英文翻譯：原文.pdf

Eur. J. Mech. B/Fluids 18 (1999) 7837921999 ditions scientifiques et mdicales Elsevier SAS. All rights reservedThree-dimensional mixing in Stokes flow: the partitioned pipe mixer problem revisitedV. V. M e l e s h koa, O.S. Galaktionova;b, G.W.M. Petersb;*, H.E.H. MeijerbaInstitute of Hydromechanics, National Academy of Sciences, 252057 Kiev, UkrainebDutch Polymer Institute, Eindhoven Polymer Laboratories, Eindhoven University of Technology, P.O. Box 513, 5600 MB Eindhoven,The Netherlands(Received 24 April 1998; revised 22 February 1999; accepted 7 March 1999)Abstract The velocity field and mixing behaviour in the so-called partitioned pipe mixer were studied. Starting with the same physical model as inprevious studies, an exact analytical solution was developed which yields a more accurate description of the flow than the previously used approximatesolution. Also, the results are in better accordance with the reported experimental data. 1999 ditions scientifiques et mdicales Elsevier SASStokes flow / laminar distributive mixing / static mixers1. IntroductionThe aim of the present paper is to study the three-dimensional creeping flow in an infinitely long cylindricalpipe with internal walls, that divide the pipe into a sequence of semicircular ducts. Such a system, called thepartitioned pipe mixer (PPM) was introduced by Khakhar et al. 1 as a prototype model for the widely usedKenics static mixer (Middleman 2).In the Kenics mixer each element is a helix, twisted on a 180, plate; elements are arranged axially withina cylindrical tube so that the leading edge of an element is at right angles to the trailing edge of the previousone. Computational fluid dynamics tools make a straightforward numerical simulation of this kind of three-dimensional flow feasible (Avalosse and Crochet 3, Hobbs and Muzzio 4, Hobbs et al. 5). However, suchsimulations do require significant computational resources, especially when studying the effect of varyingparameters on the mixing process. Therefore, simplified analytical models, that give the possibility of fastsimulations of the process (or mimic its features closely enough), are still useful.The PPM model of the essentially three-dimensional flow was highly idealized, nevertheless retaining themain features of the flow under study. The model involves two superimposed, independent, two-dimensionalflow fields: a cross-sectional (rotational) velocity field and a fully developed axial Poiseuille profile in everysemicircular duct. This gives two independent two-dimensional boundary problems instead of the three-dimensional problem. The solution proposed by Khakhar et al. 1 for the cross-sectional velocity field wasonly an approximate one. There exists, however, exact analytical solutions in a closed form.In the present paper we use these exact solutions to examine the mixing properties in this three-dimensionalmixer. Important differences in some mixing patterns were obtained, and our results resemble more closely theavailable experimental results of Kusch and Ottino 6.* Correspondence and reprints: Department of Mechanical Engineering, Building W.h. 0.119, Eindhoven University of Technology, P.O. Box 513,5600 MB Eindhoven, The Netherlands; e-mail: gerritwfw.wtb.tue.nl784 V. V. M el eshko et al .2. Velocity field in PPMConsider the interior of an infinite cylinder 06r6a; 06 62 ; jzj0; 06 6 =2with the constant tangential velocity V applied at the plane D0.Figure 1(a) shows contour levels of the stream function (6). The cross-sectional flow exhibits a single vortexcell with one elliptic stagnation point at (0:636a; =2).The previous studies (Khakhar et al. 1, Ottino 7) suggested the approximate one-term solution of theboundary problem (1), (4):9D4Va3 ra21 rasin2; D.11=3/1=21 0:915; (16)which has been obtained by a variational method. This expression (16), however, does not satisfy both thegoverning biharmonic equation (1) and the no-slip condition at the moving boundary! It turns out that thetangential velocity at the boundary rDa varies as .4=3/V sin2instead of being constant V. Therefore, thevelocity is overestimated (up to 33% at the circular boundary) in some zones far from the flat boundary, and it isartificially smoothed near corners. The contour plot of the stream function according to one-term solution (16)is presented in figure 1(b).The solution of boundary problem (2), (5) for the fully developed axial flow in a semicircular duct reads(Ottino 7):vzD16 28hvzi1XkD1ra2k 1ra2sinT.2k 1/ U.2k 1/f4 .2k 1/2g; (17)wherehvziD8 24 21pza2is the average axial velocity. Using straightforward transformations and tables of infinite sums (Prudnikov etal. 12), we can present expression (17) in a closed form:vzD2 28hvzi(r2a2sin2Craarsin 14r2a2a2r2sin.2 /lnr2C2arcos Ca2r22arcos Ca2C122 r2a2a2r2cos.2 /arctan2arsin a2r2); (18)which is preferable for numerical simulations of the advection process. It is worth mentioning that the firstthree terms of the infinite sum (17) used in Khakhar et al. 1 and Ottino 7 provide reasonable accuracy withEUROPEAN JOURNAL OF MECHANICS B/FLUIDS, VOL. 18,N5, 1999Three-dimensional mixing in Stokes flow 787Figure 2. Contour plots of the axial velocity vz: solid lines correspond to the exact expression (18), dotted lines correspond to three-terms approximationof (17).maximum errors (compared to exact expression (18) that are within a few percent. In figure 2 the contourlines of vz, defined by (18) are shown as a solid lines, while the same contours for three-term approximationof (17) are plotted as dotted lines. Despite this approximation the shape of the contours is rather similar, thediscrepancy amounts up to 7% of the average velocityhvzi, reaching a maximum not far from the corner points,where the velocity vzis underestimated. Increasing the number of terms in (17) to one hundred, reduces therelative error to less then 0:005%, but, it will take much more computer time to simulate the passive tracersadvection.3. Chaotic mixing in PPMThe motion of a passive individual (Lagrangian) particle is described by the advection equationsdrdtDvr.r; /; rd dtDv.r; /;dzdtDvz.r; /; (19)with the velocity field on the right hand side of (19) defined by (6) and (18). The initial conditions arerDr0; D 0;zD0attD0.Here the variable is obviously defined asD8:; 2kL6z.2kC1/L; 06 6 ; 2kL6z.2kC1/L; 2 ;=2;.2kC1/L6z.2kC2/L; =26 63 =2;C =2;.2kC1/L6z.2kC2/L; 06 =2;3 =2;.2kC1/L6z.2kC2/L; 3 =2 2 ;(20)where kD0; 1; 2;:System (19) describes a steady motion of an individual particle along the streamline in each compartment.However, as the flow is three-dimensional and spatially periodic, it can exhibit chaotic behaviour (Aref 13,Section 5.4).In Khakhar et al. 1 the single non-dimensional parameter , the mixing strengthD4VL3 hvzia; (21)EUROPEAN JOURNAL OF MECHANICS B/FLUIDS, VOL. 18,N5, 1999788 V. V. M el eshko et al .was introduced to completely describe the behaviour of such a system. Although the parameter has noparticular meaning for the exact solution (6), the value of is used to compare our results with those of theliterature.Poincar mapping was applied to reveal the zones of regular and chaotic motion. The Poincar maps wereconstructed by taking an initial point .r0; 0/at the levelzD0 and recording the coordinates of the intersectionsof the trajectory with the planes znD2nL; nD0; 1; 2;:The Poincar maps for several values of were computed and analysed using both the approximate andexact solution. Here we present the resulting Poincar maps for which one single starting point was chosen inthe chaotic zone (figure 3). White regions in the plots correspond to islands. The boundaries of the islands areplotted as thin solid lines.Islands in Poincar maps correspond to the KolmogorovArnoldMoser (KAM) tubes in the flow. The fluidcaptured in such a tube will only travel inside, not mixing with the rest of the fluid outside the tube. Theinfluence of the KAM tube on mixing can be characterized by the relative flux carried by the tube compared tothe total flux through the mixer. So, for the islands both their area and the flux carried by corresponding KAMtubes are evaluated. The flux can be computed as the integral of vzover the islands area, or, by using Stokestheorem, as a contour integral over the boundary of islands.Figures 3(a) and 3(b) present the Poincar maps for D4. For the approximate solution the eight largestislands are clearly seen (figure 3(a). They occupy about 49% of the cross-section area and carry approximately55% of the total flux. The exact solution provides a completely different system of islands (figure 3(b). Theirinfluence is considerably lower since they occupy only about 13% of the area and bear 18% of the total flux.The difference becomes even stronger for larger values of the mixing strength . Figures 3(c) and 3(d)represent the case of D8. The approximate solution provides two large islands that occupy about 13% ofthe cross-section (see figure 3(c) and bear 18% of total flux, while the islands revealed by the exact solution(figure 3(d) occupy only about 0:7% of the cross-section area. The relative flux through KAM tubes amountsin this case to approximately only 1% of total flux.In both examples presented the total area of the cross-section of the KAM tubes is significantly smaller whenthe exact solution is used. As both the approximate and exact solutions are based on the same simplified modelof the PPM, i.e. neglecting the transition effects at the joints of the mixer elements, the calculated shape of theKAM tubes should be considered with some reservations. The relative cross section of, and the relative fluxthrough these tubes are of more relevance and they can give an useful estimation of these values for practicalflows.Streaklines can serve as a tool to characterise the mixing and to visualise underlying mixing mechanisms.Kusch and Ottino 6 noted that computed streaklines, originating from a cross-section of a KAM tube, aremuch different from those experimentally observed. Computed streaklines for D8:0 and the experimentalresults obtained for D10:0 0:3 were compared to get, at least, some resemblance. They pointed out that thePPM model can hardly mimic closely the experimental results (due to the small length of dividing plateslessthan the pipe radius). However, the results of numerical simulations using the corrected velocity field (6), (18)and the right value for gives a much better agreement. Figures 3(e) and 3(f) show the Poincar maps forD10, using both solutions. In figure 3(f) the approximate contours of the two islands of period 2 are plottedwith solid lines. These contours were used to reveal the shape of the correspondent KAM tubes (see figure 4(c).Contours were represented by closed polygons and the vertices of these polygons were then tracked numericallythrough four mixing elements, showing the outer boundary of the KAM tube. The other two images in figure 4represent the numerical (a) and experimental (b) results from Kusch and Ottino 6, respectively. As for theexperimental results the actual value of mixing strength was D10:0 0:3, we calculated the KAM tubeEUROPEAN JOURNAL OF MECHANICS B/FLUIDS, VOL. 18,N5, 1999Three-dimensional mixing in Stokes flow 789(a) (b)(c) (d)(e) (f)Figure 3. Poincar maps for different values of mixing strength D4 (a) and (b), D8 (c) and (d), D10 (e) and (f), respectively. Pictures inthe left column (a), (c), (e) were obtained by using approximate solution (16), (17), while those in the right column were obtained by using the exactsolution (6), (18).EUROPEAN JOURNAL OF MECHANICS B/FLUIDS, VOL. 18,N5, 1999790 V. V. M el eshko et al .(a) (b) (c)Figure 4. Computed KAM tubes for the PPM model with mixing strength parameter D10:0 (c) compared with (a) computed ( D8) and (b)experimental ( D10:0 0:3) streaklines from Kusch and Ottino 6. (Images (a) and (b) are taken from figure 9 of the cited paper, reproduced withpermission from Cambridge University Press.)shapes for the limiting values D9:7and D10:3 as well. The overall shape of the tubes does not changemuch, variation of mixing strength influences mainly the tube thickness: it is thinner for larger parameter andvice versa.Kusch and Ottino 6 did not specify explicitly the location where the dye for streakline visualization wasinjected. However, it is easy to show that when the dye is injected just a little outside the KAM tube, this isclearly visible because the dye starts to spread over the mixing elements. To illustrate this, circles were drawnaround the geometrical center of the island (see figure 3(f). Markers were evenly distributed on the boundaryof every circle and tracked through four mixing elements (two spatial periods) of the PPM. In figure 5(a) theradius of the circle was 0:03a, thus all markers were positioned well inside the KAM tube. In figure 5(b) thecircle (of radius 0:062a) touches the tube boundary. Such streaklines can be slightly deformed but are stillcaptured completely within the tubes. In figure 5(c) the initial circle was slightly larger then the island shownin figure 3(f), and thus contains markers outside the KAM tube. It is clearly seen that within just four mixingcells the markers spread over the whole cross-section of the pipe.The use of approximate numerical solution (16), (17) led Kusch and Ottino 6 to a great discrepancy withexperimental results for 10 40: experiments showed remarkably stable KAM tubes, while computationsexhibited a lot of bifurcations (see, for example, figure 10(d) from their paper). However, using the exactsolution (6), (18) relatively simple stable structures are predicted. For example, for a relatively large mixingstrength of D20, four KAM tubes of first order were found but no KAM tubes of period 2 were detected.EUROPEAN JOURNAL OF MECHANICS B/FLUIDS, VOL. 18,N5, 1999Three-dimensional mixing in Stokes flow 791(a) (b) (c)Figure 5. Traces of the markers, originally regularly spaced on circles of different radii, centered around the geometrical centers of the islands of period 2.Each circle contains 100 markers. The radii are: (a) 0:03awell inside the KAM tube, (b) 0:062atouching its boundary, (c) 0:08acircumscribingthe tube boundary.The cross section of these tubes (and, consequently, the flux associated with them) is relatively small. Theseperiodical structures are, nevertheless, stable.4. ConclusionsAlthough the flow under study is merely a prototype flow, it possesses some important features of flows inwidely used mixing devices. The comparison of an approximate and an exact solution, obtained within theframework of the same model, shows the possible major consequences of some mathematical simplifications.Such simplifications can cause large differences in the predicted systems behaviour, especially for systemsthat are supposed to exhibit chaotic properties. Here, the difference in the predicted behaviour was caused bythe use (in previous studies) of a one-term approximate solution that artificially smoothes the cross-sectionalvelocity field. The exact solution shows much better agreement with the reported experimental results.Of course, there exists an important problem regarding the abrupt transition between mixing elements andignoring developing flows at these transitions. Results of recent numerical simulations (Hobbs et al. 5) showthat, indeed, this is a major assumption: for the Kenics mixer with a finite thickness of helical screwed mixingplates, flow transitions at the abrupt entrance and exit of each element strongly affect the velocity field over upto one quarter of the element length.However, the conclusion from the results presented of the importance of an accurate description of thevelocity field in mixing flows, where even small changes can significantly alter the overall mixing behaviour ofthe system, is still applicable for real industrial situations.EUROPEAN JOURNAL OF MECHANICS B/FLUIDS, VOL. 18,N5, 1999792 V. V. M el eshko et al .AcknowledgementsThe authors would like to acknowledge support by the Dutch Foundation of Technology (STW), grant no.EWT44.3453. We also thank one of the referees for expressing the opinion that unenlightened use of thecomputer or uninformed parametrizations can lead one to nonsensical results.References1 Khakhar D.V., Franjione J.G., Ottino J.M., A case study of chaotic mixing in deterministic flows: t