betaFeedback

WorksPhysics and Astronomy

Natural modes of the two-fluid model of two-phase flow

Abstract

A physically-based method to derive well-posed instances of the two-fluid transport equations for two-phase flow, from the Hamilton principle, is presented. The state of the two-fluid flow is represented by the superficial velocity and the drift-flux, instead of the average velocities of each fluid. This generates the conservation equations of the two principal motion modes naturally: the global center-of-mass flow and the relative velocity between fluids. Well-posed equations can be obtained by modelling the storage of kinetic energy in fluctuations structures induced by the interaction between fluids, like wakes and vortexes. In this way, the equations can be regularized without losing in the process the instabilities responsible for flow-patterns formation and transition. A specific case of vertical air-water flow is analyzed showing the capability of the present model to predict the formation of the slug flow regime as trains of non-linear waves.

Two-phase flows are important transport processes both in natural and artificial domains. Examples of natural systems are the geysers, hot jets of water and steam ejected from volcanic soils. In industry, the uses of two-fluid flows are numerous, ranging from transport of petroleum mixtures in ducts, the variety of boiling heat transfer phenomena involved in energy systems, management of interfacial reactions in chemical processes, etc.

Multiple fluid mixtures are generally too complex to describe by keeping track of all interfaces and motions at both sides. We must resort then to representations of the system based on mean-field state variables and their correspondent evolution equations. The overall averaging procedure based on continuum mechanics, established long ago, provides with a consistent theoretical framework [1, 2]. However, its major theoretical development, the two-fluid model, where the phases are described as interwoven continuum fluids, yields an ill-posed mathematical problem, which, as a recent update states ,“despite extensive efforts to remedy, and despite claims to the contrary, remains to this day” [3]. Recently, strong evidence was presented indicating that the physical mechanism responsible for the problem is the Kelvin-Helmholtz instability (KHI), summoned whenever two fluids are forced to flow together shearing the pressure field [4]. This is another manifestation of KHI, one of the most important hydrodynamic instabilities, which continues drawing the attention of the fluid community [5, 6]. Current numerical regularization schemes basically suppress this instability. Alas, the price paid is that any possible predictive capacity of the physical phenomena associated to the instability, like flow pattern formation, is lost in the maneuver.

The motivation of our work is the solution of the ill-posedness problem of the two-fluid model by means of inertia-exchange mechanisms. We present a physically-based way to derive well-posed instances of the two-fluid transport equations from the first principles of field theory, which opens a promising road to focus future research efforts, especially for conflating experimental data and theoretical models. In the 80s and 90’s several researchers applied the Hamilton principle of mechanics to the one-dimensional dispersion of one fluid into the continuous bulk of another. A variety of inertial coupling models were explored and different lines of interpretation of the resulting conservation equations were investigated [7, 8, 9, 10].

Since most applications of two-fluid flows take place in long conduits, like pipes, ducts and channels, where mixtures are being transported, processed or heated, one-dimensional approaches are the most common for treating practical problems because they highly reduce computational costs. But there are also other factors, e.g., that less phenomenological parameters are required to close the equations and that most of them were developed for one-dimensional conditions.

We adopt the usual notation where the physical properties of each fluid are identified with the sub-index i=1,2i = 1, 2. By convention, i=1i = 1 is assigned to the denser fluid. All properties are assumed constant and, therefore, the fluids are incompressible. The volume fraction occupied by fluid-ii is treated as a state field of the system, and it is denoted by αi\alpha_i. The streamwise velocity of each fluid averaged over the volume occupied by the corresponding fluid is denoted by uiu_i. Mass densities are denoted by ρ\rho. Then, the following variables can be constructed with αi\alpha_i and uiu_i:

ρm=∑iαiρi(1)\rho_m = \sum_i \alpha_i \rho_i \tag*{(1)}

Mixture density

j=∑iαiui(2)j = \sum_i \alpha_i u_i \tag*{(2)}

Superficial velocity

ρmvm=∑iρiαiui(3)\rho_m v_m = \sum_i \rho_i \alpha_i u_i \tag*{(3)}

Mixture momentum

ur=u2−u1(4)u_r = u_2 - u_1 \tag*{(4)}

Relative velocity

J=urα1α2(5)J = u_r \alpha_1 \alpha_2 \tag*{(5)}

Drift flux

The superficial velocity, jj, is the velocity of the center of volume of the mixture, whereas the velocity vmv_m is the velocity of the center of mass of the mixture. The drift flux, JJ, is the volumetric flux of fluid-2 passing through a cross-section surface that moves with velocity jj.

The continuity equation of each fluid is

∂αi∂t+∂αiui∂x=0(6)\frac{\partial\alpha_i}{\partial t}+\frac{\partial\alpha_i u_i}{\partial x}=0 \tag*{(6)}

which can be combined in the two following forms:

∂αi∂t+j∂αi∂x+(−1)i∂J∂x=0(7)\frac{\partial\alpha_i}{\partial t}+j\frac{\partial\alpha_i}{\partial x}+(-1)^i\frac{\partial J}{\partial x}=0 \tag*{(7)}
∂j∂x=0(8)\frac{\partial j}{\partial x}=0 \tag*{(8)}

The kinetic energy density of a representative volume element (RVE) of a two-fluid system is:

K=12∑iρiαiui2+12∑iρiαiσi2(9)K=\frac{1}{2}\sum_i\rho_i\alpha_i u_i^2+\frac{1}{2}\sum_i\rho_i\alpha_i\sigma_i^2 \tag*{(9)}

where:

σi2=⟨ui2⟩−ui2(10)\sigma_i^2=\langle u_i^2\rangle-u_i^2 \tag*{(10)}

is the mean quadratic deviation of the velocity of fluid-ii, and ⟨ui2⟩\langle u_i^2\rangle is the square of the velocity module of each fluid averaged over the volume occupied by the corresponding fluid.

At this level of description, the state of the system is represented by the fields uiu_i and αi\alpha_i. However, we are interested in describing the two-fluid flow with variables representing the global motion of the mixture and the relative motion between fluids. To do so, we express the Lagrangian in terms of the superficial velocity, jj, and the drift flux, JJ. Eq. (9) then becomes:

K=12ρmj2+12ΓJ2−ΔρjJ(11)K=\frac{1}{2}\rho_m j^2+\frac{1}{2}\Gamma J^2-\Delta\rho jJ \tag*{(11)}

where:

Γ=∑iρiαi+∑iρiαiγi(αi)(12)\Gamma=\sum_i\frac{\rho_i}{\alpha_i}+\sum_i\rho_i\alpha_i\gamma_i(\alpha_i) \tag*{(12)}
γi(αi)=1αi2(1−αi)2(σi2ur2)(13)\gamma_i(\alpha_i)=\frac{1}{\alpha_i^2(1-\alpha_i)^2}\left(\frac{\sigma_i^2}{u_r^2}\right) \tag*{(13)}

In principle, σi2\sigma_i^2 may have a turbulent component proportional to ui2u_i^2, but this would be dissipative in nature, and so it will be incorporated later as a dissipative force. The part of σi2\sigma_i^2 that interests us is related to the velocity redistribution due to the interaction between fluids, and in general is proportional ur2u_r^2 with a coefficient that depends on the volume fraction αi\alpha_i of the correspondent fluid [7, 9]. This term is analogous to the added mass of immersed bodies. The Lagrangian of the two-fluid system is then written as:

L=K+∑nφnRn(14)L = K + \sum_n \varphi_n R_n \tag*{(14)}

where RnR_n are admissibility conditions that restrain the variation of the state variables, αi\alpha_i, jj and JJ; and φn\varphi_n are the corresponding Lagrange multipliers. In our case, the restrictions are the continuity Eqs. (7) and the confinement relation:

∑iαi=1(15)\sum_i \alpha_i = 1 \tag*{(15)}

Then:

L=12ρmj2+12ΓJ2−Δρ jJ+∑iφi[∂αi∂t+j∂αi∂x+(−1)i∂J∂x]+φα∑iαi(16)L = \frac{1}{2}\rho_m j^2 + \frac{1}{2}\Gamma J^2 - \Delta\rho\, jJ + \sum_i \varphi_i \left[\frac{\partial\alpha_i}{\partial t} + j\frac{\partial\alpha_i}{\partial x} + (-1)^i\frac{\partial J}{\partial x}\right] + \varphi_\alpha\sum_i \alpha_i \tag*{(16)}

The corresponding Euler-Lagrange equations may now be cast in two modes [17]. For the center of mass of the mixture:

∂Mm∂t+∂Σm∂x=0(17)\frac{\partial\mathcal{M}_m}{\partial t} + \frac{\partial\Sigma_m}{\partial x} = 0 \tag*{(17)}

where:

Mm=ρmvm(18)\mathcal{M}_m = \rho_m v_m \tag*{(18)}
Σm=ρmvm2+ρ1ρ2α1α2ρmJ2+12J2∑iρidγid(1/αi)+p(19)\Sigma_m = \rho_m v_m^2 + \frac{\rho_1\rho_2}{\alpha_1\alpha_2\rho_m}J^2 + \frac{1}{2}J^2\sum_i \rho_i\frac{d\gamma_i}{d(1/\alpha_i)} + p \tag*{(19)}

The first term of Σm\Sigma_m is the global momentum flux of the mixture, which behaves as a pseudo-fluid with density ρm\rho_m and velocity vmv_m. The second term is the momentum flux due to the relative motion without inertial coupling between fluids. The third term is the momentum flux introduced by the inertial coupling between fluids. The last term of Σm\Sigma_m is the pressure, which is recognized as the scalar field p=−φαp=-\varphi_\alpha. If γi=0\gamma_i=0, Ishii’s drift-flux mixture momentum balance equation is recovered [1].

For the relative motion between fluids, we have:

∂MJ∂t+∂ΣJ∂x=0(20)\frac{\partial\mathcal{M}_J}{\partial t} + \frac{\partial\Sigma_J}{\partial x} = 0 \tag*{(20)}

where:

MJ=ΓJ−jΔρ(21)\mathcal{M}_J = \Gamma J - j\Delta\rho \tag*{(21)}
ΣJ=−12Γ′J2+ΓJj(22)\Sigma_J = -\frac{1}{2}\Gamma'J^2 + \Gamma Jj \tag*{(22)}

Henceforth the prime indicates derivative respect to α2\alpha_2, and Δρ=ρ1−ρ2\Delta\rho=\rho_1-\rho_2. The length scale of this equation is of the order of the pipe diameter, and the characteristic time is much shorter than the transit time along the conduit. Eqs. (20)–(22) have some interesting associations. By imposing j=0j=0, Eqs. (20)–(22) coincide with the equation found in previous works for that condition [9, 10]. Furthermore, if j=0j=0 and ρ2=0\rho_2=0, then MJ=\mathcal{M}_J= ρ1J/α1\rho_1 J/\alpha_1 and ΣJ=(ρ1/2)(J/α1)2\Sigma_J = (\rho_1/2)(J/\alpha_1)^2, and Eq. (20) reduces to the Burger’s equation. Also, in stratified flow, a gravitational term appears, which for j=0j = 0 and ρ2=0\rho_2 = 0 leads to the shallow water equation of Saint-Venant.

Perturbing and linearizing Eqs. (20)–(22) yields the following dispersion relation:

ωk=jo+JoΓo(−Qo′±12QoQo′′)(23)\frac{\omega}{k} = j_o + J_o\Gamma_o\left(-Q'_o \pm\sqrt{\frac{1}{2}Q_oQ''_o}\right) \tag*{(23)}

Where, the sub-index oo refers to the steady-state and Qo=1/ΓoQ_o = 1/\Gamma_o. Since QoQ_o is always positive, if Qo′′<0Q''_o < 0 the system is elliptic for any wave number. In particular, for k=∞k = \infty (i.e., zero wave length) the perturbation growth rate is infinite. This sort of ultraviolet divergence is the infamous ill-posedness problem, which makes it impossible to achieve numerical convergence. The necessary condition of a well-posed two-fluid theory reduces to:

Qo′′≥0(24)Q''_o \ge0 \tag*{(24)}

or equivalently:

2Γo′2−ΓoΓo′′≥0(25)2\Gamma_o^{\prime2} - \Gamma_o\Gamma_o'' \ge0 \tag*{(25)}

Eq. (25) is the same condition found for constant jj [9]. Our result shows that the criterion given by Eq. (24) is general for the incompressible two-fluid model, holding also if jj varies in time. This is a significant result, for it proves that the cause of the ill-posedness problem lies in the relative fluid motion mode of the system, particularly in the term 12J2\frac{1}{2}J^2 of the Lagrangian. Furthermore, Eq. (25) is equivalent to the threshold of the Kelvin-Helmholtz instability in two-layers shallow water [11], appearing here as an extreme manifestation that triggers the growth of any wave length, thus rendering the numerical solution dependent on the spatial resolution.

It should be mentioned that in horizontal flows, gravity and surface tension can make the system hyperbolic up to certain values of JJ while stratified conditions hold [12, 13]. Those special conditions are not difficult to introduce into Eq. (23) to extend the criterion. Nevertheless, in most practical cases, the role played by the magnitude QQ appears to be the crux of the matter for a well-posed model, in the sense that numerical convergence to the differential limit be achievable. However, such well-posed models can still be unstable due to the interaction between inertia and drag. This is a mechanism which is well known in the global motion of the mixture, especially when boiling is included: the so-called density-wave instabilities. In the relative-motion mode between fluids, the dynamic interplay between the interfacial drag and the inertia coupling can also lead to destabilizing well-posed mechanisms.

Next we will present a simple model of inertial coupling to show the procedure. Consider the RVE depicted in Figure 1. A bubbly cluster or a wide bubble of the lighter fluid, occupies a fraction αb\alpha_b of the channel cross section along a length fraction xbx_b of the RVE. The remaining portion of the RVE is a slug filled with the heavier fluid moving streamwise with mean velocity usu_s. Note that the volume fraction of fluid-2 is:

Representative volume element used in the model of inertial coupling

Figure 1. Representative volume element used in the model of inertial coupling. The bubble cluster occupies a mean fraction αb\alpha_b of the channel cross-section.

α2=αbxb(26)\alpha_2 = \alpha_b x_b \tag*{(26)}

Disregarding fluctuations having zero-average streamwise velocity, like vortexes, the kinetic energy density in the RVE is given by:

K=12ρ1(1−xb)us2+12ρ1xb(1−αb)uf2+12ρ2α2u22(27)K = \frac{1}{2}\rho_1(1-x_b)u_s^2 + \frac{1}{2}\rho_1 x_b(1-\alpha_b)u_f^2 + \frac{1}{2}\rho_2\alpha_2u_2^2 \tag*{(27)}

where ufu_f is the average velocity of fluid-1 in sector xbx_b. Using Eqs. (9)-(13), we find that the inertial coupling is given by [8, 10]:

γ1=αb−1+α1α12(1−α1)(1−αb)(28)\gamma_1 = \frac{\alpha_b-1+\alpha_1}{\alpha_1^2(1-\alpha_1)(1-\alpha_b)} \tag*{(28)}

Now suppose that there are zero-average velocity fluctuations in fluid-1, like undulating coherent structures at the back and amid the bubble cluster. Although these fluctuations do not affect u1u_1, they do contribute to ⟨u12⟩\langle u_1^2\rangle. In certain ways, this would be similar to the thermodynamic degradation of mechanical energy into thermal energy, although in this case the pseudo-thermalization takes place between the scale of the tube length and the scale of the tube diameter. A simple model for this mechanism is to assume that the energy of these wavy fluctuations is proportional the kinetic energy of the streamwise velocity dispersion in the RVE, represented by Eq. (28), and to the room available for the development of these perturbations, i.e., α1\alpha_1. If the coupling coefficient of proportionality is a constant parameter ww, then:

γ1=(1+wα1)(αb−1+α1)α12(1−α1)(1−αb)(29)\gamma_1 = \frac{(1+w\alpha_1)(\alpha_b-1+\alpha_1)}{\alpha_1^2(1-\alpha_1)(1-\alpha_b)} \tag*{(29)}

The inertial coupling is then characterized by the control parameters ww and αb\alpha_b. The parameter ww represents the ratio of the kinetic energy of zero-average fluctuations of fluid1 divided by the quadratic deviation of its streamwise velocity. The parameter αb\alpha_b characterizes local axial asymmetries of the distribution of fluid-2 in the RVE.

We can verify that:

Γ′2−12ΓΓ′′=wρ1(1−αb)(1−α1)3[ρ2+ρ1(αb1−αb)(1+w)](30)\Gamma'^{2}-\frac{1}{2}\Gamma\Gamma''=\frac{w\rho_{1}}{(1-\alpha_{b})(1-\alpha_{1})^{3}}\left[\rho_{2}+\rho_{1}\left(\frac{\alpha_{b}}{1-\alpha_{b}}\right)(1+w)\right] \tag*{(30)}

which is always positive, rendering the system hyperbolic and thus well-posed. Higher order models of inertial coupling can be addressed by refining the generation mechanisms of coherent structures. It should be noted that, in spite of being hyperbolic, the system may still be linearly unstable, because of the interaction between drag and inertial forces.

To verify the performance of the derived modal two-fluid equations, we consider an air-water upward flow in a circular conduit of length L=4L = 4 m. Constant superficial velocity jj is assumed, and therefore Eqs. (17) and (20) decouple. In this way, the study focuses on the inter-fluid instability, leaving aside the influence of the acceleration of the global mixture. On account of Eq. (8), this condition is easy to achieve experimentally by controlling the fluids injection at the inlet.

Sources of gravity and interfacial drag are added in the drift-flux momentum equation as:

∂J∂t+∂ΣJ∂x=gΔρ−CDDρ1ur∣ur∣(31)\frac{\partial J}{\partial t}+\frac{\partial\Sigma J}{\partial x}=g\Delta\rho-\frac{C_D}{D}\rho_{1}u_r\lvert u_r\rvert \tag*{(31)}

where D=30D = 30 cm is the tube diameter, and:

CD=exp⁡(1.32+8.4α2−42α22+45α23−15.4α24)(32)C_D=\exp(1.32+8.4\alpha_{2}-42\alpha_{2}^{2}+45\alpha_{2}^{3}-15.4\alpha_{2}^{4}) \tag*{(32)}

is an empirical interfacial drag coefficient we produced from experimental data [17].

Eq. (31) together with the continuity Eq. (7) are discretized in space using a fourth order estimator of the spatial derivatives, and the resulting set of ordinary differential equations in time is solved by means of a backward differentiation scheme. For convenience, periodic boundary conditions are imposed to the spatial domain, which is an idealized representation of fully developed flows.

To better visualize the non-linear behavior of the system, it is useful to define the system variable:

Δα2=∫0L/2α2 dx−∫L/2Lα2 dx(33)\Delta\alpha_{2}=\int_{0}^{L/2}\alpha_{2}\,dx-\int_{L/2}^{L}\alpha_{2}\,dx \tag*{(33)}

representing the imbalance of gas volume in the conduit, Δα2=0\Delta\alpha_{2}=0 indicating that the gas mass is sheared in exact halves between x∈(0,L/2)x\in(0,L/2) and x∈(L/2,L)x\in(L/2,L).

Figure 2 shows the phase-plane evolution of Δα2\Delta\alpha_2 and the simultaneous value of α2\alpha_2 at x=L/2x=L/2, for the case with background ⟨α2⟩=0.5\langle\alpha_2\rangle=0.5. It can be seen that the initial small perturbation grows until non-linear mechanisms act containing the amplitude of the oscillation. The dynamics has the properties of a soliton, in the sense that the asymptotic state is independent of the initial perturbation. In the phase-plane, the asymptotic trajectory describes a limit cycle. Figure 3 and Figure 4 depict the limit cycles obtained for different background gas fractions ⟨α2⟩\langle\alpha_2\rangle between 0.3 and 0.55, and the corresponding snapshots of spatial profiles α2(x)\alpha_2(x). For ⟨α2⟩\langle\alpha_2\rangle between 0.3 and 0.4, the profiles present narrow spikes of gas concentrations, which can be construed as small clusters of bubbles. For ⟨α2⟩\langle\alpha_2\rangle between 0.45 and 0.55, the regions of higher gas fraction become longer, which can be construed as Taylor bubbles occupying the center of the conduit cross section surrounded by a thin film of liquid. These bubbles are separated by slugs of low gas concentration, typical of the slug flow-pattern regime. This result is in agreement with experimental observations [14, 15]. Furthermore, by increasing the resolution of the spatial discretization, 99% convergence was achieved with 200 nodes.

Phase-space evolution of the simultaneous values of $\Delta\alpha_2$ and $\alpha_2$

Figure 2. Phase-space evolution of the simultaneous values of Δα2\Delta\alpha_2 and α2\alpha_2 at x=L/2x=L/2, for the reference case: L=4L=4 m, D=30D=30 cm, j=0.5j=0.5 m/s, ⟨α2⟩=0.5\langle\alpha_2\rangle=0.5, αb=0.9\alpha_b=0.9, w=0.4w=0.4. The transient starts with a small perturbation around the steady-state values. The system develops an oscillation with growing amplitude, approaching asymptotically to a limit cycle.

Limit cycles for the reference parameters and different background $\langle\alpha_2\rangle$

Figure 3. Limit cycles for the reference parameters (αb=0.9\alpha_b=0.9, w=0.4w=0.4) and different background ⟨α2⟩\langle\alpha_2\rangle. The cycles up to ⟨α2⟩=0.4\langle\alpha_2\rangle=0.4 are associated to bubble clustering, higher ⟨α2⟩\langle\alpha_2\rangle are associated to slug-flow pattern.

Gas fraction profiles for different background gas fractions

Figure 4. Gas fraction profiles for the reference parameters (αb=0.9\alpha_b = 0.9, w=0.4w = 0.4) and different background ⟨α2⟩\langle\alpha_2\rangle. Profiles up to ⟨α2⟩=0.4\langle\alpha_2\rangle= 0.4 are associated to bubble clustering, higher ⟨α2⟩\langle\alpha_2\rangle are associated to slug-flow pattern.

In the resulting scenario, the control of the character of the equations (elliptic, hyperbolic, or parabolic) is achieved by the kinetic energy of the coherent (or pseudo-turbulent) structures of the fluid fluctuations, like wakes, vortexes, localized flow gradients, etc. This is performed via the inertial coupling function γi(αi)\gamma_i(\alpha_i) given by Eq. (13). It is interesting to reckon that similar expressions of inertial coupling in mixtures of continuous media were found by using the principle of virtual powers [16]. Different spatial configurations of the fluids are associated to different inertial coupling functions, which in turn yield different energy partitions. In principle, the system will “chose” the configurations that minimize the action. Therefore, the new methodology regularizes the equations using a physically based procedure. Essentially, the method consists of producing parametric families of inertial coupling functions, providing enough flexibility for the system to find equilibrium configurations that match the experimental observations.

A brief comment about the physical interpretation of the inertial coupling function is offered as an educated conjecture. Notice that the actual effect of the inertia coupling can be construed as a modification of the apparent density of the fluids, as if a fraction of the mass of each fluid were exchanged in the process. Viewed from this perspective, regarding that ultimately the inertial terms translate into force terms in the momentum transport equations, the inertial coupling resembles in a certain way the physics of elementary particles, where the interactions take the form of force carriers. Suggestively, the present modal two-fluid field theory permits the construction of inertial exchange models in a progressive way, much as the way of the perturbative levels in which elementary force-carrier particles appear. Actually, note that the function Γ\Gamma can be (loosely) seen to play a role similar to the propagator in field theory. By way of illustration, Figure 5 lists the inertial coupling function Γ\Gamma obtained at different levels of perturbation for the model given by Eqs. (12), (28) and (29). By the same token, other forms of perturbation of the velocity fields can be further introduced and tested against experimental observations, analogous to how the experiments in particle colliders provide insight to field theories. Among the mechanisms that are promising candidates for future studies are: bubble entrainment, collision, breakup and coalescence, interaction with solid structures, interfacial deformation and waves, vortexes inside or in the wake of bubbles, among others.

Inertial function configurations and corresponding coupling functions

Figure 5. Inertial function Γ\Gamma in successive levels of detail of kinetic energy. The conservation equations are elliptic with Γ1\Gamma_1, parabolic with Γ2\Gamma_2, and hyperbolic with Γ3\Gamma_3.

Acknowledgements

The authors want to thank Alexander López de Bertodano for his interesting discussions. This work was partially supported by the Argentine National Atomic Energy Commission.

*[email protected]

#[email protected]

Supplementary Material for “Natural modes of the two-fluid model of two-phase flow”

Derivation of the Euler-Lagrange equations of the modal two-fluid model

We start with the Lagrangian given by:

L=12ρmj2+12ΓJ2−Δρ jJ+∑iφi[∂αi∂t+j∂αi∂x+(−1)i∂J∂x]+φα∑iαi(34)L = \frac{1}{2}\rho_m j^2 + \frac{1}{2}\Gamma J^2 - \Delta\rho\,jJ + \sum_i \varphi_i \left[\frac{\partial\alpha_i}{\partial t} + j\frac{\partial\alpha_i}{\partial x} + (-1)^i\frac{\partial J}{\partial x}\right] + \varphi_\alpha\sum_i \alpha_i \tag*{(34)}

which accounts for the kinetic energy in terms of jj, JJ and αi\alpha_i, and the admissibility conditions of those variables with the Lagrange multipliers φi\varphi_i and φα\varphi_\alpha.

The following gauging is made to simplify the derivation:

L→L−∑i[∂αiφi∂t+j∂αiφi∂x+(−1)i∂Jφi∂x](35)L \to L - \sum_i \left[\frac{\partial\alpha_i\varphi_i}{\partial t} + j\frac{\partial\alpha_i\varphi_i}{\partial x} + (-1)^i\frac{\partial J\varphi_i}{\partial x}\right] \tag*{(35)}

so, the Lagrangian becomes:

L=12ρmj2+12ΓJ2−Δρ jJ−∑iαi∂φi∂t−j∑iαi∂φi∂x−∑i(−1)i∂φi∂x+φα∑iαi(36)\begin{aligned} L ={}& \frac{1}{2}\rho_m j^2 + \frac{1}{2}\Gamma J^2 - \Delta\rho\,jJ - \sum_i \alpha_i\frac{\partial\varphi_i}{\partial t} - j\sum_i \alpha_i\frac{\partial\varphi_i}{\partial x} - \sum_i (-1)^i\frac{\partial\varphi_i}{\partial x} \\ &+ \varphi_\alpha\sum_i \alpha_i \tag*{(36)} \end{aligned}

The corresponding Euler-Lagrange equations are:

∂L∂j=ρmj−Δρ J−∑iαi∂φi∂x=0(37)\frac{\partial L}{\partial j} = \rho_m j - \Delta\rho\,J - \sum_i \alpha_i\frac{\partial\varphi_i}{\partial x} = 0 \tag*{(37)}
∂L∂J=ΓJ−Δρ j−∑i(−1)i∂φi∂x=0(38)\frac{\partial L}{\partial J} = \Gamma J - \Delta\rho\,j - \sum_i (-1)^i\frac{\partial\varphi_i}{\partial x} = 0 \tag*{(38)}
∂L∂αi=12ρij2+12ΓiJ2−∂φi∂t−j∂φi∂x+φα=0(39)\frac{\partial L}{\partial\alpha_i} = \frac{1}{2}\rho_i j^2 + \frac{1}{2}\Gamma_i J^2 - \frac{\partial\varphi_i}{\partial t} - j\frac{\partial\varphi_i}{\partial x} + \varphi_\alpha= 0 \tag*{(39)}

where Γi≡∂Γ/∂αi\Gamma_i \equiv\partial\Gamma/\partial\alpha_i. Eqs. (SM-4)-(SM-6) form a linear algebraic system for ∂φi/∂x\partial\varphi_i/\partial x and ∂φi/∂t\partial\varphi_i/\partial t, which can be cast in two compact 2×22 \times2 sets as:

Mϕ=β(40)\mathbf{M}\boldsymbol{\phi} = \boldsymbol{\beta} \tag*{(40)}

where:

M=(α1α2−11)(41)\mathbf{M} = \begin{pmatrix} \alpha_1 & \alpha_2 \\ -1 & 1 \end{pmatrix} \tag*{(41)}

and β=(a,b)T\boldsymbol{\beta} = (a,b)^T for ϕ=(∂φ1∂t,∂φ2∂t)T\boldsymbol{\phi} = \left(\frac{\partial\varphi_1}{\partial t},\frac{\partial\varphi_2}{\partial t}\right)^T, β=(c,d)T\boldsymbol{\beta} = (c,d)^T for ϕ=(∂φ1∂x,∂φ2∂x)T\boldsymbol{\phi} = \left(\frac{\partial\varphi_1}{\partial x},\frac{\partial\varphi_2}{\partial x}\right)^T; with:

a=−12ρmj2+jJΔρ+12ΓmJ2+φα(42)a=-\frac{1}{2}\rho_m j^2+jJ\Delta\rho+\frac{1}{2}\Gamma_m J^2+\varphi_\alpha \tag*{(42)}
b=12Δρj2+12Γ′J2−ΓJj(43)b=\frac{1}{2}\Delta\rho j^2+\frac{1}{2}\Gamma'J^2-\Gamma Jj \tag*{(43)}
c=jρm−JΔρ=ρmvm(44)c=j\rho_m-J\Delta\rho=\rho_m v_m \tag*{(44)}
d=ΓJ−Δρj(45)d=\Gamma J-\Delta\rho j \tag*{(45)}

and the following magnitudes are defined to simplify notation:

Γ′=∑i(−1)iΓi(46)\Gamma'=\sum_i(-1)^i\Gamma_i \tag*{(46)}
Γm=∑iαiΓi(47)\Gamma_m=\sum_i\alpha_i\Gamma_i \tag*{(47)}

The solutions are straightforward to obtain as:

ϕ=M−1β=(1−α21α1)β(48)\phi=\mathbf{M}^{-1}\boldsymbol{\beta}= \begin{pmatrix} 1 & -\alpha_2 \\ 1 & \alpha_1 \end{pmatrix}\boldsymbol{\beta} \tag*{(48)}

which finally gives:

∂ϕi∂t=a−(−1)i(1−αi)b(49)\frac{\partial\phi_i}{\partial t}=a-(-1)^i(1-\alpha_i)b \tag*{(49)}
∂ϕi∂x=c−(−1)i(1−αi)d(50)\frac{\partial\phi_i}{\partial x}=c-(-1)^i(1-\alpha_i)d \tag*{(50)}

Cross differentiating and subtracting Eqs. (SM-(16) and (SM-(17), yields to the master transport equation for each fluid:

∂Ui∂t−∂Ξi∂x=0(51)\frac{\partial U_i}{\partial t}-\frac{\partial\Xi_i}{\partial x}=0 \tag*{(51)}

where:

Ui=ρij−[Δρ−(−1)i(1−αi)Γ]j(52)U_i=\rho_i j-\left[\Delta\rho-(-1)^i(1-\alpha_i)\Gamma\right]j \tag*{(52)}
Ξi=jJ[Δρ−(−1)i(1−αi)Γ]+12ΓiJ2+φα(53)\Xi_i=jJ\left[\Delta\rho-(-1)^i(1-\alpha_i)\Gamma\right]+\frac{1}{2}\Gamma_iJ^2+\varphi_\alpha \tag*{(53)}

Equation of the momentum mixture

Multiplying Eqs. (SM-(18) by αi\alpha_i and adding them up for both fluids, yields:

∑iαi∂Ui∂t−∑iαi∂Ξi∂x=0(54)\sum_i\alpha_i\frac{\partial U_i}{\partial t}-\sum_i\alpha_i\frac{\partial\Xi_i}{\partial x}=0 \tag*{(54)}

Using Eq. (SM-(19), the first term of Eq. (SM-(18) gives:

∑iαi∂Ui∂t=ρmdjdt−Δρ∂J∂t+∑i(−1)iαi∂∂t[(1−αi)ΓJ](55)\sum_i \alpha_i \frac{\partial U_i}{\partial t}=\rho_m\frac{dj}{dt}-\Delta\rho\frac{\partial J}{\partial t}+\sum_i(-1)^i\alpha_i\frac{\partial}{\partial t}\left[(1-\alpha_i)\Gamma J\right] \tag*{(55)}

Using Eq. (3), Eq. (SM-(22) can be written as:

∑iαi∂Ui∂t=∂ρmvm∂t+j∂ρmvm∂x+∑i(−1)iαi∂∂t[(1−αi)ΓJ](56)\sum_i\alpha_i\frac{\partial U_i}{\partial t}=\frac{\partial\rho_m v_m}{\partial t}+j\frac{\partial\rho_m v_m}{\partial x}+\sum_i(-1)^i\alpha_i\frac{\partial}{\partial t}\left[(1-\alpha_i)\Gamma J\right] \tag*{(56)}

The second term of Eq. (SM-(21) gives:

∑iαi∂Ξi∂x=jΔρ∂J∂x−j∑i(−1)iαi∂∂x[(1−αi)ΓJ]+12∑iαi∂∂x(ΓJ2)+∂φα∂x(57)\sum_i\alpha_i\frac{\partial\Xi_i}{\partial x}=j\Delta\rho\frac{\partial J}{\partial x}-j\sum_i(-1)^i\alpha_i\frac{\partial}{\partial x}\left[(1-\alpha_i)\Gamma J\right]+\frac{1}{2}\sum_i\alpha_i\frac{\partial}{\partial x}\left(\Gamma J^2\right)+\frac{\partial\varphi_\alpha}{\partial x} \tag*{(57)}

Combining Eqs. (SM-(21) to (SM-(24):

∂ρmvm∂t+j∂ρmvm∂t+∑i(−1)iαi(∂∂t+j∂∂x)[(1−αi)ΓJ]−jΔρ∂J∂x−12∑iαi∂∂x(ΓJ2)−∂φα∂x=0(58)\begin{aligned} \frac{\partial\rho_m v_m}{\partial t}+j\frac{\partial\rho_m v_m}{\partial t}+\sum_i(-1)^i\alpha_i\left(\frac{\partial}{\partial t}+j\frac{\partial}{\partial x}\right)\left[(1-\alpha_i)\Gamma J\right]-j\Delta\rho\frac{\partial J}{\partial x}\\ -\frac{1}{2}\sum_i\alpha_i\frac{\partial}{\partial x}\left(\Gamma J^2\right)-\frac{\partial\varphi_\alpha}{\partial x}=0 \tag*{(58)} \end{aligned}

Using Eq. (7), the third term of Eq. (SM-(25) reduces to:

∑i(−1)iαi(∂∂t+j∂∂x)[(1−αi)ΓJ]=(∂∂t+j∂∂x)[ΓJ∑i(−1)iαi(1−αi)]−ΓJ∑i(−1)i(1−αi)(∂αi∂t+j∂αi∂x)=(∂∂t+j∂∂x)[ΓJ∑i(−1)iαi(1−αi)]+ΓJ∂J∂x∑i(1−αi)=(∂∂t+j∂∂x)[ΓJα1α2∑i(−1)i]+12Γ∂J2∂x=12∂ΓJ2∂x−12J2∂Γ∂x=12∂ΓJ2∂x−12J2∑iΓi∂αi∂x(59)\begin{aligned} \sum_i(-1)^i\alpha_i\left(\frac{\partial}{\partial t}+j\frac{\partial}{\partial x}\right)\left[(1-\alpha_i)\Gamma J\right] &=\left(\frac{\partial}{\partial t}+j\frac{\partial}{\partial x}\right)\left[\Gamma J\sum_i(-1)^i\alpha_i(1-\alpha_i)\right]\\ &-\Gamma J\sum_i(-1)^i(1-\alpha_i)\left(\frac{\partial\alpha_i}{\partial t}+j\frac{\partial\alpha_i}{\partial x}\right)\\ &=\left(\frac{\partial}{\partial t}+j\frac{\partial}{\partial x}\right)\left[\Gamma J\sum_i(-1)^i\alpha_i(1-\alpha_i)\right]+\Gamma J\frac{\partial J}{\partial x}\sum_i(1-\alpha_i)\\ &=\left(\frac{\partial}{\partial t}+j\frac{\partial}{\partial x}\right)\left[\Gamma J\alpha_1\alpha_2\sum_i(-1)^i\right]+\frac{1}{2}\Gamma\frac{\partial J^2}{\partial x}=\frac{1}{2}\frac{\partial\Gamma J^2}{\partial x}-\frac{1}{2}J^2\frac{\partial\Gamma}{\partial x}\\ &=\frac{1}{2}\frac{\partial\Gamma J^2}{\partial x}-\frac{1}{2}J^2\sum_i\Gamma_i\frac{\partial\alpha_i}{\partial x} \tag*{(59)} \end{aligned}

Combining Eqs. (SM-(25) and (SM-(26):

∂ρmvm∂t+j∂∂x(ρmvm−ΔρJ)+12∂ΓJ2∂x−12∑i[J2Γi∂αi∂x+αi∂∂x(ΓJ2)]−∂φα∂x=0(60)\begin{aligned} \frac{\partial\rho_m v_m}{\partial t}+j\frac{\partial}{\partial x}\left(\rho_m v_m-\Delta\rho J\right)+\frac{1}{2}\frac{\partial\Gamma J^2}{\partial x}-\frac{1}{2}\sum_i\left[J^2\Gamma_i\frac{\partial\alpha_i}{\partial x}+\alpha_i\frac{\partial}{\partial x}\left(\Gamma J^2\right)\right]-\frac{\partial\varphi_\alpha}{\partial x}\\ =0 \tag*{(60)} \end{aligned}

The summation in the fourth term of Eq. (SM-(27) reduces to:

∑i[J2Γi∂αi∂x+αi∂∂x(ΓiJ2)]=∂∂x(J2∑iαiΓi)=∂∂x(ΓmJ2)(61)\sum_i \left[J^2\Gamma_i\frac{\partial\alpha_i}{\partial x}+\alpha_i\frac{\partial}{\partial x}\left(\Gamma_iJ^2\right)\right]=\frac{\partial}{\partial x}\left(J^2\sum_i\alpha_i\Gamma_i\right)=\frac{\partial}{\partial x}\left(\Gamma_mJ^2\right) \tag*{(61)}

Combining Eqs. (1), (3), (SM-27) and (SM-28) yields:

∂ρmvm∂t+∂ρmvm2∂x+∂∂x{[12(Γ−Γm)−Δρ2ρm]J2}=∂φα∂x(62)\frac{\partial\rho_m v_m}{\partial t}+\frac{\partial\rho_m v_m^2}{\partial x}+\frac{\partial}{\partial x}\left\{\left[\frac{1}{2}\left(\Gamma-\Gamma_m\right)-\frac{\Delta\rho^2}{\rho_m}\right]J^2\right\}=\frac{\partial\varphi_\alpha}{\partial x} \tag*{(62)}

Using the expressions of Γ\Gamma and Γm\Gamma_m, Eqs. (SM-13) and (SM-14), the bracket in the third term of the r.h.s. of Eq. (SM-29) can be written as:

12(Γ−Γm)−Δρ2ρm=∑iρiαi−Δρ2ρm−12∑iρiαi[γi−d(αiγi)dαi]=ρ1ρ2α1α2ρm−12∑iρiαi2dγidαi.(63)\begin{aligned} \frac{1}{2}\left(\Gamma-\Gamma_m\right)-\frac{\Delta\rho^2}{\rho_m}=\sum_i\frac{\rho_i}{\alpha_i}-\frac{\Delta\rho^2}{\rho_m}-\frac{1}{2}\sum_i\rho_i\alpha_i\left[\gamma_i-\frac{d(\alpha_i\gamma_i)}{d\alpha_i}\right] \\ &=\frac{\rho_1\rho_2}{\alpha_1\alpha_2\rho_m}-\frac{1}{2}\sum_i\rho_i\alpha_i^2\frac{d\gamma_i}{d\alpha_i}. \tag*{(63)} \end{aligned}

Combining Eqs. (SM-29) and (SM-30), finally yields:

∂ρmvm∂t+∂∂x[ρmvm2+ρ1ρ2α1α2ρmJ2+12J2∑iρidγid(1/αi)]=−∂p∂x(64)\frac{\partial\rho_m v_m}{\partial t}+\frac{\partial}{\partial x}\left[\rho_m v_m^2+\frac{\rho_1\rho_2}{\alpha_1\alpha_2\rho_m}J^2+\frac{1}{2}J^2\sum_i\rho_i\frac{d\gamma_i}{d(1/\alpha_i)}\right]=-\frac{\partial p}{\partial x} \tag*{(64)}

Note that the pressure is recognized as the scalar field p=−φαp=-\varphi_\alpha in Eq. (SM-29). The emergence of the pressure field in this fashion is typical in variational formulations of incompressible problems.

Equation of the relative motion between fluids

From Eqs. (SM-19) and (SM-20), we verify that:

∑i(−1)iUi=ΓJ−jΔρ(65)\sum_i(-1)^i\mathcal{U}_i=\Gamma J-j\Delta\rho \tag*{(65)}
∑i(−1)iΞi=12Γ′J2−ΓJj(66)\sum_i(-1)^i\Xi_i=\frac{1}{2}\Gamma'J^2-\Gamma Jj \tag*{(66)}

Hence, the transport equation of the relative motion between fluids is obtained by subtracting Eqs. (SM-18), which gives:

∂Mj∂t+∂Σj∂x=0(67)\frac{\partial\mathcal{M}_j}{\partial t}+\frac{\partial\Sigma_j}{\partial x}=0 \tag*{(67)}

where:

Mj=ΓJ−jΔρ(68)\mathcal{M}_j=\Gamma J-j\Delta\rho \tag*{(68)}
Σj=−12Γ′J2+ΓJj(69)\Sigma_j=-\frac{1}{2}\Gamma'J^2+\Gamma Jj \tag*{(69)}

Conservative forces

Conservative forces can be included by subtracting in the Lagrangian the potential energy from which these forces derive. For upward vertical flow, the gravitational potential is given by:

G=gρmx(70)G = g\rho_m x \tag*{(70)}

This introduces the following term in the derivatives of the Lagrangian:

∂(−G)∂αi=−ρigx(71)\frac{\partial(-G)}{\partial\alpha_i} = -\rho_i g x \tag*{(71)}

which adds a force gρmg\rho_m to the mixture momentum equation, and a force gΔρg\Delta\rho to the equation of the relative motion.

Interfacial drag

Non-conservative interactions between fluids are modeled by means of a drag force DJ\mathcal{D}_J:

DJ=CDDρ1ur∣ur∣(72)\mathcal{D}_J = \frac{C_D}{D}\rho_1 u_r \lvert u_r\rvert \tag*{(72)}

In order to keep the values of the variables within ranges typically encountered in real situations, we produce the interfacial drag coefficient CDC_D from a set of experimental data of the relative velocity uru_r measured at different α2\alpha_2, available in the open literature [13, 14].

At steady state, the drag and buoyancy forces balance each other, and therefore we can write:

CD(α2)=gDΔρur2(α2)ρ1(73)C_D(\alpha_2) = \frac{gD\Delta\rho}{u_r^2(\alpha_2)\rho_1} \tag*{(73)}

Figure 6 shows the phenomenological function fitted together with the experimental data. The resulting drag coefficient function is:

Drag coefficient constitutive law

Figure 6. Drag coefficient constitutive law. (●) Experimental data [14, 15], (solid curve) Eq. (SM-47).

CD=exp⁡(1.32+8.4α2−42α22+45α23−15.4α24)(74)C_D = \exp(1.32 + 8.4\alpha_2 - 42\alpha_2^2 + 45\alpha_2^3 - 15.4\alpha_2^4) \tag*{(74)}

References

  1. [1]M. Ishii, Thermo-fluid dynamic theory of two-phase flow (Eyrolles, Paris, 1975).
  2. [2]D.A. Drew, S.L. Passman, Theory of multicomponent fluids (Springer, New York, 1999).DOI
  3. [3]D. Lhuillier, C.H. Chang, T.G. Theofanous, J. Fluid Mech. 731, 184–194 (2013).
  4. [4]M. López de Bertodano, W.D. Fullmer, A. Clausse, V.H. Ransom, Two-Fluid Model Stability, Simulation and Chaos (Springer, New York, 2017).DOI
  5. [5]J. Hoepffner, R. Blumenthal, S. Zaleski, Phys. Rev Lett. 106, 104502 (2011).
  6. [6]P.M. Lushnikov, N.M. Zubarev, Phys. Rev. Lett. 120, 204504 (2018).
  7. [7]J.A. Geurst, Physica A 135, 55–486 (1986).
  8. [8]G.B. Wallis, Int. J. Multiphase Flow 18, 989–1017 (1992).
  9. [9]C. Pauchon, P. Smereka, Int. J. Multiphase Flow 18, 65–87 (1992).
  10. [10]A. Clausse, Int. J. Heat Tech. 21, 13-20 (2003).
  11. [11]J.D. Ramshaw, J.A. Trapp, Nuclear Sci. Eng. 66, 93–102 (1978).
  12. [12]Lannes, D., Ming, M., in Hamiltonian Partial Differential Equations and Applications, pp. 185-234, edited by P. Guyenne, D. Nicholls and C. Sulem, (Springer, New York, 2015).
  13. [13]A. Vaidheeswaran, A. Clausse, W. D. Fullmer, R. Marino, M. Lopez de Bertodano, Chaos 29, 033121 (2019).
  14. [14]H. Cheng, J.H. Hills, B.J. Azzopardi, Int. J. Multiphase Flow 28, 1047–1062 (2002).
  15. [15]C.H. Song, H.C. No, M.K. Chung, Int. J. Multiphase Flow 21, 381–404 (1995).
  16. [16]G.A. Maugin, Continuum Mech. Thermodynamics 25, 127–146 (2013).
  17. [17]See Supplementary Material.

Paper details

vlnc:W_tcvkuuu9
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Physics and Astronomy

Not recorded for this paper

View statement
Looking for related papers…
Contents