« Membranes in ChE Education A Simple Analysis For GAS SEPARATION MEMBRANE EXPERIMENTS Ricuarp A. Davis, Orvitie C. Sanpat* University of Minnesota Duluth * Duluth, MN 55812 ‘mba apptiations fr gs separations have made rid ad M vances ove the pst dee de" Tn sore case, membrae teh nologies have been used enhanc or place mre aoa iethods of gs pian The ned fr educating undergraduate hea Calengineeig students about membrane based sepuratns hasnt gone noticed: Never editions of ppulrspantins textbooks have aed Chapter on membranes with scons on gs pemetion Earlier, Davis and Sandall®! described an undergraduate laboratory mem- brane exprient ar analysis or separating fhe components of it I renin relvantoday aon ppoich to rovidng stents wih hans on experience wih this important ecology. The experimental objec. ves eladedan inverse mass aster analysis of expesimental data for key membrane transport paraosers, The ginal alas involved oy inga st of diferent specks balances and fing te resus to expe tment daa by iran, tial and-eorecnigus. They found tat tc tuercal methods reset implement ter analysis were beyond the ‘cope ofthe undergrads chemi engineering Iboratry experience, Conquer, they provided stents with True BASIC programs tht wereure to olvete model equations, Unfcunatly the progrars were Tite tothe spectic membrane configuration inthe laboratory. St dents ce unable slr alemalv designsig the vali mod els without modifying the progans, Inte meantime, several pogula, modem, computational softvare applications (ach as Exel, Mathcad, Matlab or Polymaty have emerged that provide readily acesible tools for svg comple problems that nvalveronieaaiebaic and Gi ferential equation, The drawbacks inte orginal analysis, along with evelopment in compuational tol have ldo a simle alealive nals deseribed in tis paper EXPERIMENT Davis and Sandall® provided specific details of the experimental ob- Jectives, apparatus, and procedure for a commercial hollow-fiber mem= brane unit for aie separation. The Prism separator developed by Permea TGniveriy of Calorie, Sana Barbara, CA 93106 Figure 1. Prism hollow-fiber membrane apparatus, “chard A Davis i Associate Professor nth Department of Ghamecal Engines at tha Unworstyof Minnesota Dutt Heamedtis BSin Chanical Enginetng kom agham Yeung ‘Unwomtyand i PRD tom the Urner Calioraa, Santa ‘Barbara Hotoachos 9 vate of courses warszor phere ‘sar sopartona andi rent reserch rete ince roses modetng and optnaton. Ovile € Sandal 's Professor of Chemica! Enginnering at ‘ne Unveraty of Catto, Sava Barears Hols agrsauaie {he Univer oF Alberts (256 and MS) are the Uso oF Com, Bakcioy (PAD). He Yaching aed ressarch tr strani tin archers tater and aepsiaon prance © conyiit 8 Dinion of AS82008 ” Chemical Engineering Education@ Membranes in ChE Education ) Corporation, shown in Figure 1, consists of four hollow-fi- ‘ber membrane modules arranged ina series of columns. Fach module is a shell-and-tube arrangement of a bundle of hol- low-fiber membranes that are capped at the top. High-pres: sure feed air is introduced to the shell side of the fibers. The permeating gas flows through the hollow-fiber bores and is, collected in a manifold at the open end. The pressure drop, ‘across the shell side of the membrane unit was found to be. negligible."The permeate streams are open tothe atmosphere. ‘The pressure at the closed end of the fiber bores is not di- rectly measurable in the current module arrangements. Infor- ‘mation about fiber length, fiber inside diameter, and the num- ber of fibers in the Prism separator bundle is not available, ‘but a conservative estimate of the pressure build-up in the fiber bore was calculated to increase by less than nine per- cent above atmospheric pressure forthe range of experimen- tal operating conditions. For most of the experiments, the pressure build-up was estimated to be less than three percent. 4 nacre Sige Ctr) Figure 2, Schematic of single, countercurrent flow column or four columns with alternating flow patterns. mI x a Lat Mena (pact [+ Pemeae anne JP panei) Lm Rene Lya Feed >) HP iat ay Figure 3. Ideal cocarrent flow pattern Tee Pemese <—] Sai ® eed i Menraze pes a] om Pte 2000 Ly Reenite se "ad sm Figure 4. Ideal countercurrent flow pattern. inv 2008, ‘Modern gas-soparation membrane modules introduce the high-pressure feed to the bore side of the fibers to eliminate channeling and maintain a more uniform flow distribution. High-pressure feed (othe fiber bores can result in a signi cant axial pressure drop in the fibers. Although not required for this membrane module, the effects of pressure are included in the analysis for completeness, As shown in the schematic of Figure 2, the air-flow pattern consists of alternating countercurrent and cocurrent flow Uhrough dhe columns. The composition of the retentate and permeate streams was measured with oxygen analyzers. The flow rate of the retentate stream was measured with a volu- metric flow meter. The feed and permeate flow rates may be calculated by mass balances. ‘The membrane separator may be operated as four columns in series, or asa single colurnn by closing a valve on the tube connecting the retentate and feed streams between the first ‘hwo columns, The frst column operates in countercurrent flow and was used to calibrate the membrane models from a series of runs performed at various feed-flow rates and pressures. ‘The calibrated model was confirmed by favorable compari- sons of model predictions with experimental results from the four-column configuration. THEORY AND ANALYSIS A differential model of binary gas separation in the mem- brane experiment was validated by Davis and Sandall and is summarized next. For the conditions of the experiment, it cean be shown that a simplification to the equations permits an algebraic solution. The mathematical model of membrane gas separation was based on several key assumptions. First, the temperature was assumed to be constant. Further, it was assumed that all streams through the shell and permeate sides of the fibers ‘were in plug flow. The air fed to the unit as assumed to be 2 binary mixture of 79% N, and 21% O,, All four columns were assumed to have the same dimensions and specific area for tmass transfer. Finally, axial pressure drop was ignored for the fiber bore. This assumption is valid for low permeate flow or large transmembrane pressure differences where small changes in permeate pressure are negligible relative to the high feed pressure. B Differential Modet Walawender and Stem! derived the differential equations for a binary gas system in countercurrent and cocurrent plug flow patterns, shown ideally in Figures 3 and 4. Details of the derivation are available in several references." For a binary gas system, the total mole and O, species balances around the separator are 1"C Membranes in ChE Education ap =n tmp 0 Nene =XRMR FY Mp @ where n, n,, and n, are the molar flow rates of the feed, retentate, and permeate streams, respectively, and x, x,, and X, are the feed, retentate, and permeate O, mole fractions, respectively. The species balances around a differential vol- lume element in the membrane give axa) = Qo, (xP - yp4A @ af(.-s)n]= Oy, [1-9 -(1-y)pjaa “ ‘where Qj is the permeance of species j,A i the membrane surface atea, and P and p are the average retentate and per rmeate side pressures, respectively For convenience in the analysis, Eqs. (1) to (4) were com- bined into the following dimensionless equations for coun. tercurrent flow: Yer -ai-s ~(\-y)}} @ or t-2koe-v-sfo-s- a 6) (ar-y)+(I-x}e=(I-y) ) where y, is the mole fraction in the permeate at the closed end ofthe fibers. The dimensionless transport parameters are defined as 6 ) Ky=me/Qx,Ane (10) 0° =Qo, 10K, ay waning (2) where A, is the total membrane area. The ideal separa- tion factor, ¢, Was assumed constant, but the dimensionless transport parameter, K,, was defined as a function of the retentate molar flow rate. The solution to Eq, (7) was used to check the assumptions leading to the algebraic model of the next section, The countercurrent flow equations are integrated from the retentate end of the membrane, sub- ject lo the initial conditions % (3) Note the discontinuity in Eq, (6) at x= x, requires application of I'Hopital’s rule." ‘The dimensionless cocurrent flow model equations are Jer -ater-9)-s0-ne=0-y) (is) = Ho -nterv)--er- a) (is) where Kp = — (16) © Oy. AMP: The cocurrent model equations are integrated from the feed end, subject to the initial conditions xo*Fl wa’ =0 9 | io ‘The permeate composition at the capped end of the hollow fibers is calculated from the ratio of Eqs. (3) and (4) yi @"[xe-yi] vi [l=xe-(- vi] where, for countercurreat flow, x= x,. For cocurrent Flow, x = x,. Equation (18) is quadratic in y.. Note that there is an error in the denominators of Eqs. (17) and (22) of the paper by Davis and Sandal.*! The correct solution tothe quadratic equation is (is) (@-i)ocener=yfe (09) Davis and Sandall successfully used the differential model in their analysis of OJ/N, separation in the membrane mod- tule, At the time, they found that the background required to solve the model equations for a” and K, was beyond the scope of an undergraduate student in their laboratory course. Consequently, they developed True BASIC programs that ‘were provided to the stuclents to solve the model equations. Since then, advances in computational software (such as ‘Mathead) have simplified the process of solving the model (Chemical Engineering EducationMembranes in ChE Education ) equations. Undergraduate students are now able to develop their own solutions using standard numerical methods for solving systems of nonlinear equations or differential equa: tions that are readily available in these computer tools. Nevertheless, students are still required to set up a stan- dard method such as Euler's or Runge-Kutta for the initial- value problems inorder to find the values for a and K, by inverse analysis of the first column in countercurrent flow. For example, Mathcad and Polymath do not permit their in- trinsic capabilities for solving systems of first-order differ- ential equations to be treated as part of another function. An example of programming required in Mathcad forthe inverse ‘mass transfer is shown in Figure 5. This type of solution may be intimidating for undergraduate students, depending on their level of experience. This realization, along with the observa- tion that the composition profiles along the membrane were ‘approximately linear, led to the following alternative analy- sisthat avoids the initial-value problem solution requirements entirely. Ge coe feos eengte ad Mei) of.tya)= EEL Ea -nadeeey neti) ere) = fi ae bine nfed (e-mfe-rglen Belle Wren TY os 2 La-yawa-n=vit-2 (1-9) eine oes) fr sew feel = bacaniet sep Heda es) artsy e) Jeeps Aarne) Sbestatanty be ov a (neha es (eb), b= wens) Figure 5: Example of Mathoad programming for inverse inass transfer analysis for a and K, Wer 2008 W Algebraic Model Boucif, ral, presente a series solution to the binary component differenti! model Eqs. (3), (6), (14), and (15) shat requires a numerical soltion ta pao third-order poly- nomial equations. The solution tothe series equations agrees ‘with numerical solrions to the differential mode when the cut isles than 50%. The series solution does not include axial pressure effects in the feed or permeate gas, however. Hundyl and Koros! presented 2 more complete analysis of hollow-fiber membrane modules for multicomponent gas separation that includes pressure efecs. Their approach is based on a fnte-volume element model that requires itera- tive solutions to a large system of nonlinear algebraic equa tions. The finite-element approach is recommended when de- tailed information of pressure, temperature, and composition effects is required simple, alerative analysis ofthe membrane unit described here was developed that involves ony the solution toa smal system of nonlinear algebraic equations and includes pres- sure effects when necessary. The simpler-model equations are analogous to the shell-and-tube hea-exchanger design equations that are familiar to undereradute chemical engi- neering students, The following analysis assumes laminar flow and constant species permeances that are independent ofthe pressure and composition ofthe feed or permeate gas, “The Hagen-Poissuille equation is commonly used to cale- late axial pressure effects! a _ 28RT pn o de pri; where R is the eal gos constant, Tis the gas temperature, isthe gas viscosity m isthe variable molar flow fate of per rmeate gas, dis the inside fiber bore diameter, and, isthe numberof fibers na bundle. ther expressions derived from the Hagen-Poisseuille equation have been developed to ac- count for compressibility and flow in porous channels when necessary.!*!°! It has been observed that when the change in the feed mole fraction of oxygen s less than SO‘, the diferential balances ay be replaced with algebraic expressions involving the Jogarithnie mean of the transinembrane partal-pressure dif ference."! In Bg. (3), let xP-yp i) “The driving force for diffusion across the membrane, A, is assumed t0 be a Tinear funetion of the change in the molar flow on the feed side of the membrane (xn) _ (x0)y e racers @ Combine eqs. (2), (3), and (22), separate variables and inte- ”(Membranes in ChE Education > grate Yotp FB =05.(4% =p) f aa (23) or ‘ Yap = Qo, (3P—¥P hig Am (24) where the log-mean difference in O, partial pressure across the membrane is defined as py a lth = 20g (a= Pp > (Powhe Flee), P-vel,] A similar results found for aN, flux expression (1-¥p)ny = Qs, [(0-2)P-(1-y)p],,, An (26) ‘The steady-state binary-gas membrane equations can be writ- ten in dimensionless form using the average pressures, kp =XR(1-6)+¥,6 @ Yok eB = (1-8) (x Yq (28) (yp) 29=0-Oftt-0)r—(1-9} hq (29) \where the cutis defined here as the ratio of permeate-to-feed flow rates O= 4 / ny (0) Alternative forms of Eqs. (28) and (29) in terms of K, are Yok 0 = 0° (1-9) on (t-y,)kp0=[(1--(-y)],, (2) “The permeate composition at the closed end of the hollow- fiber membranes is calculated from Eq. (19) ‘The experimental separation factor was calculated from the ‘measured compositions of the permeate and rerentate streams Under conditions where the hang inthe feed composi ton exceeds 50% the log mean model ci be applied too more times as necessary across a mole sich that each ct does not exceed a 50% change in , From the previ tp. “The pressure atthe closed end ofthe fiber ore canbe calc lated by assuming thatthe permeate low ate aia une ton of uistance along the fiber ae cs whore Ls the fiber length, Equation (20) can he integrated with substitution from Eq. (34) to give an estimate for the permeate presse athe closed end ofthe fbr” 7 > SRILA, 35) iN oo 1 Solution Method ‘The algebraic model Eqs. (19) andl (27-29) represent a sys- tem with four degrees of freedom, or four equations in eight variables: Xp YypYy 8, O°, Ky, andr. The model was ini- tially calibrated by fixing x, and F and measuring x, and y., leaving y,, 8, a, and K, as unknowns in the solution. ‘The solution of the system of nonlinear algebraic equa tions requires an iterative, trial-and-error technique, such as Newton’s method. The log-mean approximation of the par- tial-prossure driving force is notoriously difficult to converge tunder these circumstances. Fortunately, there are good ap- proximations to the log-mean that avoid problems of diver- gence in the solution. The following form of the Chen ap- proximation was used"! 42—a Falaa! 4s) 69) Floudas noted that the Chen approximation to the log-mean hhas the advantage that it becomes zero if either the feed or cxit partial-pressure driving forees become zero." ‘The four-column configuration requires sequential solution 10 the countercurrent and cocurrent models. Note that n,, = 1,,and K,, = K,, between the frst and second columns, and that,, =m, and K,,= K,, between the third and fourth col- ‘ums. The feed flow rates fo each coluran are calculated from the cut for the previous column. RESULTS AND DISCUSSION ‘The experimental data of Davis and Sandall™ were used to illustrate the analysis procedure. The assumption of Eq, (22) for the log-mean approximation was evaluated by plotting representative numerical solution to Eq, (7), shown in Figure 6A linear least-squares regression of the numerical results Figure 6.[ = Numerical] oa. results to support the] > Iogatithmic-| cay assumptions for| * * x,0.21,) on %,n0.16,| ay, Chemica Enginerin cation( Membranes in ChE Education ) shows that the assumption ofa linear function for A is valid for the conditions of this laboratory experiment. A sample calculation of the single countercurrent flow: ‘model calibration using Mathcad is shown in Figure 7. The experimental data and results ofthe algebraic model are com- pared with the results from the differential model in Table 1 for a” and K,. There ate no significant differences in the results between these models. sna) [414 y oes (1-0)+yp0 x aft) Et yoke «(1 Ohedger=spomt= (rare = 0-8) fae (som Frd{kg.c.ys6)=| aus Figure 7. Example of Mathoad caloulation for Inverse ‘mass transfer analysis using the log-mean model. TABLE 1 Calibration Data" and Results for Single Countercurrent Column xgcimeutl Data ——_‘Dilleentit Model Algebraic Mode el . . Pre) igmais) x,y, | a” x, K, 377 073 018 043 | 581 31] 582 311 grr om ous 043 | 381 str | 382 31 377 103 019 a4s | 598 496 | 598 «96 377 132 019 044 | Sos 496 | 598 496 37254 020 o44 | 571 987 | 571 986 sis o@ ois oas | 593 262 | $97 261 51s 073 046 046 | G02 333 | 60s 332 513 095 017 047 | 612 439 | 614 439 Sis tsi 18 047 | 585 582 | 586 582 5is 225 019 048 | 596 921 | S96 921 oss a7 ora ods | 578 ans | set sia 653 09s 03s oa7 | sae 388 | S88 387 653132 046 a0 | 390 a1 | 593 490 653238 48 ous | 323 a7 | 374 856 oss 34 019 03 | Saross | 381 135 ‘Average 588 3.90 AA linear relationship between the retentate flow rate and K,, is caloulated for use in the remaining three column pre- ble at CONCLUSIONS ‘A membrane experiment for investigating gas separation tas been in use for over ten years in the undergraduate labo ratory atthe University of California, Santa Barbara. A simple analysis method was presented that requires only the solu- tion toa system of four algebraic equations. The simpler analy- sis is equally applicable to newer membrane configurations that introduce the high-pressure feed tothe fibec bores in or- dec to maintain betr flow patterns in the membrane mod- tule. The experimental apparatus was designed to permit single and four-column investigations of ar separation. The single column was used to calibrate the models for binary sa separation. Comparing results for the four-column op: eration validated the calibrated model. Good model and ex- perimental agreement lend confidence inthe model and val date the mode! assumptions. Students are then able o use the model to develop competing designs for gas separation and optimize their designs for maximizing efliiency of separa- tion. The advantages of the simpler approach are that stu- dents can readily set up and solve the model equations without complicated programming, Students are also able to explore alternative designs by building models and comparing the results NOMENCLATURE ‘membrane area, m= diameter, m variable fiber length, m Greek Symbols ‘experimental separation factor A transmembrane partial pressure, KPa viscosity, Noms 0 cutof feed to permeate stream Subscripts/Superseripts © closed end of fiber bore experimental fiber feed closed end of permeate stream log-mean result membrane ‘oxygen predicted resentate dimensionless or ideal parameter speozs REFERENCES. |. Panes, and RS. Chauhan, “Membranes for Gas Separation.” Prog. Poly, Se, 266), $53 2001) 2. McCabe, WLC, Smith. and P Hani. Uns Operations of Chem cal Engineering, Shed, MeGraw-ill, New York, NY (1993) 3, Geankoplis.C. Tniport Processes and Unit Operations, 3d e., Premce Hal, Englewood Cli, NJ (1993) 4. Seater, LD. and EJ. Henley, Separation Process Principles, John Wiley aad Sons, New York. NY (1998) 5. Davis, RA. and OC, Sandell "A Membrane Gas Separation Experi tment forthe Underaradate Laboratory” Chom Eng, Ey 25(1), (0 (1990) 6, Wlavendr, WP, and S.A. Stern, “Analysis of Membrane Separa- tio Parameters. I: Counter Current snd Cocuret Flow in & Single Permeation Stage,” Sep Sei 13), 383(1972) 17. Bowell, . Majumder, ad KK Sitka, “Series Solutions fora Gas Peat with Coutecurent and Coeurent Flow, Ind. Eng. Chem Fund, 23,470 (1984) ‘8 Thundyil. ML. and WJ. Kors, “Mathematical Medeting of Gas Sep: tation Permentrs: For Radiat Crossflow, Couatereurrent, and CCocurent Holow Fiber Membrane Mecils."J. Mem Sei. 125,275 7) 9, Zolantz, RR. and G.K. Fleming, “Design of Gas Permeation Sys tems” in Membrane Hndboot, WS. Ho and KK, Sinko, eds, Van Nostrand Reinbold, New Yor NY, p. 6 (1992) 10, Bruning, WS. “A General Desrpton of Flows ond Pressures in Hollow Fiber Mera Moses," Chom Eng, Si, 446), 14811989) 11, ederapie, Ws, 3. Wiliams, and B.O. Hater, “Cas Flow Dyna {ey n Hollow Fiber Membranes," AIChE J, 427), 2094 (1996) 12 Li 8, XT, and K. Li, "GaslVapour Separation Using Men branes: Elect of Prewure Drop in Lumen of Hollow Fibers" Chee Bag. Se 8, 2681 (2000) 1B, Mulder, M, Basi Principles of Membrane Technology, Kluwer Ace deme Pblisers, Dordrecht, Nethetands (1991) 14, Chen. JJ "Coraments on Improvements on Replace ar he ogitlmie Mean,” Chee Eng. Se 4210), 2488 (1987) 15, Roules, CA. Nonlinear and Mizetnteger Optimization. Oxford University Pres. Oxford (1995) 16, Vlasov, GT. and ML. Miovi, “Pressure Drops and Hyd lie Resistances ina Thrae-Phase Hollow Fier Membrane Contactor vwtn Fame Elements" Chm: Eng. Proc 49,3 (2001) Chemical Engineering Education