A., 1987, Simulation Methods for Steady State Multicomponent tillation in Packed Columns, IChemE Symposium Series, A413. Dis- Guide, Systems User Process Enterprise Ltd., London, U. K. 1997, gPROMS Jacobs, R., and Krishna, R., 1993, Multiple Solutions in Reactive Distillation for methyl Tertiary-Butyl Ether Synthesis, Ind. Eng. Chem. Res., 32,1706. Jelinek, J. and Hlavacek, V., 1976, Steady State Countercurrent equilibrium Stage Separation with Chemical Reaction by Relaxation Method, Chem. Eng. Commun., 2,79. Hauan, S., Hertzberg, T., and Lien, K. M., 1995, Why Tertiary Butyl Ether Production by Reactive Distillation May Yield Multiple Solutions, Ind. Eng. Chem. Res, 34,987. Havre, K., and Skogestad,S., 1996, Input/Output Selection and Partial ConCongress, San Francisco, USA, VOI. M, World 181. IFAC trol, 170 Hayden, J. C., and O'Connell, J. P., 1975, A Generalised Method for PredictSecond Coefficients, Virial Ind. Engng. Chem. Proc. Des. Dev., 14,209. ing Higler, A., Taylor, R. and Krishna, R., 1997, Modeling of a Reactive Separation Process Using a Non-Equilibrium Stage Model, AIChE Annual Meeting. Higler, A., Taylor, R. and Krishna, R., 1998, Modeling of a Reactive SepaProcess Using Non-Equilibrium Stage Model, Comp. Chem. Eng., ration a 22, S111. Higler, A., Taylor, R. and Krishna, R., 1999a, Nonequilibrium Modelling of Multiple Steady States in MTBE Synthesis, Chem. Reactive Distillation: Eng. Sci., 54,1389. Higler, A., Taylor, R. and Krishna, R., 1999b, The Influence of Mass Transfer Column for Mixing Performance Tray Reactive Distillation, the on of a and Chem. Eng. Sci., 54,2873. Hilborn, R. C., 1994, Chaos and Nonlinear Dynamics (An Introduction to Scientists adn Engineers), Oxford University Press Inc., USA. Holland, C.D., 1981, Fundamentals Multicomponent Distillation, Mc Graw Hill, New York. Holt, B. R. and Morari M., 1985, Design of Resilient ProcessingPlants. 4.The Effects of Right Half-Plane Zeroes on Dynamic Resilience, and 5.The Effect Sci. Chem. Eng. 1229. 40,59 Resilience., Dynamic Dead time and on of Synthesis for Framework Modular Generalised the A S., 1998, Rahim, Ismail UniverThesis, Separation Processes, Reactive Separation Nonideal and of 1998. London, sity of Equilibria Vapour-Liquid K., Lee, W. Y., 1992, Y. Lee, W., Y. with Kang, and EthylAcetic-Acid, Containing Systems EquilibriumReaction. Chemical Chem. Japan, Eng. 25,649. J. Ethyl-acetate, Water and alcohol, Simultaneous Heat A., D. O. C., Mellichamp, 1977, Sandall, F., and Kayhand, 171 Mass Transfer in Binary Distillation-II and 32,747. Experimental, Chem. Eng. Sci. Kenig, E., and Gorak, A., 1995, A Film Model-Based Approach for Simulation of Multicomponent Reactive Separation, Chem. Eng. Prog., 34,97. Kenig, E., Schneider, R., and Gorak, A., 1999, Rigorous dynamic modelling of complex reactive absorption processes,Chem. Eng. Sci., 54,1347. King, J. C., 1980, Separation Processes,McGraw-Hill, New York. Kister, H. Z., 1992, Distillation Design, McGraw-Hill, New York. Kister, H. Z., 1990, Distillation Operations, McGraw-Hill, New York. Kooijman, H. A., and Taylor, R., 1995, Nonequilibrium Model for Dynamic Simulation of Tray Distillation Columns, AIChE J., 41,1852. Kooijman, H. A., and Taylor, R., 1996, ChemSep Technical Manual, v3.01. Komatsu, H., 1977, Application of the Relaxation Method for Solving Reactive Distillation Problems, J. Chem. Eng. Japan, 10,200. Komatsu, H., and Holland, C.D., 1977, A New Method of Convergencefor Solving Reacting Distillation Problems, J. Chem. Eng. Japan, 10,292. Kreul, L. U., Gorak, A., Dittrich, C., and Barton, P.I., 1998, Dynamic CatComp. Validation, Simulation Experimental Advanced Distillation: and alytic Chem. Eng., 22, S371. Kreul, L. U., Gorak, A., and Barton, P.I., 1999, Modeling of Homogeneous Sci., Eng. Chem. 54,19. Columns, in Packed Processes Separation Reactive Film Model InA Standart, G. L., Multicomponent 1976, R., Krishna, and Maxwell-Stefan Solution General Matrix Method the to of a corporating J., 227383. AIChE Equations, 172 Krishnarnurthy, R., and Taylor, R., 1982, Calculation of Multicomponent Mass Transfer at High Transfer Rates, Chem. Eng. J., 25,47. Krishnamurthy, R., and Taylor, R., 1985, A Non-Equilibrium Stage Model Multicomponent Separation Processes, Part I 11, AIChE J, 31(3), 449. of and Kubi'c'ek, M., 1976, Algorithm 502. Dependence of Solutions of Non-linear Systems on Parameter, ACM Trans. Math. Software, 2,98. V i Kub'Cek, M., and Marek, M., 1983, Computational Methods in Bifurcation Theory and Dissipative Structures, Springer-Verlag, New York. Luyben, W. L., 1974, Process Modelling, Simulation, and Control for Chemical Engineers, McGraw-Hill, New York. Mess, A. I., 1981, Dynamics of Feedback Systems, Wiley, New York EquiChemical Calculation in Concepts Basic the L., M. 1995, Michelsen, of Systems, Separation Reactive Conference Nordic librium, presented at on Technical University of Denmark, Denmark. DeOptimal N., E. 1996, Pistikopoulos, D., J. Mohideen, M. J., Perkins, and J., AIChE 42,2251. Uncertainty, Systems Dynamics under sign of IEEE Control, Integral Systems Stability Morari, M., 1985, Robust with of Transactions on Automatic Control, 30,574. ReSeparation Stage with Nelson, P.A., 1971, Countercurrent Equilibrium J., 17,1043. AIChE action, Coefficients Transfer Mass Y., 1968, Okomoto, Onda, K., Takeuchi, H., and Japan, Eng. Chem. J. Columns, Packed in Phases Liquid Gas Between and 1,56. Denmark. DTU, Report, CAPEC Internal 1., 1999, Papaeconomou, 173 Pelkonen, S., Gorak A., Kooijman, H., and Taylor, R., 1997, Operation of a Packed Distillation Column: Modelling and Experiments, IChemE. Distillation and Absorption 97,269. Perkins, J. D., and Wong, M. P.F., 1985, AssessingControllability Chemicalof Plants, Chem. Eng. Res. Des., 63,358. Perez-Cisneros, E. S., Schenk, M., and Gani, R., 1997a, A New Approach to Reactive Distillation Modelling and Design,IChemE Symposium Series, 142, 715. Perez-Cisneros, E. S., Gani, R., and Michelsen, M. L., 1997b, Reactive SepaSystems. Part Computation 1: Physical Chemical Equilibrium, ration of and Chern. Eng. Sci., 52,527. Perez-Cisneros, E. S., Schenk, M., Gani, R., and Pilavachi, P.A., 1996, AsSimulation, Design Analysis Reactive Distillation Operations, pects of and of Comp. Chem. Eng., 20, S267. Petersen, C. K., 1989, PATH User's Guide. Department of Applied MatheStudies Centre for Studies, Nonlinear University Leeds, U. K. matical and of Pilavachi, P.A., Schenk, M., and Gani, R., 1999, Aspects of Chemical and Physical Equilibrium and Energy Requirements in Distillation, IChemE, submitted. Pilavachi, P.A., Schenk, M., P e'rez-Cisneros, E. S., and Gani, R., 1997, ModResearch, Operations, I&CE Distillation Reactive Simulation of elling and 36,3188. Powers, M. F., Vickery, D. J., Arehole, A., and Taylor, R., 1988, A Nonequilibrium Stage Model of Multicomponent Separation Processes-V.ComputaChem. Eng., Comp. for Solving Equations, 12, Model Methods the tional 1229. Gases B., The Properties M., Poling, J. 1987, Prausnitz, C., R. Reid, of and 174 Liquids, McGraw-Hill, New York-London. and Rehfinger, A., and Hoffmann, U., 1990, Kinetics Methyl Tertiary Butyl of Ether Liquid Phase Synthesis Catalyzed by Ion Exchange Resin-1. Intrinsic Rate Expression in Liquid Phase Activities, Chem. Eng. Sci., 45,1605. Rosenbrock,H. H., 1970, State-space and Multivariable Theory, Nelson, London. Rovaglio, M., and Doherty, M. F., 1990, Dynamics of Heterogeneous Azeotropic Distillation Columns, AIChE J., 36,39. Ruiz, C. A., Basualdo, M. S., and Scenna,N. J., Reactive Distillation Dynamic Simulator, Trans IChemE, 70,363. Ruiz, C.A., Cameron, I. T., and Gani, R., 1988, A Generalized Dynamic Model for Distillation Columns-111.Study of Startup Operations, Comp. Chem. Eng., 12,1. Saito, S., Michishita, T., and Maeda, S., 1971, Separation of Meta- and Para- Xylene Mixture by Distillation Accompanied by Chemical Reactions, J. Chem. Eng. Japan, 4,37. Sawistowski, H., and Pilavakis, P.A., 1979, Distillation with Chemical Reaction in a Packed Column, IChemE Syrnposium Series, 56. Sawistowski, H., and Pilavakis, P.A., 1982, Vapor-Li qui d-Equilibrium with Association in Both Phases - Multicomponent Systems Containing AceticAcid, J. Chem. Eng. Data, 27,64. Scenna, N. J., Ruiz, C.A., and Benz, S., 1998, Dynamic simulation of startChern. S719. Cornp. Eng., distillation 22, columns, up procedures of reactive Schrans, S., de Wolf, S., and Baur, R., 1996, Dynamic Simulation of Reactive Distillation: An MTBE Case Study, Com. Chem. Eng., 20, S1619. 175 Scheiber, I., and Marek, M., 1995, Chaotic Behaviour Deterministic Disof Systems, Cambridge University Press, Cambridge. sipative Seader, J. D., 1989, The Rate-Based Approach for Modelling Staged Separations, Chem. Eng. Prog., 85,41. Sennewald, K., Gehrmann, K., and Schafer, S., 1971, Column for Carrying Out Organic Chemical Reactions in Contact with Fine Particulate Catalysts, US Patent 3,579,309. Sivasubrarnanian, M. S., and Boston, J. F., 1990,The Heat and Mass Transfer Rate-Based Approach for Modelling Multicomponent Separation Processes, Com. Appl. Chem. Eng., 331. Skogestad, S., 1997, Dynamics and Control of Distillation Columns- A Critical Survey, Modelling, Identification and Control, 18,177. Skogestad, S., Lundstrom, P., and Jacobsen, E. W., 1990, Selecting the Best Distillation Control Configuration, AIChE J., 36,753. Skogestad, S., and Morari, M., 1987, Control Configuration Selection for Disti Ration-Columns, AIChE J., 10,1620. Skogestad, S., and Postlethwaite, 1., 1996, Multivariable Feed-Back Control, John Wiley & Sons Ltd., England. Sneesby,M. G., Tade, M. O., Datta, R., and Smith, T. N., 1997, ETBE Synthesis via Reactive Distillation. 2. Dynamic Simulation and Control Aspects, Ind. Eng. Chem. Res., 36,1870. Solari, H. G., Natiello, M. A., and Mundlin, G.B., 1996, Nonlinear Dynamics: from Physics Math, Trip to Two-Way a Sorensen,E., 1994, Studies on Optimal Operation and Control of Batch Distillation Columns, Phl). Thesis, University of Throndheim, Norway. Sundmacher, K., 1995,Reaktivdestillation mit katalytischen Fuellkoerperkack176 ungen ein neuer Process zur Herstellung der Kraft stffkomponente MTBE, PhD Thesis, CUTEC Institute, Clausthal-Zellerfeld, Germany. Sundmacher, K., and Hoffmann, U., 1996, Development New Catalytic of a Distillation Process for Fuel Ethers via a Detailed Nonequilibrium Model, Chem. Eng. Sci., 51,2359. Suzuki, I., Komatsu, H., and Hirata, M., 1970, Formulation and Prediction of Quaternary Vapor-Liquid Equilibria Accompanied by a Chemical Reaction, J. Chem. Eng. Japan, 3,152. Suzuki, I., Yagi, H., Komatsu, H., and Hirata, M., 1971, Calculation Mulof ticomponent Distillation Accompanied by a Chemical Reaction, J. Chem. Eng. Japan, 4,26. Stewart, W. E., and Prober, R., 1964, Matrix Calculation of Multicomponent Mass Transfer in Isothermal Systems, Ind. Eng. Chem. Fundam., 3,224. Taylor, R., Kooijman, H. A., and Hung, J. S., 1994, A Second Generation Nonequilibrium Model for Computer- Simulation of Multicomponent Separation Processes, Comp. Chem. Eng., 18,205. Taylor, R., Kooijman, H. A., and Woodman, M. R., 1992, Industrial Applications of a Nonequilibrium Model of Distillation and Absorption Operations, IChemE Symposium Series, A415. Taylor, R., and Krishna, R., 1993, Multicomponent Mass Transfer, Wiley, New York. Taylor, R., and Lucia, A., 1995, Modelling and Analysis of Multicomponent Separation Processes, AIChE Symposium Series, 91,19. Thompson, J. M. T., and Stewart, H. B., 1986,Nonlinear Dynamics and Chaos: Geometrical Methods for Engineers and Scientists, Wiley, New York. Toor, H. L., 1964, Solution of the Linearized Equations of Multicomponent Mass Transfer, AIChE J., 10,448.4 177 Ung, S., and Doherty, M. F., 1995a,Vapour-Liquid PhaseEquilibrium in Systems with Multiple Chemical Reactions, Chem. Eng. Sci., 50,23. Ung, S., and Doherty, M. F., 1995b, Vapour-Liquid Phase Equilibrium in Systems with Multiple Chemical Reactions, Ind. Eng. Chem. Res., 34,2555. Ung, S., and Doherty, M. F., 1995c, Necessaryand Sufficient Conditions for Reactive Azeotropes in Multi-Reactions Systems, AIChE J., 41,2383. Venkatararnan, S., Chan, W. K., and Boston, J. F., 1990,Reactive Distillation Using ASPEN PLUS, Chem. Eng. Prog., 86(8), 45. Vogelphol, A., 1979, Murphree Efficiencies in Multicomponent Symposium Series 2(56), 25. Wesseling, J. A., 1997, Non-equilibrium Res. Des., 75,519. Systems, IChemE Modelling of Distillation, Chem. Eng. Wesseling, J. A., and Krishna, R., 1990, Mass Transfer, Ellis Horwood, ChichEngland. ester, Whitman, W. G., 1923, A Preliminary Experimental Confirmation of the Two-Film Theory of Gas Absorption, Chem. Metall. Eng., 29,146. in Coefficients Diffusion Correlation Chang, P., 1955, C. R., Wilke, of and Dilute Solutions, AIChE J., 1 264. A ProceW., K. 1992, Mathisen, M., S., Hovd, Skogestad, Wolff, E. A., and Ulf. London, Workshop, IFAC Analysis, Controllability for dure Processes. Distillation Catalytic Study X., Xu, 1992, Zheng, Y., and on QuasiHomogeneous Processes Distillation Catalytic and 2. Simulation of A, Part 465. E 70, IChem. Trans. Model, Rate-based 178 Appendix A Properties A. 1 Ent halpy The component liquid enthalpy is calculated from a polynomial expression for the heat capacity taken from the DIPPR data bank (Daubert and Dan1986). ner, hý = T JT.IL (A, p + B, pT 2+ + CpT DpT 3+E, 4 )dT pT Mixture liquid enthalpy is calculated by means of the following equation, NC (A. 2) hL) xi(hLir - HL hL is a reference enthalpy. where r The vapour enthalpy is calculated through the liquid enthalpy and the heat (Daubert heat bank data For DIPPR the the of vaporisation of vaporisation. follows, The is is 1986) Danner, expression as again used. and vap H, Aal.I'(1 -T r) (B, &H+C, &HTr+Da. '2+EAHTr) HTr 3 Tr =T TC 179 (A-3) (A. 4) NC AH vap XiHvap (A. 5) i Mixture vapour enthalpy is calculated by, NC L- Hv A. 2 AH yi(h i vap (A. 6) Density Component density of the liquid phase is calculated with the DIPPR data bank (Daubert and Danner, 1986). Ap PLi (A. 7) P) ('+('- CTP), Bp Mixture density is calculated through, PLp - ENC i=l (A-8) PLi, pXi, p Density of the vapour phase is calculated from the equation of state (or ideal law). gas A. 3 Viscosity Viscosity for liquid and vapour phase are calculated through a suitable cor(1986), by Daubert Danner relation given and Bj,,, AjjvT v P' ':ý (I + CA,IT + Div/T2) Ejj) C,,, B,, In(T) DAIT + + + exp(A. 1 IIT (A. 9) (A. 10) Mixture viscosity is calculated through, NC YL = exp[E i=l In pixi, + 1/2 p Gij] xixj (A. 11) p O, ENC -tyi, Y, j (A. 12) i:Aj NC YV =E- i=l 180 j=l where oij [8(1+ A. 4 (A. 13) MlMj)]112 Surface Tension Surface tension for pure components is calculated with DIPPR data bank (Daubert and Danner, 1986): (1 A, * eri = Tr)Bý, + CGT (A. 14) For mixtures an approximate estimate is used: NC X, Orr with r= to r=1 -3 A. 5 Thermal (A. 15) is usualy used. Conductivity data DIPPR is for Thermal conductivity calculated with pure components bank (Daubert and Danner, 1986): Bx Ai = Mixture thermal conductivity A,xT Q\IT + D,\IT2 1 (A. 16) is calculated through the following expression, NC NC E pipjAij (A. 17) Aij = 2(Ai + Ai xivi Pi .ENC v j=l 3 (A. 18) XL X: NC Ay Y, 2 i=l NC j: j=1 yjoi3 for interaction is the Oij gas viscosity. mixure parameter where 181 (A. 19) (A. 20) A. 6 Binary Diffusion Coefficients The relationship between the Fick's diffusion coefficient and the MaxwellStefan diffusion coefficients is given by: Df = DIP Df is the Fick's coefficient, D is the Maxwell-Stefan coefficient where r and is a factor accounting for thermodynamic non-idealities. For ideal systems both coefficients are identical (IF = 1). Fick's diffusion coefficient incorporates two aspects: a) significance inverse of drag and b) the thermodynamic non-idealities. Fick's coefficients are less easy to interpretate physically than the MaxwellStefan coefficients. (Taylor and Krishna, 1993). o Vapour phase Reid ct al. (1987) and Daubert and Danner (1986) recommend the due (1966,1969) Fuller to to calculate the use of a correlation et al. diffusion coefficients and this is the one used in this work. M2)/Mlm2l + Df = CT 1.75ý[(Ml P(, ýýV +, 12 ýýV)2 (A. 21) kelvin(K), T in P M, C=1.013e-02, M2in in Pa, with and g1mol and 2/, V2 diffusion D will be in (rn S) The terms V, and the are molecular . by in the atomic are calculated summing contribution volumes and components. Liquid phase 4o For the liquid phase, Wesselingh and Krishna (1990) suggested the following model for the limiting diffusivities: Dij = (D?.Dj'i .7 182 )1/2 (A. 22) Do due the the Wilke to are calculated coefficients with where method (1955): Chang and (02M2) D'12 = 7.4e - 08 1/2 /12VJO. 6 T (A. 23) (the diffusion Do is 1 the coefficient of species where solute) present in 21S; M2 infinitely low concentration in species2(the solvent), cm is the P2 is the T is K; temperature, the the solvent, g/mol; molar mass of V, is the the of molar volume at the normal viscosity solvent, cP; and 3/Mol. boiling point of the solute, cm 183 Appendix B Results B. 1 o Simulation Results for MTBE Changes in Feed Rate The transient response to a 10% decrease in methanol feed flow rate boil-up distillate held is in Fig. B. I. with vapour and constant shown Without a decrease in the vapour boil-up input, there was a decrease in the bottom flow rate which means low MTBE production. The composition of the bottom shifted to higher MTBE concentration (actually the global isobutene conversion did not considerably increased The in temperature the accordingly as column changed). the reboiler remained in phase equilibrium. After the disturbance the its original steady state. column recovers Increasing the methanol feed flow rate in 10% makes the system unstable, going to a low conversion solution without recovering from that. The transient is shown in Fig. B. 2. The control of the stoichiometric ratio is rather complicated, specially because the feed composition will not be accurately (it will probably depends on a process upstream). The unreacted methanol is recovbottoms MTBE in the the the the and amount column, of ered with isobutene in increases the the conversion. reactive zone of methanol 184 _E, .2 E E 0 CD fi Z A:! E 8 w 10 15 Time (h) Figure B. I: Decreasing 10% methanol feed flow-rate. 0.9 0.8 s 0 t 0.7 E E 0.6( 0 1 0.5 0.4 0 CL I 0.3 LU In 0.2 0.1 - Of %nq 25 Time (h) Figure B. 2: Increasing 10% methanol feed flow-rate. 185 A compromise must be determined between isobutene conversion (increasesas methanol increases)and ether purity (which falls as methanol increases). High MTBE purity can be produced with a lower methanol but be excess, some excessshould used to suppressside reactions. Changes Feed Composition in e The isobutenes feed composition is fixed by upstream plant operations between 15 and 55% of isobutene, depending in the and usually varies (which Increasing process upstream units and catalyst are employed). the concentration of isobutene could have some effects on the reactive distillation columns operations, such as: in increase, the reaction zone with a'favourable' - reactant concentration in the reaction equilibrium, effect temperature temsection gradient reactive section and reactive of increase the reaction equilibrium, and perature with effect on duty be decreased to the maintain product specific reboiler must specifications or conditions. Fig. B. 3 shows the transient responseto a 10% of isobutene feed comby increased isobutene The increased. the global conversion position in MTBE for driving force the the purity well as as reaction additional bottoms. The bottom temperature follows the composition changes. When the isobutene composition decreasesis interesting to notice that (15% A low is the system going to step change conversion solution. decreasing isobutene feed composition) makes the system go to very low MTBE production and it will not go back to the high conversion back is to the the amount. normal composition when again, solution This effect is shown in Fig. BA. is the is 10% then If the step change only a small change produced and (Fig. B. 5). time reactive column will recover after some the increased in the The temperature and exceeded reactive section limit for the catalyst temperature degradation. 186 1 0.995 -2,0.99 .2 0.985 E E 0.98 0 t: W .8 SO. . 975 0.97 80.965 w 9 0.96 0.95E 0.95 L 0 5 Time (h) 10 15 Figure B. 3: Increasing 10% isobutene feed composition. 0.9 0.E T 0.-0 0.E 0., - n 10 5 li Time (h) feed isobutene Decreasing 15% BA: Figure composition. 187 0.965 0.96 2 ri 20.955 0 i5 E 0.95 E 0.94E 0.94 0.9319 E 8 0.9. w c12 0.921 0.91 0.91! 10 15 Time (h) Figure B. 5: Decreasing 10% isobutene feed composition. Decreasing the methanol feed flow rate will help to maintain the desired conditions. Stages Non-Reactive Reactive Effects and of 9 Increasing or decreasing the number of reactive stages from the 'opti between bad interaction the phase equilibrium and muml produced a the chemical reaction which leads to high decomposition of product on the lower reactive stages. The rectification zone of a reactive distillation column should remove from the in (light this inert reactive the case) component, n-butene, (isobutene) the to secreactive reactant unreacted recycle section and tion. from MTBE the the the reactive The stripping section should remove the as product to purify and conditions good reaction section maintain loss the the product. with of reactants prevent as well 188 Even though the separation objectives are clear, increasing in rectification or stripping stages is not necessarily good for the reaction section conditions. If the rectification zone is increased too much will result in loss of isobutene in the distillate, and if the stripping zone is increased too from be the reaction section. moved away much methanol will B. 1.1 Controllability tion for the MTBE Analysis produc- Three cases are presented: Casel The input variables (u) are given by, ID1 u= (B. 1) vj The disturbances (d) are considered as follows, F bute ne, Xiso TF (B. 2) (y) are, while the measurementsvariables [XMTBE, XMTBE, D' (B. 3) B. by, is (after G The transfer function scaling) given 2.8325 0.1254 0.0388 0.0288 (B. 4) from different is detG(s) The functional is controllable. The reactive column zero. 189 Pole zero map Xle 0 3 0 x oxx 2 x 0 x cý x 1 XX (0 -9 ..................... 1137 0( lQ, ........ X ........... 0.. -- AX : 0: -1 x c; x 0 x -2 x ox x 0 -3 0 L -4 -3 -2 01234 Real Axis x1o Figure B. 6: Poles and Zeroes for Case I (MTBE) Computation e of poles and zeroes All poles are in the left half plane (system stable). Existence of RHPfrom be Fig. B. As 6. can seen zeroes. o RGA The steady-state RGA matrix is given by, [ 1.06340 -0-0634 RGA = 1.06340 -0.0634 190 (B. 5) 5 Frequency response of RGA element 1 10 c ................:...................... 10-11 10-8 I**II*I.................................................. 10 -4 10 10-2 10 0 2 10 10 4 Frequency (rad/h) 25 20 15 10 ............... : ................ ................................ ................ ................ ................ I ................................. i........... .................... 5 10-8 10-6 10 -4 10-2 10 0 10 2 10 Frequency (rad/h) Figure B. 7: RGA element (1, I) as a function of the frequency for Case 1 (MTBE) The Ij element of the matrix can be plotted as a function of the frequency, as can be seenin Fig. B. 7. The values of RGA are closeto I high high frequencies frequency, takes very some smaller only at at any distillate is The and composition a good choice. values. pairing of Singular value o The singular values are : 0.0223 (B. 6) 2.8356 (B. 7) The singular values at every frequency are given in Fig. B. 8 191 4 Singular Values -160 -180 -200 (0 -220 Co :3 -240 CO -260 -280 " -300 1 10 le 10, le 10 5 Frequency(rad/hr) Figure B. 8: Singular values for Case I(MTBE) 192 le 10 7 Step Response From: U(l) From: U(2) 3 2.5 2 1.5 1 0.5 'a 3 0 X 10-13 0 5 4 0 0.85 1.7 2.55 3.4 0 0.85 1.7 2.55 X 10-3 3.4 x 10, Time (h.) Figure B. 9: Step response for Case 1 (MTBE). (10% step size) Analysis The condition number is large (,- 127) but not too large. The RGA values high is it the not really should not condition number are small, and since be a problem to control this column. The pairing of ditillate with the prodbottom bottom the the the the of with compostion at and uct composition logical. The best, is is the only negative effect are which physically column the RHP-zeroes which implies inverse response,as can be seenfrom Fig. B. 9. 193 Case2 The input variables (u) are given by, DI v (B-8) The disturbances (d) are considered as follows, d Xisobutene,F (B. 9) TF (y) the while measurements variables are, [ XMTBE, D TB ] The transfer function G is given by, ] 0.1223 0.00410 3.88e - 05 2.88e - 05 (B. 11) The reactive column is functional controllable. The detG(s) is different from zero. Computation of poles and zeroes 9 All poles are in the left half plane (system stable). Existence of RHPfrom Fig. B. be 10. As seen can zeroes. e RGA The steady-state RGA matrix is given by, RGA = 1.0475 -0.0475 -0.0475 1.0475 (B. 12) function be the the The 1,1 element of of a as plotted matrix can RGA tends to The Fig. B. be in 11. frequency, as can values of seen is frequencies, high that the a good 1 at pairing ul-yl which confirms choice. 194 Pole zero map x 10, 4,1 11111 0 3 0 x oxx 2 x X xP 1 X6X (0 Co .................... ...... 0 x -1 x x x oxx -2 x 0 -3 0 L -4 -3 -2 4 0 Real Axis (MTBE) Case for 2 Zeroes Poles B. Figure 10: and 195 X 105 Frequency response of RGA element 1 10 10 1 -1 lo 10 10-4 10-6 10-2 10 0 10 2 10 4 Frequency (rad/hr) 30 25 20 cn ............... L.................. 15 0 Co CL *................ .................. ........................................ ...... .................................. .................................. I- ............................................................................. 10 5 n %1 10 -8 10 -6 10 -4 10 -2 10 z 10 10 Frequency (rad/hr) Case for 2 frequency function (1, the I) of Figure B. 11: RGA element of as a MTBE 196 o Singular value The singular values are : 0.00387 or = '57= 0.122 (B. 13) (B. 14) Analysis The condition number is large 31.5), but not too large. The RGA values A indicating interactions that the are small, negative effect are not strong. be from RHP-zeroes. The Fig. B. It inverse 12. the are response can seen be in should not a problem controlling this column. Case3 The input variables (u) are given by, [D] v (B. 15) The disturbances (d) are considered as follows, x',, buten, F TF (B. 16) (y) are, while the measurements variables [TD] TB (B. 17) The transfer function G is given by, 0.12230 0.00410 0.00060 -0.0077 (B. 18) from different is detG(s) The functional is The reactive column controllable. zero. 197 Step Response 1.4 x 10-9 From: U(l) From: U(2) 1.2 1 0.8 0.6 0.4 a) CL 0.2 n X 10-13 6 54 2 0 02 60 26 X 10-3 Xle Time (hr. ) Figure B-12: Step response for Case 2 (MTBE). 198 (10% step size) x10 1.5 r- 5 Pole zero map 0 1 0.5 0 OXXx xx cb CO --0- 0 .......... .................................. - -cx- x 0x xgiL:10 - ............................. - oxx 0x -0.5 -1 0 -1.5' -3 11 -2.5 -2 -1.5 -1 -0.5 Real Axis 0 0.5 1 1.5 x lor) Figure B. 13: Poles and Zeroes for Case 3 (MTBE) Computation of poles and zeroes All poles are in the left half plane (system stable). Existence of RHPbe from B. As Fig. 13. can seen zeroes. RGA The steady-state RGA matrix is given by, RGA 0.71 0.29 0.29 0.71 199 (B. 19) Frequency response of RGA element 1 10 10 10-11 10-8 I O-b 10 -4 10-2 10 0 10 2 10 4 Frequency (rad/hr) 0 ..................................................... -50 C» 0-100 ............... ................ ................................ I ................................. cu m -150 CL ................................ I .......................................................... I I ....................... I -200 -250'lo- 10-6 10-4 10-2 0 10 2 10 4 10 Frequency (rad/hr) Figure B. 14: RGA element (1, I) as a function of the frequency for Case 3 (MTBE) The Ij element of the matrix can be plotted as a function of the frequency, as can be seen in Fig. B. 14. The values of RGA tends to I at high frequencies, which confirms that the pairing ul-yl is a good choice. o Singular value The singular values are : or 0.000316 (B. 20) '57= 0.32 (B. 21) Singular values for every frequency are given in Fig. B. 15. 200 Singular Values -150 -200 -250 cts co -300 -350 L-400 1 10 2 10 10 3 10 4 10 5 10 6 Frequency (rad/hr) Figure B. 15: Singular values for Case 3 (MTBE) Analysis The condition number is large (?-.o 1012-93). The RGA values are different form unity, indicating that the interactions are strong. Another negative effect are the RHP-zeroes. The inverse responsecan be seenfrom Fig. B. 16. 201 Step Response 1.4 From: U(2) From: U(l) x 10-9 1.2 0.8 0.6 0.40.2 0 E 10' 0 -1 -2 CSJ3 >. o -4 I-5 -6 -7 -8 0 2 60 4 26 X 10-3 x 10, Time (hr.) Figure B. 16: Step response for Case 3 (MTBE). 202 (10% step size) B-2 Acetic Acid Ethanol of with Esterification For the case of the esterification we have assumed the 0.00387D-V configuration. Casel The input variables (u) are given by, [D] (B. 22) v The disturbances (d) are considered as follows, d Xet han ol, F (B. 23) TF (y) the are, measurements variables while [TD] (B. 24) TB The transfer function G is given by, 3.282 -7.387 -0.0015 4.53e - 04 1 (B. 25) from different is detG(s) The functional The reactive column is controllable. zero. Computation * of poles and zeroes RHPExistence (system half left in All poles are the of stable). plane B. from Fig. 17. be As seen can zeroes. o RGA by, is RGA The steady-state given matrix [ 1.1726 -0.1726 RGA = 1.1726 -0.17260 203 (B. 26) Pole zero map 80 0 60 x x 40 xx0 0 x 20 00 x x 0 U) CD E cu .. 0 - -OX. OX ---... 0xXxx<)-Ox. x- ä-- 0 0)0 - 0. x. 0x0 ... X-OK-><)... 0 >o XXx 0 .- 8-x 09--)ICD... X- (lK- 00( X X- - 19 4330 00x0 x 0 -20 0x 0 xx 0 x -40 x -60 on _vv -180 0 -160 -140 -120 -100 -80 -60 -40 -20 Real Axis Figure B. 17: Poles and Zeroesfor Case 1 (Esterification) 204 0 20 Frequency response of RGA element 1 10 ...................................................................................................... i ol 10 10 -6 10 10 -4 10-2 10 0 10 2 10 4 Frequency (rad/h) 0 -2 'a -8 -101-8 10 10 -6 10 -4 10-2 10 0 10 2 10 Frequency (rad/h) Figure B. 18: RGA element (1,1) as a function of the frequency for Case 1 (Esterification) The 1,1 element of the matrix can be plotted as a function of the frequency, as can be seen in Fig. B. 18. The values of RGA tends to 1 at high frequencies, which confirms that the pairing ul-yl is a good choice. decomposition Singular value * The matrix G is decomposedinto it singular value giving, G=UE VH (B. 27) for our case, U 0.4064 -0.9137 205 0.9137 0.4064 (B. 28) 4 1 8.08 0 0 1.06e 03 vH 0.9900 05 -0.417e 05 -0-417e -0.9900 (B. 29) (B. 30) Where the matrix E gives the singular values in the main diagonal. Analysis Since there are elements in G(s) smaller than I in magnitude this suggests that may be control problems with input constraints. The 1,1 element of the gain matrix G is 3.282. Thus, increase in ul (distilan late) yields a large steady-state change in (temperature distillate), that yj of is the outputs are very sensitive to changes in ul. In increase in contrary an (vapour boil-up) U2 gives Y2 = -0-0015, a small change and in the opposite direction of that for the increase in ul (the opposite analysis to this is for U2)- If both ul andU2 (distillate and vapour-boil-up) are increasedsimultaneously, the overall steady-state change in yj is (3.282 - 0.0015) = 3.2805, which indicates that the temperature is strongly dependent on internal changes of flows. This can also be seenfrom the smallest singular value: 0.00106. From the input singular vector, 0.9137 0.4064 (B-31) is in direction, that the the to the we see effect move outputs same which lead keep both to to temperature at the will a relative short control action desired values simultaneously. due 7620), The condition number is rather large (e-%. to the very small miniO Although high RGA the the values are relatively small, mum singular value. RHPimplies the the condition number control problems as well as value of (RHP-zeros is to the to the also very close close origin are origin which zero, the most difficult to overcome). The RHP-zero also implies inverse response, from Fig. B. be 19. seen as can 206 Step Response From: U(2) From: U(l) x >- 0 10,1-11 0 0 x 10-8 -2 -4 A In 0 0.5 1 20 1.5 0.5 1 1.5 Time (h) Figure B. 19: Step response for Case 1 (Esterification). 207 (10% step size) 2 Case2 The input variables (u) are given by, [D] (B-32) v The disturbances (d) are considered as follows, dxth (B. 33) ano IIF TF while the measurements variables (y) are, [ X4D (B. 34) XIBI For this case the transfer function G is given by, 0.2440 -0.042 Computation e 03 -1.152e 1.033e - 05 (B. 35) of poles and zeros All poles are in the left half plane (system stable). Existence of RHPPoles zeros. and zeroes map is shown in Fig. B. 20 e RGA The RGA matrix is given by, RGA = [ 1.055 -0.055 -0.055 1.055 (B. 36) The 1,1 element of the matrix can be plotted as a function of the frequency, as can be seen in Fig. B. 18. The values of RGA tends to 1 at high frequencies, which confirms that the pairing ul-yl is a good choice. 208 Pole zero map 60 0 x 40 0 x XX 00 20 x 00 Co xx 0 0. ............................................. QX x mx. 0 000xx 0 (aw. 0 xc) x -20 00 XX x 0 -40 x -P. A -450 -400 -350 -300 -250 -200 Real Axis -150 -100 -50 Figure B. 20: Poles and zeroesfor Case 2 (Esterification) 209 0 50 Frequency response of RGA element 1 10 ............................... ........ I. ................. ............................... .................................. ................. ................. ... .......... ................. ..................... .................................................................... ...................... ................ .................................. .................................. ................. ................................... .................................................. ................ ................................. ................ ................................................. ................. 10 ............................................... ................... .................................. .................................................... ................................................... ................... ................................. ..................... ......................................... .............................................. .................................. ............... .................................................................................................. ................................................................................... 10 -11 10 10-6 ............... 10 -4 0 10 10-2 2 10 4 10 Frequency (rad/hr) ................................. cm 0 10 -1 ............................ ..................... --8 10 -6 10 -4 .............. - ........................... .................... ............. 10 fN............ *.......... 10-2 ................... 10 024 10 10 Frequency (rad/hr) Figure B. 21: RGA element (1,1) as a function of the frequency for Case 2 (Esterification) 210 Singular decomposition * value The matrix G is decomposedinto it singular value giving, VH (B-37) for our case, 0.9855 -0-1693 VH = -0.1693 -0.9855 (B. 38) 0.244 0 0 1.875e- 04 (B 39) . 0.47le 03 -0.999 03 0.9900 -0.47le - (B. 40) Where the matrix E gives the singular values in the main diagonal. Singular values aa function of the frequency are given in Fig. B. 22. Analysis Since there are elements in G(s) smaller than 1 in magnitude this suggests that may be control problems with input constraints. The 1,1 element of the gain matrix G is 0.2440. Thus, an increase in ul (distillate) yields a small steady-state change in yj (composition of distillate), that is the outputs are not sensitive to changes in ul. Similarly an increase in (vapour in boil-up) the a small change and gives opposite Y2 U2 = -0.0012, direction of that for the increase in ul (the same analysis to this keeps for U2)- If both ul andU2 (distillate and vapour-boil-up) are increasedsimultaneously, (0.244 0.2428, is 0.0012) in the overall steady-state change which yj = indicates that compositions are weakly dependent on internal changes of flows. From the input singular vector, 0.9855 (B. 41) -0.1693 direction. This in is the the to the that opposite effect outputs move we see desired both large keep the to lead to at composition control a action will 211 SingularValues -160 -180 -200 c12 -220 Co Z C» r_ CO -240 -260 -280 -300' 10 10-1 lo, 10 1 Frequency(rad/hr) Figure B. 22: Singular values for Case 2 (Esterification) 212 10 3 10 4 Step Response ;3 10-9 x From: U(l) Fran: U(2) ............ 2.5 2 1.5 1 0.5 V 0 4-. CL 10-10 x -5 -10 -15 -qn 0 .................... 2.5 5 7.5 100 2.5 5 7.5 10 Time (hr) Figure B-23: Step responses for Case 2 (Esterification). (10% step size) values simultaneously. The condition number is rather large (ý,- 1300), due to the very small minihigh RGA Although the the values are relatively small, mum singular value. RHPimplies the the control problems as well as condition number value of inverse be The is to the response can seen origin. also very close zero, which from Fig. B. 23. Case3 The input variables (u) are given by, [D] U= V 213 (B. 42) The disturbances (d) follows, are considered as XlF (B. 43) X4F. while the measurements variables (y) are, X4D] (B. 44) XlB For this case the transfer function G is given by, 1.4e - 02 ] 05 -2.58c - 0.1091 -0-0738 Computation 9 (B. 45) of poles and zeros All poles and zeros are in the left half plane (system Fig. B. 24. stable). o RGA The RGA matrix is given by, RGA = [ 1.0274 -0.0274 -0.0274 1.0274 (B. 46) RGa as a function of the frequency is given in fig. B. 25. Singular 4o value The singular values as a function of the frequency are given in Fig. B. 26. Analysis The condition number is not too large (r--t 165), and the RGA values are disappear if in diagonal The RHP-zero to the the unity main very close disturbance on the feed temperature is not present. Step changesare shown in Fig. B. 27. 214 Pole zero map 80 0 60 0 x 40 x xx 00 20 x x00 0x ci) 00 09 0 0 @DOK-X- CK 4e 0 (D ............................................ a) 0e9 E 0,0 0 .......... 0 x 00 x0 x -20 CýK 00 xx x -40 x 0 -60 an -%. TV -500 -400 -300 -200 Real Axis -100 0 Figure B. 24: Poles and zeros for Case 3 (Esterification) 215 100 Frequency response of RGA element 1 10 10 10-11 10-8 10-6 10 -4 10-2 10 0 10 2 10 4 Frequency (rad/hr) 1 0 ... ....................... ............... ................................................ ......... \ ............. .................................. ............ ................................. «a -2 ................ *.......... 0-3 m CL ................ -4 -5 -%j -Q 10 -8 ............... ............................. 10 -6 *, ****,, ...... .............................. 10 -4 - *, -xf............................ ****, 10 -2 10 ............................. 10 id le Frequency (rad/hr) for Case 3 frequency function (1, the I) Figure B. 25: RGA element of as a (Esterification) 216 Singular Values -160 -180 -200 (D -220 0 CO -240 U) -260 -280 ' -300 10-1 id, 101 lcý 10 Frequency(rad/hr) Figure B. 26: Singular values for Case 3 (Esterification) 217 le Step Response 6 From: U(l) x From: U(2) a) 4. - CL 0 x 10-10 -2-3-4 -5 -- ................ --- 1111 -6 0 0.5 20 1.5 0.5 1.5 Time (hr) (Esterification). Case for 3 Step Figure B. 27: responses 218 (10% step size) 2 Appendix C Optimization CA Objective Function Disturbances and Objective Function The objective function is composed by capital and operating costs. The andetermined by its internal, investment is the the cost of column, cost nualised and reboiler and condenser. minimiseCost + = mznzmzse-(Cshell Ctray + Chx + (c. Cop) 1) Ctray7 Ch., and C,, are given by, p whereCshell7 Cshell M&S ý1 3( EA+1: ) F, (101.9Dt)(2.18 + T80 )p 2Apf) 0.802(Zp - Zp+, OC. 2) P/<p where, A is a constant equal to 101.9; 800; Swift to Marshall index is the M&S equal and B is a constant equal to 2.18; bottom to the top the the equal column of H,, is the sum of and spacing at irn.; 219 Fc is the material of construction factor equal to 1.05; A is the column diameter (variable); and, Apf is a dummy variable which is a function of tray spacing and is set to 1 for a tray spacing of 2 feet. M&S Ctray 4.7D' 280 )tp . 55 Flc E 2Zp (C. 3) where, M&S is the Marshall and Swift index equal to 800; F, I is the material of construction factor equal to 1. The column diameter, Dt, its easily calculated from the internal vapour flowbe for factors the a given materials of construction will constant rate, and system. Chx M&SI =1 3[ 280 101 (2.29 + F,, .3 65 ) (Ao . cr 65) . + Ao (C. 4) where, M&S is the Marshall and Swift index equal to 800; F,, is the material of construction factor for heat-exchangerequal to 1.30; A, and A, are the condenserand reboiler areas(variables),respectively. Cop QcCwater + QrCsteam hence, directly, is The cost of steam and cooling water computed andCsteam kgmol); dollars (2.8573E-4 per the cost of cooling water kg). dollars (3.63763E-3 per cost of steam (C-5) Cwater is is the Disturbances Sinusoidal in feed composition: (wt) 0.15 sin zAA ý zAA + ZEt ZEt+ 0.15sin(wt) 220 (tOti1fl IJNN (C.6) (C. 7)