betaFeedback

WorksPhysics and Astronomy

Conditional moment methods for polydisperse cavitating flows

Abstract

The dynamics of cavitation bubbles are important in many flows, but their small sizes and high number densities often preclude direct numerical simulation. We present a computational model that averages their effect on the flow over larger spatiotemporal scales. The model is based on solving a generalized population balance equation (PBE) for nonlinear bubble dynamics and explicitly represents the evolving probability density of bubble radii and radial velocities. Conditional quadrature-based moment methods (QBMMs) are adapted to solve this PBE. A one-way-coupled bubble dynamics problem demonstrates the efficacy of different QBMMs for the evolving bubble statistics. Results show that enforcing hyperbolicity during moment inversion (CHyQMOM) provides comparable model-form accuracy to the traditional conditional method of moments and decreases computational costs by about ten times for a broad range of test cases. The CHyQMOM-based computational model is implemented in MFC, an open-source multi-phase and high-order-accurate flow solver. We assess the effect of the model and its parameters on a two-way coupled bubble screen flow problem.

Introduction

Cavitating flows arise near ship propellers [1], in hydraulic machinery [2], over spillways [3], and during medical therapies like lithotripsy and histotripsy [4, 5, 6]. In these flows, polydisperse bubble clouds are generated directly through nucleation or break-up and shedding larger vapor regions. Directly simulating the flow of these dispersions is challenging since the associated bubble dynamics frequently occur at smaller length and time scales than the background flow [7].

In the dilute limit, a strategy for overcoming this scale separation recognizes that the dynamics of each bubble are unimportant compared to statistics of the bubble population. Ensemble phase-averaging [8, 9], for example, results in equations for the continuous liquid face forced, in a two-way-coupled manner, by statistics of (typically) the bubble radius radial velocity. The associated bubble variables become stochastic, Eulerian fields. This contrasts against Lagrangian models that track, model, and average over samples of individual particles [10, 11]. Unless the bubble populations are such that any small volume (compared to the mixture-averaged flow field variations) contains a sufficient number of bubbles to describe the statistics faithfully, these models simulate at best a single realization of a random process.

The population balance equation (PBE) represents the evolution of a dispersion’s probability (or number density function) according to a set of internal variables [12, 13]. The internal variables must provide sufficient information to close each particle’s dynamic and thermodynamic evolution. The PBE is commonly used to model the particle size distribution. It depends solely on particle motion, coagulation, and breakup [14, 15, 16], wherein the internal variables are the spatial coordinates and velocities. PBE-based models have successfully modeled flowing soot during combustion processes [17], aerosol sprays [18], and more. For oscillating bubbles, the internal variables must include the bubble radius, bubble radial velocity, and sufficient information regarding the bubble contents, which can often characterize with a single equilibrium radius [7]. When appropriate, these three internal variables can be appended to those associated with relative motion, coalescence, and break up [19, 20].

The current work formulates a PBE-based model for the distribution of bubble radius, radial velocity, and equilibrium radius. Once formulated, the PBE is a PDE in independent coordinates associated with each internal variable (as well as spatiotemporal coordinates associated with the continuous flow in which the bubbles reside). Solving this six-dimensional PDE via the method of lines is intractable. Instead, the method of moments represents and evolves the number density function via its statistical moments [21]. Other approaches to solving PBEs exist, like class [22, 23], particle [24], and Monte Carlo [25, 26] methods, though these are comparatively inefficient for multivariate PBEs or large simulations [27].

The ensemble phase-averaging methods determine the governing flow equations for oscillating bubble populations. As described in § 2, we develop an Euler–Euler approach to represent the flow and the averaged bubble dynamics at the sub-grid level [10, 28]. This method is two-way-coupled between the dispersed bubbles and suspending liquid. In particular, moments of the oscillating bubble dynamics alter the effective pressure and void fraction evolution equation.

Next, in § 3, we develop and verify the conditional moment methods discussed above to close these equations. This requires a dynamical/thermodynamical model for each bubble. Even under the assumption of spherical bubbles, the most general form of such a model consists of a set of PDEs for the balance of mass, momentum, energy. Under further assumptions, however, these can be reduced to a set of ODEs for the bubble radius and radial velocity, with equilibrium radius as a parameter [29]. The set of ODEs for each particle determines the derivatives of the internal variables that close the moment transport equations [30]. Quadrature-based moment methods then invert the moment set for a quadrature rule approximating the quantities required to close the moment transport equations [31, 32]. With multiple independent variables, conditional quadrature moment methods are computationally preferable [33]. This work implements the conditional quadrature method of moments (CQMOM [33]) and its hyperbolically constrained version, CHyQMOM [34, 35, 36].

§ 4 describes our interface-capturing numerical algorithm for solving the resulting system of equations. We demonstrate the methodology in § 5 by considering an acoustically excited bubble screen. Lastly, § 6 summarizes our results and potential next steps for extending and analyzing the PBE-based method.

Model formulation

Compressible flow equations

A dilute suspension of dynamically evolving bubbles flow in a compressible liquid is considered. For simplicity, we assume no slip between the bubbles and the surrounding liquid and that the gas density is much smaller than the liquid density. Under these assumptions, the mixture-averaged form of the compressible flow equations are

∂ρ∂t+∇⋅(ρu)=0,∂ρu∂t+∇⋅(ρuu+pI)=0,∂E∂t+∇⋅(E+p)u=0,(1)\begin{aligned} \frac{\partial\rho}{\partial t} + \nabla\cdot(\rho\mathbf{u}) &= 0, \\ \frac{\partial\rho\mathbf{u}}{\partial t} + \nabla\cdot(\rho\mathbf{u}\mathbf{u} + p\mathbf{I}) &= 0, \\ \frac{\partial E}{\partial t} + \nabla\cdot(E+p)\mathbf{u} &= 0, \tag*{(1)} \end{aligned}

where ρ\rho, u\mathbf{u}, pp, and EE are the density, velocity vector, pressure, and total energy. Terms associated with the bubbles modify these quantities and transport them in space according to ensemble phase-averaging, discussed next.

Ensemble phase-averaging

The ensemble-averaged equations follow from Zhang and Prosperetti [8] and Bryngelson et al. [10]. The disperse phase has a void fraction α\alpha and is assumed to be a dilute (α≪1\alpha\ll1) population of spherical bubbles. The bubbles are represented statistically via random variables RR, R˙\dot{R}, and RoR_o corresponding to the instantaneous bubble radius, time derivative, and equilibrium bubble radius. Section 2.3 presents this bubble dynamics model in detail. The mixture-averaged pressure field is computed as

p(x,t)=(1−α)pℓ+α(R3pbw‾R3‾−ρR3R˙2‾R3‾),(2)p(\mathbf{x},t) = (1-\alpha)p_\ell+ \alpha\left( \frac{\overline{R^3p_{bw}}}{\overline{R^3}} - \rho\frac{\overline{R^3\dot{R}^2}}{\overline{R^3}} \right), \tag*{(2)}

where pbwp_{bw} are the associated bubble wall pressure and pℓ(x,t)p_\ell(\mathbf{x},t) is the liquid pressure according to the stiffened-gas equation of state [37]:

Γℓpℓ+Π∞,ℓ=11−α(E−12ρu2).(3)\Gamma_\ell p_\ell+ \Pi_{\infty,\ell} = \frac{1}{1-\alpha}\left(E-\frac{1}{2}\rho u^2\right). \tag*{(3)}

The coefficients of (3) represent water, with specific heat ratio γℓ=1+1/Γℓ=7.15\gamma_\ell= 1 + 1/\Gamma_\ell= 7.15 and stiffness Π∞,ℓ=356\Pi_{\infty,\ell} = 356 MPa [38].

The bubble number density per unit volume n(x,t)n(\mathbf{x},t) is conserved in the absence of coalescence or breakup:

∂n∂t+∇⋅(nu)=0.(4)\frac{\partial n}{\partial t} + \nabla\cdot(n\mathbf{u}) = 0. \tag*{(4)}

For the spherical bubbles considered here, nn is related to the void fraction α\alpha via

α(x,t)=43πR3‾n(x,t),(5)\alpha(\mathbf{x},t) = \frac{4}{3}\pi\overline{R^3}n(\mathbf{x},t), \tag*{(5)}

and thus the void fraction α(x,t)\alpha(\mathbf{x},t) transports as

∂α∂t+u⋅∇α=3αR2R˙‾R3‾,(6)\frac{\partial\alpha}{\partial t} + \mathbf{u}\cdot\nabla\alpha= 3\alpha\frac{\overline{R^2\dot{R}}}{\overline{R^3}}, \tag*{(6)}

where the right-hand-side represents the change in void fraction due to bubble growth and collapse. The over-barred terms appearing in (2), (5), and (6),

R3R˙2‾, R3‾, R2R˙‾, and R3pbw‾.(7)\overline{R^3\dot{R}^2},\ \overline{R^3},\ \overline{R^2\dot{R}},\ \text{and}\ \overline{R^3p_{bw}}. \tag*{(7)}

denote average quantities of the bubble dispersion. In particular, they are raw moments μlmn\mu_{lmn} with respect to a bubble number density function f(R,R˙,Ro)f(R,\dot{R},R_o),

μlmn=RlR˙mRon‾=∫ΩRlR˙mRonf(R,R˙,Ro) dRdR˙dRo,(8)\mu_{lmn}=\overline{R^l\dot{R}^mR_o^n}=\int_{\Omega}R^l\dot{R}^mR_o^n f(R,\dot{R},R_o)\,\mathrm{d}R\mathrm{d}\dot{R}\mathrm{d}R_o, \tag*{(8)}

which are computed via the methods of section 3.

Bubble dynamics model

Even for spherical models, dozens of available models employ differing assumptions about the bubble contents and simplifications of the physics. These models range from a system of two or more ODEs up to a set of PDEs (in a radial coordinate and time) describing the balance of mass, momentum, and energy of each bubble. PDE-based models are deemed intractable (at present), and we focus on ODE-based modeling. We choose one of the simplest possible models to demonstrate our methodology but note that our framework can be extended to arbitrarily complex ODE-based models.

We assume the spherical bubbles are filled with noncondensible gas and that, insofar as their dynamics are concerned, the liquid may be assumed incompressible. We assume that the gas undergoes a polytropic process during compression and expansion. Under these assumptions, the bubble radius is governed by a Rayleigh–Plesslet-like equation

RR¨+32R˙2+4Re⁡R˙R=(RoR)3γ−1Cp−2We⁡Ro[RoR−(RoR)3γ],(9)R\ddot{R}+\frac{3}{2}\dot{R}^{2}+\frac{4}{\operatorname{Re}}\frac{\dot{R}}{R}=\left(\frac{R_o}{R}\right)^{3\gamma}-\frac{1}{C_p}-\frac{2}{\operatorname{We}R_o}\left[\frac{R_o}{R}-\left(\frac{R_o}{R}\right)^{3\gamma}\right], \tag*{(9)}

which is dimensionless via the reference bubble size Ro∗R_o^* and ambient liquid pressure p0p_0, and density ρ0\rho_0. The polytropic index is γ\gamma; in examples below we use γ=1.4\gamma=1.4. In (9), Cp≡p0/pℓC_p\equiv p_0/p_\ell is the forcing pressure ratio and Reynolds and Weber numbers correspond to viscous and surface tension effects as

Re⁡≡p0ρ0Ro∗ν0andWe⁡≡p0Ro∗S,(10)\operatorname{Re}\equiv\sqrt{\frac{p_0}{\rho_0}}\frac{R_o^*}{\nu_0}\quad\text{and}\quad\operatorname{We}\equiv\frac{p_0R_o^*}{S}, \tag*{(10)}

where SS is the air–water surface tension coefficient and ν0\nu_0 is the liquid kinematic viscosity. Under these conditions, the bubble wall pressure of (2) simplifies to pbw=(Ro/R)3γp_{bw}=(R_o/R)^{3\gamma} and the last moment of (7) reduces to R3‾(Ro/R)3γ\overline{R^3}(R_o/R)^{3\gamma}.

Population balance formulation

The population balance equation (PBE)

∂f∂t+∂∂R(fR˙)+∂∂R˙(fR¨)=0(11)\frac{\partial f}{\partial t}+\frac{\partial}{\partial R}(f\dot{R})+\frac{\partial}{\partial\dot{R}}(f\ddot{R})=0 \tag*{(11)}

governs the conditional PDF f(R,R˙∣Ro)f(R,\dot{R}\mid R_o) in the absence of bubble coalescence or breakup, though this approach can naturally accommodate these effects if required. Figure 1 summarizes the PBE–quadrature approach.

Figure 1. Illustration of the quadrature-based moment method for a fully coupled bubbly flow.

Method of moments

Following the usual procedure, a finite set of raw moments μ⃗\vec{\mu} represent ff per (8) [14]. The specific moments that make up μ⃗\vec{\mu} depend on the inversion algorithm and are defined in the appendices A and B. These moments transport on the grid and evolve according to the bubble dynamics of section 2.3 as

∂nμ⃗∂t+∇⋅(nμ⃗u)=nμ⃗˙=ng⃗(12)\frac{\partial n\vec{\mu}}{\partial t}+\nabla\cdot(n\vec{\mu}u)=n\dot{\vec{\mu}}=n\vec{g} \tag*{(12)}

where

glmn=lμl−1,m+1,n+m∭ΩR¨RlR˙m−1Ronf(μ⃗) dRdR˙dRo(13)g_{lmn}=l\mu_{l-1,m+1,n}+m\iiint_{\Omega}\ddot{R}R^{l}\dot{R}^{m-1}R_{o}^{n}f(\vec{\mu})\,\mathrm{d}R\mathrm{d}\dot{R}\mathrm{d}R_{o} \tag*{(13)}

and Ω=ΩR×ΩR˙×ΩRo=(0,∞)×(−∞,∞)×(0,∞)\Omega=\Omega_R\times\Omega_{\dot{R}}\times\Omega_{R_o}=(0,\infty)\times(-\infty,\infty)\times(0,\infty). The integrand of (13) is closed via the bubble dynamics model (9) for R¨\ddot{R}, and the integral is approximated via quadrature, which is discussed next.

Conditional quadrature moment inversion

Since RoR_o is not a dynamic variable, the number density function is split as

f(R,R˙,Ro)=f(R,R˙∣Ro)f(Ro).(14)f(R,\dot{R},R_o)=f(R,\dot{R}\mid R_o)f(R_o). \tag*{(14)}

The raw moments (8) are then

μlmn≡∫ΩRof(Ro)Ronμlm(Ro) dRo(15)\mu_{lmn}\equiv\int_{\Omega_{R_o}}f(R_o)R_o^{n}\mu_{lm}(R_o)\,\mathrm{d}R_o \tag*{(15)}
≈∑i=1NRowiR^o,inμlm(R^o,i),(16)\approx\sum_{i=1}^{N_{R_o}}w_i\widehat{R}_{o,i}^{n}\mu_{lm}(\widehat{R}_{o,i}), \tag*{(16)}

with NRoN_{R_o} time-independent weights wiw_i and nodes R^o,i\widehat{R}_{o,i}. Simpson’s rule computes these nodes and weights and the accuracy in the approximation of (16) depends on NRoN_{R_o}. Other numerical integration methods are suitable and will be discussed in section 5.3. The RoR_o-conditioned moments are

μlm(R^o,i)≡∬ΩR,R˙f(R,R˙∣R^o,i)RlR˙m dRdR˙(17)\mu_{lm}(\widehat{R}_{o,i}) \equiv\iint_{\Omega_{R,\dot{R}}} f(R,\dot{R}|\widehat{R}_{o,i})R^l\dot{R}^m\,\mathrm{d}R\mathrm{d}\dot{R} \tag*{(17)}
≈∑j=1NR∑k=1NR˙[w^j,kR^jlR˙^km]R^o,i.(18)\approx\sum_{j=1}^{N_R}\sum_{k=1}^{N_{\dot{R}}}\left[\widehat{w}_{j,k}\widehat{R}_j^l\widehat{\dot{R}}_k^m\right]_{\widehat{R}_{o,i}} . \tag*{(18)}

The moment indices comprising the moment set of (18) are determined by the conditional quadrature moment method used to invert those moments for a set of quadrature points and weights (and the number of points desired) [33, 35]. In particular, μlm(R^o,i)\mu_{lm}(\widehat{R}_{o,i}) is traded for quadrature points {R^j,R˙^k}(R^o,i)\{\widehat{R}_j,\widehat{\dot{R}}_k\}(\widehat{R}_{o,i}) and weights w^j,k(R^o,i)\widehat{w}_{j,k}(\widehat{R}_{o,i}) for each i=1,…,NRoi=1,\ldots,N_{R_o} (with j=1,…,NRj=1,\ldots,N_R; k=1,…,NR˙k=1,\ldots,N_{\dot{R}}). The total moments (16) are then approximated by substituting (18) into (16) as

μlmn=∑i=1NRowiR^o,in∑j=1NR∑k=1NR˙[w^j,kR^jlR˙^km]R^o,i.(19)\mu_{lmn} = \sum_{i=1}^{N_{R_o}} w_i\widehat{R}_{o,i}^{n}\sum_{j=1}^{N_R}\sum_{k=1}^{N_{\dot{R}}}\left[\widehat{w}_{j,k}\widehat{R}_j^l\widehat{\dot{R}}_k^m\right]_{\widehat{R}_{o,i}} . \tag*{(19)}

The CHyQMOM and CQMOM algorithms are described in appendices A and B. Their implementation is verified by ensuring that the error in closing a linear set of moment transport equations is comparable to a finite-precision round-off [39].

Model performance

We consider the ability of the QBMMs to represent the statistics of the bubble dynamics. For this example, we take a monodisperse population with the same equilibrium radius Ro=1R_o=1 (and thus NRo=1N_{R_o}=1). RR and R˙\dot{R} are initialized with independent log-normal and normal distributions with variances σR2\sigma_R^2 and σR˙2\sigma_{\dot{R}}^2, respectively. The bubbles are forced by a step-change in pressure at t=0+t=0^+ from equilibrium to a value CpC_p (which is varied). This represents a stringent test of model fidelity as high values of CpC_p will induce strongly nonlinear bubble dynamics.

Figure 2 shows the evolution of the carried moment set for Cp=0.3C_p=0.3. To assess the accuracy, the CHyQMOM results are compared against Monte-Carlo solutions that used 10410^4 samples to ensure that the sampling error was at least 10-times smaller than the QBMM model-form error. Thus the Monte-Carlo solutions can be regarded as exact solutions for these comparisons. The radial moments μ10\mu_{10} and μ20\mu_{20} show how the bubbles oscillate in response to the change in external pressure, eventually reaching a statistical equilibrium [40]. These oscillations are damped rapidly for smaller bubbles due to stronger viscous effects (smaller Re). Thus, larger bubbles entail larger model-form errors as they build over time. This is most prominent for the R˙\dot{R} moments, such as μ02\mu_{02}, and smaller bubbles ((b) Ro∗=10 μmR_o^*=10\,\mu\mathrm{m}) as the velocity statistics change more quickly than the radial ones.

A QBMM model-form relative L2L_2 error is

ε≡1Nt∑i=1Nt[μ(QBMM)(ti)−μ(MC)(ti)μ(MC)(ti)]2,(20)\varepsilon\equiv\frac{1}{N_t}\sqrt{\sum_{i=1}^{N_t}\left[\frac{\mu^{(\mathrm{QBMM})}(t_i)-\mu^{(\mathrm{MC})}(t_i)}{\mu^{(\mathrm{MC})}(t_i)}\right]^2}, \tag*{(20)}

where Monte-Carlo (MC) simulations serve as surrogate truth data and tit_i are Nt=103N_t=10^3 uniformly spaced times in the time interval t∈[0,T]t\in[0,T]. Figure 3 shows ε\varepsilon for select first- and second-order moments μ\mu. We see that ε\varepsilon is smaller for CHyQMOM and CQMOM than Gaussian closure, though the difference is modest and more strongly associated with CpC_p. The larger errors for smaller CpC_p

First- and second-order moments for bubbles with Reynolds and Weber numbers corresponding to two cases

Figure 2. Example first- and second-order moments μ\mu as labeled for bubbles with Reynolds and Weber numbers corresponding to (a) Ro∗=1 μmR_o^* = 1\ \mu\mathrm{m} and (b) Ro∗=10 μmR_o^* = 10\ \mu\mathrm{m}. The dimensionless initial shape parameters are σR=σR˙=0.2\sigma_R = \sigma_{\dot{R}} = 0.2. The CHyQMOM implementation uses two quadrature nodes in each internal coordinate direction (2×22 \times2), and “Gaussian” corresponds to Gaussian closure [10]. Monte Carlo serves as an exact solution. The final time is T=13.9T = 13.9 and represents 10 natural periods of the largest bubble.

are associated with stronger bubble dynamics and the formation of non-Gaussian statistics, like skewness and kurtosis, that these closures do not represent [41]. One can represent higher-order statistics by carrying higher-order moments and thus inverting for a higher-order quadrature rule. However, this is computationally cumbersome and numerically unstable in the cases tested here.

Model accuracy was similar for all of the closure methods, but the computational time to solution is not. Figure 4 shows the total relative simulation time for the cases of Figure 3 under the same adaptive time-step tolerance. CHyQMOM simulations are about 10-times cheaper than those using CQMOM or Gaussian closures, even though the accuracy is the same. This is attributed to larger time step sizes (given the same error constraint) and a smaller carried moment set than CQMOM. Due to its low cost for the same accuracy, the fully-coupled simulation algorithm employs CHyQMOM and is discussed next.

Relative model-form errors for varying pressure ratio and closure methods

Figure 3. Relative model-form errors for varying pressure ratio CpC_p and closure methods as labeled. All cases have Re=100\mathrm{Re} = 100 and We=13.9\mathrm{We} = 13.9.

Figure 4. Total simulation times for the cases of Figure 3. A single-core of a 2.8 GHz Intel i7-7700HQ performed the computations. Only relative times are relevant, so the cheapest simulation normalizes the scale (CHyQMOM, Cp=0.8C_p = 0.8).

Interface-capturing flow solution algorithm

The QBMM approach of the previous section is solved using interface-capturing numerics. MFC, an open-source flow solver, is the basis for the implementation [42]. A brief description of the algorithm follows.

The governing equations (1), (6), and (12) combine as

∂q⃗c∂t+∇⋅F⃗=r⃗(21)\frac{\partial\vec{q}_c}{\partial t} + \nabla\cdot\vec{F} = \vec{r} \tag*{(21)}

where q⃗c={ρ,ρu,E,α,nμ⃗}\vec{q}_c = \{\rho, \rho u, E, \alpha, n\vec{\mu}\} are the conservative variables, F⃗\vec{F} are the advective fluxes, and r⃗\vec{r} are diffusive source terms. Finite volumes with uniform size Δ\Delta discretize the domain.

Flow state initialization

We initialize the flow as follows. The number density function f(R,R˙,Ro)f(R, \dot{R}, R_o) is independently distributed with log-normal, normal, and log-normal shapes in the RR, R˙\dot{R}, and RoR_o directions with expected values E[R]=E[Ro]=1\mathbb{E}[R] = \mathbb{E}[R_o] = 1 and E[R˙]=0\mathbb{E}[\dot{R}] = 0 and shape parameters σ\sigma.

The NDF ff initializes the moment set μ⃗\vec{\mu} via integration. The primitive variables q⃗p≡{ρ,u,p,α,μ⃗}\vec{q}_p \equiv\{\rho, u, p, \alpha, \vec{\mu}\} are initialized as appropriate. The bubble number density nn follows from (5), q⃗p\vec{q}_p, and μ300\mu_{300}. Lastly, the known primitive variables are used to compute the conservative ones as q⃗c={ρ,ρu,E,α,nμ⃗}\vec{q}_c = \{\rho, \rho u, E, \alpha, n\vec{\mu}\}.

Algorithm 1 Flow initialization procedure

  1. f(R,R˙,Ro)←f(R,\dot{R},R_o) \leftarrow Presumed form, uncorrelated, {μR,μR˙,μRo}\{\mu_R,\mu_{\dot{R}},\mu_{R_o}\} and {σR,σR˙,σRo}\{\sigma_R,\sigma_{\dot{R}},\sigma_{R_o}\}

  2. μ⃗←f\vec{\mu} \leftarrow f and (8)

  3. q⃗p≡{ρ,u,p,α,μ⃗}←\vec{q}_p \equiv\{\rho,u,p,\alpha,\vec{\mu}\} \leftarrow Patches

  4. n←α,μ300n \leftarrow\alpha,\mu_{300} and (5)

  5. q⃗c←n,q⃗p\vec{q}_c \leftarrow n,\vec{q}_p

Flux divergence computation

A fifth-order-accurate WENO [43] scheme reconstructs the primitive variables q⃗p\vec{q}_p and the HLLC approximate Riemann solver [44] computes the fluxes. Algorithm 2 describes this process in detail. High-order WENO reconstructions do not guarantee that the reconstructed moments are realizable, though the moment sets remained invertible in the subsequent simulations.

Algorithm 2 Algorithm for computing the flux divergence.

  1. q⃗p←q⃗c\vec{q}_p \leftarrow\vec{q}_c and (3)

  2. q⃗^p←WENO⁡(q⃗p)\hat{\vec{q}}_p \leftarrow\operatorname{WENO}(\vec{q}_p)

  3. {R^,R˙^,w^}jk←QBMM⁡(μ^)\{\hat{R},\hat{\dot{R}},\hat{w}\}_{jk} \leftarrow\operatorname{QBMM}(\hat{\mu})

  4. g⃗,{R3R˙2,R3,R2R˙,R3pbw}←{R^o,w}i,{R^,R˙^,w^}jk,i\vec{g},\{R^3\dot{R}^2,R^3,R^2\dot{R},R^3p_{bw}\} \leftarrow\{\hat{R}_o,w\}_i,\{\hat{R},\hat{\dot{R}},\hat{w}\}_{jk,i}

  5. F⃗←HLLC⁡(q⃗^p,g⃗,{⋅})\vec{F} \leftarrow\operatorname{HLLC}(\hat{\vec{q}}_p,\vec{g},\{\cdot\})

  6. ∇⋅F⃗←F⃗\nabla\cdot\vec{F} \leftarrow\vec{F}

Time stepping

The conservative variables are integrated in time using third-order-accurate SSP–RK3 time integration [45]. Once the spatial derivatives have been approximated, (21) becomes a semi-discrete system of ordinary differential equations in time. We treat the temporal derivative using a Runge–Kutta time-marching scheme for the state variables. To achieve high-order accuracy and avoid spurious oscillations, we use the third-order-accurate total variation diminishing scheme of Gottlieb and Shu [46]:

q⃗c(1)=q⃗cn~+ΔtL(q⃗cn~),q⃗c(2)=34q⃗cn~+14q⃗c(1)+14ΔtL(q⃗c(1)),q⃗cn~+1=13q⃗cn~+23q⃗c(2)+23ΔtL(q⃗c(2)),(22)\begin{aligned} \vec{q}_c^{(1)} &= \vec{q}_c^{\tilde{n}}+\Delta tL(\vec{q}_c^{\tilde{n}}),\\ \vec{q}_c^{(2)} &= \frac{3}{4}\vec{q}_c^{\tilde{n}}+\frac{1}{4}\vec{q}_c^{(1)}+\frac{1}{4}\Delta tL(\vec{q}_c^{(1)}),\\ \vec{q}_c^{\tilde{n}+1} &= \frac{1}{3}\vec{q}_c^{\tilde{n}}+\frac{2}{3}\vec{q}_c^{(2)}+\frac{2}{3}\Delta tL(\vec{q}_c^{(2)}), \tag*{(22)} \end{aligned}

where superscripts (1) and (2) indicate intermediate time-step stages, LL represents the portion of (21) that does not include the time derivative, and n~\tilde{n} is the time-step index.

Application to polydisperse bubble screens

Problem setup

We assess the statistics of bubble dynamics in an idealized model problem consisting of an acoustically excited dilute bubble screen. The bubble screen parameterization matches that of Bryngelson et al. [10], with a length of Ls=5 mmL_s=5\,\mathrm{mm}, initial void fraction αo=10−4\alpha_o=10^{-4}, median bubble equilibrium size Ro∗=10 μmR_o^* = 10\,\mu\mathrm{m} and log-normal variance σRo\sigma_{R_o}. The domain is Ld=5LsL_d = 5L_s long, and its boundaries are non-reflective via characteristic-based boundary conditions [47]. The one-way (positive x^\hat{x}-direction) sound wave p∞(t)p_\infty(t) is generated via source terms in the governing equations (21) according to Bryngelson et al. [10]. Its form is a single period of a sinusoid with peak amplitude 0.3p00.3p_0 and frequency 300 kHz.

Figure 5. Schematic of the acoustically excited bubble screen test problem. The screen is centered at x=0x = 0.

Bubble screen behavior

We start by considering a screen with fixed dynamic coordinate distributions σR=σR˙\sigma_R = \sigma_{\dot{R}}, but varying distributions of equilibrium sizes σRo\sigma_{R_o}. Polydispersity in RoR_o is integrated via Simpson’s rule 61 quadrature points for all cases. Figure 6 shows the bubble screen pressure for these cases as they evolve in time. For larger σRo\sigma_{R_o} (or broader distributions or bubble equilibrium sizes), the pressure is less oscillatory in time. A similar observation was made by Bryngelson et al. [10] for cases with no RR or R˙\dot{R} distributions. In the case of Figure 6, we instead observe high-frequency oscillations in addition to the long-wavelength behaviors associated with the impinging pressure wave p∞p_\infty. The origin of these oscillations is discussed next.

Bubble-screen-centered pressure for varying $R_o$ log-normal distributions

Figure 6. Bubble-screen-centered pressure for varying RoR_o log-normal distributions with shape parameter σRo\sigma_{R_o} and fixed σR=σR˙=0.2\sigma_R = \sigma_{\dot{R}} = 0.2.

Figure 7 shows the dynamics associated with a bubble screen in varying degrees of statistical disequilibrium, represented via different σR\sigma_R and σR˙\sigma_{\dot{R}}. Figure 7 (a) fixes σR˙\sigma_{\dot{R}} and varies σR\sigma_R. We observe the shorter-wavelength oscillatory behavior, observed in Figure 6, becoming more prominent for larger σR\sigma_R. These wavelengths are commensurate with the mean bubble natural frequencies, which superimpose the longer wavelength acoustics associated with the impinging p∞p_\infty wave. Figure 7 (b) shows a smoother pressure profile for larger σR˙\sigma_{\dot{R}}. Phase-cancellation between the larger waves associated with broader σR˙\sigma_{\dot{R}} distributions and those of the σR\sigma_R distributions may account for this behavior. Notably, these behaviors are qualitatively similar to those associated with varying RoR_o distribution widths. Thus, parameterizing an RoR_o distribution based on single-probe pressure measurements is insufficient.

Bubble-screen-centered pressure before, during, and after excitement due to an acoustic wave

Figure 7. Bubble-screen-centered pressure before, during, and after excitement due to an acoustic wave. The bubbles are polydisperse with log-normal RoR_o distribution (σRo=0.2\sigma_{R_o}=0.2) and Re=103\mathrm{Re}=10^3. Variations in (a) σR\sigma_R and (b) σR˙\sigma_{\dot{R}} are shown about a σR=σR˙=0.2\sigma_R=\sigma_{\dot{R}}=0.2 representative state.

Closure errors

We quantify the moment closure error, εc\varepsilon_c, via the mismatch in bubble screen pressure p(t,x=0)p(t,x=0) due to truncated RoR_o integration. Otherwise, the definition of εc\varepsilon_c has the same form as ε\varepsilon of (20), though the truth values are changed from Monte-Carlo simulations (which are unavailable) to high-resolution NRo=401N_{R_o}=401 simulations as

εc≡1Nt∑i=1Nt[p(QBMM)(ti,x=0)−p(NRo=401)(ti,x=0)p(NRo=401)(ti,x=0)]2.(23)\varepsilon_c \equiv\sqrt{\frac{1}{N_t}\sum_{i=1}^{N_t}\left[\frac{p^{(\mathrm{QBMM})}(t_i,x=0)-p^{(N_{R_o}=401)}(t_i,x=0)}{p^{(N_{R_o}=401)}(t_i,x=0)}\right]^2}. \tag*{(23)}

This choice is made for two reasons. First, extending the QBMM method to additional RR and R˙\dot{R} quadrature points (or moments, equivalently) was found numerically unstable for most bubble cavitation problems. One approach to addressing this specific problem is introducing a recurrent neural network to correct the quadrature points and weights [48], though we do not discuss it further here. Second, we will see that the closure errors are most strongly associated with NRoN_{R_o}, and thus focus on the influence of this parameter.

Figure 8 shows the NRoN_{R_o} closure errors associated with variations in all three PDF directions (panels a–c). For varying σR\sigma_R (a) and σRo\sigma_{R_o} (c) we see that increasing variance σ⋅\sigma_{\cdot} results in larger closure errors εc\varepsilon_c. This appears to be associated with the larger pressure oscillations p(x=0,t)p(x=0,t) observed for such cases. For figure 8 (b), we see the reverse trend, with larger σR˙\sigma_{\dot{R}} corresponding to smaller closure errors, though this effect is small. Indeed, this effect matches that of the RR- and RoR_o-direction effects, where larger σR˙\sigma_{\dot{R}} results in smoother pressure histories. We also investigated the effect of Re\mathrm{Re}, which results in smoother pressure profiles for smaller Re\mathrm{Re}, and found the same trend.

Relative closure error for increasing number of $R_o$-direction quadrature points

Figure 8. Relative closure error εc\varepsilon_c (defined in (23)) for increasing number of RoR_o-direction quadrature points NRoN_{R_o}. Variations in (a) σR\sigma_R, (b) σR˙\sigma_{\dot{R}}, and (c) σRo\sigma_{R_o} are shown. Unless labeled otherwise, cases have the baseline σR=σR˙=σRo=0.2\sigma_R=\sigma_{\dot{R}}=\sigma_{R_o}=0.2.

Figure 9. Closure error associated with σR=σR˙=σR0=0.2\sigma_R = \sigma_{\dot{R}} = \sigma_{R_0} = 0.2 for different R0R_0-direction quadrature rules as labeled.

Compelled by the closure error associated with NR0N_{R_0} dominating the total simulation error, we also implemented Gaussian–Hermite and Gauss–Legendre quadrature methods in the R0R_0-direction in an attempt to reduce the error. Figure 9 shows these closures errors ϵc\epsilon_c for increasing NR0N_{R_0}. We see that these alternative quadrature rules do not significantly change the closure error, despite being optimal for a given NR0N_{R_0}. This is because the integrand becomes highly oscillatory without viscous damping, as bubbles of different equilibrium radii R0R_0 become out of phase [40]. Long-time simulations without significant viscous effects, broadly polydisperse bubble populations (large σR0\sigma_{R_0}) will require treatment of these developing oscillations. For example, a time-stationary analysis could be used to approximate the R0R_0-moments, so long as the bubble dynamics are fast compared to the fluid flow [40]. We will investigate this in future work.

Conclusion

A fully coupled numerical method for simulating sub-grid cavitating bubble dispersions was presented. The approach represents and evolves the statistics of the bubble dynamic variables. This work built upon the well-established framework of quadrature-based moment methods: a governing set of moment transport equations were derived, effectively representing the underlying statistics. Those moments were inverted for a quadrature rule to close the fully coupled disperse flow equations.

Our results showed that only four quadrature points, two in each of the bubble dynamics coordinates, can be sufficient to represent the statistics of the monodisperse bubbles. This model was particularly accurate for weak pressure forcing, and thus nearly linear bubble dynamics. It was also accurate for highly damped dynamics, characteristic of low Reynolds numbers, as non-Gaussian statistics cannot present themselves. This result contrasts against a previous approach that assumed Gaussian statistics, which had demonstrably worse performance and higher computational costs [41]. The method would require more quadrature points for cases with more complicated dynamics, such as larger forcing.

An acoustically excited dilute bubble screen problem demonstrated the model’s behavior in a coupled flow. Modeling the statistics in the bubble dynamics variables resulted in qualitatively different behavior in the screen region. For example, increasing the distribution breadth in the bubble radius coordinate resulted in short-wavelength pressure oscillations of increasing magnitude superimposing the background response to the impinging acoustic wave. Thus, modeling the RR–R˙\dot{R} distributions is potentially critical to modeling cavitating bubble clouds and certainly important if bubbles are ever in such statistical disequilibrium.

Finally, we found that for broadly polydisperse populations (large σR0\sigma_{R_0}), the closure error in the equilibrium radius coordinate R0R_0 was most important. Our results also show that phase-cancellation can modestly reduce some computational costs associated with resolving the RoR_o coordinate. Still, RoR_o-direction quadrature dominates the solution cost of these polydisperse bubble populations, even in the face of more sophisticated quadrature rules like Gauss–Hermite (corresponding to the underlying log-normal distribution). However, one can apply existing approaches if time-separation is achieved, or a novel method based on, e.g., neural networks, can also treat this. We will investigate these in-depth in future work.

Acknowledgements

We thank Alexis Charalampopoulos and Themis Sapsis for invaluable discussions regarding the presented method. The US Office of Naval Research supported this work under grant numbers N0014-17-1-2676 and N0014-18-1-2625. Computations were performed via the Extreme Science and Engineering Discovery Environment (XSEDE) under allocations TG-CTS120005 (PI Colonius) and TG-PHY210084 (PI Bryngelson), supported by National Science Foundation grant number ACI-1548562.

CHyQMOM moment inversion

This appendix describes the 2D, 2×22 \times2 (4) node CHyQMOM algorithm of [34]. We note that Patel et al. [35] provides the 3D version of this algorithm, though it is not necessary for our bubble dynamics problem. Algorithm 3 is the full 2D CHyQMOM algorithm, which references the 1D 2-node HyQMOM algorithm 4. The optimal moment set for this case is

μ⃗={μ00,μ10,μ01,μ20,μ02,μ11},(24)\vec{\mu} = \{\mu_{00},\mu_{10},\mu_{01},\mu_{20},\mu_{02},\mu_{11}\}, \tag*{(24)}

which is also the input to algorithm 3.

Algorithm 3 CHyQMOM 2×22 \times2. μ\mu are the input moments, DD are normalized moments, CC are central moments, and ww, xx and yy are the weights and node locations in the first and second coordinate directions (corresponding to RR and R˙\dot{R} in the main text).

  1. procedure CHyQMOM4(μ\mu)

  2. Dij←μij/μ00D_{ij} \leftarrow\mu_{ij}/\mu_{00}

  3. C20←D20−D102C_{20} \leftarrow D_{20} - D_{10}^{2}

  4. C02←D02−D012C_{02} \leftarrow D_{02} - D_{01}^{2}

  5. C11←D11−D10D01C_{11} \leftarrow D_{11} - D_{10}D_{01}

  6. ρ,x′=HyQMOM2⁡({1,0,C20})\rho, x' = \operatorname{HyQMOM2}(\{1,0,C_{20}\})

  7. yj′=xj′C11/C20y'_j = x'_j C_{11}/C_{20}

  8. μavg2=C02−∑jρjyj′2\mu_{\mathrm{avg}}^{2} = C_{02} - \sum_j \rho_j y_j'^{2}

  9. ρ~,x~′=HyQMOM2⁡({1,0,μavg2})\tilde{\rho}, \tilde{x}' = \operatorname{HyQMOM2}(\{1,0,\mu_{\mathrm{avg}}^{2}\})

  10. for i,j∈[1,2]i,j \in[1,2] do

  11. wij←μ00ρiρ~jw_{ij} \leftarrow\mu_{00}\rho_i\tilde{\rho}_j

  12. xij←D10+xj′x_{ij} \leftarrow D_{10} + x'_j

  13. yij←D01+yj′+x~i′y_{ij} \leftarrow D_{01} + y'_j + \tilde{x}'_i

  14. end for

  15. return w,x,yw,x,y

  16. end procedure

Algorithm 4 The two-node 1D HyQMOM algorithm. Nomenclature follows algorithm 3, with ww and xx serving as dummy weights and node locations.

  1. procedure HyQMOM2(μ\mu)

  2. C2←(μ0μ2−μ12)/μ02C_2 \leftarrow(\mu_0\mu_2-\mu_1^2)/\mu_0^2

  3. w1=w2←μ0/2w_1 = w_2 \leftarrow\mu_0/2

  4. x1←μ1/μ0+C2x_1 \leftarrow\mu_1/\mu_0 + \sqrt{C_2}

  5. x2←μ1/μ0−C2x_2 \leftarrow\mu_1/\mu_0 - \sqrt{C_2}

  6. return w,xw,x

  7. end procedure

B CQMOM moment inversion

This appendix has the same form as appendix A, though it instead describes the 2D CQMOM algorithm of [33]. Algorithm B is the full 2D CQMOM algorithm, referencing Wheeler’s method for computing optimal 1D quadrature nodes and weights 6. The optimal moment set for this case is

μ~={μ00,μ10,μ01,μ20,μ02,μ11,μ30,μ03,μ12,μ13},(25)\tilde{\boldsymbol{\mu}} = \{\mu_{00}, \mu_{10}, \mu_{01}, \mu_{20}, \mu_{02}, \mu_{11}, \mu_{30}, \mu_{03}, \mu_{12}, \mu_{13}\}, \tag*{(25)}

which is also the input to algorithm 5.

Algorithm 5 2D CQMOM algorithm with dummy coordinate direction yy conditioned on xx. V\mathbf{V} is a Vandermonde matrix and W\mathbf{W} is a diagonal weight matrix.

  1. procedure CQMOM(μ\boldsymbol{\mu})

  2. w(x),x←Wheeler⁡({μ00,μ10,…,μ2Nx−1,0})\mathbf{w}^{(x)}, \mathbf{x} \leftarrow\operatorname{Wheeler}(\{\mu_{00}, \mu_{10}, \ldots, \mu_{2N_x-1,0}\})

  3. for i,j∈[1,…,Nx]i,j \in[1,\ldots,N_x] do

  4. Vij=xji−1V_{ij} = x_j^{i-1}

  5. Wii=wi(x)W_{ii} = w_i^{(x)}

  6. end for

  7. for i∈[0,…,2Ny−1]i \in[0,\ldots,2N_y-1] do

  8. xi′←VWxi′=μ0:Nx−1,i\mathbf{x}'_i \leftarrow\mathbf{V}\mathbf{W}\mathbf{x}'_i = \mu_{0:N_x-1,i}

  9. wij(y),yij←Wheeler⁡(xi′)w_{ij}^{(y)}, y_{ij} \leftarrow\operatorname{Wheeler}(\mathbf{x}'_i)

  10. end for

  11. wij=wi(x)wij(y)w_{ij} = w_i^{(x)} w_{ij}^{(y)}

  12. return w,x,y\mathbf{w},\mathbf{x},\mathbf{y}

  13. end procedure

Algorithm 6 Wheeler’s algorithm for optimal quadrature points x\mathbf{x} and weights w\mathbf{w} given a moment set μ\boldsymbol{\mu} [32, 49]. J\mathbf{J} is a Jacobi matrix, and EVD refers to the eigenvalue decomposition.

  1. procedure Wheeler(μ\boldsymbol{\mu})

  2. for i∈[1,…,2N]i \in[1,\ldots,2N] do

  3. σ1,i←μi−1\sigma_{1,i} \leftarrow\mu_{i-1}

  4. end for

  5. a0←μ1/μ0a_0 \leftarrow\mu_1/\mu_0

  6. b0←0b_0 \leftarrow0

  7. for i∈[2,…,N]i \in[2,\ldots,N] do

  8. for j∈[i,…,2N−i+1]j \in[i,\ldots,2N-i+1] do

  9. σi,j←σi−1,j+1−ai−2σi−1,i−1\sigma_{i,j} \leftarrow\sigma_{i-1,j+1}-a_{i-2}\sigma_{i-1,i-1}

  10. end for

  11. ai−1←σi,i+1/σi,i−σi−1,i/σi−1,i−1a_{i-1} \leftarrow\sigma_{i,i+1}/\sigma_{i,i}-\sigma_{i-1,i}/\sigma_{i-1,i-1}

  12. bi−1←σi,i/σi−1,i−1b_{i-1} \leftarrow\sigma_{i,i}/\sigma_{i-1,i-1}

  13. end for

  14. for i∈[2,…,N−1]i \in[2,\ldots,N-1] do

  15. Ji,i←ai−1J_{i,i} \leftarrow a_{i-1}

  16. Ji+1,i=Ji,i+1←biJ_{i+1,i}=J_{i,i+1} \leftarrow\sqrt{b_i}

  17. end for

  18. QΛQ−1←EVD⁡(J)\mathbf{Q}\boldsymbol{\Lambda}\mathbf{Q}^{-1} \leftarrow\operatorname{EVD}(\mathbf{J})

  19. for i∈[1,…,N]i \in[1,\ldots,N] do

  20. xi←Λiix_i \leftarrow\Lambda_{ii}

  21. wi←μ0Qi,12w_i \leftarrow\mu_0 Q_{i,1}^{2}

  22. end for

  23. return w,x\mathbf{w},\mathbf{x}

  24. end procedure

References

  1. [1]S. S. Cook, Erosion by water-hammer, Proc. Royal Soc. London A 119 (1928) 481–488.
  2. [2]R. Arndt, Cavitation in fluid machinery and hydraulic structures, Annu. Rev. Fluid Mech. 13 (1981) 273–326.DOI
  3. [3]H. T. Falvey, Cavitation in chutes and spillways, US Department of the Interior, Bureau of Reclamation Denver, 1990.
  4. [4]R. O. Cleveland, O. A. Sapozhnikov, M. R. Bailey, L. A. Crum, A dual passive cavitation detector for localized detection of lithotripsy-induced cavitation in vitro, J. Acoust. Soc. Am. 107 (2000) 1745–1758.pubmed.ncbi.nlm.nih.gov/10738826
  5. [5]Y. A. Pishchalnikov, O. A. Sapozhnikov, M. R. Bailey, J. C. Williams, R. O. Cleveland, T. Colonius, L. A. Crum, A. P. Evan, J. A. McAteer, Cavitation bubble cluster activity in the breakage of kidney stones by lithotripter shockwaves, J. Endourol. 17 (2003) 435–446.pubmed.ncbi.nlm.nih.gov/14565872
  6. [6]A. D. Maxwell, T.-Y. Wang, C. A. Cain, J. B. Fowlkes, O. A. Sapozhnikov, M. R. Bailey, Z. Xu, Cavitation clouds created by shock scattering from bubbles during histotripsy, J. Acoust. Soc. Am. 130 (2011) 1888–1898.DOI
  7. [7]C. E. Brennen, Cavitation and bubble dynamics, Oxford University Press, 1995.
  8. [8]D. Z. Zhang, A. Prosperetti, Ensemble phase-averaged equations for bubbly flows, Phys. Fluids 6 (1994).DOI
  9. [9]D. Z. Zhang, A. Prosperetti, Averaged equations for inviscid disperse two-phase flow, J. Fluid Mech. 267 (1994) 185–219.DOI
  10. [10]S. H. Bryngelson, K. Schmidmayer, T. Colonius, A quantitative comparison of phase-averaged models for bubbly, cavitating flows, Int. J. Mult. Flow 115 (2019) 137–143.DOI
  11. [11]J. CapeceLatro, O. Desjardins, An Euler–Lagrange strategy for simulating particle-laden flows, J. Comp. Phys. 238 (2013) 1–31.
  12. [12]A. D. Randolph, M. A. Larson, Theory of particulate processes, San Diego, Academic Press, 1987.
  13. [13]D. Ramkrishna, Population Balances, Academic Press, New York, USA, 2000.
  14. [14]R. O. Fox, Computational models for turbulent reacting flows, Cambridge University Press, 2003.DOI
  15. [15]S. Rigopoulos, Population balance modelling of polydispersed particles in reactive flows, Prog. Energy Combust. Sci. 36 (2010) 412–443.DOI
  16. [16]K. F. Lee, R. I. Patterson, W. Wagner, M. Kraft, Stochastic weighted particle methods for population balance equations with coagulation, fragmentation and spatial inhomogeneity, J. Comp. Phys. 303 (2015) 1–18.DOI
  17. [17]M. E. Mueller, G. Blanquart, H. Pitsch, A joint volume-surface model of soot aggregation with the method of moments, Proc. Combust. Inst. 32 1 (2009) 785–792.DOI
  18. [18]A. Sibra, J. Dupays, A. Murrone, F. Laurent, M. Massot, Simulation of reactive polydisperse sprays strongly coupled to unsteady flows in solid rocket motors: Efficient strategy using Eulerian multi-fluid methods, J. Comp. Phys. 339 (2017) 210–246.DOI
  19. [19]P. Carrica, D. Drew, F. Bonetto, R. Lahey, A polydisperse model for bubbly two-phase flow around a surface ship, Int. J. Mult. Flow 25 (1999) 257–305.DOI
  20. [20]J. C. Heylmun, B. Kong, A. Passalacqua, R. Fox, A quadrature-based moment method for polydisperse bubbly flows, Comp. Phys. Comm. (2019).DOI
  21. [21]H. M. Hulburt, S. Katz, Some problems in particle technology. A statistical mechanical formulation, Chem. Eng. Sci. 19 (1964) 555–574.
  22. [22]M. J. Hounslow, R. L. Ryall, V. R. Marshall, A discretized population balance for nucleation, growth, and aggregation, AIChE J. 34 (1988) 1821–1832.DOI
  23. [23]M. Vanni, Approximate population balance equations for aggregation breakage processes, J. Colloid Interface Sci. 221 (2000) 143–160.pubmed.ncbi.nlm.nih.gov/10631014
  24. [24]R. I. Patterson, W. Wagner, M. Kraft, Stochastic weighted particle methods for population balance equations, J. Comp. Phys. 230 (2011) 7456–7472.
  25. [25]H. Zhao, A. Maisels, T. Matsoukas, C. Zheng, Analysis of four monte-carlo methods for the solution of population balances in dispersed systems, Powder Technol. 173 (2007) 38–50.DOI
  26. [26]D. E. Rosner, R. McGraw, P. Tandon, Multivariate population balances via moment and Monte Carlo simulation methods: an important sol reaction engineering bivariate example and “mixed” moments for the estimation of deposition, scavenging, and optical properties for populations of nonspherical suspended particles, Indust. Eng. Chem. Res. 42 (2003) 2699–2711.DOI
  27. [27]A. Zucca, D. L. Marchisio, M. Vanni, A. A. Barresi, Validation of bivariate DQMOM for nanoparticle processes simulation, AiChE J. 53 (2007) 918–931.DOI
  28. [28]K. Ando, T. Colonius, C. E. Brennen, Numerical simulation of shock propagation in a polydisperse bubbly liquid, Int. J. Mult. Flow 37 (2011) 596–608.DOI
  29. [29]M. S. Plesset, A. Prosperetti, Bubble dynamics and cavitation, Annu. Rev. Fluid Mech. 9 (1977) 145–185.
  30. [30]M. Frenklach, S. Harris, Aerosol dynamics modeling using the method of moments, J. Colloid Interface Sci. 118 (1987) 252–261.DOI
  31. [31]R. McGraw, Description of aerosol dynamics by the quadrature method of moments, Aerosol Sci. Technol. 27 (1997) 255–265.DOI
  32. [32]D. L. Marchisio, R. O. Fox, Computational models for polydisperse particulate and multiphase systems, Cambridge University Press, 2013.
  33. [33]C. Yuan, R. O. Fox, Conditional quadrature method of moments for kinetic equations, J. Comp. Phys. 230 (2011) 8216–8246.DOI
  34. [34]R. O. Fox, F. Laurent, A. Vié, Conditional hyperbolic quadrature method of moments for kinetic equations, J. Comp. Phys. 365 (2018) 269–293.DOI
  35. [35]R. G. Patel, O. Desjardins, R. O. Fox, Three-dimensional conditional hyperbolic quadrature method of moments, J. Comp. Phys. X 1 (2019) 100006.DOI
  36. [36]R. Fox, F. Laurent, Hyperbolic quadrature method of moments for the one-dimensional kinetic equation, arXiv:2103.10138 (2021).arxiv.org/abs/2103.10138
  37. [37]R. Menikoff, B. J. Plohr, The Riemann problem for fluid-flow of real materials, Rev. Mod. Phys. 61 (1989) 75–130.DOI
  38. [38]K. Maeda, T. Colonius, Eulerian–Lagrangian method for simulation of cloud cavitation, J. Comp. Phys. 371 (2018) 994–1017.DOI
  39. [39]S. H. Bryngelson, T. Colonius, R. O. Fox, QBMMlib: A library of quadrature-based moment methods, SoftwareX 12 (2020) 100615.DOI
  40. [40]T. Colonius, R. Hagmeijer, K. Ando, C. E. Brennen, Statistical equilibrium of bubble oscillations in dilute bubbly flows, Phys. Fluids 20 (2008).DOI
  41. [41]S. H. Bryngelson, A. Charalampopoulos, T. P. Sapsis, T. Colonius, A Gaussian moment method and its augmentation via LSTM recurrent neural networks for the statistics of cavitating bubble populations, Int. J. Mult. Flow 127 (2020) 103262.DOI
  42. [42]S. H. Bryngelson, K. Schmidmayer, V. Coralic, J. C. Meng, K. Maeda, T. Colonius, MFC: An open-source high-order multi-component, multi-phase, and multi-scale compressible flow solver, Comp. Phys. Comm. (2020) 107396.DOI
  43. [43]G.-S. Jiang, C.-W. Shu, Efficient implementation of weighted eno schemes, J. Comp. Phys. 126 (1996) 202–228.
  44. [44]E. Toro, M. Spruce, W. Speares, Restoration of the contact surface in the HLL-Riemann solver, Shock waves 4 (1994) 25–34.DOI
  45. [45]S. Gottlieb, C. W. Shu, E. Tadmor, Strong stability-preserving high-order time discretization methods, SIAM Rev. (2001).DOI
  46. [46]S. Gottlieb, C.-W. Shu, Total variation diminishing Runge–Kutta schemes, Math. Comput. 67 (1998) 73–85.DOI
  47. [47]P. O. Thompson, W. C. Cummings, S. J. Ha, Sounds, source levels, and associated behavior of humpback whales, Southeast Alaska, J. Acoust. Soc. Am. 80 (1986) 735–740.pubmed.ncbi.nlm.nih.gov/3760326
  48. [48]A.-T. Charalampopoulos, S. H. Bryngelson, T. Colonius, T. P. Sapsis, Hybrid quadrature moment method for accurate and stable representation of non-Gaussian processes and their dynamics, arXiv:2110.01374 (2021).DOI
  49. [49]J. C. Wheeler, Modified Moments and Gaussian Quadratures, Rocky Mt. J. Math. 4 (1974) 287–296.DOI

Paper details

vlnc:W_bcweshh9
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