A fast, decomposed pressure correction method for an intrusive stochastic multiphase flow solver
Introduction
Methods of uncertainty quantification (UQ) can be placed into two categories: intrusive and non-intrusive. The non-intrusive category includes approaches such as Monte Carlo [1], collocation methods [2], and non-intrusive polynomial chaos (PC) [3]. The latter two essentially improve on a Monte Carlo by presenting a better way to select the input values so not as many simulations need to be run. In all cases a standard solver can be utilized, which is run many times with parameters selected from a distribution of inputs to compile a database of simulation results. This database is then used to calculate useful statistics of the system in question.
Unlike non-intrusive flow solvers, intrusive UQ methods require a change in the fundamental structure of the solver resulting from modified equations created by the inclusion of stochastic (random) variables. An intrusive solver is created with stochastic variables which store information as a function of added uncertainty dimensions. Stochastic variables developed as a function of uncertainty may take several forms, including PC [4] and Karhunen–Loeve expansions [5], [6]. Each of these methods offer their own advantages. The advantage of PC lies in the ability to utilize any number of uncertainty dimensions, the availability of a number of orthogonal basis function families, and the straightforward integration of continuous PC variables into many systems of differential equations. Assuming stable intrusive and non-intrusive UQ schemes, computational cost comparisons are based on the time it takes to generate a reliable source of statistical information on the system being modeled.
Considering UQ applications to multiphase flow dynamics, application of intrusive UQ methods to gas-liquid flows is a developing field. Le Maître et al. [7], [8] first developed the stochastic Navier–Stokes equations for single-phase incompressible flows utilizing a PC expansion. Since these works, several studies have implemented a PC-based approach to single-phase flows for a variety of test cases [9], [10], [11]. Previous work by Turnquist and Owkes [12] provided the first intrusive UQ method for gas-liquid multiphase flows named the multiUQ framework. The current work builds on this previous work by reducing the computational cost.
In either intrusive or non-intrusive UQ methodology, much computational expense is devoted to solving the pressure Poisson equation. To numerically solve the incompressible Navier–Stokes equations, the standard pressure correction method (SPCM), first introduced by Chorin [13], is a commonly used approach. With the SPCM, time is discretized so that at every time step the convective, viscosity, and any source terms are evaluated and used to predict the velocity without the pressure term. Continuity (or mass conservation) is then used to enforce a divergence free condition at the next time step, while also creating an elliptic Poisson equation to solve for pressure. The approach makes it possible to solve the Navier–Stokes equations with imposed boundary conditions at reasonable computational expense. Further work has been done to expand the method, including improvements to order [14] and application to unstructured grids [15]. When using the SPCM in a multiphase scenario, the pressure Poisson equation becomes coupled to density, which adds computational cost and limits the possible algorithms used to solve. In an effort to counter this cost and following the work of Dong and Shen [16], Dodd and Ferrante [17] proposed a decomposed pressure correction method (DPCM) which would allow for using a fast Fourier transform (FFT) based solver. While numerical errors are added to the model, computational cost is reduced; certainly the trade off is worth consideration.
Given the computational cost improvements of a DPCM in the deterministic setting, it seemed reasonable to apply this methodology to the multiUQ framework [12]. Because of the coupled nature of non-linear terms in the stochastic Navier–Stokes equations due to the use of PC variables, the simulation expense grows at an exponential rate. This so-called curse of dimensionality increases the computational cost very rapidly for intrusive UQ. However, the same curse also affects non-intrusive methods. For example, the use of a Monte-Carlo [1] approach with two or more uncertain variables requires a way to compare the effect of one uncertain variable on another, compounding the number of simulations run to get convergent statistics. Due to this problem, understanding the interaction in uncertainty between multiple variables in a multiphase system is extremely expensive. Use of accurate and cost effective numerical techniques will bring these analyses within reach.
This narrative seeks to develop a more efficient pressure correction approach for stochastic multiphase flows by applying the DPCM to the multiUQ framework outlined in Turnquist and Owkes [12]. A mathematical development of the stochastic DPCM is introduced, followed by a derivation of the numerical methods. We then present test cases which illustrate the computational improvement over previously published methods and the error associated with the density decoupled approach. Finally, we close with a summary of the results and a discussion of where this work will fit in moving forward.
Section snippets
Mathematical development
Since the focus of this work is to develop an efficient pressure solver for stochastic multiphase flows, we begin with a development of the stochastic equations for fluid motion. Assuming the fluids are incompressible, this motion can be explained by the Navier–Stokes equations, wherefor velocity time specific volume (for density ), pressure dynamic viscosity and surface tension force where is the interface normal vector, is the
Numerical methodology
Computations are done on a two-dimensional rectangular domain with a structured Cartesian mesh. Scalar values such as pressure level set density and viscosity are held at the cell center. Subscripts on denote discrete spatial indexing in the and directions, respectively, while superscripts denote time discretization. Vector components of velocity surface tension and continuous normal vector are held at the cell walls. Second-order finite difference operators are
Test cases and computational assessment
Two test cases are used to evaluate the accuracy and efficiency of the proposed method, while a third is used to test the method on a more complicated scenario. First, an oscillating droplet case is used as there exists an analytic solution providing the oscillation period [22]. This case tests the ability of the surface tension force to drive flow. Second, a damped surface wave, which also has an analytic solution [23], is used to judge the accuracy of the interplay between viscosity and
Conclusions
We have presented a modified pressure correction method, extending the density decoupled approach of Dodd and Ferrante [17] to an intrusive stochastic multiphase solver. Deviations from the standard pressure correction method arise due to differences between the estimated pressure and the actual pressure . We propose a method to reduce these deviations by imposing a semi-Lagrangian extrapolation method for a better initial estimate of then further improve by iterating over the
CRediT authorship contribution statement
Brian Turnquist: Writing - original draft, Writing - review & editing, Methodology, Software, Conceptualization. Mark Owkes: Writing - review & editing, Supervision, Funding acquisition, Conceptualization.
Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Acknowledgments
This material is based upon work supported by the National Science Foundation under grant nos. 1511325 and 1749779. Computational efforts were performed on the Hyalite High-Performance Computing System, operated and supported by University Information Technology Research Cyberinfrastructure at Montana State University.
References (32)
- et al.
A spectral collocation method for the Navier–Stokes equations
J Comput Phys
(1985) - et al.
Polynomial chaos expansion for subsurface flows with uncertain soil parameters
Adv Water Resour
(2013) - et al.
Stochastic 2D incompressible Navier–Stokes solver using the vorticity-stream function formulation.
J Appl Math
(2013) A second-order accurate pressure-correction scheme for viscous incompressible flow
SIAM J Sci Stat Comput
(1986)- et al.
A continuum method for modeling surface tension
J Comput Phys
(1992) - et al.
Direct numerical simulations of gas–liquid multiphase flows
(2011) - et al.
Design of the Hypre Preconditioner Library
Tech. Rep.
(1998) Motion of two superposed viscous fluids
Phys Fluids
(1981)- et al.
A finite-volume HLLC-based scheme for compressible interfacial flows with surface tension
J Comput Phys
(2017) - et al.
A coupled level-set and reference map method for interface representation with applications to two-phase flows simulation
J Comput Phys
(2019)
Surface tension and viscosity with lagrangian hydrodynamics on a triangular mesh
J Comput Phys
A balanced force refined level set grid method for two-phase flows on unstructured flow solver grids
J Comput Phys
A localized re-initialization equation for the conservative level set method
J Comput Phys
The Monte Carlo method
J Am Stat Assoc
A non-intrusive polynomial chaos method for uncertainty propagation in CFD simulations
44th AIAA Aerospace sciences meeting and exhibit. aerospace sciences meetings
Cited by (5)
Temporal atomization of a transcritical liquid n-decane jet into oxygen
2022, International Journal of Multiphase FlowmultiUQ: A software package for uncertainty quantification of multiphase flows
2021, Computer Physics CommunicationsA volume-of-fluid method for variable-density, two-phase flows at supercritical pressure
2022, Physics of Fluids