tte Cee Don’t Gamble With Physical Properties For Simulations Finding good values for inadequate or missing physical property parameters is the key toa successful simulation. And this depends upon choosing the right estimation methods. Erie . Carlson, Aspen Technolgy, ne hhemical engineers use process simulation to perform a variety ‘of important work. This. work ranges from calculations of rmass- and energy balances of large flow- sheets to prediction of the performance of process alternatives that can save millions of dollars. An engineer very quickly can define a complex flowsheet and all the process conditions. Desktop computers now allow rating, sizing, optimization, and dynamic calculations that previously required large mainframe computers. In the past, these simulations were often built by a group of experts, including physical propery expert. Now, simulators such as ASPEN PLUS, ChemCAD IIL, HYSIM, PRO If, and SPEEDUP are easi er to use and more powerful than the standalone programs of the past. Today, a single engineer can set up the basic simu- lation specifications, including the physi cal properties, in very little time. Missing of inadequate physical prop. erties, however, can undermine the accu- racy of a model or even prevent you from performing the simulation. That some re- Quired information is missing is not an oversight in the simulator. After all, for ‘most compounds, physical property pa- rameters are not known for every thermo ynamie model or for all temperature or pressure ranges, Models have built-in as. sumptions and practical limits that should apply. In this article, we will provide practi- cal tips and techniques to help you accu rately describe the physical properties needed in a simulation. As an engineer, you always will have to make assump- tions in terms of physical properties, however. The goal of this article is to out line the appropriate assumptions and to provide techniques when properties are The five important tasks Successfully describing the physical properties to be used in a simulation in- volves five tasks: 1. selecting the appropriate physical property methods: 2. validating the physical properties: 3. describing nondatabank — compo- nents (chemical species or compound) and missing parameters 4 obtaining and using physica) prop- enty daa: and estimating any parameters. Tt can be argued that these tasks are ‘not sequential and, to some degree, they are concurrent, During simulation devel- ‘opment, however, you will need to visit each area to be confident chat your simu- lation is as accurate as possible — so that important decisions can be made based on the results of your simulations. missing property Selecting the appropriate physical property methods This essential first step will affect all subsequent tasks in developing accurate physical properties in your simulation Indeed, the choice of the physical proper ty models for a simulation can be one of the most important decisions for an engi- net. Several factors need 10 be consi ‘CHEMICAL ENGINEERING PROGRESS * OCTOBER 1995 © 35SUCCEEDING AT SIMULATION ered, and no single method can han. dle all systems. Table 1 lists some thermodynamic models available in siraulators, ‘The four factors that you should consider when choosing property methods are ‘the nature of the properties of interest, + the composition of the mixture ‘the pressure and temperature range: and + the availability of parameters, ‘To ease the selection of the right physical propertly methods, we sug: gest using the decision trees shown in Figures 3. These trees are based on the four factors for selecting property methods, and can be used when the chemical components and approxi= ‘mate temperature and pressure ranges are known, While these diagrams are simplifications, they do. show the basic steps of the decision-making Process, while the notes in the sidebar amplify some of the key points. The nature of the properties of in- terest. A question that you may ask yourself when starting a simulation is ‘Does the choice of physical property methods matter?” The answer is an emphatic YES. The choice can strongly affect the prediction of the simulation. You should be selecting a collection of methods that will best predict the properties or results of in- terest 10 you, Because many chemical process simulations include distillation, strip- ping, oF evaporation, one important potential consideration for the choice of physical propery models is vaporiliquid equilibrium (VLE). This is te area in which the most physical property work is focused in chemical engineering. Liguid/iquid equilibri- tum (LLE) also becomes éaportant in processes such as solvent extraction and extractive distillation, Another critical consideration is pure-component and mixture en- thalpy, Enthalpies and heat capacities are important for unit operations such a heat exchangers, condensers, dis- tillation columns, and reactors. PRM Cue eae au EUG au eau Equation-ol State Modols -AetvityCoetciom Models Benedict-Nebb-Fubin(BWR)-Lee-Staling lectayee NATL Hayder-0'Conne For Huggins Hydrogen fvrie equation of tte for NATL hexameration® Seatchard-Midebrand deat gas ow" vuniguac Leo Kaslr UX) LUNIFAC Lee-kesier Piockor Peng-Rabinson (PR) Pemutbed-Hard-Chsin Von Lar Wison Prosative SK Special Models Redleh-Kwong IRK) {PI sourwater method Redich-kwong-Soave (KS) Broun Kap KS or PR with Wong Sander mixing ue avo Sandor KS or PRwith mociied Huron Vida? mix- | Grayson Ste ingle | Kenteisendera Sancnex-Laconbe for polymers | steam Tatos “Not used forth guid phase > Sea Figure? Blctolte Sxetoye NATL \ — i Reghenxwang:Soave, | _ Gromer sutecor | ap Braunk-10 <4 Vacuum | Pseudocomponents Vee | | | Figure 1. The frst steps or selecting physical property methods. 36 + ccTOBER 195 * CHEMICAL ENGINEERING PROGRESS——__—— — ——_—— Navigating the decision trees Here are some pointers to help you navigate the decison tees that ap- pear as Figures 1-2 What ae psevdocompongnts? in many applications where ony non polarmelecules are present uch sin hydracarton processing and re fining the madre f 30 complex that instead of representing iby allthe known constituent, tis esse to group the constants by some wel propery such as boiing por, nhs way, amisture of hundreds of con teats ean be reduced to 300” fewer. The properties of these groped Consttuents, called peeudocomponens, ace represented by an average boing point specie gravity, and molecular weigh. 1 you do not use peewlocomponents, the consituens shuld be describes by a molecu Tar formula and are refered to as ral components, Why are lecroine matures diferent? Becrayte mires include components tht are charged moleculs lions) or that form sas. Some simulstors allow calculaton of electrolyte reaction quiron wath phase equioriam. This i 8 vary samara method and its usage covers Inany applications such a8 saustie scrubtiag, neuvazion, acd pro: duction, and sa eeeiptetion, The noridealty of elecahyte sohtions, ‘usualy containing water, an be observed in boing pont elevation, sal ing out of gates thats, edsing satsto the soluon to change the slut iy of gests. end sak preciptaton. The most common eectoye math os are the Piaer model, and the modied-NATL actity coefficient model of Chen and coworkers. Sone elctoytes, ke formic acid and ‘eet ait are wry weak and an elacrate metho is nat equed Which type af method shoud be chosen for motures eonaning polar eamponents bu no electrolytes? There ar two graups of matods — based on acti coetcins or equations of state Use actiy-coeh Seien-tased methods when pressures are ow 10 medium (pial ast ‘than 10 boro 180 peal and # no components are near eid point. Ae ‘wy coeticint modes also olen are used to accurately preict non ideal quia beviosueh a8 for VLE and for LLE. in cones, equation of-state methods excel in thee aby to represent gate and exapoate wth temperetre and pressure up to and above the mature cial point Now, however, methods relying on cubic equations of stat with predictive mixing rules efectvely combine the strengths af the two method, (See Table 2) Far higher pressures (and yomperatures) hese Special equation of state ace beer ax they were developed to applyto ‘wider rage of temperatures, Thasematods incorporate sett coet ficients in the ealeuition of companant interactions represents by ex cess Gibbs free energy. Mos ofthe latar use a UNIFAC-basedactvty oefciet madel as the detau, but you can use ay activity coefficient. [A simulation pressures les than 10 atm and were there are no near ertieal components, for the best resus use the Wisoa, NATL or LUNIQUAG binary paramters tat may be availble in bitin databanks, or fitting parameters to experimental data (# avaiele using activity cootficien models. These parameters may have Been determined at di ferent temperatures, pressures and compositions than you are smltig, ‘though, so you may not obtain the est possti accutcy.f neracton parameters ae rot avaiable, however you can use the UNIERC method When should UNIFAC be used7UNIFAC an otber UNIPAC- based ac- tivity coufiien models re predictive approaches that use strvcturl sfoupsm extmate componem interactions. From structural information bout organic components usualy avaliable in the built-in databank, LUNIFAC ig abl to predict the activity coetficionts asa funeton of com- positon and temperature. You ran make use of UNIFAC when you do not have experimental data or binary parameters 0” When an approximate value is aecopable Vor instance, fr a component wth ow Brit In recent yeors, there have Been improvements to UNIFAC (ee Table 3) ‘hatcan beter predict VLE, heat ot mixing, an LLE over a wider temper~ ature range. Recent extensions to UNIFAC proposed for molecules such 35 relrigerants and sugars maybe useful and you can ads the groups land parameters to your simulation. Simulators may have che tity 10 Generate binary interaction pavametrs for Wilson, UNIQUAC, or NATL from UNIRAC Not all canponents can be describe using UNIFAC, howeves, and otal grup interactions are availabe. Examples of components that do fot have UNIFAC groups include metals, organometals, and phosphates So, we highly recommend aways doing 2 search for avaiable data on binary o ternary systems fires. ‘How should the vapor ahase Ge treat? The choice of the VIE metod using an actity coecient model also requires a choie of mode forthe vagr phase properties if vapor ahase association sob Served (ati the case of acoic acid), then the vapor phase model shouldbe Hayden.('Connel or Nothaagel. A system comaining hyo: ‘gen furide may require» special model to recesent the high degree of {szociaton dus to hydrogen Bonding. Association ie de vapor phase an have a strong affect on psse equilibria and ently. ‘hon should daaults be overiden for ether physical property ‘methods? Pretetion of density. enthalpy, and viscosty aso ae impor tant in sinuatrs, and you shout automatically accapt the detett rmathods. Checkthe simulator documentation forte default method and ining rules. Vapor densiy i calculated by an equation of state or the ide! 92 Jaw, More liquid dersives canbe coleuated by an equation of stat, 2 samperature-dependent model suchas tha of Racket, or bya tempera ture and pressire-dependent model sych asthe COSTALD. For psuedo- ‘components, 8h American Petroleum Insitute (APH) method typical i ‘employed. The Racket madel is recommenced for general use. 'Vagor enthalpy usualy is calelated via an ideal gas assumption or ‘an equaton of state, The equatin-oF state medncds calculate a Spr ture trom ideality ale ew vapor enthalpy deparure For components “sui as acai aid, the Hayden Connell models best, and wil caleu- Inte a largerthap-nomal vapor enthalpy departure Ugud enrages are ealevlated by a variety of methods, tthe simu- lato utes the idol gas as te relerence state, then the pure-component ‘iguidenthaipy is caleulated from the ideal gos enthalpy and qua en ‘halpy departure, This can be writen as He sain (it~ Hm) 0 were His the pure-componentliaua enthalpy His th doa gas en {halpy and (H"~ 9g he iui enthalpy deparre. Ths departre cludes the heat of vaporization, the vapor enhaiy deparure from the ‘dea! pressure tothe saturation pressure, andthe lui pressure coe rcton from the saturation pressure to the real pressure Simuors ‘380 allow separate calculations for» fig entlpyciecty from the lquitheat-copaciy gaimomial. For tome components, the methed in a. twill ot be accurate enough for lquié-nea-capacty predictions. ‘This ean be varyimportant you ate exporing our property information te anther prosram auch as one fr rigorous heat-exchanger design You can use the later qud-heat-capacity (Cp) method to rprove the ae curacy of uid haat capactes. Viscosity is nother imporant property 4r sng of piping, ump, host exchangers and itilaion columns. Thare are various vapor and Aqui methoss fr ealeulting scosty and, generally, the Darameter re ‘uicements for these methods re substantial CHEMICAL ENGINEEANG PROGRESS © OCTOBER =96 © 37SUCCEEDING AT SIMULATION i pe as [7 ane thee Variances [vo _ WHSON. NATL uNrouAE, + Sha These Variances yar UNA a> ‘“?, Non-slectolves io _ UNIFAC and its Ne, basasins P> 1obar M Figure Proceeding jor polar and nonce componenis. — Petobar ise a0 Foure3) > Sehwarentuber Renan, Phorsks wan WS, PR or KS wth V2 PSK, + Por ks win MV? BD vevtncus fo Source: ‘ison, NATL, UNIOUAG, ‘¢ UNBAC win special EOS forhexamess Wiison ‘ison, NETL, UNIQUAG, PIRES. irat with Hayden 0'Conne Hoxamers im the — > onesagee rat te en Una, te + MAE wand i ates | & Dagres of Pehmration “UN one eine soucen Figure 3. Options for vapor-phase caleulations with activity-coefficient models. 3B = OCTOBER 1996 + CHEMICAL ENGINEERING PROGRESS Table 2. Examples of special Gao Prdicive SRK (PSAK} PR with modted Huron-Vidab2 ing rule PR with Panagotoplous mixing rule PAiwith Wong-Sandler mixing rule KS wh medod Huron-Vidal2 ining rule KS with Panagiotopcous mixing ule KS with Wong: Sanlar maxing ule Table 3. UNIFAC revisions poco Me Predicts Dormunésmodiied | VUE ALE 7 GNFAC 885) Kioiverextnsion | VLE of furinatee fist) hydrocarbons yogby-modiies | YE Me case UMTEAC (986 (13) | Entoipy unirac, WE ue (gen fi ‘ercyrevsions | VLE (ssn ‘leit ution activity cotficiont In addition, density, viscosity, pH, and thermal conductivigy may’ be es- sential for other process calculations. “Transport properties are important when doing equipment sizing calcula- tions, Also, processes such as metallur- ‘gy and mining will require calcufations fos phase equilibria including solids, The composition of the mixture. Composition will influence all proper- ties, due to the way mixture properties are calculated. It will affect phase equilibria greatly because of the inter- action of the components in the mix- ture. Usually, the interaction in the liguid phase is the more important be~ cause of the close proximity of the motecules in that phase, The nature of the vapor phase also can be significant if the components form complexes. The important intermolecular forces are electrostatic, induction, attraction, and repulsion between nonpolar com- ponents, and chemical forces such as hydrogen bonding. A good overview of these forces is given in Ref. |Vapor Molerection | © -uque Materacton ‘The magnitude of the electrostatic simulator to report the dipole mo- water, at 1 atm, The azeotrope is ac~ and induction forces is related to the ments of databsnk components as one _curately predicted at approximately Polarity of the components. Compo- measurement of polarity. In general, 0.7 mole fraction of acetonitrile. Fi nents such as water, acetone, mixtures of nonpolar components ure 5 presents VLE for & mixture of formaldehyde, and methyl chloride will exhibit less nonidea! bebvior. wo slightly polar compounds, are strong dipoles, Many polar com- ‘Figures 4-7 illustrate the elfect of toluene and phenol, at 1 atm. The de- pounds are associative, and form polarity on binary vapor/liquid equi- _viation from ideality is shown by complexes or dissociate into ions, _libria, Figure 4 shows the predicted comparing the predicted curve from Components like ethane and n-hep- and experimental VLE of two highly an ideal liguid assumption to that ‘ane are nonpolar, You can use your polar components, acetonitrile and from a method predicting nonideality TT = Vor atactens a ad stan 1 Figure 6 (above). VLE of eyclohexanevbenzene system at 1 atm. Wi Figure 7 (right). VLE of ethanelpropylene system at 40°F. CHEMICAL ENGINEERING PROGRESS * OCTOBER 1996 + 39SUCCEEDING AT SIMULATION (the nonrandom worliquid activity coefficient model (NRTL) and Redlich-Kwong equation of state for the vapor phase), Figure 6 depicts the VLE of a mixture of cyclohexane and foenzene at 1 atm, Here, the interac- tion of seemingly similar molecules with a difference in boiling point of less than 1°C causes an azeotrope at 2 composition of about 0.54 mole frac tion of benzene. A mixture such as ethane and propylene (Figure 7) is an almost ideal one, and does not deviste such from Raoult's law. Mixtures of nonpolar and polar compounds, such as water and hydso carbons, often will form two liquid phases that are very immiscible. Fig tures 8 and 9 show examples of misci- bie and immiscible systems of lig- vuidiquid equilibria, respectively, at 1 atm. In Figure 8, cyclohexanol is im- miscible in the water phase but the organic phase contains up 10 0.50 mole fraction water (0.10 mass frac tion water), Figure 9 shows the high degree of immiscibiticy ia both the organic and water phases for a mix ture of benzene and water where there is less than 0.06% by mole ben: zene (0.3% by mass). Because of this bbchavior, some simulators have a spe= cial property method to treat the water phase as organie-tree (also called Free-Water) Most simulators offer collections of property methods in predefined sets based upon methods that fre- quently are used for certain types of mixtures. Usually the sets are identi- fied by the method used for phase equilibria. When these sets use an equation-of-state model, the same ‘model is used for many properties, in- cluding those for phase equilibria. The pressure and temperature range. This is especially important in choosing the method to perform phase equilibria calculations. Meth- ods that are based on Raoult’s law or that use activity coefficients are not accurate at high pressure or when the temperature is above the erties! tem- perature of a component, You can use Henry's law when you have light gases in subcritical solvents, but it generally is not recommended for concentrations of solute greater than ‘5%. In general, equations of state are better suited 1 predict VLE over a wide temperature or pressure range, especially at high temperature and pressure The availability of parameters. Without sufficient pure-component and binary parameters, you will be ‘unable to calculate pure-component or mixture properties. You must choose among obtaining and using experi- ‘mental or literature data, estimating parameters, oF choosing a fess rigor us method. This should be investigat- efor all physical property methods in cluding those shown in Figures 1-3. 40 + OCTOBER 1096 + CHEMICAL ENGINEERING PROGRESS M Figure 8 (lft. LLE of exelobexanoliwater system at fate Wi Figure 9. (above). LLE of benzene/water system at I aim. Validating the physical properties AA necessary step in any simulation project is validation of the physical properties. This involses reporting, tab- Uulating, or ploting pure-component and rixture properties and comparing the results t0 known data or expected be- havior This isan important step in any simulation and shouldbe performed for dltabank a5 well 5 ondatabank com- Ponents. Simulators can provide these Calculated properties in tabular and plot format. This is @ useful tool for under- standing how pare-component and mix- ture properties, such as density, feat ca- pacity, and excess properties, vary with ‘temperature, pressure, and composition, and how they behave when exttapolat ed, Similarly, such results ean be used 0 ‘generate plots of VLE and LE to com- pare to diagrams in the literature and ac- ‘ual field daca. Some simulators have the capability to generate residue curves for distillation of temary mixtures, The residue plot capability also isa powerful ‘0 for distillation analysis Use the tabulation and plotting tools to determine the cause of dis- csepancies in properties. I a mixture property is incorrect, investigate if a single component is the cause by re- poring pure-component properties. Another useful technique is to com- pare the same flowshest or property- table results while using different physical property methodsBy default, most phase equilibria calculations are performed assuming vapor and Tiguid phases. If your pro- cess involves two liquid phases (VLLE), be sure to specify three: phase calculations, If not, you will et incorrect results. As a part of the Validation, you also should check that your property methods do not falsely predict two liquid phases. ‘Simulators let you specify that only fone phase is present in a steam or @ unit operation. If vapor and liquid phases are possible, however, you should use the two-phase specification, Nondatabank components and missing parameters When you want to simulate non- databank Components or have compo. nents for which parameters are miss- create a list of parameters that are missing, You should detail this infor- mation when communicating the as- sumptions of the simulation to other users oF your management, Certain propery parameters always are required for a simulation. These ccan include molecular weight, vapor pressure, and ideal-gas heat capacity constants, The need for other parame- ters depends upon your choice of physical property methods. The simu- Jator manuals should include the infor- ‘mation about the parameter require ments (7). There also are parameters that will be required for calculating the heat of reactions or the reaction equi- librium constants, This includes the beat of formation and the Gibbs free energy of formation of all components that participate in the reactions, check for different ordering of atoms. For instance, ammonia can be de- scribed as H,N instead of NH,, Ref. 2 contains a formula index of organic, compounds and is a good resource for ‘alternative names. (Once you have determined the pa- rameter requirements that ase not sat- isfied, the next stage should be ob- taining and using physical property data, Obtaining and using physical property data Sources of data. To provide par rameters for nondatabank compo- tients oF to do regression for pure component and binary parameters, you will aced 10 search for available ‘data. Such data may be found in a va- riety of Sources, including data-com- ing, ask yourself the fol- pilation references, and- lowing: ‘books, journals, and inter 13s this 2 major compo. Techniques to remove or minimize ral data collections rent in the mixture? If itis, While most steams. in ‘minor, can I take it out of the impact of specific parameters simulations contain mix- the simulation? + Does the component take pant in VLE? + Is the component non- volatile? + Is it polar or nonpolar? + Will reaction (including decom- position) cause this component to be depleted? ‘= What properties need to be accu rate for the chosen property methods? These questions will Help you to identify the parameters that are need= ed based on your choice of physical property methods. If these parameters are not available or cannot be deter- mined through literature search, re- gression, or estimation, then you will have to reevaluate your choice of physical property methods or obtain data by measurement ‘You should determine what the pa- rameters will default to if the simula tor does not find any available. It is dangerous to assume that the physical property parameters were available just Decause the simulator did not give you an error message. Use the simulator manuals and on-tine help to should be used with caution. You can use your judgment about the importance of parameter to set nominal values for unimportant prop. erties. For example, if you know that 2 component is very nonvolatile and are using Antoine's equation for vapor pressure (In P= A+ BAT+C)), you can set the value of parameters A, Band C to -100, 0, and 0, respective. ly. (Fis temperature.) This will assign the vapor pressure used in Raoult's Law a very small value, almost zero (3 x 10%, This and similar tech- niques to remove or minimize the im= pact of specific parameters should be used with caution, however. If you can't find a component in the simulator's databanks, make sure you check for synonyms. For exam- ple, methoxy benzene may be listed as, ‘methyl phenyl ether or anisole. A. 00d approach is to search for the component using its formula. When selecting the component by formula, tures, accurate property cal- culations are not possible ‘without accurate pure-com: ponent properties. The im- portance of pure compo- nent data should not be underestimat- ‘ed as they are the basis for both pure ‘component and mixture properties. For instance, pure component proper ties such as vapor pressure will be used in phase equilibria calculations Table 4 contains common sources for pure component properties, while Table 5 lists common sources for mixture properties, The recommended order of data search is 1. critically evaluated data sources, 2. nonevaluated sources; 3. experimental measurements: and 4. estimation techniques. Binary parameters for phase equi- libria. Because of the large number of binary pairs in even a simulation of only ten components, we recom- mend ranking the components so as to prioritize the pairs and focus the literature search and measurement ef- forts on the most important parame: ‘CHEMICAL ENGINEERING PROGRESS * OCTOBER 196 + 41SUCCEEDING AT SIMULATION ‘CRC Handbook of Chemistry and Physes (Bester DIPPR Date Complain Eneyclopeia of Comical, Grup, ond Biologicals Encyclopedia of Poymer Science and Engineering {ESDU Validated Engineering Data Index Handbook ot Tharmophysial Properties of Gases and Liuids JIANAF Thermochenies! ables, Lange's Handbook of Chemistry Perrys Chemical Engineers Hanbook Propertas of Gases and Lids Selectod Values of Properties of Chemical ‘Compounds (TAC 4 \opor Pressure of Pure Substances * Parts ofthese sources ae avaible oti rom tionl or Technical Databases Services, Ine-(TDS) ters. First, divide the components into three groups: high, medium, and Tow priority. Base the priority on eri- teria. such as composition, and the purity specifications of the process if a component purity is specified, that component is important even if it appears only in low concentrations, Second, pair the components. into high/high, high/medium, high/low, mediun/medium, medium/low, and Tow/ow groups. Search the available sources, including in-house ones, for any data for all groups. JF certain component pairs are known to be- have ideally, they can be excluded from the search, Next, use the UNI- FAC method for the missing pairs in the medium/medium, medium/low, and low/low categories. UNIFAC is not recommended, however, for any pairs that include the components of high priority, A secondary fiterarure search can be used to find binary data, for similar compounds and those pa- rameters then substituted. Propose experimental work if any binary pa- rameter data are still missing or if data regression exposes data as inad- equate (3). Qe (OCTOBER 1996 « Table 4. Examples and reliability of sources Siri ks valued — | Gover ricaly "| Nonercsty crite Nonerteaty Nonercaly Nonercaly Noneically ceticaty Yes Nonestcaly ves Nonerteaty es Noneically Yes Crtclly Yes Nonertcaty ws IALOS information Services, STN Interna: Regressing data Data regression is a powerful tool for engineers not just to make the best of available data, but also to ana- lyze the goodness of fit of a physical property model to the data. Most sim blators include a data regression fea. ture. Examples of commonly re. sgressed data include binary VLE and LE, vapor pressure, heat of vapor- ization, density, and heat capacity. Data regression finds the best fit of parameter estimates to the experimen tal data. The best fit is represented by finding the lowest value of an obje tive function while matching the phase equilibrium or other constraints One common regression technique is called Maximum Likelihood Estima- tion. The objective function for this method is: 5 wy Eaer - C10," Q Where jis a data group, Cj" and Cf are measured and estimated ‘variables, spectively, such as temperature, pres- sure, composition, or heat capacity, 6, is the standard deviation or the error in the measurement of the variable, and. ww; is the weighting of the data group. ‘CHEMICAL ENGINEERING PROGRESS ‘When fiting phase equilibria data, the regression algorithm attempts 10 re- dace che objective function while the physical property method is being used fo check thatthe components moet the constraints of phase equilibria The work of successful regres- sion involves selecting the right phys- ieal property model and parameters, representing the data properly, choos- ing appropriate standard deviations of the data, and starting with suitable initial estimates of the parameters The following are general guidelines for data regression ‘+ Make sure that you are regress- ing the right parameters. Use the same physical property method and builtin databank that you will be using in the simulation. Choose ps- rameters that have impact onthe data being used. For example, when using an equation-of-state method such as Peng-Robinson or Redlich-Kwong: Soave, you should determine the acentric factor, w. But, if you are using an activity coefficient method, you should determine wo or more Constants for che Antoine model. * Estimate as few parameters as possible, There is'a tendency to use a Targe number of parameters when fit ting a model to data such as tempera- ture-dependent properties or binary phase equilibria. Try to regress the Gata with as few parameters as possi- ble. If the regression results report thatthe standard deviation ofthe esti mated parameters is ofthe same order ‘of magnitude as the values ofthe pa- rameters, you may be estimating t00 many parameters for your given data The larger the température range of your data, the more parameters that you can estimate. * Watch out for incompiee data. A regression may yield poor results if there are missing data points, particu- larly composition data. For example, some authors do not report all com. Positions in VLE. of immiscible LLE. You may need 10 estimate the missing compositions so that phase equilibrium can be calculated for al components. Find out How your sim-ufator handles missing data to best eal with incomplete data. * Specify the right number of phas- es. A regression will yield incorrect, results if the number of phases is not specified correctly. This is a common problem in VLLE systems. For some Titerature data, the number of phases is hard to interpret because of the pre- sentation of the data or lack of de scription, Often in VELE data, only a total liquid composition is reported even though two liquid phases were present, The author may be reporting a heterogeneous azeotrope — an, azeotrope where the vapor composi- tion equals the total liguid composi- tion bur two liquid phases are present. When doing the regression of a het- erogeneous azeotrope, divide the data into two groups, the VLE data and the VLLE data. This will ensure that the correct phase equilibria is considered. In cegressions such as this, it is im- portant to use the property tabulation and plotting features of the simulator to check that the parameter estimates correctly reproduce the original data. + Use a model's full functionality. A physical property model may be used to calculate several properties. For ex- ample, you can use binary excess-en- thalpy (#,) daca and binary VLE or LE data to determine binary parame ters for activity coefficient models. For ‘equation-of-state models, you simulta- neously can use liguid- and vapor heat capacity, vapor pressure, and heat of ‘vaporization data. If data are available for these properties, use these data to- ‘gether to estimate the parameters. Data ‘groups of different types cam be used together in the same regression, + If necessary, regress parameters even if values are available in the darabank. The physical property pa~ sameters found in the builtin pure- ‘component and binary databanks gen- erally are very reliable. You may find, however, that you need to determine new parameters to replace the data- dank values for your application, Check the built-in parameters to en- sure that the recommended tempera ‘ure, pressure, and composition range Sources ‘Activity Cosficonts at infinite Dion, DECHEMA Chemistry Series Binary VLE Dats fle, OIPPR. ‘Dortmund Oatatank superset of DECHEMA data collection” Heats of Mixing Date Colecton,DECHEMA Chemistry Series Liquid- Liquid Eouibbrom Dace Catton, OECHEMA Chemisty Sores Phase Equilvia and Enthalpies of Electrolyte Solutions, DECHEMA Chemisty Seves ‘Vapor Lui Equa Data Collection, DECHEMA Chemisty Series Vapor-Liqui Equilibrium Data foc Electrolyte Solutions, BECHEMA Chemistry Series ‘Vepr- Laud Eguibriun for Matures of Low Boing Substances, DECHEMA Chemistry Series Selected Values of Chemical Compounds, Texas A&M University Soli-iuid Equi Data Colin, DECHEMA Chemisty Series * Online databanks js not outside the range of your simu- lation, For example, vapor pressure parameters may not have been deter- mined at temperatures below the nor- mal boiling point, Most physical prop- eny models extrapolate outside the femperature bounds reasonably well — but at some compromise in accura- cy. The parameter values also may apply to a very wide sange of tempera- ture and thus not provide as 260d a fit if you only need a narrow range in the simulation. For phase equilibria caleu- lations, 10 improve the accuracy of VLE or LLE predictions, you may ‘ant to use ternary or quafemary data to fine-tune binary parameters that may be available in the simulator, * Check that the parameters repro duce the data. The simulator will re- port qualitative results of the regres- sion, including the residuals (experi- ‘mental minus estimated variables), Use the property tabulation or ploting fea tures t0 reproduce the data at the speci- fied conditions. This can be performed in the same regression run. Check that the comect number of phases is pre: dicted by allowing two-liquid-phase calculations for the property table or plot. In addition, your simulator may have an option where you can evaluate the fit using the existing paramevers and model with experimental data ‘without doing a regression * Remove components not in phase equilibria. If components that are solids or ions do not appear ina Prk phase, you can remove them from the phase’ equilibria constraints. This is useful in VLE. * Generate equilibrium data. If ‘you have binary parameters for an ac- tivity coefficient or equation-of-state ‘model, your simulator may be able to generate VLE or LLE data for regres- sion using these parameters. You can regress these “data” with another physical property model, This allows consolidation of known parameters into a single property method. + Fit other daca. Yout simulator may have a data fiting feature that can be used for plant data. This method may not be as useful for predictive simulation, though, if the data are not from a wide variety of conditions. Estimating missing Property parameters Property estimation usualy is done after a data search is performed, to supply missing property parameters. You can use built-in estimation meth- ds to fill in some gaps in your physi- cal-propery-parameter requirements. ‘Simulators include one or more esti- ‘mation methods for each of the most ‘common parameters. There are (wo types of estimation methods for pure component parameters: structural ‘group, and corresponding states, Structural group methods are based fon the idea that contributions of the ats or structural groups of the compo- nent are additive for properties such as ‘CHENNCAL ENGIVEERING PROGRESS * OCTOBER 1996 + A3SUCCEEDING AT SIMULATION * at atm, 25° torial gos. Table 7. Estimated properties for propyl phenyl ether (Cl,.0). Estimated Value 196.1887 86872 Nim | 3088,350, Property Name Estimation Method Formula ‘Abrose Ambrose Fedors Bonson Mojacar weight Critica temperature tcl pressure Critica volume Standard heat of formation” miKmole o4sza73 ixmole | 125530000 at an, 26° for idea as. ‘normal boiling point, critical tempera Comesponding states methods are ture, critical pressure, ideal-gas heat ca- based on empirical mathematical re- pacity, and standard heat of formation. tionships among properties. For ex- Some methods, such as that of Benson, ample, the Letsou-Stiel method. re- Contain additional corrections for next: ates liquid viscosity to critical tem- nearest-neighbor atoms or for rings. perature, critical pressure, and acen- ‘Stuctural group contributions are dewer- ric factor. These methods most likely mined by taking an average contritution will be inaccurate when used. for ‘based on known physical constants of compounds unlike those upon which many organic compounds. Because the the correlation was based, Benson, Joback (J0), and other siruc- A. good approach for both group tural-group methods are based mainly contribution and corresponding states ‘on data for organic compounds, they methods is to check the accuracy of cannot be used for inorganics, including as_many methods as possible for ‘metals, of ions. In addition, structural compounds for which properties are group methods do not accurately repre- known and which are_ structurally sent very lange organic molecules (that similar to the compound you are esti- is, ones with a molecular weight > 200) mating. The following example such as proteins, New group-contrita- shows the use of this concept. tion methods like that of Constantinou Estimating the properties of propyl ‘and Gani (4) potentially may provide phenyl ether. Let's say that you are better estimations for organics. Other modeling a process containing propyl possibly useful methods are proposed in phenyl ether (PPE), also called the lteranare but may apply t9 only ce —propyloxy benzene, The only data tain families of components. you have are its boiling point 44 + OCTOBER 196 + CHEMICAL ENGINEERING PROGRESS B Proper ame | we foe | ‘iia erpernue r ws wns | tt beck (189.9°O), density at 25°C (0.9474 gion), and motecular structure ‘You want to estimate the proper- ties of PPE using the most appropri- ate methods, Step 1. Determine the best estima tion methods for 2 similar phengt ether. Select sther compound(s) chem- ically similar to PPE for which you have experimental property data, (Of course, the more similar compounds you can use, the greater your confi- dence that you are selecting the most appropriate methods.) In this case, for simplicity, let's choose only phenetol: cy ock,cH, Data for phenetol is available from the DIPPR data collection (5). Use the simulator’s built-in methods {© estimate properties for phenetol. ‘Then, compare the results of the various methods with the experimentally deter ‘mined values to identify which methods sive the best estimates for this class of ‘compounds. Table 6 lists the results for the different methods for phenetol ‘You can see that the Ambrose method gives the best overall predic- tions for critical temperature and pres- sure, the Fedors method for critical vat- ume, and the Joback method for stan- ‘dard heat of formation for phenetol. So,we will use these methods to predict the corresponding properties for PPE. Step 2. Enler the available daca and structure for PPE, Enter normal boiling point and molecular structure of PPE, and specify the methods that {gave the best predictions for phenetol. Step 3. Examine the estimation re- sults for PPE, These appear in Table 7. One area of property estimacion that is more difficult is differentiating the properties of stereo isomers. Some zgroup-contribution methods have cor- rections for ortho, meta, and para configurations, but few have built-in corrections for optical isomers. The separation of these isomers ia a chemt- ‘cal process is based on their slightly different properties — relative volatil- ity in distillation is one example. Employing simpler methods Tn addition to structural group and corresponding states methods, another useful esimation approach is provided by series and family plots. Series plots Took atthe values of a property such as ‘ormal boliing point with increasing ‘molecular weight or cafbon number for compounds ina series that differ by ne substitient group, such 2s the CH, unit in alkanes. Figure 10 is a series plot for the normal boiling point of m= alkylbenzenes. Family plots ate simi- lor, but che number of groups is larger. For example, Figure 11 shows a family Plot of the critical pressure of ‘ethylttydrogenichlorasilanes. You can use these plots to predict properties by extending the cue oF to check your data for errors (6). To cteate a seful series of family plot, however yOu must be careful about the eompo- ents included. ‘When accuracy is nor tical consid- er the simple but pawerfl technique of component substitution. In this, you use the properties of anotes, similar compo- ent forall properties oF the component of tres that you do not know. A simi lar component is one that has compa ible volatility (vapor presse) density an heat capacity. This is useful if the component is nonvolatile oF i not in- volved in phase equlitra, For example, you have a small amount of a non- volatile component in a stream that is at 100°C and | atm, You can access the properties of a nonvolatile component, say CH, (molecular weight = 282.55, and boiling poine = 343.78°C), instead of estimating properties. This method is ‘ery efficient if you do not need accurate properties of the component, Take care, though, if you use this appreach and one of the UNIFAC actvity-coeficient methods, as you may change the as sumptions made abou the liquid phase, Another technique to simplify a mixture of similar components is to represent them with a single compo- nent. This is a useful technique when components are not known exactly. For instance, Component C,+ could repsesent hydrovarboas of 5 carbon atoms and greater, Estimating binary parameters ‘You caa escimate binary parame- ters for Wilson, NRTL, and UNI- QUAC activity-coefficient models using two approaches: UNIFAC and infinite-dilution activity coefficiens UNIFAC-estimated binary parameters usually do not provide enough accu- zacy and, So, only ate recommended for early stages of physical propery data investigation and to “fill in the blanks" for components with media oF low priorities. Better binary parameters can be estimated using infnite-dilution ac tivity coefficient data. (Some simula- tors may include this feature under their regression tools.) This method is better because itis based on the com- Ponents of interest, unlike the group contribution method, which averages 1B Figure 10 oft) Series plot of normal boiling Point for nll Bencenes. Figure 11 (below), Family plot of enical pressure of ‘methylthydrogen) hlorosilanes, (CHEMICAL ENGINEERING PROGRESS * OCTOBER IS=6 © 45SUCCEEDING AT SIMULATION SS TT Literature Cited 1. Red RC, J. M. Prausnitz, and BLE, Paling, “The Propeies of Gases and Liquids” hed, McGesw Hil, New York (980), 2 "CRC Handbook of Chemisry and Physic” D Lide, ed, CRC Press, Boca Raton, FL (1999), Dewan, ALK, and M.A. Moore, ‘Methodology 10 Develop an ASPEN PLUS Model a Stell Development Com pny? pfesened at ASPEN. WORLD, Cartridge, MA (1998) (Availbe fom ‘Aspen Technology, la.) 4. Constantia, Lay and R. Gani, "New Group Contin Method for Esimat ing. Properies “of Pure Compounds” AICKE J, 40(10),p, 1697 (1998), 5S Daubert, TE, RP. Danner, H. M. Sibu, and C. C, Stebbins, "Priel and “Theanodynani Properies of Pare Chem- jeals: Data Compiltion” Design Insite for Physic) Property Dat (DIPPR), [AICHE, New York (1989-onmad, 6 Smith, A. La “Fal Plots for Evatt ing Physial Properties of Organoslcon Compounds” AICKE J, 40 Q), . 373 1998, 17. “ASPEN PLUS User Guide” Vl. 1. Re. lease 9, Aspen Technology. In, Camm bridge. MA (1995. % Gmebling, J JL, and M. Scher, "A Modified UNIFAC Model. 2. Present Pa. rameter Matix and Resuls for Difeent Thermodynamic Properties” AEC Res, 32,9 178 1993), the effect of group interactions from different components. Estimation of physical properties can get you started in a simulation problem — but you should do an ex- hhaustive literature search to. find missing pure-component and binary parameters Tis important to enter any known, parameters before doing property es- ‘mation. First, experimental data generally are more accurate than esti- mated values. Second, corresponding States estimation methods require fother physical constants as_inpor Using an experimental vaiue will i jprove the prediction of these property parameters, Otherwise, the error in estimating parameters such as normal boiling point, critical temperature, 9, Hansen H Ka P. Rasmessen, A. Fre: Aeastund, M. Seer, and J. Gmebling, "VaporLiquid Equilibria by UNIFAC Group Contribation. 5. Revision and Ex. tension?” [AEC Res, 30 (10), p. 2382 (990). 10, Joback K. Gu and B.C: Reid, “Estima tom of Pure Component Properies fom Group-Contibations” Chem. Eng. Com. sma 57, 233 (1987), 1. Kleiber, Mz “An Exteasion 10 he UNI. FAC Groap Assignment for Prediction of ‘Vapor Ligud Equiiea of Mixtres Con- taining Refrgeras,” Fluid Phase Eu bri 107, p, 161 (1995). 12, Magnussen, Py P. Rasmussen, and A. Fredensiund, “UNIEAC Parameter Pre= Aiton Table for Prediction of Liquid Liquid Equilibria” JREC Proc. Dev. 20, 331 0980), , Rasmussen, and A. Fee "A Moiied Group-Contit tion Method for Prediction of Phase Equ- iba and Heats of Mixing.” [AEC Res, 26,» 2.214 (198). fi 14, Dertmond Databank, VLE Dat. System. No. 980, Unis, of Dortmund, Germany (1995), 15, Dorunund Databank, VLE Data, System No. 4,385, Univ. of Dortmund, Germany (2995). 16, Dortmund Databank, VLE Data, System 1No. 752, Unis, of Dormend, Germany (4995. and critical pressure propagates to other property parameters Documenting what you've done Simulation projects often have a long life at a company. New users may come along and be unfamiliar with the assumptions and recom- mended use of the simulation. You may find that you need to revisit @ simulation a year or mote later. Doc- umenting the data sources, the range of applicability, 2nd physical property assumptions is extremely important. This can be incorporated using the comment or descriptions fields in the simulator. Include a statement about any properties that were not well de- fined or components that should not 46 + OCTOBER 196 + CHENICAL ENGINEERING PROGRESS Acknowledgment "The techniques and guidelines resected in thsaicle ae the reals of experience elp- Ing simulation Users solve engines rob Jets T woald Ike to tank be following leagues at Aspen Teebloay fr vice fd Sfoumaton: Valentin Debec™, Marcelo Marches, il" Mosk, Andrea Tafoodan, and Suplar Watamsi, bbe added given the physical property method employed — for instance, electrolytes when an equation-of- state method is being used. Keep track of the references for data and list them in the simulation, if possi- ble. Include comments about proper: ties, such as densities or heats of mix= ing, that were not of interest oF not validated in che simulation. Keep the estimation, regression, and simulation files together. If possible, create a file containing all pure-component and binary parameters including those ac- cessed in the built-in databanks. This ‘way you will be able 10 reproduce ‘your results in the future with upcom- ing simulation-software releases. Keeping the right perspective ‘The physical propery system of the simulator isnot a black box, but a well developed set of rules and rela- tionships that can execute very com- plex calculations very quickly. It does not replace that most useful of all tools of a chemical engineer — com mon sense. Always use your judg ‘ment to evaluate simulation errors or suspicious results to find their source, ‘That way, you'll make the best use of your simulator, and avoid unneces- ‘sary mistakes. foo] .6.CARLSON af angina As “achnny, ne. Comba, Mt (errs a Fax 61757 tmetcarson@aspetch com. e's ‘arposile or povsra echnical gud ted raing ihe areas psa propeies nd eat eden to rocee simon tare He eceied #8 fom tho Un of acheter and an MS rm Ne Casing Site Un eh chan einai. Has Somber of ICE 2