Anda di halaman 1dari 21

599 Supplemental Material

600 1. Refined near-interface fluid transport property blending model


601 In the base OpenFOAM interFoam solver [21], and many earlier VOF simulation studies [15], [16], [36],
602 1-weighted arithmetic averaging of liquid- and vapor-phase values is employed to approximate fluid
603 transport properties (, k) in the vicinity of the interface, where 1 ranges between 0 and 1.

604 = + 1 for = , , (S1)


605 This approach can yield accurate results if near-interface transport resistances are small compared with
606 overall resistances in the flow. For example, for film condensation (see Section 3.1), the majority of
607 thermal resistance is due to conduction through the many cells spanning the liquid film. The diffuse
608 blending of liquid and vapor thermal conductivity in the 3-4 cells in the interface region only affects the
609 solution slightly, and the effect diminishes with increasing mesh resolution. However, for cases with thin
610 thermal and momentum boundary layers near the interface (e.g., rising bubble condensation, Section
611 3.2), careful evaluation of and k is needed [49].

612 In the developed solver, a runtime selectable switch is provided that applies more precise blending of
613 and k on cell faces (denoted with f subscript), depending on the relative orientation of the interface
614 ( = ) to the cell faces ( ). This accounts for the fact that the liquid and vapor transport
615 resistances in a cell can act in series or parallel, depending on the orientation of the interface. The
616 formulation of Marschall et al. [49] is applied for viscosity blending.

! !
617 , = , ,
,
+ ,
# +$1 , , %& , + $1 , % ' (S2)
"

618 A similar model is derived for thermal conductivity on cell faces

! !
619 , = , ,
,
+ ,
# + $1 , , %& , + $1 , % ' (S3)
( ("

620 These models are evaluated using the rising bubble condensation case described in Section 3.2. In this
621 problem, a thin boundary layer forms near the bubble nose, making accurate evaluation of transport
622 properties near the interface critical. Here, the bubble diameter is 0.23 mm, fluid properties are those
623 listed in Section 3.2, and the phase-change source terms )*+, and * ,+, are turned off to facilitate
624 comparison with analytical and single-phase simulation results. Simulation results are averaged from t =
625 0.10 to 0.15 s after startup. Coarse (first radial cell r = 11.0 m) and fine mesh case (first r = 3.8 m)
626 results for bubble rise velocity and average heat flux are presented in Table S1 with the arithmetic
627 averaging and improved transport property blending models.

628 The analytical bubble rise velocity model of [58], [59]1 (Eqn. S4) and condensation heat flux model of
629 [54] (Eqn. 11) are presented for comparison.
4
01232 5 !5"
630 -./. = (S4)
6

1
Refs. 58 66 are provided in Supplemental Material Section 9: Supplemental Material References

21
631

632 Additionally, heat flux results are presented from a high-resolution steady-state single-phase liquid
633 simulation, assuming an isothermal (Tsat) slip condition at the bubble interface. The single-phase
634 comparison study is evaluated using COMSOL [20] with specified liquid inlet velocities. The COMSOL
635 mesh has 15 thin layers of cells around the bubble to resolve the momentum and temperature
636 boundary layers. Four simulations were performed in COMSOL with varying mesh resolution. The first
637 boundary layer cell thicknesses for these cases were: 0.092, 0.067, 0.057, and 0.043 m. For an inlet
638 velocity of 0.0342 m s-1 (Rebub = 3.3), the resulting average bubble heat fluxes were 9657.7, 9646.3,
639 9643.1, and 9642.3 W m-2. The close agreement between the three finer mesh cases (within 0.04%)
640 indicate converged results. The mesh with r= 0.057 m is presented in Fig. S1. The heat flux results for
641 the r= 0.057 m mesh are presented with average VOF simulation bubble rise velocities in Table S1.

642
643 Figure S1 a. Mesh used for single-phase simulation of rising bubble heat transfer, b. Detail view of
644 near-bubble interface region

645 Table S1 VOF simulation results for bubble rise velocity and heat flux with the base and refined fluid
646 transport model blending models. Results are compared with single-phase liquid simulation results and
647 analytical correlations (relative VOF simulation errors in parentheses).
Bubble Rise VOF Heat Single Phase Analytical
VOF Case Velocity Flux Simulation Heat Flux Correlation Heat
(Ubub, m s-1) (W m-2) at Ubub (W m-2) Flux (W m-2)
Arithmetic average fluid transport property blending
r = 11.0 m 0.0218 (-36%) 11,491 8,525 (+35%) 8,174 (+41%)
r = 3.8 m 0.0400 (+17%) 12,134 10,091 (+20%) 9,514 (+28%)
Refined fluid transport property blending
r = 11.0 m 0.0402 (+18%) 10,796 10,106 (+7%) 9,527 (+13%)
r = 3.8 m 0.0367 (+7%) 10,578 9,843 (+7%) 9,298 (+14%)

22
Hadamard [58] and
0.0342 - 9,646 9,125
Rybczynski [59]
648
649 For the coarse VOF mesh, arithmetic averaging of fluid viscosity near the interface results in significant
650 under prediction of the analytical bubble rise velocity (0.0218 m s-1 vs. 0.0342 m s-1, -36% error). The
651 corresponding fine mesh case with yields closer agreement (0.0400 m s-1, + 17%). Improved agreement
652 is found with the refined transport property blending model for both cases: +18% and +7% error,
653 respectively. Similarly, the base transport property blending model significantly over predicts the single-
654 phase simulation heat flux with the VOF result bubble rise velocities: +35% error (coarse mesh) and 20%
655 error (fine mesh). With the refined transport blending model, the relative heat flux errors are 6.8%
656 (coarse mesh) and 7.5% (fine mesh). These results indicate that the refined transport property blending
657 models can improve VOF simulation accuracy for problems with strong near-interface transport effects.

658

659 2. VOF Discretization of near-interface temperature gradient discontinuities


660 The adopted VOF method without geometric reconstruction yields a diffuse interface. As a result, sharp
661 discontinuities in flow fields (temperature, pressure, velocity, etc.) are spread out over 3-4 cells.
662 Accurate overall flow results can still be obtained with sufficient mesh resolution near the interface. This
663 is demonstrated for two simulation cases: smooth falling-film evaporation (see Section 3.1) and rising
664 bubble condensation (Section 3.2).

665 The smooth falling-film evaporation case is identical to the falling-film condensation case described in
666 Section 3.1, except that the wall temperature is greater than the saturation temperature (Tw = Tsat + 1).
667 Two meshes are considered with average film-region cell sizes of x = 10.1 m and 6.0 m in the wall
668 normal direction, respectively. The wall-normal direction temperature profiles at t = 0.1 s are presented
669 in Fig. S2. The theoretical solution of Nusselt predicts a linear temperature profile in the film (from Tw at
670 the wall to Tsat at the interface), and isothermal behavior in the vapor bulk (at Tsat). This behavior is
671 found in the simulation results for both cases. The discontinuity in the temperature gradient at the
672 interface is captured within ~3 cells in both meshes. The transition is visibly sharper with the fine mesh
673 (Fig. S2). This result indicates that the employed VOF formulation can capture temperature gradient
674 discontinuities due to phase change, provided that the mesh is sufficiently fine near the interface.

675

676

23
677

678 Figure S2 Wall-normal-direction temperature profile for smooth falling film evaporation study with
679 coarse (x = 10.1 m) and fine (x = 6.0 m) meshes. Detail view of interface region showing sharper
680 resolution of the discontinuous temperature gradient at the interface with the fine mesh.

681 In a second study, the temperature profiles near the interface of a rising condensing bubble are
682 assessed. Axisymmetric results obtained using the developed VOF code are compared with high-
683 resolution steady-state single-phase liquid simulation results (described in Supplemental Material
684 Section 1). The VOF case employs the geometry, and fluid conditions described in Section 3.2, with no
685 phase change or volume dilatation source terms to facilitate comparison. The VOF mesh is refined near
686 the central axis, with an inner cell r = 5.1 m (bubble radius = 230 m). The simulation is initiated from
687 rest (initial Tliquid = Tsat - 1 K, Tvapor = Tsat), and results are averaged from t = 0.1 0.15 s. The single-phase
688 comparison case is evaluated using COMSOL with an inlet liquid velocity of 0.0356 m s-1 (the average rise
689 velocity from the phase-change simulation). The COMSOL mesh has 15 thin layers of cells around the
690 bubble to resolve the momentum and temperature boundary layers. The first boundary layer cell
691 thickness is 0.057 m, which was found to yield converged results for this case in Supplemental Material
692 Section 1.

693 For this case, the COMSOL study gives an average interface heat flux of 9,752 W m-2. The OpenFOAM
694 simulation yields an average heat flux of 10,453 W m2 (7% greater). Both values are higher than the
695 value of 9,217 W m-2 obtained with the analytical correlation of Ranz and Marshall [54] (Eqn. 11, here Re
696 = 3.3). The axial temperature profiles through the bubble nose and tail and radial temperature profile
697 through the bubble center are presented in Fig. S3 from the VOF and COMSOL studies. The greatest
698 temperature deviation is observed closest to the bubble nose (0.15 K) where the thermal boundary layer
699 is thinnest. Here, the single-phase simulation mesh is about 100 finer than the VOF mesh, and can thus
700 resolve the temperature profile more accurately. In the tail region, the temperature difference peaks at
701 0.07 K near the bubble, and reduces to about 0.03 K in the wake. On the side of the bubble, the
702 temperature difference reaches a maximum value of 0.13 K. In the nose and side regions, the thermal
703 boundary layer thicknesses are comparable (not fully captured in tail region). Considering that the
704 single-phase study mesh is much finer, these results indicate reasonable agreement. Overall, this study

24
705 demonstrates the ability of the solver to capture discontinuous temperature gradients and relatively
706 sharp temperature variations resulting from boundary layers near the liquid-vapor interface.

707

708
709 Figure S3 a. Schematic of bubble configuration and coordinate system, b. Comparison of VOF study
710 and single-phase liquid study axial temperature profiles (along z, r = 0), c. Comparison of VOF and single-
711 phase liquid radial temperature profiles (along r, z = 0)

712

713 3. Expanded Descriptions of Phase Change Models


714 InterfaceEquilibrium This is an updated version of the formulation of Rattner and Garimella [28].
715 In this model, localized phase change thermal terms are applied to cells that contain the liquid vapor
716 interface. The source terms are specified such that interface cells return to the saturation
717 temperature at the end of each time step (i.e., 7* +, = 89+ ;<). Because this model determines
718 the phase change rate based on the simulation time step size, it was demonstrated to yield mesh-
719 independent results without tuning a rate parameter, as in the formulations of [15], [24], [25].
720 Interface-containing cells are identified by a simple graph-scan over the mesh, which has a low
721 computational cost compared with methods that geometrically reconstruct the interface. This
722 approach yields a two-cell thick layer of interface cells that straddle a threshold value of 1. The
723 numerical interface thickness thus decreases proportionately with increasing mesh resolution.
724 Different thresholds can be specified for condensation (higher 1) and evaporation (lower 1), which
725 has been shown to reduce numerical interface diffusion. As the interface detection method only
726 operates on cell connectivity, it can also be applied to unstructured meshes [28]. The model also
727 supports a user-specified under-relaxation factor, which can improve numerical stability.

728 In [28], this formulation was demonstrated to yield solutions for the Stefan problem that converged
729 to the exact analytical solution for film thickness with increasing mesh resolution (error proportional
730 to y0.84). Additionally it was found to match the Nusselt solution for smooth falling film heat flux to
731 within 2% for 50 Re 1600. In [31], a similar formulation was shown to yield heat transfer
732 coefficients within 8% of analytical correlation results for film boiling, and within 20% of the Nusselt

25
733 solution for film condensation. Onishi et al. [31] also successfully applied this formulation to model
734 the operation of a looped thermosiphon.

735 InterfaceEquilibrium_SplitDilatation This model is a modified version of the above model


736 motivated by the approach of Hardt and Wondra [30]. Here the liquid- and vapor- portions of the
737 dilatation source term ( v& pc ) are separately, and iteratively shifted by a few cells thicknesses to the
738 corresponding sides of the interface [60]. Specifically, the liquid and vapor components of v& pc are
739 transported between neighboring cells based on the relative dot product of the phase-fraction
740 gradient ( ) and cell-face normal vectors. This procedure is performed over four passes in each
741 time step, ensuring that the v& pc field acts away from the region of high magnitude. While this
742 process increases the effective interface thickness, it reduces smearing of the interface due to
743 application of dilatation on interface cells [60]. As in the InterfaceEquilibrium model, accurate phase
744 change results can be obtained with a sufficiently fine mesh near the interface.

745 In the present study, this formulation has been shown to yield linear convergence to the analytical
746 solution for film thickness for the Stefan problem (error proportional to y1.05, Supplemental
747 Material Section 4.1) and agreement to within 2% of the Nusselt solution for smooth falling film
748 condensation and evaporation heat flux over a range of mesh resolutions (Supplemental Material
749 Section 4.2). Good heat flux agreement has also been obtained with a simple analytical model for
750 small rising bubble condensation (within 9%, Supplemental Material Section 4.3). The formulation
751 was previously applied to study Taylor flow evaporation, and was used to estimate the magnitude of
752 the wake-region heat transfer enhancement effect [60].

753 InterfaceEquilibrium_NoDilatation This is a modified version of InterfaceEquilibrium model in


754 which the dilatation rate v& pc is set to 0. This model is specifically suited for cases where the phase-
755 change momentum effects are negligible. For example, in most film condensation analyses, the
756 momentum transfer due to vapor-impingement on the interface is neglected. This formulation can
757 improve numerical stability because it partially decouples the momentum and thermal energy
758 equations. Additionally, it can relax CFL time step constraints for cases where the impingement
759 velocity of condensing vapor exceeds the characteristic flow velocity. This approximation has been
760 successfully employed in many prior computational studies of phase change phenomena [16], [24],
761 [30], [35].

762 In the present study, this formulation demonstrated reasonable heat flux agreement with high-
763 resolution single-phase simulations of fixed volume rising bubble condensation (within 7.5% over a
764 range of grid resolutions, Supplemental Material Section 4.3). It has previously been applied to
765 laminar-wavy falling-film condensation [28], and heat flux results were found to lie within the range
766 of available empirical correlations for 75.4 < Re < 1447.8.

767 EmpiricalRateParameter The model of Yang et al. [15] applies a user specified empirical
768 parameter to determine the rate of phase change (Eqn. 1). The rate parameters rL and rV (see Eqn. 1)
769 represent the thermal time constant of mesh cells, and thus must be tuned for different studies and
770 meshes [28]. This approach applies phase change source terms throughout the domain, and not just

26
771 in a narrow interface region, as in the InterfaceEquilibrium formulation. However, if the rate
772 parameter is tuned correctly, phase change effects are confined to a relatively thin region near the
773 interface [27]. This formulation is attractive because of its relative simplicity and because the diffuse
774 phase-change source terms yield relatively stable results. However, very high values of rL and rV can
775 cause oscillatory temperature fields [15]. Many phase-change simulation studies have been
776 published using this model, including those of [23][25], [61][63].

777 Fang et al. [24] adopted this formulation to study vapor venting through a membrane in a 3D
778 microchannel geometry using VOF approach. They observed that the liquid velocity profile matched
779 very well with the analytical solution along the entire geometry. They concluded that the model can
780 accurately predict the capillary transport dynamics within a porous structure. In another study,
781 Alizadehdakhel et al. [61] utilized this model to investigate two-phase liquid-gas mixture and co-
782 current condensation/evaporation within a thermosyphon. They also conducted experimental
783 studies at corresponding conditions to verify the computational results. They reported that the
784 numerical predictions for temperature profile in the thermosiphon agreed well with experimental
785 observations. Wilson et al. [27] validated a formulation based on this approach for the Stefan
786 problem, and applied the methodology to simulate solvent injection into a heated porous medium.

787 noPhaseChange This placeholder model disables phase-change source terms, yielding sensible-
788 only heat transfer results. This model is useful for debugging case setups, and can help determine
789 whether phenomena are due to phase-change effects or other causes.

790

791 4. Mesh sensitivity and validation studies


792 Three mesh sensitivity studies are performed to evaluate the convergence behavior and accuracy of the
793 solver and phase change models. These cases can be accessed from the solver package by running the
794 command: git checkout MeshSensitivity.

795 Horizontal film condensation


796 In the horizontal film condensation problem (Stefan problem), a liquid film condenses and grows on an
797 isothermal surface at Tw < Tsat. The analytical solution for film thickness at time t after startup is:

! /6
( E "
798 => < = ?2< A5B D A6 + B D K (S5)
C C, FGHI !FJ

799 This study was performed with the fluid properties (isobutane at 25C) and operating conditions
800 summarized in Table S2.

801

802

803

804

27
805

806 Table S2 Fluid properties and case parameters for the horizontal film condensation case
Fluid Property Liquid Value Vapor Value
Dynamic viscosity (, kg m-1 s-1) 1.5 10-4 7.7 10-6
Density (, kg m-3) 550.6 9.1
Thermal conductivity (k, W m-1 K-1) 0.089 0.017
Specific heat (cp, kJ kg-1 K-1) 2.45 1.82
Surface tension (, kg s-2) 0.01
Enthalpy of phase change (iLV, kJ kg-1) 329.4
Other Case Parameters
Wall temperature subcool (Tsat TL, K) 5
807

808 A 1-D domain is employed with a subcooled horizontal lower wall (Tsat - 5 K), and an open boundary 1.0
809 mm above the wall. The domain is initially set to all vapor with T = Tsat. The condensation process is
810 simulated for 1.0 s, with logging performed every 50 simulation time steps (< 0.8 ms in all cases).
811 Meshes are refined near the wall with the closest cell thickness being 1/3 of the furthest cell values.

812 Five cases are evaluated with cell counts increasing by factors of 1.3 from 77 to 220 cells. Film thickness
P
813 growth for these cases is presented in Fig. S4a. Integrated film thickness error (M = NT |=P Q => |R<)
814 is presented in Fig. S4b for these cases. A least squares fit to these errors yields M = 0.36 .TY for
815 average cell thickness . This result indicates approximately linear convergence to the exact solution
816 with increasing mesh resolution.

817

818 Figure S4 a. Film thickness growth from VOF simulation with varying mesh resolution. Analytical
819 solution presented for reference. b. Convergence of integrated film thickness error to 0 with increasing
820 mesh resolution.

821

822 Smooth falling film condensation and evaporation


823 The smooth falling film condensation problem described in Section 3.1 is also employed for mesh
824 sensitivity analysis. The domain setup, boundary and initial conditions, fluid properties, and post-

28
825 processing procedure described in Section 3.1 is employed. Studies are performed for both
826 condensation (Tw = Tsat 5 K) and evaporation (Tw = Tsat + 5 K) with film-region wall-normal-direction cell
827 sizes increasing by a factor of 1.3 from 3.6 m to 13.2 m. Time averaged wall-heat-flux values and
828 discrepancies with the analytical solution of Nusselt [1] are presented in Table S3.

829 Table S3 Average wall heat flux values and relative deviations from the Nusselt solution for smooth
830 falling film condensation and evaporation
Mesh Resolution Condensation Evaporation
(m) Heat flux (W m-2) Relative Error Heat flux (W m-2) Relative Error
13.2 16,960 +0.3% 17,179 +1.0%
10.1 16,801 -0.7% 17,147 +0.7%
7.8 16,721 -1.2% 17,205 +0.9%
6.0 16,643 -1.7% 17,115 +0.4%
4.6 16,578 -2.1% 17,110 +0.2%
3.6 16,857 -0.4% 17,079 +0.1%
831
832 In all cases, the relative errors are less than 2.1%. This indicates that the solution has little sensitivity to
833 mesh resolution, with the average wall heat flux only varying by ~2.5% over a four-fold change in cell
834 size. Wall-heat-flux results oscillate slightly with grid resolution. This may be due to the precise location
835 of the interface position within the mesh cells, because the selected phase change model
836 (InterfaceEquilibrium_SplitDilatation) assumes negligible thermal resistance in the two layers of cells
837 containing the interface. The condensation heat flux values are slightly lower than evaporation values
838 due to the corresponding increase in film thickness and thermal resistance along the domain length.

839 Rising bubble condensation


840 Finally, a mesh sensitivity study is performed for the rising bubble condensation case described in
841 Section 3.2. A series of studies are first performed using the InterfaceEquilibrium_NoDilatation phase
842 change model (without )*+, and * ,+, source terms) to facilitate comparison with analytical and single-
843 phase simulation results for steady fixed-size bubbles. The innermost radial cell size is increased by a
844 factor of 1.3 between cases, from 3.8 m to 11.0 m. Heat transfer coefficient results are compared
845 with those from the single-phase liquid heat transfer simulation described in Supplemental Material
846 Section 1 and the analytical model of Ranz and Marshall [54] at time-averaged VOF study rise velocities
847 (Table S4).

848
849
850
851
852
853
854

29
855 Table S4 VOF simulation results for bubble rise velocity and heat flux without )*+, and * ,+, source
856 terms. Results are compared with single-phase liquid simulation results and the analytical correlation of
857 Ranz and Marshall [54] (relative VOF simulation errors in parentheses).
Bubble Rise VOF Heat Single Phase Analytical
VOF Inner Radial
Velocity Flux Simulation Heat Flux Correlation Heat
Cell Width (r, m)
(Ubub, m s-1) (W m-2) at Ubub (W m-2) Flux (W m-2)
11.0 0.0402 10,796 10,105 (+6.8%) 9,527 (+13.3%)
8.5 0.0338 10,132 9,612 (+5.4%) 9,163 (+10.6%)
6.5 0.0339 9,661 9,619 (+0.4%) 9,105 (+6.1%)
5.0 0.0356 10,453 9,751 (+7.2%) 9,220 (+13.4%)
3.8 0.0367 10,578 9,840 (+7.5%) 9,298 (+13.8%)
858

859 These results indicate that, over a ~3 range of mesh resolutions, the average VOF bubble rise velocities
860 and heat fluxes vary by approximately 10% and 6% of the mean value, respectively. This somewhat
861 oscillatory behavior with varying mesh resolution has been reported previously, and can be attributed to
862 the interFoam interface compression scheme [56]. These results provide an estimate for the grid
863 sensitivity of the solver for this particular phase change problem. Overall, the bubble rise velocity agrees
864 reasonably well with the analytical solution of Hadamard [58] and Rybczynski [59] (0.0342 m s-1). VOF
865 simulation heat fluxes match the high-resolution single-phase results relatively closely (<7.5% error).
866 Reasonable heat flux agreement is found with the correlation of Ranz and Marshall [54] (<14% error).

867 A series of grid sensitivity studies are also conducted with )*+, and * ,+, source terms turned on, as in
868 Section 3.2 (using the InterfaceEquilibrium_SplitDilatation model). For these cases, results are averaged
869 from t = 0.05 0.15 s after startup, and time averaged heat flux deviations from the model of Ranz and
870 Marshall [54] are reported (evaluated at instantaneous VOF study velocities) (Table S5).

871 Table S5 VOF simulation results for bubble rise velocity and heat flux with )*+, and * ,+, source terms.
872 Results are compared with the analytical correlation of Ranz and Marshall [54]
VOF Heat Analytical Time Averaged
VOF Inner Radial
Flux Correlation Heat Absolute Relative
Cell Width (r, m)
(W m-2) Flux (W m-2) Error
11.0 10,490 10,805 8.9%
8.5 10,211 10,723 7.0%
6.5 10,368 10,612 5.7%
5.0 10,817 10,826 5.5%
3.8 11,497 11,219 5.5%
873

874 Relatively close agreement with the analytical model is obtained for these cases with changing bubble
875 volume (<9% in all cases). As in the fixed bubble-size studies, the time-averaged heat flux oscillates
876 somewhat with mesh resolution. Again, these results provide an estimate of the degree of mesh
877 sensitivity for this phase change process, with a heat flux range of about 12% of the mean value for the
878 considered cases.

30
879 5. Verification study: Effects of spurious currents due to surface tension force discretization
880 In interface capturing simulations, the discretization of the surface tension force is known to modify the
881 velocity field near the interface. These spurious or parasitic currents arise due to force imbalances
882 from the theoretically sharp capillary force being distributed over multiple cells containing the diffuse
883 interface. This effect has been extensively studied for adiabatic VOF solvers [39], [56], [64], [65], and a
884 number of improved surface tension force implementations that do not require interface
885 reconstruction have been proposed to reduce this effect (e.g., [42], [46]). These artificial velocity field
886 effects are known to scale with relative surface tension force strength. In heat transfer processes,
887 spurious currents can modify convective transport near the interface, and are of particular concern for
888 low-velocity, high Prandtl number cases (Pr = 9\ ). For such conditions, transport due to spurious
889 currents may be significant compared with relatively weak conduction.

890 To evaluate this effect, a study is performed for the model problem of phase-change heat transfer for a
891 2D droplet of one fluid (diameter D = 50 m) in a square container of a second fluid (side length W = 100
892 m, Fig. S5a). To isolate parasitic current effects, gravitational acceleration is set to zero, and material
893 properties are matched for both fluids ( = 1000 kg m-3, = 0.001 kg m-1 s-1, k = 0.1 W m-1 K-1). The latent
894 heat of phase change is 100 kJ kg-1, and surface tension (), specific heat (cp and thus Pr), and the mesh
895 resolution are varied between cases. The InterfaceEquilibrium phase change model is employed with no
896 phase-fraction generation or phase-change dilatation source (i.e., latent heating only). Studies are
897 performed using the default OpenFOAM implementation of the surface tension force model of [40] and
898 the Sharp Surface Tension (SST) force model of [42]. The default surface tension force model is known
899 to yield relatively large spurious currents. The SST model is employed with an interface sharpening
900 factor of Cpc = 0.5, a thresholding factor to eliminate surface tension forces on mesh faces below 10-5 of
901 the mean non-zero value, and 100% filtering of surface tension force components tangential to the
902 interface (definitions in [42]).

903 One quadrant of this domain is simulated, with symmetry planes specified on the interior boundaries,
904 and no-slip zero-normal-gradient pressure constraints on the outer walls. The walls are set 1 K below
905 Tsat. The domain is meshed with square cells (Nx = Ny = 20, 30, or 45 in tested cases). The temperature
906 field is initialized to Tsat in the droplet and Tsat 1 K in the surrounding fluid. The velocity field is
907 initialized to zero. Time steps are limited to ensure a maximum Courant number of 0.1 and a maximum
908 grid Fourier number (conduction time scale) of 0.4. The studies are evaluated for 0.04 s, and total latent
909 heating rates (Qpc,sim) and volume-averaged velocity magnitudes (Uavg,sim) are averaged from t = 0.02
910 0.04 s. The latent heating rate is compared with the steady state theoretical value: ]+,,> =
911 ;2^ln 1.08bc =0.204 W m-1. The theoretical average velocity should be zero, as flow only arises
912 from spurious currents. Results are summarized in Table S6.

913

914

915

916

31
917 Table S6 Summarized results from spurious current heat transfer study

Qpc,sim (W m-1), (% absolute error) Uavg,sim (m s-1)


(kg s-2) Pr
Nx,y = 20 Nx,y = 30 Nx,y = 45 Nx,y = 20 Nx,y = 30 Nx,y = 45
Sharp Surface Tension force model (based on [42])
0.010 10 0.207 (1.7%) 0.211 (3.5%) 0.205 (0.4%) 1.66 10-5 2.46 10-5 3.68 10-5
0.010 50 0.212 (4.0%) 0.213 (4.5%) 0.205 (0.7%) 1.73 10-5 2.71 10-5 3.29 10-5
0.100 10 0.210 (3.1%) 0.209 (2.6%) 0.204 (0.2%) 5.01 10-5 4.49 10-5 3.06 10-5
0.100 50 0.216 (6.0%) 0.211 (3.3%) 0.205 (0.6%) 4.45 10-5 3.96 10-5 4.64 10-5
Default OpenFOAM implementation (based on [40])
0.010 10 0.219 (7.5%) 0.216 (5.8%) 0.209 (2.5%) 3.69 10-3 3.14 10-3 5.20 10-3
0.010 50 0.235 (15.3%) 0.222 (8.9%) 0.223 (9.1%) 3.69 10-3 3.14 10-3 5.20 10-3
0.100 10 *0.145 (28.9%) 0.217 (6.5%) *0.169 (17.0%) *2.82 10-2 9.09 10-3 *3.01 10-2
0.100 50 *0.315 (54.7%) 0.224 (9.9%) *0.260 (27.6%) *2.81 10-2 9.09 10-3 *3.00 10-2
918 *In these cases, spurious currents essentially removed the bubble from the domain before the end of
919 the simulation.

920 The Sharp Surface Tension force model yielded low spurious currents (O(10-5) m s-1) and close
921 agreement with the analytical result for heat transfer. The observed temperature distributions matched
922 the expected smooth conduction result (Fig. S5b). Here the correct pressure rise was observed across
923 the droplet interface (d = 2ec = 400, 4000 Pa for = 0.01, 0.10 kg s-2) with minimal artificial
924 advection from spurious currents. For the fine grid cases (Nx = Ny = 45), all latent heat transfer rates
925 were within 1% of the analytical value, even with increased surface tension (0.01 0.10 kg s-2 and
926 Prandtl number 10 50).

927 The default surface tension force model based on the formulation of [40] yielded higher spurious
928 current magnitudes (O(10-3) for = 0.01 kg s-2, O(10-2) for = 0.10 kg s-2). For some high surface tension
929 cases (marked with * in Table S6), the resulting velocity field effectively removed the droplet from the
930 domain before the end of the simulation. At low (0.01 kg s-2) and Pr (10), relatively close agreement
931 was obtained with the analytical heat transfer rate (2.5% error with high grid resolution). However, at
932 high Pr (50), the effect of transport from spurious currents is amplified (Fig. S5c), yielding greater
933 thermal transport errors (9.1%). Errors are generally greater for the high surface tension cases.

934 Based on these results, the default surface tension force model is acceptable for low Pr and low capillary
935 pressure () cases. It is also expected to be acceptable in cases when physical advection is sufficient to
936 dominate numerical advection. The more computationally expensive SST model is recommended for
937 other cases where spurious currents can significantly alter results.

32
938

939 Figure S5 a. Spurious current analysis case geometry and setup. b-c. Temperature field results at t =
940 0.020 s for Nx,y = 45, = 0.10 kg s-2, Pr =50 with (b) sharp surface tension force model and (c) default
941 surface tension force model.

942

943 6. Verification study: Artificial heat transfer due to numerical phenomena


944 The VOF formulation employed in this solver yields a gradual variation of fluid transport properties
945 across the diffuse interface (typically 3-4 cells thick). This may lead to incorrect inter-phase heat transfer
946 rates for cases with large thermal conductivity differences between the two phases. Additionally
947 numerical diffusion or convection-driven mixing of the two phases may cause artificial inter-phase
948 heat transfer. These effects are expected to be less significant in cases with thermally driven phase
949 change, because the interface temperature approaches (Tsat), which is independent of the temperature
950 distributions on either side of the interface. This effectively isolates the thermal transport in the two
951 phases. Here, two example studies without phase change are performed to evaluate the effect of
952 artificial inter-phase heat transfer.

953 Artificial heat transfer for stratified flow


954 In this problem, two fluids (A and B) flow in a stratified fashion through a 2D, L = 0.2 m long domain with
955 uniform axial velocity U = 0.1 m s-1. The inlet temperature of fluid A is T = 10 K higher than fluid B. Both
956 fluids have density = 1000 kg m-3, = 10-3 kg m-1 s-1, specific heat cp = 2000 J kg-1 K-1, and surface
957 tension 0.005 kg s-2. The thermal conductivity of fluid A is varied between cases, and that of fluid B is
958 fixed at k = 1.0 W m-1 K-1. As the fluids flow through the domain, thermal boundary layers form on both
959 sides of the interface. Neglecting axial conduction, the exact solution for heat transfer rate (per unit
960 thickness) is:
/4
(h lm5BC
961 7* f = 2 g ; A1 /4 /4 Dk (S6)
(h i(j n(o

962 A 10 mm wide portion of the domain is evaluated using the developed VOF solver (5 mm thick on each
963 side of the interface, Fig. S6a). No thermal phase change model is employed, and the SST surface tension
964 force is selected. This domain is discretized with 50 cells in the axial direction (y = 4 mm), and 200 cells
965 in the interface-normal (x) direction. Uniform grading is applied in the x-direction on both sides of the
966 interface so that the middle cell thicknesses are 12.7 m and the outermost cells are 127 m. The

33
967 studies are evaluated for 10 s (5 flow cycles, p- = 2 s). The simulation heat transfer rate is evaluated
968 based on the average outlet enthalpy flow rate of fluid B from t = 5 10 s (averaged over all time steps).

969 A representative temperature distribution is presented in Fig. S6b (for kA = 0.359 W m-1 K-1, kB = 1.0 W m-
1 -1
970 K ). As can be observed in this figure, a thermal boundary layer develops on both sides of the
971 interface, increasing in thickness in the flow direction. Here, the boundary layer thickness is visibly
972 greater in the more conductive fluid (B) by the outlet. A series of cases are evaluated with decreasing
973 Fluid A conductivity (1.0 W m-1 K-1 to 1.29 10-5 W m-1 K-1). A power-law curve fit to the relative deviation
974 of simulation heat transfer rate from the analytical result (Eqn. S6), indicates error scales as k-0.76 (Fig.
975 S6c). The relative deviation is below 5% for kA/kB > 2.8 10-4, and reaches 53% for kA/kB = 1.3 10-5. For
976 this problem, the numerical error in heat transfer rate becomes significant when the mesh resolution
977 near the interface is of similar order to the thermal boundary layer thickness. Here, xint = 12.7 m, and

978 for kA = 2.8 10-4 W m-1 K-1, the thermal boundary layer thickness is of the order qr = k s p$89+ -% =
979 16.7 m. This result also indicates, that for this model problem, in the limit as the conductivity of phase
980 A approaches 0, the heat transfer rate decreases (slightly faster than predicted with the analytical
981 model).

982 This case was reevaluated with the conduction terms removed from the thermal energy transport
983 equation. The average absolute heat transfer rate was found to be 0.0045 W m-1, corresponding to an
984 effective thermal conductivity of 1.5 10-11 W m-1 K-1 in both phases using Eqn. S6. This result indicates
985 that the degree of artificial heat transfer is relatively small for this model problem.

986

987

988 Figure S6 a. Phase fraction distribution in co-flow model problem. b. Temperature plot in domain with
989 detail view of interface region near outlet (with thicker thermal boundary layer in more conductive Fluid
990 B region), c. Comparison of simulation and analytical heat transfer rates across the interface with
991 varying phase A thermal conductivity, and relative error.

34
992 Artificial heat transfer for dam break problem
993 The second example considers a 2D dam break problem, which can cause artificial heat transfer due to
994 the convective mixing of the two phases. Here, a 50 mm square domain is considered with walls on all
995 four sides, with a 30 mm square of liquid initially at rest in the lower right corner. The initial liquid
996 temperature is 1 K higher than that of the gas phase. During the simulation, gravity causes the liquid
997 region to break and slosh until it settles on the bottom of the domain due to viscous dissipation (Fig.
998 S7a). This simulation is evaluated with the conduction transport terms removed from the thermal
999 energy equation. Thus, any change in temperature of the liquid or gas would result from artificial
1000 conduction.

1001 The properties of the two phases are: L = 100 kg m-3, V = 10 kg m-3, L = 0.050 kg m-1 s-1, V = 0.005 kg
1002 m-1 s-1, cp,L = cp,V = 1000 J kg-1 K-1, and = 0.1 kg s-2. A higher order convection scheme (limited cubic
1003 interpolation) and a second-order accurate backward time discretization scheme are employed to
1004 reduce numerical diffusion. The SST surface tension force model is employed, and no phase-change heat
1005 transfer model is applied. The domain is meshed with a uniform square grid (50 50 cells up to 400
1006 400 cells).The case is evaluated for 10 s, which is found to be sufficient for the liquid layer to settle on
1007 the bottom of the domain. The amount of artificial heat transfer is quantified as the total energy
1008 increase of the vapor phase, which is proportional to the initial temperature difference between the
1009 phases.

1010 Overall heat transfer values for these cases are presented in Fig. S7b. Refining the mesh from = 1000
1011 m to 360 m (first four cases) causes the overall heat transfer to reduce approximately linearly with
1012 mesh resolution. The improvement rate decreases with further refinement of the mesh. This may be
1013 due in part to the complexity of the model problem. As the mesh is refined, numerical diffusion is
1014 reduced in the momentum equation, allowing the development of capillary waves on the interface (see
1015 Fig. S7a) and higher shear rates. Additionally, some small pockets of vapor are trapped under the liquid
1016 layer, and rise to the top of the domain in close contact with the higher temperature liquid. These are
1017 not resolved with coarser meshes. Such effects may increase inter-phase mixing and artificial heat
1018 transfer rates. Overall, this analysis demonstrates that artificial inter-phase heat transfer due to
1019 convective mixing can be reduced with mesh refinement, but diminishing gains may be found for cases
1020 with strong small-scale transport mechanisms (e.g., capillary waves, near grid-scale bubbles/droplets).

35
1021

1022

1023 Figure S7 a. Phase fraction and temperature distributions for high-resolution dam break example case,
1024 with detail view of capillary waves on interface. b. Overall artificial inter-phase heat transfer variation
1025 with mesh resolution.

1026

1027 7. Rising Bubble Condensation at Increased Reynolds Numbers


1028 In Section 3.2, a demonstration case was presented of condensation heat transfer for a small vapor
1029 bubble rising in a quiescent liquid medium. In that case, the peak Reynolds number was approximately
1030 7.3. Here, results are presented for a series of cases with Re up 94, demonstrating the ability of this
1031 solver to support heat transfer flows with stronger advection effects.

1032 The case setup is similar to that presented in Section 3.2, except that the bubble and domain sizes are
1033 doubled to yield higher Re flows (Dbub = 920 m, cylindrical domain radius 3.0 mm, cylindrical domain
1034 axial length 4.0 mm). Fluid properties are summarized below (Table S7). The liquid viscosity is varied
1035 between cases to yield different bubble rise velocities and Reynolds numbers.

36
1036 Table S7 Fluid properties and setup parameters for the bubble condensation case at varying Re
Fluid Property Liquid Value Vapor Value
Dynamic viscosity (, kg m-1 s-1) (varies) 5.0 10-5
Density (, kg m-3) 900 10
Thermal conductivity (k, W m-1 K-1) 1.00 0.02
Specific heat (cp, kJ kg-1 K-1) 2.0 2.5
Surface tension (, kg s-2) 0.00747
Liquid Prandtl number (PrL) (varies)
Enthalpy of phase change (iLV, kJ kg-1) 2000
Saturation temperature (Tsat, K) 100
Other Case Parameters
Initial bubble diameter (m) 920
Liquid subcool (Tsat TL, K) 1
1037

1038 As in Supplemental Material Section 1, the InterfaceEquilibrium phase change model is employed with
1039 )*+, and * ,+, set to 0. This represents the case of a rising vapor bubble at low phase change rates,
1040 capturing interfacial heat transfer without the second order effects of changing bubble volume. This
1041 enables comparison with steady-state analytical and single-phase heat transfer results. Due to the thin
1042 thermal boundary layers at these higher Re cases, applying the phase-change source terms centered on
1043 the 1 = 0.5 level of the diffuse interface would significantly distort the temperature profile in the
1044 thermal boundary layer. Thus, these terms are offset slightly inwards (to the vapor side), and are
1045 centered on 1 = 0.3.

1046 The domain is meshed with Nr = 150 cells and Nz = 240 cells (similar radial grading as in Section 3.2). The
1047 VOF simulations are computed for 0.25 s, and bubble rise velocity (Ubub,sim) and overall heat transfer rate
1048 (Qsim) are averaged for each time step after a startup period (t = 0.15 0.25 s).

1049 The selected fluid properties and operating conditions correspond to a Bond number (Bo = 8 vc 6e)
1050 of 1.0. Liquid viscosity is varied for the five cases to yield Morton numbers (Mo = v x 8e y ) of 10-3,
1051 10-4, 10-5, 10-6, and 10-7. Overall heat transfer rates for each case are compared with those predicted
1052 using the model of Ranz and Marshall [54] (Eqn. 11), and a single-phase liquid-only finite element study
1053 (FEM, described in Supplemental Material Section 1) at the average VOF bubble rise velocity (Table S8).

1054 Table S8 Summarized results from high Reynolds number bubble condensation study

Liquid Morton Prandtl Bubble Reynolds VOF Heat Correlation Heat FEM Heat
viscosity Number Number Velocity Number Transfer Rate Transfer Rate Transfer Rate
(kg m-1 s-1) (m s-1) (mW) (mW, relative error) (mW, relative error)
13.99 10-3 10-3 27.97 0.0448 2.65 15.70 14.25 (+10.1%) 16.37 (-4.1%)
7.86 10-3 10-4 15.73 0.0683 7.19 18.06 17.32 (+4.3%) 19.34 (-6.6%)
4.42 10-3 10-5 8.85 0.0966 18.08 22.44 20.91 (+7.3%) 22.75 (-1.4%)
2.49 10-3 10-6 4.97 0.1284 42.76 24.94 25.02 (-0.3%) 26.50 (-5.9%)
1.40 10-3 10-7 2.80 0.1584 93.78 28.54 29.35 (-2.8%) 30.46 (-6.3%)
1055

37
1056 For the three lower Re cases (Re = 2.65, 7.19, 18.08), the VOF simulation heat transfer rates are slightly
1057 higher than predicted with the analytical correlation of [54] (+4.3 +10.1% relative errors), and slightly
1058 lower than predicted with the FEM studies (-1.4 -6.6%). For the two higher Re cases (Re = 42.76,
1059 93.78), the VOF simulation results slightly under-predict both the analytical correlation (-0.3%, -2.8%)
1060 and FEM studies (-5.9%, -6.3%). However, in these cases, the VOF bubble shape becomes oblate
1061 ellipsoidal rather than spherical, as assumed in the correlation and FEM studies. This transition can be
1062 observed in Fig. S8, and begins to be observable for Mo = 10-5 (third case). This finding qualitatively
1063 agrees with the spherical-to-oblate ellipsoidal transition range of Mo ~ 10-4 10-6 for Bo ~ 1 reported by
1064 Bhaga and Weber [66].

1065 For all cases here, the varying term in the correlation of [54] is significantly larger than the constant term
1066 (Nu = 2 + 0.6Re /6 Pr /y, 0.6Re /6 Pr /y = 3.0 8.2 here).

1067 It should be noted that the VOF study was performed in a 2D axisymmetric domain, and the FEM study
1068 assumed axisymmetric and steady flow. Thus, 3D wake patterns and vortex shedding, which may occur
1069 at high Re, were not considered here.

1070 Based on the relatively close agreement between the VOF heat transfer predictions and the analytical
1071 model and FEM values, this analysis demonstrates that this solver can be employed for phase-change
1072 heat transfer processes with significant advective transport contributions.

1073

1074 Figure S8 Liquid-phase temperature distributions and bubble interface profiles (1 = 0.3 isosurface) for
1075 condensing rising vapor bubbles at varying Morton and Prandtl numbers (Bo = 1). Transition from
1076 spherical to oblate ellipsoidal bubble profile begins around Mo = 10-5.

1077

1078

1079

38
1080 8. Parallelization study and performance
1081 Strong and weak parallel performance studies were performed using a slightly modified version of the
1082 phase change solver on the US National Energy Research Scientific Computing Center (NERSC) Edison
1083 cluster. This study explores the model problem of vapor absorption into a falling-film liquid mixture
1084 flowing over cooled rectangular tubes. This configuration is of particular interest for waste-heat
1085 activated absorption refrigeration systems. Weak scaling studies (increasing the domain size) are
1086 particularly informative for this case because the liquid film can grow unstable, yielding 3D effects such
1087 as film break-up and jetting (Fig. S9). The appropriate domain size required to capture such phenomena
1088 can be difficult to predict apriori, therefore, weak scaling validation analyses must be performed in
1089 actual production scientific studies to demonstrate convergence of results.

1090

1091 Figure S9 Simulation of vapor absorption into a falling-film mixture flowing over cooled rectangular
1092 tubes. Film breakup progression. Liquid colored by concentration (increasing as fluid flows down
1093 through the domain).

1094 The strong scaling analyses were performed on a two tube-row problem with 2.6 million cells (12 mm
1095 long tube sections). Results were collected for studies with 24- to 192-way parallelism (Fig. S10a). Peak
1096 performance and computing efficiency was obtained with the 48-way parallel case, with 5.1 billion
1097 mesh-cell-time-steps per wall-clock hour (wch) and 107 million mesh-cell-time-steps per processor-hour.
1098 This corresponds to approximately 54,000 mesh cells per processor.

1099 A weak scaling analysis was conducted for increasingly large domains (12 to 192 mm long, 2.6 to 42
1100 million mesh cells) with approximately 54,000 mesh cells per parallel process. Net computing
1101 performance was found to increase with mesh size and parallelism for all cases, reaching 30.9 billion cell
1102 time steps per wch for the 768-way parallel case (Fig. S10b). Computing efficiency was found to
1103 decrease sub-linearly with problem size, with a 20% reduction when increasing from 384 to 768 parallel

39
1104 processes. This result indicates that the developed code will likely scale relatively efficiently up to a few
1105 thousand-way parallelism. Beyond this point, the underlying MPI-based OpenFOAM [21] parallelization
1106 framework may have to be modified, potentially with a hybrid MPI/OpenMP approach.

1107

1108

1109 Figure S10 Parallel scaling performance and efficiency of the falling-film mixture absorption
1110 simulation. a. Strong scaling results for fixed mesh size (2.6 million cells), b. Weak scaling performance
1111 for 54,000 mesh cells per parallel process.

1112 The code performance can be described using a simple model that assumes that the simulation time is
1113 divided between computation and communication. The computation time is assumed to scale linearly
1114 with per-process mesh size. The communication time is assumed to scale at a fractional rate with the
1115 degree of parallelism due to decreased process locality. The resulting equations for performance (P: cell-
1116 time-steps per wall-clock hour) and efficiency (E: cell-time-steps per processor hour) are presented
1117 below and are graphed in Fig. S6.

n
1118 P= (S7)
(2.42 10 )(n k )+ (7.62 10 5 )k 0.42
9

n
1119 E= (S8)
(2.42 10 )n + (7.62 10 )k
9 5 1.42

1120 Here n is the total mesh cell count and k is the number of parallel MPI tasks. For fixed mesh sizes
1121 (strong-scaling), this model predicts peak performance with approximately 50,000 cells per processor.
1122 For increasing problem size (weak scaling) performance increases approximately as n0.65, and efficiency
1123 reduces as n-0.35.

1124 This sub-linear scaling performance is typical of incompressible CFD codes, in which a Poisson equation
1125 must be solved iteratively and globally to obtain the pressure and velocity fields at each time step. This
1126 step is the most computationally expensive in the solution algorithm, and is difficult to parallelize
1127 because it necessitates substantial inter-process and inter-node communication. Overall, these results

40
1128 demonstrate relatively efficient operation for large engineering-scale simulations with tens of millions of
1129 mesh cells and 100 1,000-way MPI parallelism.

1130

1131 9. Supplemental Material References


1132 [58] J. S. Hadamard, Motion of liquid drops (viscous), Comp. Rend. Acad. Sci. Paris, vol. 154, pp. 17351755,
1133 1911.
1134 [59] W. Rybczynski, On the translatory motion of a fluid sphere in a viscous medium, Bull. Acad. Sci., Cracow,
1135 Ser. A, p. 40, 1911.
1136 [60] A. S. Rattner, Single-pressure absorption refrigeration systems for low-source-temperature applications -
1137 Chapter 5, Georgia Institute of Technology, Atlanta, GA, 2015.
1138 [61] A. Alizadehdakhel, M. Rahimi, and A. A. Alsairafi, CFD modeling of flow and heat transfer in a
1139 thermosyphon, Int. Commun. Heat Mass Transf., vol. 37, no. 3, pp. 312318, Mar. 2010.
1140 [62] E. Da Riva and D. Del Col, Effect of Gravity During Condensation of R134a in a Circular Minichannel,
1141 Microgravity Sci. Technol., vol. 23, no. 1, pp. 8797, 2011.
1142 [63] H. Lee, C. R. Kharangate, N. Mascarenhas, I. Park, and I. Mudawar, Experimental and computational
1143 investigation of vertical downflow condensation, Int. J. Heat Mass Transf., vol. 85, pp. 865879, 2015.
1144 [64] M. Williams, D. Kothe, and E. Puckett, Accuracy and convergence of continuum surface tension models,
1145 Fluid Dyn. Interfaces, pp. 294305, 1998.
1146 [65] D. J. E. Harvie, M. R. Davidson, and M. Rudman, An analysis of parasitic current generation in Volume of
1147 Fluid simulations, Appl. Math. Model., vol. 30, no. 10, pp. 10561066, 2006.
1148 [66] D. Bhaga and M. E. Weber, Bubbles in viscous liquids: shapes, wakes and velocities, J. Fluid Mech., vol.
1149 105, pp. 6185, 1981.

41

Anda mungkin juga menyukai