betaFeedback

WorksPhysics and Astronomy

A Volume of Fluid Immersed Boundary Method for Industrial Polymer Mixing

Abstract

This work develops advanced numerical methods for free-surface simulations of polymer mixing processes, integrating a Volume of Fluid (VOF) interface-capturing approach with a non-conforming Immersed Boundary (IB) method to model two-phase flows of highly viscous polymer melts and air within partially filled rotating mixing devices, implemented within the Finite Volume OpenFOAM library. To overcome severe numerical instabilities arising from the strong viscosity contrast between polymer melts and air, a block-coupled scheme providing fully implicit viscous diffusion treatment is integrated into the VOF-IB framework, relaxing time-step stability constraints and substantially reducing computational cost with respect to standard segregated solvers. The resulting BC-VOF-IB solver is applied to industrially relevant geometries of single- and twin-screw extruders, yielding physically consistent predictions of velocity and pressure fields under partial filling conditions. While further developments, most notably the inclusion of thermal effects, remain necessary, the proposed framework represents a meaningful step toward bridging academic CFD research and the practical demands of industrial polymer processing.

Introduction

Polymer mixing processes constitute a large family of industrial processes aimed at homogenizing polymeric compounds to achieve desired chemical and physical properties for applications ranging from automotive components to medical devices [24].

In practice, mixing occurs in the melt state, with polymers that are heated above their melting point and blended with other polymers or additives such as fillers, pigments, plasticizers, and stabilizers [56, 80]. These operations are carried out using continuous machines, such as single-screw extruders (SSE) or twin-screw extruders (TSE) or batch systems, e.g., internal mixers. In all mixers, rotating screws or rotors subject the material to high shear and temperature, converting mechanical energy into heat through viscous dissipation, promoting diffusive and convective transport of the components [38, 75]. The quality of the final product, including its mechanical, thermal, rheological and optical properties, depends critically on how well components are distributed (components spreading) and dispersed (components rupture) during mixing. Poor mixing leads to inhomogeneities such as agglomerates or uneven filler distribution, resulting in defective or under-performing parts.

Computational fluid-dynamics (CFD) simulations are essential for understanding, designing and optimizing polymer mixing processes, as laboratory experiments alone are often costly, time-consuming and limited in their ability to resolve internal flow patterns, temperature fields and species distributions [53, 55, 4]. Numerical models allow for the reconstruction of three-dimensional velocity, pressure, stress and temperature fields, as well as local mixing indices that are otherwise inaccessible experimentally.

However, the numerical simulation of polymer mixing processes poses several challenges. First, the non-Newtonian rheology of polymer melts requires advanced constitutive equations to capture shear-thinning and elastic effects, introducing nonlinearities and potential numerical instabilities. Second, the multi-physics nature of the problem, including variable temperature fields, viscous dissipation, chemical reactions, and mass transfer, must be solved simultaneously. Third, complex three-dimensional geometries with narrow gaps and moving parts demand careful mesh generation, often relying on unstructured, moving, or non-conforming grids. Lastly, multiple fluid phases can coexist in the mixing apparatus under partial filling conditions, where the polymer melt interacts with air or another gas, leading to free-surface dynamics that must be captured using proper multiphase models. Ad-hoc interface-tracking or interface-capturing algorithms should be adopted, which typically require the solution of additional equations and with stricter numerical stability constraints, thus increasing the overall computational cost when compared to single phase simulations.

This work addresses these challenges by proposing a novel numerical approach for free-surface simulation of polymer mixing processes, exploiting a non-conforming method to handle complex geometries. Depending on the physical regime under consideration, several numerical methods are available in the literature to perform simulations of multiphase flows, which are typically classified according to the interface topology as either dispersed flows, where discrete bubbles, droplets, or particles are suspended in a continuous phase, or segregated flows, where phases are separated by a well-defined interface. In the present context, a segregated flow regime is assumed, consisting of two immiscible incompressible fluid phases coexisting within the same spatial domain and separated by an interface. Segregated flows are typically addressed by interface-resolving approaches, where the exact position and deformation of the boundary are tracked or captured as part of the solution [54, 57]. These methods are broadly divided into interface tracking and interface capturing [26, 77, 54, 95]. Interface tracking methods, such as Front-Tracking (FT) [84], moving mesh or Arbitrary Lagrangian-Eulerian (ALE) [36, 99, 86], conform the mesh to the moving interface and offer high accuracy, but become computationally demanding under large deformations [84, 36]. In contrast, interface capturing methods rely on an Eulerian description, where the interface is implicitly identified by a scalar indicator function evolving on a fixed grid [76, 78, 89]. Volume of Fluid (VOF) [37], Level Set (LS) [79, 66, 68, 17, 81] and Phase-Field (PF) [42, 3] are the most prominent approaches.

In this work, the VOF method is adopted, where the interface is captured by solving a transport equation for the phasic volume fraction, with the main advantage being its inherent mass conservation [77, 57, 52, 8]. In this work, VOF is adopted as it represents a well-established and extensively documented approach within the OpenFOAM framework [82], the software used throughout this study, in line with its widespread adoption in both academic and industrial studies [30, 77]. When simulating multiphase flows interacting with moving solid bodies of arbitrarily complex shapes, additional numerical challenges arise, as the inherent complexities of multiphase systems must be combined with the accurate and efficient treatment of a time-dependent computational domain.

Non-conforming approaches, such as Immersed Boundary Methods (IBM), offer a practical solution, allowing to solve the flow on a fixed background grid while accounting for solid boundaries through forcing terms or velocity corrections [58, 40, 44]. Few works are available in the literature that combine IBM with VOF [78, 71, 50, 39] or Level Set [10, 32] to obtain multiphase simulations within domains with complex boundaries.

Applications of multiphase methods to polymer processing have grown in relevance, given the prevalence of partially filled configurations in industrial mixers and extruders [80, 75, 56], which feature complicated geometries with rotating boundaries, as well as highly viscous non-Newtonian material rheologies, typically modeled by generalized Newtonian constitutive laws or viscoelastic models [51, 12, 67, 34]. In the context of batch processes, VOF combined with moving-mesh techniques has been used to study fill factor effects and viscous heating in the processing of rubber and polymers [2, 73, 88]. Concerning continuous processes, VOF with a cut-cell approach has been applied for simulations of single-screw extruders [63, 48], while the only known application of VOF-IBM to a continuous mixing geometry concerns a twin-screw kneader [78]. The Level Set method has also been used to model gas injection in a co-rotating twin-screw extruder [46].

The scarcity of works simultaneously addressing free-surface flows, rotating complex geometries, and non-Newtonian rheology, particularly for continuous mixing processes, motivates the development of our non-conforming interface-capturing framework. This work builds on an existing non-conforming Immersed Boundary (IB) method, developed through the consolidated partnership between Pirelli Tyre S.p.A. and the MOX Laboratory at Politecnico di Milano [60, 62], extending the framework by integrating a Volume of Fluid (VOF) approach to simulate the two-phase flow of polymer melt and air in partially filled mixing devices. The proposed methods are implemented in the in-house polymer mixing library, built on the OpenFOAM C++ CFD library [82, 43, 94] (OpenFOAM-10 version from The OpenFOAM Foundation), which is a Finite Volume (FV)-based software widely used for industrial CFD applications.

The technical challenges addressed are two-fold. First, the accurate treatment of moving contact lines, where the two-phase interface intersects immersed boundaries, poses significant geometric and algorithmic difficulties. Secondly, strong viscosity contrasts between polymer melt and air lead to numerical instabilities in classical segregated VOF solvers.

The main contributions of the work are: (i) a non-conforming VOF-IB solver capable of handling moving contact lines along non-conforming moving boundaries; (ii) a block-coupled (BC) scheme for the momentum equation that employs a fully implicit treatment of the viscous diffusion term, to improve stability under high viscosity contrasts, yielding the BC-VOF and BC-VOF-IB solvers. A graphical overview of the different VOF solvers, distinguishing between segregated and block-coupled (BC), as well as conforming and IB approaches, is presented in Figure 1.

Figure 1. Classification of the investigated Volume of Fluid (VOF) solvers. Green outlines denote the original implementations developed in the present work.

The formulation of the two-phase Navier-Stokes equations is presented in Section 2. Afterwards, the formulation of the VOF-IB method is discussed in Section 3, by first providing an overview of the underlying non-conforming IB approach [62]

Mathematical Model

We consider a spatial domain Ω⊂Rd\Omega\subset\mathbb{R}^{d} containing two immiscible fluid phases, each occupying the volume Ωk\Omega_k, k=1,2k = 1, 2, separated by the two-phase interface Σt\Sigma_t and such that Ω=Ω1∪Ω2\Omega= \Omega_1 \cup\Omega_2 and we consider a time-interval I⊂R+\mathcal{I} \subset\mathbb{R}^{+}. We denote by Σ=∣Σ∣nΣ\boldsymbol{\Sigma} = \lvert\Sigma\rvert\mathbf{n}_{\Sigma} the surface normal vector associated to the interface separating the two phases, directed from phase 1 to phase 2, with nΣ=n12=−n21\mathbf{n}_{\Sigma} = \mathbf{n}_{12} = -\mathbf{n}_{21} unit normal vector and ∣Σ∣\lvert\Sigma\rvert surface area of the two-phase interface (see Figure 2).

Figure 2. Two-dimensional sketch of the domain Ωt\Omega_t.

The Volume of Fluid governing equations rely on a single-field (or mixture) approach that allows to write a unique set of incompressible Navier-Stokes equations, valid in the whole domain Ω\Omega.

Specifically, the derivation stems from two sets of incompressible Navier-Stokes equations valid in the two phase domains, equipped with proper interfacial coupling conditions on Σ\Sigma. Then, by means of a conditional averaging procedure that avoids the otherwise unfeasible local microscopic description of the two-phase interface, the two-phase Navier-Stokes equations in single-field formulation are obtained, endowed with an advection equation for the two-phase interface evolution. The rigorous derivation is based on the works proposed in [23, 92, 35, 76, 54], while in the following we the main definitions and the final form of the two-phase VOF equations are reported.

The phase indicator function χ(t,x)\chi(t,\mathbf{x}) associated to phase 1 is defined as

χ(t,x)={1if x∈V1, t∈I,0if x∈V2, t∈I,(1)\chi(t,\mathbf{x}) = \begin{cases} 1 & \text{if } \mathbf{x} \in V_1,\ t \in\mathcal{I}, \\ 0 & \text{if } \mathbf{x} \in V_2,\ t \in\mathcal{I}, \end{cases} \tag*{(1)}

where V⊂ΩV \subset\Omega, such that V=V1∪V2V = V_1 \cup V_2, is a control volume contained in Ω\Omega.

The phase volume fraction is therefore computed as the volume average of χ\chi over each control volume VV:

α=χ‾=1∣V∣∫Vχ dV=∣V1∣∣V∣.(2)\alpha= \overline{\chi} = \frac{1}{\lvert V\rvert}\int_V \chi\,\mathrm{d}V = \frac{\lvert V_1\rvert}{\lvert V\rvert}. \tag*{(2)}

This allows to define the two-phase mixture density and viscosity as the weighted average of the properties of the two fluids using the phase volume fraction as the weighting factor, such that ρ=αρ1+(1−α)ρ2\rho= \alpha\rho_1 + (1-\alpha)\rho_2 and μ=αμ1+(1−α)μ2\mu= \alpha\mu_1 + (1-\alpha)\mu_2, respectively.

Remark 1. In the context of free-surface simulations of polymeric flows, we will always consider the highly viscous polymeric material is to be fluid 1 and air will be indicated as fluid 2. As mentioned in Section 1, we model the polymeric viscosity by means of generalized Newtonian constitutive laws that relate the viscosity μ1\mu_1 to the shear rate γ˙\dot{\gamma} [85]. In this work, we consider a Power Law model [67] μ1(γ˙)=Kγ˙n−1\mu_1(\dot{\gamma}) = K\dot{\gamma}^{n-1}, where KK [Pa⋅sn\mathrm{Pa}\cdot\mathrm{s}^{n}] is the consistency factor, equal to the material viscosity when γ˙=1\dot{\gamma} = 1 [s−1\mathrm{s}^{-1}], and nn is the power law exponent. Polymers usually feature a shear-thinning behavior, meaning that viscosity increases as γ˙\dot{\gamma} increases, corresponding to n<1n < 1.

A modified pressure, prgh=p−ρg⋅xp_{rgh} = p - \rho\mathbf{g}\cdot\mathbf{x}, is defined by removing the hydrostatic pressure from the total pressure pp, aiming at improving the numerical stability and simplifying the imposition of boundary conditions [76]. The gravitational acceleration is denoted by g\mathbf{g}, while x\mathbf{x} indicates the position vector.

The two-phase VOF Navier-Stokes equations read:

∂ρu∂t+∇⋅(ρu⊗u)−∇⋅(μ(∇u+∇Tu))+∇prgh=fVOFin Ωt, t∈I,∇⋅u=0in Ωt, t∈I,∂α∂t+∇⋅(αu)+∇⋅(α(1−α)uc)=0in Ωt, t∈I,u=gDon ΓtD, t∈I,σn=gNon ΓtN, t∈I,α=gαon Γtin, t∈I,u=u0in Ω0,α=α0in Ω0.(3)\begin{aligned} \frac{\partial\rho u}{\partial t} + \nabla\cdot(\rho u \otimes u) - \nabla\cdot\left(\mu(\nabla u + \nabla^{T}u)\right) + \nabla p_{rgh} &= f_{\mathrm{VOF}} && \text{in } \Omega_t,\ t \in\mathcal{I}, \\ \nabla\cdot u &= 0 && \text{in } \Omega_t,\ t \in\mathcal{I}, \\ \frac{\partial\alpha}{\partial t} + \nabla\cdot(\alpha u) + \nabla\cdot\left(\alpha(1-\alpha)u_c\right) &= 0 && \text{in } \Omega_t,\ t \in\mathcal{I}, \\ u &= g_D && \text{on } \Gamma_t^D,\ t \in\mathcal{I}, \\ \sigma n &= g_N && \text{on } \Gamma_t^N,\ t \in\mathcal{I}, \\ \alpha&= g_\alpha&& \text{on } \Gamma_t^{\mathrm{in}},\ t \in\mathcal{I}, \\ u &= u_0 && \text{in } \Omega_0, \\ \alpha&= \alpha_0 && \text{in } \Omega_0. \tag*{(3)} \end{aligned}

Equation (3a) is the two-phase momentum balance, where an additional source term appears at right-hand-side, whose expression is given by

fVOF=−∇ρ(g⋅x)+fσ(α)=−∇ρ(g⋅x)−σ∇⋅(∇α∥∇α∥)∇α.(4)f_{\mathrm{VOF}} = -\nabla\rho(g \cdot x) + f_\sigma(\alpha) = -\nabla\rho(g \cdot x) - \sigma\nabla\cdot\left(\frac{\nabla\alpha}{\lVert\nabla\alpha\rVert}\right)\nabla\alpha. \tag*{(4)}

The first term results from the substitution of ∇p\nabla p with ∇prgh\nabla p_{rgh}. On the other hand, the second term accounts for the effect of surface tension, that is a microscopic force acting at the two-phase interface, arising from the imposition of the momentum balance at the free-surface. Specifically, the Continuous Surface Force (CSF) proposed by Brackbill [7] is employed, allowing to treat the surface tension contribution as a volume source localized at the interface, defined as

fσ=σκ∇α=−σ∇⋅(∇α∥∇α∥)∇α,(5)f_\sigma= \sigma\kappa\nabla\alpha= -\sigma\nabla\cdot\left(\frac{\nabla\alpha}{\lVert\nabla\alpha\rVert}\right)\nabla\alpha, \tag*{(5)}

with κ\kappa denoting the interface curvature, computed as κ=−∇⋅nΣ=−∇⋅(∇α∥∇α∥)\kappa= -\nabla\cdot n_\Sigma= -\nabla\cdot\left(\frac{\nabla\alpha}{\lVert\nabla\alpha\rVert}\right).

Equation (3b) is the mass balance, under the assumption of incompressible fluid phase. Equation (3c) is the advection equation for the phase volume fraction α\alpha, used to follow the free-surface evolution. An additional nonlinear transport contribution is added for numerical purposes, introducing a compressive velocity ucu_c that acts to compress the interface thickness counteracting the numerical diffusion [76, 93, 64, 65]. Indeed, the compressive velocity is set to be in the direction of ∇α\nabla\alpha, namely, normal to the interface, and proportional to the velocity field and it only acts in the vicinity of the free-surface thanks to the product α(1−α)\alpha(1-\alpha). The most common definition for the compressive velocity, that is also employed in this work is

uc=min⁡{cα∥u∥,max⁡(∥u∥)}∇α∥∇α∥,(6)u_c = \min\{c_\alpha\lVert u \rVert,\max(\lVert u \rVert)\}\frac{\nabla\alpha}{\lVert\nabla\alpha\rVert}, \tag*{(6)}

where the constant parameter is usually chosen in the range 1<cα<41 < c_\alpha< 4 [54].

Concerning boundary conditions, the domain frontier is subdivided into a Neumann and a Dirichlet portion: ∂Ω=ΓD∪ΓN\partial\Omega= \Gamma^D \cup\Gamma^N. Additionally, the inflow boundary Γin:={x∈∂Ω:u⋅n<0}\Gamma^{\mathrm{in}} := \{x \in\partial\Omega: u \cdot n < 0\} is identified, denoting by nn the outward unit normal vector to ∂Ω\partial\Omega, in order to impose boundary conditions for the advection equation. Specifically, the inflow condition for α\alpha is set to 1 on the whole boundary if only fluid 1 is injected from the inlet, and, vice-versa, gα=0g_\alpha= 0 if only fluid 2 is entering the domain. On the other hand, when both phases are present on the inflow boundary, gαg_\alpha will be defined as a given Heaviside function.

A VOF Non-Conforming Immersed Boundary Solver

The current section presents the Volume Of Fluid Immersed Boundary numerical solver, that we denote by the acronym VOF-IB, in which we combine the immersed boundary approach proposed in [62], that was initially only available for single-phase flows, with the VOF method, for the numerical approximation of Equations (3) in a non-conforming framework.

The spatial discretization of Equations (3) is based on the Finite Volume method [28], that has been widely adopted in the CFD community since mid–1970s.

The solution domain Ω⊂R3\Omega\subset\mathbb{R}^3 is subdivided in NN control volumes (CVs) KiK_i such that

Ω=⋃i=1NKi,Ki∩Kj=∅∀Ki,Kj.\Omega= \bigcup_{i=1}^{N} K_i,\qquad K_i \cap K_j = \varnothing\quad\forall K_i,K_j.

We denote by Th\mathcal{T}_h the polyhedral tessellation of the computational domain. We indicate with Fh\mathcal{F}_h the set of faces of Th\mathcal{T}_h, with Fi={F:Ki∩F≠∅}\mathcal{F}_i = \{F : K_i \cap F \ne\varnothing\} the set of faces of cell KiK_i and with Ki\mathcal{K}_i the set of cells KjK_j, j≠ij \ne i that share a face with KiK_i. The set of faces Fh\mathcal{F}_h can be subdivided as Fh=Fh,I∪Fh,B\mathcal{F}_h = \mathcal{F}_{h,I} \cup\mathcal{F}_{h,B}, denoting with Fh,I\mathcal{F}_{h,I} and Fh,B\mathcal{F}_{h,B} the sets of internal and boundary faces, respectively. Moreover, let Fij∈Fh,IF_{ij} \in\mathcal{F}_{h,I} be the face shared by cells KiK_i and cell KjK_j, nijn_{ij} be its unit normal vector directed from KiK_i to KjK_j and Sij=Sijnij\boldsymbol{S}_{ij}=S_{ij}\boldsymbol{n}_{ij} be the surface area vector, where SijS_{ij} is the area of FijF_{ij}. Let us denote by xi\boldsymbol{x}_i the barycenter of cell KiK_i and by fij\boldsymbol{f}_{ij} the barycenter of face FijF_{ij}. Moreover, we indicate with dij=∥xj−xi∥\boldsymbol{d}_{ij}=\lVert\boldsymbol{x}_j-\boldsymbol{x}_i\rVert the vector connecting the centroids of cells KiK_i and KjK_j, and with xij\boldsymbol{x}_{ij} the intersection of vector dij\boldsymbol{d}_{ij} with face FijF_{ij}.

The Immersed Boundary Method

In the present section, the non-conforming IB method of our in-house polymer mixing library, that was first proposed and implemented by Negrini [60, 62], is briefly recalled.

The non-conforming IB approach allows to employ a fixed background mesh, avoiding the complications related to the generation of conforming computational grids aligned with arbitrarily complex wall boundaries. Triangulated STL surfaces representing the solid bodies immersed into the computational grid are used to mark the domain DOFs as either fluid, solid or IB cells. Subsequently, the immersed boundary conditions are enforced at IB DOFs by means of a Weighted Least Squares (WLS) interpolation.

Let ΣIB\Sigma_{\mathrm{IB}} be a closed triangulated surface representing the immersed boundary. Then, consider Ω~∈R3\widetilde{\Omega}\in\mathbb{R}^{3} such that Ω⊂Ω~\Omega\subset\widetilde{\Omega} and ΣIB∩Ω~≠∅\Sigma_{\mathrm{IB}}\cap\widetilde{\Omega}\ne\varnothing. We indicate with Σ‾IB\overline{\Sigma}_{\mathrm{IB}} the volume enclosed by ΣIB\Sigma_{\mathrm{IB}} and assume that the fluid flows outside of Σ‾IB\overline{\Sigma}_{\mathrm{IB}}, in Ω~∖Σ‾IB\widetilde{\Omega}\setminus\overline{\Sigma}_{\mathrm{IB}}.

Using the triangulated surface, the elements of Th\mathcal{T}_h are subdivided into a solid set TS={Ki∈Th:xi∈Σ‾IB}\mathcal{T}_{S}=\{K_i\in\mathcal{T}_h:\boldsymbol{x}_i\in\overline{\Sigma}_{\mathrm{IB}}\}, IB set TIB={Ki∈Th∖TS:Ki∈Kj for some Kj∈TS}\mathcal{T}_{\mathrm{IB}}=\{K_i\in\mathcal{T}_h\setminus\mathcal{T}_{S}:K_i\in\mathcal{K}_j\text{ for some }K_j\in\mathcal{T}_{S}\} and fluid set TF={Ki∈Th∖(TS∪TIB)}\mathcal{T}_{F}=\{K_i\in\mathcal{T}_h\setminus(\mathcal{T}_{S}\cup\mathcal{T}_{\mathrm{IB}})\}. We recall that Kj\mathcal{K}_j is the set of all cells KiK_i sharing a face with KjK_j. Thus, the IB set is composed by all cells of Th\mathcal{T}_h that have at least one solid cell among their neighbors.

The projections of IB cell centers on surface ΣIB\Sigma_{\mathrm{IB}} are the IB points, whose set is denoted by PIB={xIB,i}i:Ki∈TIB\mathcal{P}_{\mathrm{IB}}=\{\boldsymbol{x}_{\mathrm{IB},i}\}_{i:K_i\in\mathcal{T}_{\mathrm{IB}}}. The unit normal vectors at the IB points directed inwards TS\mathcal{T}_{S} are the IB normals nIB\boldsymbol{n}_{\mathrm{IB}}.

Analogously, the solid, IB and fluid face sets are defined as ΓS={Fij∈F:Ki∈TS∧Kj∈TS}\Gamma_{S}=\{F_{ij}\in\mathcal{F}:K_i\in\mathcal{T}_{S}\wedge K_j\in\mathcal{T}_{S}\}, ΓIB={Fij∈F:Ki∈TIB⊕Kj∈TIB}\Gamma_{\mathrm{IB}}=\{F_{ij}\in\mathcal{F}:K_i\in\mathcal{T}_{\mathrm{IB}}\oplus K_j\in\mathcal{T}_{\mathrm{IB}}\} and ΓF={Fij∈F:Ki∈TF∧Kj∈TF}\Gamma_{F}=\{F_{ij}\in\mathcal{F}:K_i\in\mathcal{T}_{F}\wedge K_j\in\mathcal{T}_{F}\}, respectively, where ⊕\oplus is the logical operator representing the exclusive or and it is used in the definition of ΓIB\Gamma_{\mathrm{IB}} to define the IB face set in order to exclude those faces shared by two cells KiK_i and KjK_j that are either both belonging to TIB\mathcal{T}_{\mathrm{IB}}, or neither of them is.

Moreover we can distinguish two subsets such that ΓIB=ΓIB,int∪ΓIB,ext\Gamma_{\mathrm{IB}}=\Gamma_{\mathrm{IB,int}}\cup\Gamma_{\mathrm{IB,ext}} defining the IB internal face set as ΓIB,int={Fij∈F:Ki∈TIB∧Kj∈TF}\Gamma_{\mathrm{IB,int}}=\{F_{ij}\in\mathcal{F}:K_i\in\mathcal{T}_{\mathrm{IB}}\wedge K_j\in\mathcal{T}_{F}\} and the IB external face set as ΓIB,ext={Fij∈F:Ki∈TIB∧Kj∈TS}\Gamma_{\mathrm{IB,ext}}=\{F_{ij}\in\mathcal{F}:K_i\in\mathcal{T}_{\mathrm{IB}}\wedge K_j\in\mathcal{T}_{S}\}.

For each IB cell Ki∈TIBK_i\in\mathcal{T}_{\mathrm{IB}} a stencil Si\mathcal{S}_i is defined in order to construct a WLS approximation operator, selecting cells in TIB∪TF\mathcal{T}_{\mathrm{IB}}\cup\mathcal{T}_{F} whose barycenter satisfy the spatial distance, connectivity distance and field of view criteria. The reader is referred to [60, 62] for a detailed description on the construction of Si\mathcal{S}_i.

Figure 3 gives an exemplified representation of the geometrical elements that have just been introduced.

Schematic representation of solid, IB and fluid cell and face sets, IB points, IB normals and IB stencil

Figure 3. Schematic representation of solid, IB and fluid cell and face sets, IB points, IB normals and IB stencil (adapted from [60]).

Without loss of generality, we construct the WLS approximant for a scalar variable ψ\psi for which either Dirichlet or Neumann conditions are imposed on ΣIB\Sigma_{\mathrm{IB}}. Given the basis of polynomials of degree pp in a point x=[x,y,z]T\boldsymbol{x}=[x,y,z]^{\mathsf{T}}, Px=[1,x,y,z,…,xp,yp,zp]T\boldsymbol{P}_{x}=[1,x,y,z,\ldots,x^{p},y^{p},z^{p}]^{\mathsf{T}}, we can define a polynomial function fIB,if_{\mathrm{IB},i} that approximates the value ψi\psi_i at each ii–th IB cell center as

fIB,i(x)=PxTβi.(7)f_{\mathrm{IB},i}(\boldsymbol{x})=\boldsymbol{P}_{x}^{\mathsf{T}}\boldsymbol{\beta}^{i}. \tag*{(7)}

The coefficients βi=[β0i,…,βNcoeffsi]T\boldsymbol{\beta}^{i}=[\beta^{i}_{0},\ldots,\beta^{i}_{N_{\mathrm{coeffs}}}]^{\mathsf{T}} are determined by solving a WLS minimization problem on the values ψj\psi_j stored in the centroids of cells Kj∈SiK_j\in\mathcal{S}_i associated to each Ki∈TIBK_i\in\mathcal{T}_{\mathrm{IB}}, plus the assigned IB value, namely, the linearly reconstructed IB condition from the nodes of the triangulated surface to the IB points.

Non-conforming Dirichlet conditions

We suppose to prescribe on ΣIB\Sigma_{\mathrm{IB}} the Dirichlet condition ψ∣ΣIB=gD\psi|_{\Sigma_{\mathrm{IB}}}=g_D. The optimal coefficients βi,∗\boldsymbol{\beta}^{i,*} for each IB cell Ki∈TIBK_i\in\mathcal{T}_{\mathrm{IB}}, provided as solution of the WLS problem, allow to compute the vector of linear combination coefficients si=Pxiβi,∗=\boldsymbol{s}^{i}=\boldsymbol{P}_{x_i}\boldsymbol{\beta}^{i,*}= [s~IBi,s~1i,…,s~ni][\tilde{s}_{\mathrm{IB}}^{i}, \tilde{s}^{i}_{1}, \ldots, \tilde{s}^{i}_{n}] that allow to define the IB interpolation operator SIBS_{\mathrm{IB}}, such that, for each Ki∈ThK_i \in\mathcal{T}_h, the corrected variable ψ~i\tilde{\psi}_i reads [62]

ψ~i=[SIB(g,Ψ)]i=SIBi(g,Ψ)={sIBi(gD,IB)i+∑j=1nsjiψjif Ki∈TIB,ψiif Ki∈TF,gS,iif Ki∈TS,(8)\tilde{\psi}_i = [S_{\mathrm{IB}}(\boldsymbol{g}, \boldsymbol{\Psi})]_i = S_{\mathrm{IB}}^{i}(\boldsymbol{g}, \boldsymbol{\Psi}) = \begin{cases} s_{\mathrm{IB}}^{i}(\boldsymbol{g}_{D,\mathrm{IB}})_i + \sum_{j=1}^{n}s_{j}^{i}\psi_j & \text{if } K_i \in\mathcal{T}_{\mathrm{IB}},\\ \psi_i & \text{if } K_i \in\mathcal{T}_{F},\\ g_{S,i} & \text{if } K_i \in\mathcal{T}_{S}, \end{cases} \tag*{(8)}

that can be compactly written as Ψ~=SIB(gD,Ψ)=Sgg+SΨ\tilde{\boldsymbol{\Psi}} = S_{\mathrm{IB}}(\boldsymbol{g}_{D},\boldsymbol{\Psi}) = S_g\boldsymbol{g} + S\boldsymbol{\Psi}, introducing matrices SS, Sg∈RN×NS_g \in\mathbb{R}^{N \times N} and the vector of the IB and solid data g∈RN\boldsymbol{g} \in\mathbb{R}^{N}. Vector gD,IB\boldsymbol{g}_{D,\mathrm{IB}} collects the IB values, namely, the vector collecting the linear reconstruction of gDg_D from the nodes of the triangulated surface to the IB points.

Non-conforming Neumann conditions

The imposition of Neumann boundary conditions on non-conforming boundaries is handled analogously to the Dirichlet case. Suppose a Neumann boundary condition is prescribed on ΣIB\Sigma_{\mathrm{IB}}, such that (∇ψ⋅n)∣ΣIB=gN(\nabla\psi\cdot\boldsymbol{n})|_{\Sigma_{\mathrm{IB}}} = g_N. We denote by gN,IB\boldsymbol{g}_{N,\mathrm{IB}} the vector of piecewise linear reconstructions of gNg_N from the triangulation to the IB points, the Neumann condition at each IB point becomes ∇ψ(xIBP,i)⋅nIB,i=(gN,IB)i\nabla\psi(\boldsymbol{x}_{\mathrm{IBP},i}) \cdot\boldsymbol{n}_{\mathrm{IB},i} = (\boldsymbol{g}_{N,\mathrm{IB}})_i.

The WLS minimization problem slightly changes in order to impose a gradient value at the IB point. The optimal coefficients βi,⋆\boldsymbol{\beta}^{i,\star} are then used to obtain an approximation of the gradient at IB external faces f∈ΓIB,extf \in\Gamma_{\mathrm{IB},\mathrm{ext}}, that is where the Neumann condition is actually imposed at the algebraic level. For each face f=Fij∈ΓIB,extf = F_{ij} \in\Gamma_{\mathrm{IB},\mathrm{ext}}, with barycenter fij\boldsymbol{f}_{ij} and face normal nij\boldsymbol{n}_{ij}, linear combination coefficients sijs^{ij} can be computed, such that

gN,Fij=SIBij(gN,IB,Ψ)=sIBij(gN,IB)i+∑k=1nskijψk.(9)g_{N,F_{ij}} = S_{\mathrm{IB}}^{ij}(\boldsymbol{g}_{N,\mathrm{IB}},\boldsymbol{\Psi}) = s_{\mathrm{IB}}^{ij}(\boldsymbol{g}_{N,\mathrm{IB}})_i + \sum_{k=1}^{n}s_{k}^{ij}\psi_k. \tag*{(9)}

Remark 2. We employ the aforementioned approach [62] to impose Dirichlet immersed boundary conditions for the velocity field. Indeed, at rotating solid walls (screws), that are usually handled with the IB method, a known rigid motion is prescribed at each IB point located at xIBP,i\boldsymbol{x}_{\mathrm{IBP},i}. On the other hand, homogeneous Neumann conditions are typically imposed for volume fraction at solid walls. It is worth mentioning that this approach is not used to impose the immersed boundary Neumann conditions for pressure, to avoid stability issues arising with anisotropic grids. Instead, the IB constraint is imposed on pressure within the PIMPLE solution algorithm presented in the next section, following the approach proposed in [60, 62, 40].

The VOF-IB Numerical Solver

We start deriving the Finite Volume discretization of (3), obtained by integrating on control volumes, applying the divergence theorem and approximating volume and surface integrals.

Concerning the time discretization, let {tn}n=1,…,NT\{t^n\}_{n=1,\ldots,N_T} be a sequence of time-steps in the time-interval [0,T][0,T], with tn+1−tn=Δtt^{n+1}-t^n=\Delta t for n=0,…,NT−1n=0,\ldots,N_T-1. Moreover, we indicate with {tk}k=0,…,Nsc−1\{t^k\}_{k=0,\ldots,N_{\mathrm{sc}}-1} the discretization of the sub-interval [tn,tn+1][t^n,t^{n+1}], where tk+1−tk=Δtsc=ΔtNsct^{k+1}-t^k=\Delta t_{\mathrm{sc}}=\frac{\Delta t}{N_{\mathrm{sc}}}, needed for the sub-cycle loop performed for the solution of the advection equation (3c), required to satisfy stability constraints on the local interface Courant number.

The Courant number, also known as the Courant–Friedrichs–Lewy (CFL) number [14] is defined as Co=UΔth\mathrm{Co}=\frac{U\Delta t}{h}, where UU is the characteristic speed of the flow, that can be taken as U=∥u∥L∞(Ω)U=\lVert\boldsymbol{u}\rVert_{L^\infty(\Omega)} [29], and hh is the characteristic mesh size. The above definition arises from stability analysis of numerical methods for hyperbolic partial differential equations, particularly concerning explicit finite difference schemes, where typically the requirement is to have Co≤1\mathrm{Co}\leq1.

At the discrete level, the local Courant number is computed in the FV context as

Coh=max⁡Ki∈ThΔt∣Ki∣∑f∈Fi∣ϕf∣(10)\mathrm{Co}_h = \max_{K_i \in\mathcal{T}_h}\frac{\Delta t}{\lvert K_i\rvert}\sum_{f\in\mathcal{F}_i}\lvert\phi_f\rvert \tag*{(10)}

with ϕf=uf⋅Sf\phi_f=\boldsymbol{u}_f\cdot\boldsymbol{S}_f denoting the flux across faces.

Similarly, a local interface Courant number is defined, restricted at the two-phase interface:

Coα,h=max⁡Ki∈ThξiΔt∣Ki∣∑f∈Fi∣ϕf∣(11)\mathrm{Co}_{\alpha,h} = \max_{K_i \in\mathcal{T}_h}\xi_i\frac{\Delta t}{\lvert K_i\rvert}\sum_{f\in\mathcal{F}_i}\lvert\phi_f\rvert \tag*{(11)}

where for each Ki∈ThK_i \in\mathcal{T}_h:

ξi={1if 0.01<αi<0.99,0if αi<0.01∨αi>0.99.\xi_i = \begin{cases} 1 & \text{if } 0.01 < \alpha_i < 0.99,\\ 0 & \text{if } \alpha_i < 0.01 \lor\alpha_i > 0.99. \end{cases}

The interface Courant number is employed in OpenFOAM to control the stability of the interface transport equation. It is a heuristic CFL-like constraint that accounts for the additional flux introduced by the nonlinear compression term that is non zero only near the interface, explaining the multiplication by the mask ξi\xi_i.

Non-conforming VOF advection equation

The numerical techniques employed for the solution of the advection equation eq:3c in OpenFOAM [76, 94, 16] rely on a combination of high-resolution interpolation schemes, like vanLeer [87], and a specialized flux-limiting algorithm known as MULES (Multidimensional Universal Limiter with Explicit Solution) [16], that allow to maintain interface sharpness and physical boundedness of α\alpha between 0 and 1.

Equation eq:3c is semi-discretized in time using the aforementioned sub-timestep Δtsc\Delta t_{\mathrm{sc}}, by means of a first order scheme and it is integrated on Ki∈ThK_i \in\mathcal{T}_h. Therefore, at each kk–th iteration of the sub-cycling procedure, the following holds:

∫Kiαk+1−αkΔt+∫∂Kiα⋆un⋅n dS+∫∂Kiαk(1−αk)ucn⋅n dS=0.\int_{K_i} \frac{\alpha^{k+1}-\alpha^k}{\Delta t} + \int_{\partial K_i} \alpha^{\star} \mathbf{u}^n \cdot\mathbf{n}\,\mathrm{d}S + \int_{\partial K_i} \alpha^k(1-\alpha^k)\mathbf{u}_c^n \cdot\mathbf{n}\,\mathrm{d}S = 0.

The velocity field is taken from the previous global time-step tnt^n, so uk=un\mathbf{u}^k = \mathbf{u}^n for all k=0,…,Nsc−1k=0,\ldots,N_{\mathrm{sc}}-1.

Approximation of the volume integral using the mid-point quadrature formula and of the surface integrals by summing face fluxed across all faces of KiK_i, entails:

αKik+1+Δt∣Ki∣∑f∈Fiαf⋆ϕfn=αKik−Δt∣Ki∣∑f∈Fiαfk(1−αfk)ϕcfn(12)\alpha_{K_i}^{k+1} + \frac{\Delta t}{|K_i|}\sum_{f\in\mathcal{F}_i}\alpha_f^\star\phi_f^n = \alpha_{K_i}^k - \frac{\Delta t}{|K_i|}\sum_{f\in\mathcal{F}_i}\alpha_f^k(1-\alpha_f^k)\phi_{cf}^n \tag*{(12)}

where ϕf=uf⋅Sf\phi_f=\mathbf{u}_f\cdot\mathbf{S}_f is the flux of velocity through face ff, while ϕcf\phi_{cf} denotes the interpolated compression flux at face ff, such that

ϕcf=uc,f⋅Sf=min⁡(∣ϕf∣α∥Sf∥,max⁡(∣ϕf∣∥Sf∥))(nΣf⋅Sf)(13)\phi_{cf}=\mathbf{u}_{c,f}\cdot\mathbf{S}_f=\min\left(\frac{|\phi_f|}{\alpha\|\mathbf{S}_f\|},\max\left(\frac{|\phi_f|}{\|\mathbf{S}_f\|}\right)\right)(\mathbf{n}_{\Sigma f}\cdot\mathbf{S}_f) \tag*{(13)}

with uc,f\mathbf{u}_{c,f} face-interpolated compression velocity (6) and nΣf\mathbf{n}_{\Sigma f} the interface unit normal vector interpolated at face ff, that is computed as

nΣf=(∇α)f∥(∇α)f∥(14)\mathbf{n}_{\Sigma f}=\frac{(\nabla\alpha)_f}{\|(\nabla\alpha)_f\|} \tag*{(14)}

and (∇α)f(\nabla\alpha)_f is determined using a central differencing (CD) scheme.

The approximation of face fluxes pertaining the advection terms is carried out by means of the MULES approach [16], that is used to reduce the numerical diffusion induced by the upwind (UD) scheme at the interface by introducing higher order schemes for face-interpolation, localized at the interfacial region to keep a low computational cost. Specifically, the advection fluxes Φf,up\Phi_{f,\mathrm{up}} and Φf,c\Phi_{f,\mathrm{c}} are defined as:

Φf,up⋆=(αf⋆)upϕfn(15)\Phi_{f,\mathrm{up}}^\star=(\alpha_f^\star)_{\mathrm{up}}\phi_f^n \tag*{(15)}
Φf,c⋆=αf⋆ϕfn−(αf⋆)upϕfn+αf⋆(1−αf⋆)ϕcfn(16)\Phi_{f,\mathrm{c}}^\star=\alpha_f^\star\phi_f^n-(\alpha_f^\star)_{\mathrm{up}}\phi_f^n+\alpha_f^\star(1-\alpha_f^\star)\phi_{cf}^n \tag*{(16)}

where we denote by (αf⋆)up(\alpha_f^\star)_{\mathrm{up}} the face-interpolated value of α\alpha obtained by means of an upwind scheme. On the other hand, the face interpolated volume fraction αf⋆\alpha_f^\star is computed blending UD and CD schemes [16, 5] in order to achieve a compromise between boundedness and accuracy, as follows:

αf⋆=(1−λα)(αf⋆)up+λα(αf⋆)cd.(17)\alpha_f^\star=(1-\lambda_\alpha)(\alpha_f^\star)_{\mathrm{up}}+\lambda_\alpha(\alpha_f^\star)_{\mathrm{cd}}. \tag*{(17)}

Various high resolution schemes are implemented in OpenFOAM for the computation of the blending factor λα∈[0,1]\lambda_\alpha\in[0,1], the most commonly used being the VanLeer scheme [87].

Face fluxes ϕf\phi_f and ϕcf\phi_{cf} are computed at first global time iteration using central differencing for the face interpolation of the initial velocity condition, while at subsequent time instants the value of velocity fluxes is directly inherited from the solution of the pressure-velocity coupled problem. This becomes clearer after introducing the PIMPLE algorithm.

An explicit Forward Euler (FE) scheme is used for the solution of (12).

αKik+1=αKik−Δt∣Ki∣∑f∈Fi[Φf,upk+λMΦf,ck](18)\alpha_{K_i}^{k+1}=\alpha_{K_i}^k-\frac{\Delta t}{|K_i|}\sum_{f\in\mathcal{F}_i}\left[\Phi_{f,\mathrm{up}}^k+\lambda_M\Phi_{f,\mathrm{c}}^k\right] \tag*{(18)}

The delimiter parameter λM\lambda_M is defined in order to be equal to 1 in the vicinity of the interface, while it is 0 elsewhere.

At the end of each sub-cycle, the mixture properties are updated. Then, the mass flux through cell-faces is computed in order to obtain the total mass flux corresponding to the global time-step as the sum of the sub-cycle mass fluxes, that will be subsequently needed for the solution of the momentum equation:

(ρfϕf)n+1=(αfn+1ρ1+(1−αfn+1)ρ2)ϕfn=∑k=0Nsc−1(αfϕf)k+1(ρ1−ρ2)+ϕfnρ2=∑k=0Nsc−1(Φf,upk+λMΦf,ck)(ρ1−ρ2)+ϕfnρ2.(19)\begin{aligned} (\rho_f\phi_f)^{n+1}=(\alpha_f^{n+1}\rho_1+(1-\alpha_f^{n+1})\rho_2)\phi_f^n=\sum_{k=0}^{N_{\mathrm{sc}}-1}(\alpha_f\phi_f)^{k+1}(\rho_1-\rho_2)+\phi_f^n\rho_2 \\ &=\sum_{k=0}^{N_{\mathrm{sc}}-1}(\Phi_{f,\mathrm{up}}^k+\lambda_M\Phi_{f,\mathrm{c}}^k)(\rho_1-\rho_2)+\phi_f^n\rho_2. \tag*{(19)} \end{aligned}

Remark 3. An inner loop, denoted as α\alpha–correction loop, is performed within each temporal sub-cycle to handle the nonlinearity introduced by the compression term, whose explicit treatment might not guarantee mass conservation if the equation was solved only once within each sub time-step.

At the numerical level, zero normal gradient conditions are prescribed for α\alpha at walls. Since the IB surface typically identifies physical walls, we restrict ourselves to the case of homogeneous Neumann IB conditions prescribed on ΣIB\Sigma_{\mathrm{IB}}:

∇α⋅n∣ΣIB=0.\nabla\alpha\cdot\mathbf{n}\big|_{\Sigma_{\mathrm{IB}}}=0.

The FE scheme consists of a unique explicit time advancement, thus, no linear system is solved, since the new solution is just a correction of the volume fraction coming from the previous iteration.

The update of α\alpha is performed using (17) without any change with respect to the conforming case. Then, the resulting field α\alpha is forced to satisfy the Neumann IB condition by modifying its value in correspondence of solid cells DOFs. Specifically, since the IB Neumann condition is imposed at IB external faces, given a dead cell Kj∈TSK_j \in\mathcal{T}_{S} and an IB cell Ki∈TIBK_i \in\mathcal{T}_{\mathrm{IB}}, the following relation holds for face Fij∈ΓIB,extF_{ij} \in\Gamma_{\mathrm{IB,ext}}:

αjk+1=αi⋆±gN,Fij∥dij∥,(20)\alpha_j^{k+1}=\alpha_i^\star\pm g_{N,F_{ij}}\lVert\mathbf{d}_{ij}\rVert, \tag*{(20)}

with gN,Fijg_{N,F_{ij}} computed using (9).

Non-conforming momentum equation

The semi-discretization in time of the momentum equation (3a) is obtained applying a first-order backward Euler scheme, linearizing the equation by treating the convective flux explicitly. The transposed velocity gradient in the viscous term is typically treated explicitly as well, allowing to decouple the equations for the three velocity components. The coupling with the advection equation for α\alpha is handled in a segregated manner, using the updated field αn+1\alpha^{n+1} into the discretized momentum equation.

Integrating the semi-discretized equation over control volume Ki∈ThK_i \in\mathcal{T}_h and applying the divergence theorem, entails:

∫Kiρn+1un+1−unΔt dV+∫∂Kiρn+1(un+1⊗un)n dS−∫∂Kiμn+1(∇un+1+∇Tun)n dS+∫∂Kiprghn+1n dS=∫Ki−∇pn+1(g⋅x)+σκn+1∇αn+1 dV,(21)\int_{K_i}\rho^{n+1}\frac{\mathbf{u}^{n+1}-\mathbf{u}^{n}}{\Delta t}\,\mathrm{d}V+\int_{\partial K_i}\rho^{n+1}(\mathbf{u}^{n+1}\otimes\mathbf{u}^{n})\mathbf{n}\,\mathrm{d}S-\int_{\partial K_i}\mu^{n+1}(\nabla\mathbf{u}^{n+1}+\nabla^{\mathsf{T}}\mathbf{u}^{n})\mathbf{n}\,\mathrm{d}S+\int_{\partial K_i}p_{\mathrm{rgh}}^{n+1}\mathbf{n}\,\mathrm{d}S=\int_{K_i}-\nabla p^{n+1}(\mathbf{g}\cdot\mathbf{x})+\sigma\kappa^{n+1}\nabla\alpha^{n+1}\,\mathrm{d}V, \tag*{(21)}

where the superscript n+1n+1 is explicitly written in ρn+1\rho^{n+1}, μn+1\mu^{n+1} and κn+1\kappa^{n+1} to highlight their dependence on αn+1\alpha^{n+1}.

Approximating the volume integrals using the mid-point quadrature formula and the surface integrals by summing face fluxed across all faces of KiK_i leads to:

ρin+1∣Ki∣Δtuin+1+∑f∈Fi(ρfϕf)n+1ufn+1−∑f∈Fiμfn+1(∇~f⊥un+1)⋅Sf+∑f∈Fiprgh,fn+1Sf=∣Ki∣R({(g⋅xf)∇~f⊥ρn+1}f∈Fi)+∣Ki∣R({σκfn+1∇~f⊥αn+1}f∈Fi)+ρin+1∣Ki∣Δtuin+∑f∈Fiμfn+1(∇hun)f⋅Sf.(22)\frac{\rho_i^{n+1}|K_i|}{\Delta t}\mathbf{u}_i^{n+1}+\sum_{f\in\mathcal{F}_i}(\rho_f\phi_f)^{n+1}\mathbf{u}_f^{n+1}-\sum_{f\in\mathcal{F}_i}\mu_f^{n+1}(\widetilde{\nabla}_{f}^{\perp}\mathbf{u}^{n+1})\cdot\mathbf{S}_f+\sum_{f\in\mathcal{F}_i}p_{\mathrm{rgh},f}^{n+1}\mathbf{S}_f =|K_i|\mathcal{R}\left(\left\{(\mathbf{g}\cdot\mathbf{x}_f)\widetilde{\nabla}_{f}^{\perp}\rho^{n+1}\right\}_{f\in\mathcal{F}_i}\right)+|K_i|\mathcal{R}\left(\left\{\sigma\kappa_f^{n+1}\widetilde{\nabla}_{f}^{\perp}\alpha^{n+1}\right\}_{f\in\mathcal{F}_i}\right) +\frac{\rho_i^{n+1}|K_i|}{\Delta t}\mathbf{u}_i^n+\sum_{f\in\mathcal{F}_i}\mu_f^{n+1}(\nabla_h\mathbf{u}^n)_f\cdot\mathbf{S}_f. \tag*{(22)}

The mass flux (ρfϕf)n+1(\rho_f\phi_f)^{n+1} in the convective term is computed after solving the advection equation by means of (18).

The face-interpolated normal gradient of velocity is approximated as [43]:

∇~f⊥u=uj−ui∥df⋅nf∥+(∇hψn)f⋅(nf−n⊥),(23)\widetilde{\nabla}_{f}^{\perp}\mathbf{u}=\frac{\mathbf{u}_j-\mathbf{u}_i}{\lVert\mathbf{d}_f\cdot\mathbf{n}_f\rVert}+(\nabla_h\psi^{n})_f\cdot(\mathbf{n}_f-\mathbf{n}_{\perp}), \tag*{(23)}

where the second term accounts for mesh non-orthogonality and (∇hψ)f(\nabla_h\psi)_f is explicitly computed using

(∇hψn)f=wij(∇h,iψn)+(1−wij)(∇h,jψn),(24)(\nabla_h\psi^n)_f=w_{ij}(\nabla_{h,i}\psi^n)+(1-w_{ij})(\nabla_{h,j}\psi^n), \tag*{(24)}

with wijw_{ij} interpolation weights [43]. Definition (22) is used also to compute the gradient of density and volume fraction fields. On the other hand, the transposed gradient of velocity is explicitly computed using (23).

The source terms related to surface tension and hydrostatic pressure are originally computed as numerical fluxes stored at face centers. However, cell-centered values are required in the momentum equations [83]. Therefore, in OpenFOAM VOF solvers a reconstruction operator R\mathcal{R} is employed to obtain cell-centered values of the source term [91, 90, 1, 83], namely

ψi≈R({ψf}f∈Fi)=(∑f∈Finf⊗Sf)−1(∑f∈Finf⊗ψf).(25)\psi_i\approx\mathcal{R}\left(\{\psi_f\}_{f\in\mathcal{F}_i}\right)=\left(\sum_{f\in\mathcal{F}_i}\mathbf{n}_f\otimes\mathbf{S}_f\right)^{-1}\left(\sum_{f\in\mathcal{F}_i}\mathbf{n}_f\otimes\psi_f\right). \tag*{(25)}

Equation (21) is valid for each Ki∈ThK_i \in\mathcal{T}_h and it can be compactly written in matrix form as

AUn+1+BTpn+1=F+FVOF,(26)A U^{n+1} + B^{\mathsf{T}}p^{n+1} = F + F_{\mathrm{VOF}}, \tag*{(26)}

with Un+1∈R3NU^{n+1} \in\mathbb{R}^{3N} and pn+1∈RNp^{n+1} \in\mathbb{R}^{N} denoting the cell-centered FV variables associated to un+1u^{n+1} and prghn+1p_{\mathrm{rgh}}^{n+1} respectively; A∈R3N×3NA \in\mathbb{R}^{3N \times3N} is the matrix containing coefficients coming from discretization of temporal, convective and viscous diffusion terms and B∈R3N×NB \in\mathbb{R}^{3N \times N} stores the coefficients related to the pressure gradient discretization. The source terms F∈R3NF \in\mathbb{R}^{3N} and FVOF∈R3NF_{\mathrm{VOF}} \in\mathbb{R}^{3N} are given by:

Fi=ρin+1∣Ki∣Δtuin+∑f∈Fiμfn+1(∇hTun)f⋅Sf;F_i = \frac{\rho_i^{n+1}|K_i|}{\Delta t}u_i^n + \sum_{f \in\mathcal{F}_i}\mu_f^{n+1}(\nabla_h^{\mathsf{T}}u^n)_f \cdot S_f;
FVOF,i=∣Ki∣R({(g⋅xf)∇~f⊥ρn+1}f∈Fi)+∣Ki∣R({σκf∇~f⊥αn+1}f∈Fi).F_{\mathrm{VOF},i} = |K_i|\mathcal{R}\left(\left\{(g\cdot x_f)\widetilde{\nabla}^{\perp}_f\rho^{n+1}\right\}_{f\in\mathcal{F}_i}\right) + |K_i|\mathcal{R}\left(\left\{\sigma\kappa_f\widetilde{\nabla}^{\perp}_f\alpha^{n+1}\right\}_{f\in\mathcal{F}_i}\right).

Remark 4. Within the source term FF we include also all correction terms for skewness and non-orthogonality that are usually treated explicitly.

Now, suppose to have an IB surface ΣIB\Sigma_{\mathrm{IB}} representing a physical wall such that u∣ΣIB=gDu|_{\Sigma_{\mathrm{IB}}}=g_D. The IB interpolation operator is computed such that (8) holds for the IB constrained velocity U~\widetilde{U}. If Ki∈TFK_i \in\mathcal{T}_F, then U~i=Ui\widetilde{U}_i=U_i, therefore all the rows of the system associated to fluid DOFs remain unchanged. On the other hand, for each IB or solid cell Ki∈TS∪TIBK_i \in\mathcal{T}_S \cup\mathcal{T}_{\mathrm{IB}} the matrix coefficients are changed in order to satisfy U~i=SIB(g,U)\widetilde{U}_i=S^{\mathrm{IB}}(g,U).

We introduce the IB constrained matrix A~∈R3N×3N\widetilde{A}\in\mathbb{R}^{3N\times3N} and source term F~∈R3N\widetilde{F}\in\mathbb{R}^{3N}, such that the IB constrained momentum equation would read

A~U~+χBTp=F~+χFVOF,(27)\widetilde{A}\widetilde{U}+\chi B^{\mathsf{T}}p=\widetilde{F}+\chi F_{\mathrm{VOF}}, \tag*{(27)}

denoting by χ\chi the IB mask such that

χi={1if Ki∈TF0if Ki∈TIB∪TS(28)\chi_i = \begin{cases} 1 & \text{if } K_i \in\mathcal{T}_F \\ 0 & \text{if } K_i \in\mathcal{T}_{\mathrm{IB}}\cup\mathcal{T}_S \end{cases} \tag*{(28)}

The constrained matrix A~\widetilde{A} is constructed imposing

A~ij={Aijif i=j, ∀Ki∈Th,0if Ki∈TIB∪TS, Kj∈Ki,Aijif i≠j, Ki,Kj∉TIB∪TS.(29)\widetilde{A}_{ij} = \begin{cases} A_{ij} & \text{if } i=j,\ \forall K_i\in\mathcal{T}_h,\\ 0 & \text{if } K_i\in\mathcal{T}_{\mathrm{IB}}\cup\mathcal{T}_S,\ K_j\in K_i,\\ A_{ij} & \text{if } i\ne j,\ K_i,K_j\notin\mathcal{T}_{\mathrm{IB}}\cup\mathcal{T}_S. \end{cases} \tag*{(29)}

While the source term is given by

F~i={AiiSiIB(g,U)if Ki∈TIB∪TS,Fi−AijSjIB(g,U)∀Ki∈Kj, Kj∈TIB∪TS,Fiotherwise.(30)\widetilde{F}_i = \begin{cases} A_{ii}S_i^{\mathrm{IB}}(g,U) & \text{if } K_i\in\mathcal{T}_{\mathrm{IB}}\cup\mathcal{T}_S,\\ F_i-A_{ij}S_j^{\mathrm{IB}}(g,U) & \forall K_i\in K_j,\ K_j\in\mathcal{T}_{\mathrm{IB}}\cup\mathcal{T}_S,\\ F_i & \text{otherwise.} \end{cases} \tag*{(30)}

IB constrained pressure-velocity problem

The continuity equation (3b) is also integrated over KiK_i, the divergence theorem is applied and the resulting surface integral is approximated, leading to:

∫Ki∇⋅u dV=∫∂Kiu⋅n dS≈∑f∈Fiuf⋅Sf=0,\int_{K_i}\nabla\cdot u\,dV=\int_{\partial K_i}u\cdot n\,dS\approx\sum_{f\in\mathcal{F}_i}u_f\cdot S_f=0,

that is compactly written in matrix form as:

BU=0.(31)BU=0. \tag*{(31)}

Putting (25) and (30) together we obtain the conforming pressure-velocity coupled algebraic system:

\begin{bmatrix} A & B^{\mathsf{T}}\\ B & 0 \end{bmatrix} \begin{bmatrix} U\\ p \end{bmatrix} = \begin{bmatrix} F+F_{\mathrm{VOF}}\\ 0 \end{bmatrix}. \tag*{(32)}\end{bmatrix}

Remark 5. The present work is based on the collocated Finite Volume method implemented in OpenFOAM, meaning that all variables are stored at cell centers, unlike staggered grids. However, the collocated approach is known to suffer from spurious pressure modes, often denoted as checkerboard oscillations, where unphysical pressure fluctuations occur due to a weak numerical coupling between velocity and pressure fields [28]. The typical technique employed to stabilize the problem is the Rhie-Chow (RC) interpolation [59, 98]. As demonstrated in [61], the addition of the RC stabilization term ensures the validity of the discrete inf-sup condition.

At the algebraic level, the stabilized system becomes:

\begin{bmatrix} A & B^{\mathsf{T}} \\ B & -C \end{bmatrix} \begin{bmatrix} U \\ p \end{bmatrix} = \begin{bmatrix} F + F_{\mathrm{VOF}} \\ 0 \end{bmatrix}. \tag*{(33)}\end{bmatrix}

with the Rhie-Chow stabilization matrix CC given by

C=R(D−1)−BD−1BT(34)C = R(D^{-1}) - BD^{-1}B^{\mathsf{T}} \tag*{(34)}

where DD denotes the diagonal part of AA and R(D−1)R(D^{-1}) is a short hand to indicate a FV discretized Laplacian operator with diffusivity coefficient D−1D^{-1}, following the notation of [61], where further details on the assembly of matrix CC are reported. From now on, we will refer to (33) as the FV algebraic formulation associated to the momentum and continuity Equations (3) (3a) and (3b).

The imposition of the IB constraint for velocity on the momentum equation, leads to the non-conforming pressure-velocity problem, that reads:

\begin{bmatrix} \widetilde{A} & \chi B^{\mathsf{T}} \\ B & -C \end{bmatrix} \begin{bmatrix} U \\ p \end{bmatrix} = \begin{bmatrix} \widetilde{F} + \chi F_{\mathrm{VOF}} \\ 0 \end{bmatrix}. \tag*{(35)}\end{bmatrix}

The VOF-IB PIMPLE Algorithm

The VOF-IB PIMPLE solution Algorithm 1 is formulated in order to obtain the non-conforming two-phase interIbFoam solver, starting from the reference conforming interFoam solver [16] of OpenFOAM-10 and employing a similar procedure to the one proposed in [60] for single-phase non-conforming solvers.

The pressure-velocity coupling is handled by the majority of the CFD community using projection schemes, that consist of segregated solution algorithms of predictor-corrector kind that allow to solve velocity and pressure fields sequentially, avoiding to solve the computationally expensive monolithic problem altogether. The basic idea is to determine a pressure field that ensures mass conservation, by re-writing the system of equations in terms of a momentum equation for velocity and a Poisson equation for pressure and solving them sequentially [59, 28].

OpenFOAM solvers are based on SIMPLE (Semi Implicit Method for Pressure Linked Equations) algorithm [70, 69], formulated for the steady-state Navier-Stokes equations, and its transient version, denoted as PISO (Pressure Implicit with Splitting of Operators) [41]. PIMPLE is a combination of the two aforementioned schemes, allowing better accuracy and numerical stability.

First of all, matrix A~\widetilde{A} is split into its diagonal D~\widetilde{D} and off-diagonal H~\widetilde{H} components such that

[H~(U)]P=F~P−∑N∈KPA~NUN,[\widetilde{H}(U)]_{P} = \widetilde{F}_{P} - \sum_{N \in\mathcal{K}_{P}} \widetilde{A}_{N} U_{N},

so that the momentum equation can be written as

D~U=H~(U)+χ(FVOF−BTp)(36)\widetilde{D}U = \widetilde{H}(U) + \chi\left(F_{\mathrm{VOF}} - B^{\mathsf{T}}p\right) \tag*{(36)}

using the continuity equation BUm+1−Cpm+1=0BU^{m+1} - Cp^{m+1} = 0, together with (36), we obtain:

R(D~−1)p=B(D~−1H~(U)+D~−1FVOF).(37)R(\widetilde{D}^{-1})p = B\left(\widetilde{D}^{-1}\widetilde{H}(U) + \widetilde{D}^{-1}F_{\mathrm{VOF}}\right). \tag*{(37)}

Then, we would have a corrected velocity U=D~−1H~(U)+D~−1FVOF−D~−1BTpU = \widetilde{D}^{-1}\widetilde{H}(U) + \widetilde{D}^{-1}F_{\mathrm{VOF}} - \widetilde{D}^{-1}B^{\mathsf{T}}p, satisfying the incompressibility constraint. However, UU has to be consistent with the IB condition:

U=SIB(g,U)=Sgg+SU,U = S_{\mathrm{IB}}(g,U) = S_{g}g + SU,

thus, we substitute the expression for UU:

U=Sgg+SD~−1H~(U)+SD~−1FVOF−SD~−1BTp.U = S_{g}g + S\widetilde{D}^{-1}\widetilde{H}(U) + S\widetilde{D}^{-1}F_{\mathrm{VOF}} - S\widetilde{D}^{-1}B^{\mathsf{T}}p.

But D~−1H~(U)\widetilde{D}^{-1}\widetilde{H}(U) already satisfies the IB constraint, such that:

D~−1H~(U)=SIB(g,D~−1H~(U))=Sgg+SD~−1H~(U),\widetilde{D}^{-1}\widetilde{H}(U) = S_{\mathrm{IB}}\left(g,\widetilde{D}^{-1}\widetilde{H}(U)\right) = S_{g}g + S\widetilde{D}^{-1}\widetilde{H}(U),

leading to

U=D~−1H~(U)+SD~−1FVOF−SD~−1BTp.(38)U = \widetilde{D}^{-1}\widetilde{H}(U) + S\widetilde{D}^{-1}F_{\mathrm{VOF}} - S\widetilde{D}^{-1}B^{\mathsf{T}}p. \tag*{(38)}

Finally, we impose again the validity of the incompressibility constraint:

0=BU−Cp=BD~−1H~(U)+BSD~−1FVOF−BSD~−1BTp−Cp=BD~−1H~(U)+BSD~−1FVOF−R(SD~−1)p\begin{aligned} 0 &= BU - Cp = B\widetilde{D}^{-1}\widetilde{H}(U) + BS\widetilde{D}^{-1}F_{\mathrm{VOF}} - BS\widetilde{D}^{-1}B^{\mathsf{T}}p - Cp \\ &= B\widetilde{D}^{-1}\widetilde{H}(U) + BS\widetilde{D}^{-1}F_{\mathrm{VOF}} - R(S\widetilde{D}^{-1})p \end{aligned}

obtaining a modified pressure equation that is consistent with the IB constraint:

R(SD~−1)p=B(D~−1H~(U)+SD~−1FVOF).(39)R(S\widetilde{D}^{-1})p = B(\widetilde{D}^{-1}\widetilde{H}(U) + S\widetilde{D}^{-1}F_{\mathrm{VOF}}). \tag*{(39)}

Within the PIMPLE projection scheme that handles the pressure-velocity coupling, Equation (38) provides a pressure field that ensures a corrected velocity field, computed using (37) that satisfies both incompressibility and IB constraint. The resulting VOF-IB PIMPLE solution procedure is reported in Algorithm 1.

Algorithm 1: VOF-IB-PIMPLE

Initialize with U0,prgh0,α0U^{0},p_{\mathrm{rgh}}^{0},\alpha^{0}

/* ----------------------------------------- TIME Loop ----------------------------------------- */

for n=0,…,NT−1n = 0,\ldots,N_{T}-1 do

/--------------------------------------- PIMPLE Loop ---------------------------------------/

for l=0,…,Nouter−1l = 0,\ldots,N_{\mathrm{outer}}-1 do

// 1. Volume Fraction Equation

/------------------------------------ SUB-CYCLE Loop ------------------------------------/

for k=0,…,Nsc−1k = 0,\ldots,N_{\mathrm{sc}}-1, with α0=αl\alpha^{0}=\alpha^{l} do

/* -------------------------------- α\alpha-Correction Loop -------------------------------- */

for q=0,…,Nα,corr−1q = 0,\ldots,N_{\alpha,\mathrm{corr}}-1, with α0=αk\alpha^{0}=\alpha^{k} do

Compute αiq+1=αiq−Δt∣Ki∣∑f∈Fi[Φf,upq+λMΦfq]\alpha_{i}^{q+1}=\alpha_{i}^{q}-\frac{\Delta t}{\lvert K_{i}\rvert}\sum_{f\in F_{i}}\left[\Phi_{f,\mathrm{up}}^{q}+\lambda_{M}\Phi_{f}^{q}\right]

Impose IB constraint on Kj∈TS∩KiK_{j}\in T_{S}\cap K_{i}, Ki∈TIBK_{i}\in T_{\mathrm{IB}}: αjq+1=αiq+1±gN,Fij∥dij∥\alpha_{j}^{q+1}=\alpha_{i}^{q+1}\pm g_{N,F_{ij}}\lVert d_{ij}\rVert;

end

αk+1←αNα,corr\alpha^{k+1}\leftarrow\alpha^{N_{\alpha,\mathrm{corr}}}

Update ρk+1=αk+1ρ1+(1−αk+1)ρ2\rho^{k+1}=\alpha^{k+1}\rho_{1}+(1-\alpha^{k+1})\rho_{2}

Update μk+1=αk+1μ1+(1−αk+1)μ2\mu^{k+1}=\alpha^{k+1}\mu_{1}+(1-\alpha^{k+1})\mu_{2}

Store mass flux (ρfϕf)k+1=(αfϕf)k+1(ρ1−ρ2)+ϕflρ2(\rho_{f}\phi_{f})^{k+1}=(\alpha_{f}\phi_{f})^{k+1}(\rho_{1}-\rho_{2})+\phi_{f}^{l}\rho_{2}

end

αl+1←αNsc\alpha^{l+1}\leftarrow\alpha^{N_{\mathrm{sc}}}

Sum mass fluxes (ρfϕf)l+1=∑k(ρfϕf)k+1(\rho_{f}\phi_{f})^{l+1}=\sum_{k}(\rho_{f}\phi_{f})^{k+1}

// 2. Momentum Predictor (optional)

Impose IB constraint on (25): A→A~A\rightarrow\widetilde{A} and F→F~F\rightarrow\widetilde{F}

Solve A~Ul+12=F~+χ(FVOF−BTpl)\widetilde{A}U^{l+\frac{1}{2}}=\widetilde{F}+\chi\left(F_{\mathrm{VOF}}-B^{\mathsf{T}}p^{l}\right)

/-------------------------------------------- PISO Loop --------------------------------------------/

for m=0,…,Ninner−1m=0,\ldots,N_{\mathrm{inner}}-1, with U0=UlU^{0}=U^{l} or U0=Ul+12U^{0}=U^{l+\frac{1}{2}} do

// 3. Pressure equation

Solve R(SD~−1)pm+1=B(D~−1H~(Um)+SD~−1FVOF)R(S\widetilde{D}^{-1})p^{m+1}=B(\widetilde{D}^{-1}\widetilde{H}(U^{m})+S\widetilde{D}^{-1}F_{\mathrm{VOF}})

// 4. Pressure relaxation

Compute pm+1←(1−βp)pm+βppm+1p^{m+1}\leftarrow(1-\beta_{p})p^{m}+\beta_{p}p^{m+1}

// 5. Velocity correction

Compute Um+1=D~−1H~(Um)−SD~−1BTpm+1+SD~−1FVOFU^{m+1}=\widetilde{D}^{-1}\widetilde{H}(U^{m})-S\widetilde{D}^{-1}B^{\mathsf{T}}p^{m+1}+S\widetilde{D}^{-1}F_{\mathrm{VOF}}

end

Ul+1←UNinnerU^{l+1}\leftarrow U^{N_{\mathrm{inner}}}

pl+1←pNinnerp^{l+1}\leftarrow p^{N_{\mathrm{inner}}}

end

Un+1←UNouterU^{n+1}\leftarrow U^{N_{\mathrm{outer}}}, pn+1←pNouterp^{n+1}\leftarrow p^{N_{\mathrm{outer}}}, αn+1←αNouter\alpha^{n+1}\leftarrow\alpha^{N_{\mathrm{outer}}}

end

Issues with high-viscosity contrasts

The standard VOF solver of OpenFOAM implements a segregated approach not only for the pressure-velocity coupling, handled by means of the PIMPLE projection scheme, but also for the momentum equation itself that couples the three velocity components. Indeed, the viscous diffusion term is split into an implicit component pertaining the gradient of velocity and an explicit one concerning the transposed gradient. This semi-implicit treatment allows to solve each velocity component sequentially and independently of the others. From the algebraic point of view, matrix A∈R3N×3NA\in\mathbb{R}^{3N\times3N} of the FV system (25) becomes a block-diagonal matrix, such that (31) has actually the form:

\begin{bmatrix} A_{x} & 0 & 0 & B^{x} \\ 0 & A_{y} & 0 & B^{y} \\ 0 & 0 & A_{z} & B^{z} \\ B^{x} & B^{y} & B^{z} & 0 \end{bmatrix} \begin{bmatrix} U_{x} \\ U_{y} \\ U_{z} \\ p \end{bmatrix} = \begin{bmatrix} (F+F_{\mathrm{VOF}})_{x} \\ (F+F_{\mathrm{VOF}})_{y} \\ (F+F_{\mathrm{VOF}})_{z} \\ 0 \end{bmatrix}. \tag*{(40)}\end{bmatrix}

This allows to solve three smaller systems of dimension N×NN \times N sequentially. Hence, the advantage of this approach is its easy implementation as well as the lower computational cost for the solution of the resulting finite-volume algebraic system. On the other hand, handling part of the diffusive term explicitly leads to a severe stability restriction on time-step Δt\Delta t that depends quadratically on the mesh size hh [18]:

Δt≤Δtν=Re⁡h26.(41)\Delta t \leq\Delta t_{\nu} = \operatorname{Re}\frac{h^{2}}{6}. \tag*{(41)}

where Re⁡=ULρ1μ1\operatorname{Re} = \frac{UL\rho_{1}}{\mu_{1}} is the Reynolds number, with UU and LL characteristic velocity and length, respectively. At low Reynolds regimes (Re⁡≪1\operatorname{Re} \ll1) where viscous effects are dominant, as in the case of polymer mixing applications, and using high mesh resolutions, condition (40) becomes particularly severe.

Moreover, multiphase flows with high viscosity contrasts complicate numerical stability even more [25, 74], due to strong inter-component velocity couplings appearing from the transposed gradient term in the vicinity of the two-phase interface [29, 74]. Radical variations in material properties lead to ill-conditioned linear systems: matrix AA, arising from the discretization of the unsteady, convective, and viscous diffusion terms, becomes highly stiff due to the sharp jump in viscosity across the interface [47].

A Block-Coupled VOF-IB Solver

An alternative numerical scheme is proposed, denoted as block-coupled approach (BC), where the whole viscous diffusion term ∇⋅(μ(∇u+∇Tu))\nabla\cdot\left(\mu(\nabla\mathbf{u} + \nabla^{\mathsf{T}}\mathbf{u})\right) of the momentum equation is treated implicitly, in order to eliminate the viscous time-step restriction (40).

The BC method was initially proposed for applications to finite elasticity problems by Cardiff et. al [11]. In the context of two-phase Navier-Stokes equations, the implicit discretization scheme used in [11] for the Cauchy stress tensor is adapted to the viscous stress tensor.

Method Derivation

We write the semi-discretized momentum equation, integrated over control volume Ki∈ThK_i \in\mathcal{T}_h:

∫Kiρn+1uin+1−uinΔt dV+∫∂Ki∑j=13ρn+1uin+1ujnnj dS−∫Ki∑j=13∂∂xj(μn+1∂uin+1∂xj−μn+1∂ujn+1∂xi) dV+∫∂Kiprghn+1ni dS=∫Ki−(g⋅x)∂ρ∂xi+σκn+1∂αn+1∂xi dV,(42)\int_{K_i} \rho^{n+1}\frac{u_i^{n+1}-u_i^n}{\Delta t}\,\mathrm{d}V +\int_{\partial K_i}\sum_{j=1}^{3}\rho^{n+1}u_i^{n+1}u_j^n n_j\,\mathrm{d}S -\int_{K_i}\sum_{j=1}^{3}\frac{\partial}{\partial x_j}\left(\mu^{n+1}\frac{\partial u_i^{n+1}}{\partial x_j}-\mu^{n+1}\frac{\partial u_j^{n+1}}{\partial x_i}\right)\,\mathrm{d}V +\int_{\partial K_i}p_{\mathrm{rgh}}^{n+1}n_i\,\mathrm{d}S =\int_{K_i}-(\mathbf{g}\cdot\mathbf{x})\frac{\partial\rho}{\partial x_i}+\sigma\kappa^{n+1}\frac{\partial\alpha^{n+1}}{\partial x_i}\,\mathrm{d}V, \tag*{(42)}

associated to the ii–th velocity component, so that the inter-component couplings introduced by the implicit treatment of the transposed gradient term are clearly visible: the ii–th equation valid on CV KiK_i (41) can not be de-coupled from the other two components, since there is an explicit dependence on the unknowns ujn+1u_j^{n+1}, j≠ij \ne i at time tn+1t^{n+1}.

Focusing only on the viscous diffusion term, in [11] the following relation is proved to hold:

∫Ki∇⋅(μ∇u+μ∇Tu) dV=∫∂Ki2μ(∇un)n⏟Tn+μ(∇ut)n+μ∇tun⏟Tt dS.(43)\int_{K_i}\nabla\cdot\left(\mu\nabla\mathbf{u}+\mu\nabla^{\mathsf{T}}\mathbf{u}\right)\,\mathrm{d}V =\int_{\partial K_i}\underbrace{2\mu(\nabla u_n)n}_{T_n} +\underbrace{\mu(\nabla u_t)n+\mu\nabla_tu_n}_{T_t}\,\mathrm{d}S. \tag*{(43)}

where un=(n⊗n)u\mathbf{u}_n=(\mathbf{n}\otimes\mathbf{n})\mathbf{u} and ut=(I−n⊗n)u\mathbf{u}_t=(\mathbf{I}-\mathbf{n}\otimes\mathbf{n})\mathbf{u} are the normal and tangential components to ∂Ki\partial K_i, respectively, and ∇t=(I−n⊗n)∇\nabla_t=(\mathbf{I}-\mathbf{n}\otimes\mathbf{n})\nabla is the tangential gradient operator.

Particular care is dedicated to the discretization of the last term:

∫∂Kiμ(∇tun) dS≈∑f∈Fiμf(∇tun)f⋅Sf≈∑f∈Fiμf(1∣Sf∣∑e∈Efme(un)eLe)⋅Sf.(44)\int_{\partial K_i}\mu(\nabla_tu_n)\,\mathrm{d}S \approx\sum_{f\in\mathcal{F}_i}\mu_f(\nabla_tu_n)_f\cdot\mathbf{S}_f \approx\sum_{f\in\mathcal{F}_i}\mu_f\left(\frac{1}{\lvert\mathbf{S}_f\rvert}\sum_{e\in\mathcal{E}_f}m_e(u_n)_eL_e\right)\cdot\mathbf{S}_f. \tag*{(44)}

where the face-interpolated tangential gradient is computed by means of a Finite Area (FA) method [11]. We denote by Ef\mathcal{E}_f the set of edges of face ff and by LeL_e the length of edge ee. Moreover, indicating with e^\hat{\mathbf{e}} the unit vector directed along edge ee, the unit bi-normal vector can be computed as me=e^×nf\mathbf{m}_e=\hat{\mathbf{e}}\times\mathbf{n}_f.

The evaluation of unu_n at the edge middle point, indicated as (un)e(u_n)_e, is obtained as a linear combination of the values in the edge nodes:

ue=up1+up22(45)u_e=\frac{u_{p_1}+u_{p_2}}{2} \tag*{(45)}

where the subscripts p1p_1 and p2p_2 identify the extrema (nodes) of edge ee.

In turn, upiu_{p_i} needs to be expressed in terms of cell-centered values that are the actual degrees of freedom of collocated Finite Volumes. For this purpose, Cardiff et al. [11] utilize a weighted least squares (WLS) method to approximate the point field values with a linear fit. The WLS interpolation stencil is defined for each vertex pip_i of the computational grid, consisting of all point-neighbors, i. e. all cells containing vertex pip_i, whose set we denote by Ppi\mathcal{P}_{p_i}. Moreover, we define the set of point-cell neighbors of the two vertices of edge ee as Pe=Pp1∪Pp2\mathcal{P}_e = \mathcal{P}_{p_1} \cup\mathcal{P}_{p_2}, as reported in Figure 5. If a node belongs to the boundary of the domain, boundary face-centers are included in the stencil instead of the neighbor cell centers.

Sketch of point-cell neighbors sets associated to vertices $p_1$ and $p_2$ of edge $e$.

Figure 5. Sketch of point-cell neighbors sets associated to vertices p1p_1 and p2p_2 of edge ee.

Figure 4. Geometric representation of a polyhedral cell Ki∈ThK_i \in\mathcal{T}_h, showing face f∈Fif \in\mathcal{F}_i with its unit normal vector nf\mathbf{n}_f, and an edge e∈Efe \in\mathcal{E}_f having vertices p1p_1 and p2p_2, tangent vector e^\hat{\mathbf{e}} and bi-normal vector me\mathbf{m}_e.

The result is that, fixed a control volume Ki∈ThK_i \in\mathcal{T}_h, in order to approximate the tangential gradient at each face f∈Fif \in\mathcal{F}_i, the extended stencil of all point-neighbors of cell KiK_i is considered, consisting of all cells KjK_j (or boundary faces if KiK_i is a boundary cell) sharing a vertex with KiK_i, that we denote as Kip\mathcal{K}_i^p (see Figure 6). We call Kip\mathcal{K}_i^p the extended stencil associated to cell KiK_i opposed to the usual compact stencil Ki\mathcal{K}_i (Figure 6) collecting face-neighbors of cell KiK_i, that is usually employed in OpenFOAM Finite Volume schemes.

Figure 6. Comparison between compact and extended stencils for a simple 3×33 \times3 two-dimensional grid (adapted from [11]).

Once the WLS approximator is obtained, using field values uKiu_{K_i} for Ki∈PpiK_i \in\mathcal{P}_{p_i} as observation data, each point field value upi\mathbf{u}_{p_i} can be expressed as

upi=∑j=1Niγjuj,(46)\mathbf{u}_{p_i} = \sum_{j=1}^{N_i} \gamma_j \mathbf{u}_j, \tag*{(46)}

where Ni=card⁡(Ppi)N_i = \operatorname{card}(\mathcal{P}_{p_i}) and γj\gamma_j are interpolation weights, whose values can be obtained after solving the WLS minimization problem.

Therefore, the discretization of the tangential gradient term, starting from (43), becomes:

∫∂Kiμ(∇tun) dS≈∑f∈Fiμf(∇tun)f⋅Sf=∑f∈Fi∑e∈Ef∑P∈Peμf2LeγP(me⊗nf)uP.(47)\int_{\partial K_i} \mu(\nabla_t u_n)\,dS \approx\sum_{f\in\mathcal{F}_i} \mu_f(\nabla_t u_n)_f\cdot\mathbf{S}_f = \sum_{f\in\mathcal{F}_i}\sum_{e\in\mathcal{E}_f}\sum_{P\in\mathcal{P}_e} \frac{\mu_f}{2}L_e\gamma_P(\mathbf{m}_e\otimes\mathbf{n}_f)\mathbf{u}_P. \tag*{(47)}

Consequently, the FV approximation of the implicit viscous term (42), under the simplifying assumption of orthogonal mesh, reads:

∫Ki∇⋅(μ∇u+μ∇Tu) dV=∫∂Ki2μ(∇nun)n⏟Tn+μ(∇tun)n+μ∇tun⏟Tt dS≈∑f∈Fi2μf∣Sf∣(nf⊗nf)uN−uP∥df∥+∑f∈Fiμf∣Sf∣(I−nf⊗nf)uN−uP∥df∥+∑f∈Fi∑e∈Ef∑P∈Peμf2LeγP(me⊗nf)uP.(48)\begin{aligned} \int_{K_i}\nabla\cdot\left(\mu\nabla\mathbf{u}+\mu\nabla^{\mathsf{T}}\mathbf{u}\right)\,dV &=\int_{\partial K_i}\underbrace{2\mu(\nabla_n u_n)\mathbf{n}}_{\mathcal{T}_n}+\underbrace{\mu(\nabla_t u_n)\mathbf{n}+\mu\nabla_t u_n}_{\mathcal{T}_t}\,dS\\ &\approx\sum_{f\in\mathcal{F}_i}2\mu_f\lvert\mathbf{S}_f\rvert(\mathbf{n}_f\otimes\mathbf{n}_f)\frac{\mathbf{u}_N-\mathbf{u}_P}{\lVert\mathbf{d}_f\rVert} +\sum_{f\in\mathcal{F}_i}\mu_f\lvert\mathbf{S}_f\rvert(\mathbf{I}-\mathbf{n}_f\otimes\mathbf{n}_f)\frac{\mathbf{u}_N-\mathbf{u}_P}{\lVert\mathbf{d}_f\rVert}\\ &\qquad+\sum_{f\in\mathcal{F}_i}\sum_{e\in\mathcal{E}_f}\sum_{P\in\mathcal{P}_e}\frac{\mu_f}{2}L_e\gamma_P(\mathbf{m}_e\otimes\mathbf{n}_f)\mathbf{u}_P. \tag*{(48)} \end{aligned}

The first two terms in (47) correspond to the normal gradient discretization, using standard FV schemes that only include the degrees of freedom belonging to the compact stencil KiK_i (Figure 6a). On the other hand, the last row of (47) is the tangential gradient discretized by means of the FA-WLS approach [11], where the three encapsulated sums involve the degrees of freedom belonging to every point-neighbor KjK_j of cell KiK_i belonging to the extended stencil KieK_i^{e} (Figure 6b).

Boundary face centers become additional DOFs since they are included as unknowns in the FV algebraic system, being point-neighbors of boundary cells, additional equations for boundary faces are attached to the linear system (see [11]).

Remark 6. We assumed to be working with an orthogonal mesh, but when working with real industrial applications, the orthogonality of the computational grid is almost never guaranteed. Non-orthogonality corrections are actually accounted for in the discretization of normal gradients, leading to additional tangential gradient contributions that are handled with the same FA-WLS method employed to obtain (46). Further details can be found in [11].

The block-coupled algebraic system has the same form of (21), but the structure and dimension are different due to the new viscous term discretization. We introduce the new block-coupled matrix Ac∈R3NE×3NEA^c\in\mathbb{R}^{3N_E\times3N_E}, where NE=card⁡(Th)+card⁡(Fh,B)=N+NBFN_E=\operatorname{card}(\mathcal{T}_h)+\operatorname{card}(\mathcal{F}_{h,B})=N+N_{BF}, containing unsteady, convective and viscous diffusion term contributions.

Numerical Results

The present section reports the numerical results obtained to assess the performance and reliability of the proposed modeling approach. As a preliminary step, the methodology is validated against the well-established benchmark test case [15]. The analysis is then extended to more complex and realistic configurations, focusing on continuous mixing devices of industrial interest.

Dog-bone injection molding test case

The dog-bone injection molding test case [15] is considered to assess the performances of segregated and block-coupled VOF-IB solvers, as well as to compare the solution obtained with conforming and non-conforming approaches.

The test case consists of a three-dimensional dog-bone shaped channel (Figure 7), that is initially full of air, in which a material is injected and a two-phase front advances.

Figure 7. Physical domain of the dogbone injection molding test case.

In order to perform simulations with the VOF-IB solver, both a conforming and a non-conforming grids are generated, that are reported in Figure 8. The background mesh for the IB case is a rectangular prism and two triangulated surfaces are constructed representing the two curved walls as immersed solid bodies. The IB grid refinement is chosen in such a way that the two grids have the same mesh size in the restricted portion of the channel.

Conforming and non-conforming computational grids for the dogbone benchmark case

Figure 8. Conforming (top) and non-conforming (bottom) computational grids for the dogbone benchmark case. The conforming grid has 64 704 cells, while the IB grid has 126 720 elements.

The boundary conditions imposed to the system of two-phase Navier-Stokes equations (3) take the following form:

{u=uin, ∇prgh⋅n=0, α=1at x=0, y∈[−0.01,0.01], z∈[−0.004,0],(∇u)n=0, prgh=0, ∇α⋅n=0at x=0.17, y∈[−0.01,0.01], z∈[−0.004,0],u=0, ∇prgh⋅n=0, ∇α⋅n=0elsewhere.\begin{cases} \mathbf{u}=\mathbf{u}_{\mathrm{in}},\ \nabla p_{\mathrm{rgh}}\cdot\mathbf{n}=0,\ \alpha=1 & \text{at }x=0,\ y\in[-0.01,0.01],\ z\in[-0.004,0],\\ (\nabla\mathbf{u})\mathbf{n}=0,\ p_{\mathrm{rgh}}=0,\ \nabla\alpha\cdot\mathbf{n}=0 & \text{at }x=0.17,\ y\in[-0.01,0.01],\ z\in[-0.004,0],\\ \mathbf{u}=0,\ \nabla p_{\mathrm{rgh}}\cdot\mathbf{n}=0,\ \nabla\alpha\cdot\mathbf{n}=0 & \text{elsewhere.} \end{cases}

where a uniform inlet velocity profile is prescribed, such that uin=(uin,x,0,0)\boldsymbol{u}_{in} = (u_{in,x}, 0, 0) with uin,x=0.12 m/su_{in,x} = 0.12\ \mathrm{m/s}.

A high viscosity non-Newtonian material is considered, following a Power law, whose parameters are calibrated to describe the rheology of a low density polyethylene [45]:

  • Fluid 1: μ1=Kγ˙n−1\mu_1 = K\dot{\gamma}^{n-1}, with K=16 630 Pa⋅sK = 16\,630\ \mathrm{Pa \cdot s}, n=0.334n = 0.334 and limiting kinematic viscosities νmax⁡=5⋅103 m2/s\nu_{\max} = 5 \cdot10^3\ \mathrm{m^2/s} and νmin⁡=10−2 m2/s\nu_{\min} = 10^{-2}\ \mathrm{m^2/s}, while density is set to ρ1=785 kg/m3\rho_1 = 785\ \mathrm{kg/m^3}.

  • Fluid 2 (air): μ2=2.6⋅10−5 Pa⋅s\mu_2 = 2.6 \cdot10^{-5}\ \mathrm{Pa \cdot s}, ρ2=0.73 kg/m3\rho_2 = 0.73\ \mathrm{kg/m^3}.

The viscosity ratio between the two fluids reaches the order of 10810^8 when ν1=νmax⁡\nu_1 = \nu_{\max}, corresponding to the limiting value for γ˙→0\dot{\gamma} \to0.

First, two simulations using the conforming VOF (segregated) and BC-VOF (block-coupled) two-phase solvers are performed.

The VOF solver of OpenFOAM (interFoam) applied to such a high viscosity contrasts between the two phases fails to provide stable numerical results in reasonable computational times. On the other hand, the new BC-VOF solver (UCInterFoam) reveals to be more robust, as expected, producing a reliable reliable solution.

Adaptive time-stepping is enabled for both solvers, setting Δtmax⁡=0.1 s\Delta t_{\max} = 0.1\ \mathrm{s} and initial time-step Δt0=10−3 s\Delta t^0 = 10^{-3}\ \mathrm{s} and Δt0=10−5 s\Delta t^0 = 10^{-5}\ \mathrm{s}, in the block-coupled and segregated setup, respectively. So that Δt\Delta t is allowed to vary throughout the simulation time under the constraint of keeping the Courant numbers max⁡(Coh)\max(\mathrm{Co}_h) and max⁡(Coα,h)\max(\mathrm{Co}_{\alpha,h}) under user-defined thresholds. In particular, the upper bounds on the Courant numbers are set to be max⁡(Coh)<1\max(\mathrm{Co}_h) < 1 and max⁡(Coα,h)<0.4\max(\mathrm{Co}_{\alpha,h}) < 0.4 for the segregated solver, on the other hand, the block-coupled solver is able to afford milder restrictions allowing to impose max⁡(Coh)<50\max(\mathrm{Co}_h) < 50 and max⁡(Coα,h)<2\max(\mathrm{Co}_{\alpha,h}) < 2. Indeed, when trying to set the latter thresholds in interFoam the simulation fails due to the Courant number that suddenly reaches the order of 10310^3 and keeps increasing, thus leading to unfeasible Δt\Delta t, that cause the simulation to crash.

The axial velocity field obtained at consecutive time-instants with both VOF and BC-VOF solvers is reported in Figure 9. Numerical instabilities arising in the lighter phase when simulating with the VOF solver, especially at final time-steps when the free-surface approaches the domain restriction. Indeed, we observe at time t=0.75 st = 0.75\ \mathrm{s} that the velocity field inside the gaseous phase degenerates in fig:9a, contrarily to the BC-VOF solution that provides a physical behavior in fig:9b.

Figure 9. Comparison of velocity field obtained with either the segregated or the block-coupled solver for the injection molding simulation of highly viscous non-Newtonian Power law fluid with. Axial velocity profile is plotted on a slice in the x−yx-y plane for both fluid phases at three successive time-instants, t=0.25 st = 0.25\ \mathrm{s}, t=0.5 st = 0.5\ \mathrm{s} and t=0.75 st = 0.75\ \mathrm{s}.

Afterwards, we tested the ability of the non-conforming BC-VOF-IB solver (UCInterIbFoam) to provide reliable results in comparison with the conforming BC-VOF counterpart.

Figures 10, 11 and 12 show the numerical solutions of axial velocity, pressure and dynamic viscosity fields, respectively, comparing BC-VOF (UCInterIbFoam) with BC-VOF (UCInterFoam) results. In particular, the advancement of the two-phase interface is reported at three different time-instants of the simulations. The results obtained with the non-conforming solver are quite satisfying, indeed, the interface advancement is qualitatively similar to the conforming solution. Some discrepancies are observed, especially in the last snapshot at time t=0.75t = 0.75 s during which the interface passes through the restriction where the two lateral walls are not planar. Here we observe that the IB two-phase interface is slowed down and at the same time a viscosity peak appears in Figure 12 that in the conformal case is not present (Figure 12).

Figure 12. Two-phase front advancement of highly viscous non-Newtonian Power law material for three successive time-instants, t=0.25t = 0.25 s, t=0.5t = 0.5 s and t=0.75t = 0.75 s, with dynamic viscosity field. Comparison between block-coupled conforming and IB simulations on a slice in the x−yx - y plane occupied by fluid 1.

Figure 10. Two-phase front advancement of highly viscous non-Newtonian Power law material for three successive time-instants, t=0.25t = 0.25 s, t=0.5t = 0.5 s and t=0.75t = 0.75 s, with axial velocity field. Comparison between block-coupled conforming and IB simulations on a slice in the x−yx - y plane occupied by fluid 1.

Figure 11. Two-phase front advancement of highly viscous Newtonian material for three successive time-instants, t=0.25t = 0.25 s, t=0.5t = 0.5 s and t=0.75t = 0.75 s, with pressure field. Comparison between block-coupled conforming and IB simulations in the whole domain occupied by fluid 1.

A quantitative comparison of the velocity and viscosity fields along the vertical centerline and of the pressure field along the horizontal centerline is reported in Figure 13 at time 1 s when the interface is approaching the outlet. The field samples at centerlines reveal a slight accuracy reduction of the IB approach with respect to the conforming solution when introducing the viscosity nonlinear term.

Plots of axial velocity, dynamic viscosity, and pressure profiles

Figure 13. Plots of axial velocity and dynamic viscosity profiles along vertical centerline (left and center, respectively), and pressure profile along horizontal centerline (right) at time t=1t = 1 s for the two-phase flow of a highly viscous non-Newtonian Power law material, comparing conforming and IB VOF block-coupled solver.

The performance of the three solvers is presented in Table 1, where we observe a substantial increase of the allowed Δt\Delta t when using BC-VOF and BC-VOF-IB. The minimum Δt\Delta t reached by the VOF solver is unfeasible, leading to a computational cost of 7 hours versus the 16 minutes CPU time1 of the BC-VOF solver. On the other hand, the non-conforming BC-VOF-IB solver is more computational expensive (58 minutes of CPU time) than BC-VOF, but it still allows to obtain a significant gain with respect to the original segregated approach.

min Δt\Delta t [s]avg Δt\Delta t [s]max Δt\Delta t [s]max(Coh)(\mathrm{Co}_h)max(Coα,h)(\mathrm{Co}_{\alpha,h})CPU time
VOF5.25⋅10−65.25 \cdot10^{-6}2.21⋅10−42.21 \cdot10^{-4}1.32⋅10−31.32 \cdot10^{-3}16.620.497h
BC-VOF1.19⋅10−31.19 \cdot10^{-3}3.86⋅10−33.86 \cdot10^{-3}6.25⋅10−36.25 \cdot10^{-3}9.823.0816min
BC-VOF-IB9.73⋅10−49.73 \cdot10^{-4}2.48⋅10−32.48 \cdot10^{-3}5.06⋅10−35.06 \cdot10^{-3}11.186.1258min

Table 1. Comparison of minimum, average, maximum Δt\Delta t, maximum Courant number and CPU time for the highly viscous non-Newtonian Power law dogbone case simulated with the segregated and the block-coupled solvers. Three-dimensional conforming and IB computational grids are composed of 64 704 and 126 720 cells, respectively, and both cases are run on 4 processors on the DMAT HPC cluster.

The trends of Δt\Delta t and maximum Courant numbers, max(Coh)(\mathrm{Co}_h) and max(Coα,h)(\mathrm{Co}_{\alpha,h}), during the whole simulation time are reported in Figures 14a and 14b, respectively. The smallest values of Δt\Delta t, corresponding to the peaks of Co, are attained when the phase front is passing through the restriction of the dogbone-shaped domain, for t∈(0.2,0.3)t \in(0.2, 0.3) s and t∈(0.7,0.8)t \in(0.7, 0.8) s.

Continuous mixing devices

The current section is devoted to the presentation of the free-surface numerical simulations obtained with the block-coupled two-phase conforming (BC-VOF) and non-conforming (BC-VOF-IB), applied to realistic geometries employed in the industrial context of polymer mixing processes. To the best of our knowledge, the results reported here represent a novel contribution to the literature on polymer mixing simulations, as few works [63, 20, 21, 22, 97, 73, 89] currently address such complex configurations under partial filling conditions.

The CFD simulations are run on the HPC cluster of the Department of Mathematics of Politecnico di Milano, composed of 40 computing nodes having 96 CPUs (AMD EPYC 7413 24-Core Processor) each and 512 GB RAM per node.

Here, we focus on the family of continuous mixing processes, so called because the material is continuously fed from a feeding hopper with a constant flow rate into the extruder device, consisting of a stationary barrel containing one or more screws that rotate at a predefined angular velocity. The screw rotation allows the material to be mixed and transported towards the outlet. The mechanical forces applied to the polymer by the barrel and the screws cause viscous friction that, together with the heat coming from thermo-regulation of the solid walls, heats and melts the material, as a consequence of the viscosity decrease due to a temperature increase. The material is continuously discharged at the end of the barrel, and exits the extruder through a die.

We introduce the drag flow rate, denoted as Q0Q_0, as the flow rate generated solely by the rotation of the screw, that only depends on the screw shape and velocity. It can be computed assuming there is no die at the end of the barrel causing a pressure resistance and in the absence of an imposed inlet flow rate. Different combinations of inlet flow rate QinQ_{\mathrm{in}}, imposed as a parameter of the process, and Q0Q_0 lead to either a stationary fully filled or partially filled (also referred to as starve-fed) scenario. In particular, for values of Qin>Q0Q_{\mathrm{in}} > Q_0 the material is expected to fully fill the extruder domain because the screw cannot move forward enough material as fast as it is fed and a negative pressure gradient is generated from inlet to outlet. On the contrary, when Qin<Q0Q_{\mathrm{in}} < Q_0 the inlet flow rate is lower than the screw’s pumping capacity and the device will be only partially filled.

Figure 14. Comparison of the block-coupled solver performance with respect to the segregated solver for the simulation of the highly viscous non-Newtonian Power law dogbone case.

The strong assumption on all the simulations that will be presented in this Chapter is the restriction to isothermal processes. That is due to the fact that our two-phase solvers currently do not include the solution of the energy equation. The results obtained on real geometries do not have a quantitative experimental counterpart for comparison, but qualitative evaluation and considerations are carried out based on the outcomes expected experimentally given the working conditions of the process.

The Single Screw Extruder

A preliminary test of the free-surface flow of a high-viscosity Newtonian material flowing in a simplified SSE geometry is provided. The presence of only one rotating screw allows the simulation of the process inside the SSE with both conforming and IB solvers. Indeed, the conforming simulation can be performed by solving the two-phase Navier-Stokes equations (3) in non-inertial reference frame, solved with respect to the relative velocity of a moving observer that rotates with the screw.

The SSE is the simplest continuous mixing device considered in this study, as it consists of a single screw rotating inside a stationary barrel. The SSE simplified geometry is displayed in Figure 15. The barrel length and diameter are LB=64L_B = 64 mm and dB=26d_B = 26 mm, respectively. The maximum and minimum screw diameters are dS,max⁡=25d_{S,\max} = 25 mm and dS,min⁡=15d_{S,\min} = 15 mm, resulting in a minimum clearance between screw and barrel equal to a=dB−dS,max⁡2=0.5a = \frac{d_B-d_{S,\max}}{2} = 0.5 mm. The screw length is LS=62L_S = 62 mm so that a gap of 2 mm is left between the end of the screw and the physical outlet, that does not coincide with the entire barrel section, but it has a reduced diameter dout=10d_{\mathrm{out}} = 10 mm.

Geometry of the simplified SSE with restricted outlet

Figure 15. Geometry of the simplified SSE with restricted outlet.

Two computational grids are generated to perform simulations using both conforming and IB approaches. Figure 16 reports a comparison of the two meshes.

Figure 16. Body-fitted and non-conforming computational grids for the simplified SSE.

The conforming grid shown in Figure 16 is generated starting from a cylindrical annulus with outer radius Rout=dB2R_{\mathrm{out}} = \frac{d_B}{2} and inner radius Rin=dS,max⁡2R_{\mathrm{in}} = \frac{d_{S,\max}}{2}. Subsequently, the grid is perturbed in order to transform the inner circular axial sections into the elliptic screw section.

In contrast, the non-conforming grid is much easier to generate. It is constructed using a cylindrical annulus of outer radius Rout=dB2R_{\mathrm{out}} = \frac{d_B}{2} and inner radius Rin=7.5 mm<dS,min⁡2R_{\mathrm{in}} = 7.5\ \mathrm{mm} < \frac{d_{S,\min}}{2}. The triangulated STL surface of the screw is used to generate the IB mask and subdivide the cells in fluid, IB and dead cells (see Figure 17). The resulting non-conforming grid of Figure 16b is necessarily finer with respect to the body-fitted counterpart. Progressive refinement in the radial direction is adopted to ensure an adequate resolution of the narrow gap between screw and barrel. It it worth noting that the use of the IB approach allows to work with more orthogonal and less skewed mesh elements.

Figure 17. Subdivision of the non-conforming background grid into IB cells (green), fluid cells (blue) and dead cells (red).

For the simulations we adopt the physical properties of a high Newtonian viscosity material, μ1=1232 Pa⋅s\mu_1 = 1232\ \mathrm{Pa}\cdot\mathrm{s}, and density ρ1=880 kg/m3\rho_1 = 880\ \mathrm{kg}/\mathrm{m}^3, concerning fluid 1, while fluid 2 is assigned to air density and viscosity. This choice is not intended to represent a realistic polymeric material. Rather, the objective of the present test is to assess the robustness of the BC-VOF and BC-VOF-IB solvers in handling a two-phase flow with a strong viscosity contrast, μ1μ2∼105−106\frac{\mu_1}{\mu_2} \sim10^5 - 10^6, within a domain featuring complex moving boundaries.

A preliminary single-phase simulation is run in order to compute the drag flow rate Q0Q_0 associated to a chosen screw velocity of 60 rpm. Afterwards, a free-surface simulation is setup reproducing a total filling scenario, by imposing a constant inlet flow rate Qin>Q0Q_{\mathrm{in}} > Q_0. The SSE device is initially full of air and the highly viscous material is injected from the inlet patch (the red surface of Figure 15) with a uniform axial velocity uˉin,z\bar{u}_{\mathrm{in},z} given by uˉin,z=Qinρ1∣Sin∣\bar{u}_{\mathrm{in},z} = \frac{Q_{\mathrm{in}}}{\rho_1 \lvert S_{\mathrm{in}}\rvert}, with ∣Sin∣\lvert S_{\mathrm{in}}\rvert denoting the surface area of the inflow boundary. Navier-slip conditions are imposed at barrel walls, allowing the viscous material to advance more freely along the solid boundary.

Velocity magnitude and pressure fields obtained with UCInterFoam and UCInterIbFoam are reported in Figure 18 and Figure 19, respectively, for the highly viscous Newtonian phase that is progressively transported in the simplified SSE geometry.

Figure 18. Comparison between body-fitted (left) and non-conforming (right) numerical results of the velocity magnitude of fluid 1 reported on a slice along the x−zx-z plane (view from above) at time instants t=0.8t = 0.8 s, 1.61.6 s, 2.42.4 s, 55 s.

Figure 19. Comparison between body-fitted (left) and non-conforming (right) numerical results of the pressure field of fluid 1 reported on a slice along the x−zx-z plane (view from above) at time instants t=0.8t = 0.8 s, 1.61.6 s, 2.42.4 s, 55 s.

Moreover, Figure 20 displays the two-phase front advancement in the whole SSE domain. The results obtained with the two approaches present some discrepancies, especially concerning the pressure field, as one can notice also from the plots of average pressure values on axial sections in Figure 21.

Solution with BC-VOF conforming solver using SRF tools

(a) (a) Solution with BC-VOF conforming solver using SRF tools.

Solution with BC-VOF-IB solver using SRF tools

(b) (b) Solution with BC-VOF-IB solver using SRF tools.

Figure 20. Comparison between body-fitted (left) and non-conforming (right) numerical results showing the two-phase front evolution at time instants t=0.8t = 0.8 s, 1.61.6 s, 2.42.4 s, 55 s.

Plots of mass flow rate, average pressure and filling ratio over axial sections at three time instants

Figure 21. Plots of mass flow rate, average pressure and filling ratio over axial sections at time-instants t=0.8t = 0.8 s, 2.42.4 s, 1010 s. The dotted gray line positioned at z=60z = 60 mm indicates where the screw ends. Red and blue dotted lines are Q0,confQ_{0,\mathrm{conf}} and Q0,IBQ_{0,\mathrm{IB}}, respectively.

In Figure 21 we present the evolution of mass flow rate QQ (left column), average pressure p‾rgh\overline{p}_{\mathrm{rgh}} (central column) and filling ratio denoted by FF (right column), computed for fluid 1 along axial sections, comparing the results obtained with the body-fitted and the non-conforming grids, using UCInterFoam and UCInterIbFoam, respectively. The latter solver is employed in two different configurations: the first one solves the equations in non-inertial reference frame (SRF tools of OpenFOAM), so the IB surface of the screw is fixed over time; secondly, dynamic immersed boundary tools are used, meaning that the momentum equation is solved in the absolute reference frame leading to a moving IB mask that follows the screw rotation.

At each axial section S~\widetilde{S} we define

  • The mass flow rate: Qconf=∫S~confαρ1uz dSQ_{\mathrm{conf}} = \int_{\widetilde{S}_{\mathrm{conf}}} \alpha\rho_1 u_z\,\mathrm{d}S and QIB=∫S~IBχIBαρ1uz dSQ_{\mathrm{IB}} = \int_{\widetilde{S}_{\mathrm{IB}}} \chi_{\mathrm{IB}} \alpha\rho_1 u_z\,\mathrm{d}S;

  • The average pressure: p‾rgh,conf=∫S~confαprgh dS∫S~confα dS\overline{p}_{\mathrm{rgh,conf}} = \frac{\int_{\widetilde{S}_{\mathrm{conf}}} \alpha p_{\mathrm{rgh}}\,\mathrm{d}S}{\int_{\widetilde{S}_{\mathrm{conf}}} \alpha\,\mathrm{d}S} and p‾rgh,IB=∫S~IBχIBαprgh dS∫S~IBχIBα dS\overline{p}_{\mathrm{rgh,IB}} = \frac{\int_{\widetilde{S}_{\mathrm{IB}}} \chi_{\mathrm{IB}} \alpha p_{\mathrm{rgh}}\,\mathrm{d}S}{\int_{\widetilde{S}_{\mathrm{IB}}} \chi_{\mathrm{IB}} \alpha\,\mathrm{d}S};

  • The filling ratio: Fconf=∫S~confα dS∫S~confdSF_{\mathrm{conf}} = \frac{\int_{\widetilde{S}_{\mathrm{conf}}} \alpha\,\mathrm{d}S}{\int_{\widetilde{S}_{\mathrm{conf}}} \mathrm{d}S} and FIB=∫S~IBχIBα dS∫S~IBχIB dSF_{\mathrm{IB}} = \frac{\int_{\widetilde{S}_{\mathrm{IB}}} \chi_{\mathrm{IB}} \alpha\,\mathrm{d}S}{\int_{\widetilde{S}_{\mathrm{IB}}} \chi_{\mathrm{IB}}\,\mathrm{d}S};

Despite the aforementioned discrepancies, both solvers provide physically reasonable solutions. Indeed, prescribing Qin>Q0Q_{\mathrm{in}} > Q_0 leads to a pressure decrease from inlet to outlet, due to the fact that the flow is dominated by the inlet mass flow rate that moves the material forward, since the screw is not rotating with a sufficiently high velocity able to contrast that pressure drop. Additionally, in the proximity of the screw ending at z=0.06z = 0.06 m (vertical dotted line in Figure 21) we observe a stronger pressure gradient due to the presence of the outlet restriction, that allows the material to be discharged from the die. As expected, at the final time t=10t = 10 s, the material eventually fills the whole domain leading to a filling ratio F∼1F \sim1 on all sections.

The Twin Screw Extruder

The real geometry of a TSE is addressed, characterized by the presence of two screws positioned side-by-side, providing improved mixing efficiency and process control compared to the simpler SSE. Specifically, we consider a co-rotating intermeshing TSE [75], where both screws rotate in the same direction and the flights of one screw interpenetrate the channels of the other, creating the so-called self-wiping effect, where the screws constantly clean each other, thus preventing material from sticking and degrading. The main advantage of the TSE lies in its enhanced mixing capability, thanks to the presence of kneading blocks, that are particular screw elements shifted at angles of 45∘45^\circ or 90∘90^\circ, specifically designed to shear, stretch, and fold the polymer.

The geometry of the TSE device used for the simulations is reported in Figure 22. We denote by ℓ\ell the characteristic length being the horizontal distance (along xx-axis) between the axes of the two screws. Then, the total width of the barrel is LB,x=2ℓL_{B,x}=2\ell and its axial length is LB,z=2.8ℓL_{B,z}=2.8\ell, where ℓ\ell has a value of few centimeters. The gap between screws and barrel is of the order of millimeters, as well as the gap between the two screws. The screws are divided into three different portions: the first and last elements consist of transport modules while the middle section is characterized by kneading modules devoted to mixing rather than transport.

Computational domain and corresponding non-conforming grid for the TSE test case

Figure 22. (a) Computational domain: inlet (red), outlet (blue), barrel (green). (b) Computational grid: view from above. (c) Computational grid: front view. Computational domain and corresponding non-conforming grid for the TSE test case, with subdivision into IB cells (green), fluid cells (blue) and dead cells (red).

A non-conforming grid with respect to the screws is generated (Figure 22) in order to use the IB approximation [62, 60]. Indeed, the use of a conforming grid would be unfeasible in this case, requiring the adoption of dynamic mesh functionalities. The computational mesh is composed of 2 767 800 elements, ensuring to have enough fluid cells in the intermeshing zone between the two screws.

Fluid 1 is a high-viscosity power law polymeric material, with coefficients of the rheological law being calibrated based on experimental data provided by the industrial partner, flowing inside the TSE device whose screws are rotating at 100 rpm.

As it is done for the SSE case, a preliminary single-phase isothermal simulation is first carried out to determine the drag flow rate Q0Q_0 and to set two-phase simulations accordingly.

Free-surface Simulation of Fully Filled TSE

The first two-phase simulation inside the TSE geometry of the highly viscous non-Newtonian polymer is obtained injecting the polymeric material from the inlet patch at the initial time with a constant inlet flow rate Qin>Q0Q_{\mathrm{in}}>Q_0, with Q0Q_0 coming from the single-phase simulation. The domain is initially full of air, the screw velocity is kept at 100 rpm and no-slip conditions are imposed at barrel for both phases.

The block-coupled BC-VOF-IB solver UCInterIbFoam is employed for the simulation.

The results obtained at consecutive times are displayed in Figure 23, where the velocity magnitude and the pressure fields are plotted on a slice of the x−zx-z plane (view from above), while the evolution of the two-phase front in the whole TSE domain is reported in Figure 24a together with the corresponding axial distributions of average pressure and filling ratio in Figure 24b.

(a) Velocity field.

(b) Pressure field.

Figure 23. Case of inlet flow rate Qin>Q0Q_{\mathrm{in}} > Q_0: velocity magnitude and pressure field of fluid 1, obtained with the BC-VOF-IB solver, reported on a slice along the x−zx-z plane (top view) at time instants t=0.5t = 0.5 s, 11 s, 22 s, 33 s.

The same behavior we observed for the simplified SSE when imposing Qin>Q0Q_{\mathrm{in}}>Q_0 is reproduced for the TSE. At the final simulation time the device is totally filled of polymer, indeed F∼1F\sim1 on all axial sections, and a negative pressure gradient originates from inlet to outlet.

Two-phase front evolution at three time instants

(a) Two-phase front evolution.

Average pressure and filling ratio over axial sections

(b) Average pressure and filling ratio over axial sections. Grey dotted lines represent start and ending of kneading modules.

Figure 24. Case of inlet flow rate Qin>Q0Q_{\mathrm{in}} > Q_0: temporal evolution of the two-phase interface (left) and plots of average pressure and filling ratio along axial sections (right) is reported for times t=1t = 1 s, 22 s, 33 s.

Free-Surface Simulation of Partially Filled TSE

The second setup for the two-phase simulation inside the TSE device is aimed at reproducing a partially filled scenario. This entails the imposition of an inlet flow rate Qin<Q0Q_{\mathrm{in}} < Q_0 for the polymeric material that is injected from the inlet patch. All the other simulation settings remain unchanged with respect to the previous case.

Figure 25 shows the numerical results for the velocity magnitude and the pressure fields plotted on a slice of the x−zx-z plane (view from above). We represent also the two-phase front advancement in the TSE domain in Figure 26a, where at the last time-step, after the material front has surpassed the kneading elements, more polymer accumulates in that portion of the device since the kneading modules are devoted to mixing rather than transport, as it is expected from experimental evidence. Plots of average pressure and filling ratio along axial slices are reported in Figure 24b, where we observe an opposite behavior for pressure with respect to the previous full filling case. Here, prescribing an inlet flow rate Qin<Q0Q_{\mathrm{in}} < Q_0, entails a pressure increase from inlet to outlet as soon as the material first reaches the outflow boundary.

Figure 25. Case of inlet flow rate Qin<Q0Q_{\mathrm{in}} < Q_0: velocity magnitude and pressure field of fluid 1, obtained with the BC-VOF-IB solver, reported on a slice along the x−zx-z plane (top view) at time instants t=2t = 2 s, 4 s, 8 s, 10 s.

Conclusions

The central contribution of the present work is a Volume of Fluid Immersed Boundary (VOF-IB) method, which couples a VOF interface-capturing solver with a non-conforming immersed boundary approach, enabling two-phase flow simulations in arbitrarily complex geometries avoiding the need of body-fitted meshes. This combination was previously unavailable in OpenFOAM-10 [82] and it provides a consistent treatment of moving contact lines and triple-point dynamics at immersed boundaries.

The critical bottleneck of standard VOF solvers in the presence of large viscosity contrasts, leading to increasingly severe time-step stability constraints arising from the explicit treatment of the transposed velocity gradient term, is overcome by introducing a fully implicit treatment of the viscous diffusion term, implementing a block-coupled approach [11] in the VOF-IB framework. Systematic comparisons on an injection molding benchmark demonstrate that this approach reduces computational times from hours to minutes while simultaneously improving solution accuracy and stability.

Subsequently, free-surface simulations of partially filled continuous mixing devices are performed, concerning single-screw and twin screw extruder geometries, capturing fill ratios and average pressure trends in agreement with physical expectations.

Figure 26. (a) Two-phase front evolution. (b) Average pressure and filling ratio over axial sections. Grey dotted lines represent start and ending of kneading modules. Case of inlet flow rate Qin<Q0Q_{\mathrm{in}} < Q_0: temporal evolution of the two-phase interface (left) and plots of average pressure and filling ratio along axial sections (right) is reported for times t=2t = 2 s, 44 s, 66 s, 88 s, 1010 s.

The methods developed here lay a rigorous foundation for future extensions, such as the inclusion of the energy equation to simulate non-isothermal flows, which are known to play a significant role in polymer mixing through viscous heating and temperature-dependent viscosity laws [51, 80, 49, 2]. Another natural extension concerns the introduction of viscoelastic constitutive models [6, 27, 33, 72, 13], enabling a more faithful rheological description beyond the generalized Newtonian assumption. Further developments may also involve compressible air-phase formulations [31, 96, 9] to capture entrapped bubble dynamics under realistic processing conditions. Finally, further systematic validation against literature benchmarks [19, 21, 22] and industrial experimental data remains essential to consolidate the reliability and predictive capability of the proposed framework in real-world applications.

Acknowledgments

This research is part of the activities of Dipartimento di Eccellenza 2023-2027. EC, NP, MV are members of the INdAM group GNCS “Gruppo Nazionale per il Calcolo Scientifico” (National Group for Scientific Computing). The authors acknowledge financial support from the Ministerial Decree no. 352 of 9th April 2022, based on the NRRP - funded by the European Union - NextGenerationEU - Mission 4 ”Education and Research”, Component 2 ”From Research to Business”, Investment 3.3, and by the company Pirelli Tyre S.p.A.

References

  1. [1]H. J. Aguerre, C. I. Pairetti, C. M. Venier, S. Márquez Damián, and N. M. Nigro. An oscillation-free flow solver based on flux reconstruction. Journal of Computational Physics, 365:135–148, 2018.
  2. [2]I. Ahmed and A. J. Chandy. 3D numerical investigations of the effect of fill factor on dispersive and distributive mixing of rubber under non-isothermal conditions. Polymer Engineering & Science, 59(3):535–546, 2019.
  3. [3]D. M. Anderson, G. B. McFadden, and A. A. Wheeler. Diffuse-interface methods in fluid mechanics. Annual Review of Fluid Mechanics, 30:139–165, 1998.
  4. [4]S. Babnik, V. Erklavec Zajec, B. Oblak, B. Likozar, and A. Pohar. A review of computational fluid dynamics (CFD) simulations of mixing in the pharmaceutical industry. Biomedical journal of scientific & technical research, 27(3):20732–20736, 2020.
  5. [5]E. Berberović. Numerical Investigation of Free Surface Flows including Interface Heat and Mass Transfer and Contact Line Phenomena. PhD thesis, TU Darmstadt, 2010.
  6. [6]A. Bonito, M. Picasso, and M. Laso. Numerical simulation of 3D viscoelastic flows with free surfaces. Journal of Computational Physics, 215(2):691–716, 2006.DOI
  7. [7]J. Brackbill, D. Kothe, and C. Zemach. A continuum method for modeling surface tension. Journal of Computational Physics, 100, 1992.
  8. [8]A. Caboussat, V. Maronnier, M. Picasso, and J. Rappaz. Numerical simulation of three dimensional free surface flows with bubbles. In Challenges in Scientific Computing-CISC 2002: Proceedings of the Conference Challenges in Scientific Computing Berlin, October 2–5, 2002, pages 69–86. Springer, 2003.
  9. [9]A. Caboussat, M. Picasso, and J. Rappaz. Numerical simulation of free surface incompressible liquid flows surrounded by compressible gas. Journal of Computational Physics, 203(2):626–649, 2005.
  10. [10]A. Calderer, S. Kang, and F. Sotiropoulos. Level set immersed boundary method for coupled simulation of air/water interaction with complex floating structures. Journal of Computational Physics, 277:201–227, 2014.
  11. [11]P. Cardiff, Ž. Tuković, H. Jasak, and A. Ivanković. A block-coupled finite volume methodology for linear elasticity and unstructured meshes. Computers & structures, 175:100–122, 2016.
  12. [12]R. P. Chhabra. Non–Newtonian fluids: an introduction. In Rheology of complex fluids, pages 3–34. Springer, 2010.
  13. [13]R. Comminai, F. Pimenta, J. Spangenberg, and M. A. Alves. Experimental and numerical investigation of the die swell in 3D printing processes. Micromachines, 14(2):329, 2023.
  14. [14]R. Courant, K. Friedrichs, and H. Lewy. Über die partiellen differenzengleichungen der mathematischen physik. Mathematische annalen, 100(1):32–74, 1928.
  15. [15]D. A. de Miranda, W. K. Rauber, M. Vaz Jr, M. V. C. Alves, F. H. Lafratta, A. L. Nogueira, and P. S. B. Zdanski. Analysis of numerical modeling strategies to improve the accuracy of polymer injection molding simulations. Journal of Non-Newtonian Fluid Mechanics, 315:105033, 2023.
  16. [16]S. S. Deshpande, L. Anumulu, and M. F. Trujillo. Evaluating the performance of the two-phase flow solver interfoam. Computational Science & Discovery, 5(014016), 2012.
  17. [17]D. A. Di Pietro, S. L. Forte, and N. Parolini. Mass preserving finite element implementations of the level set method. Applied Numerical Mathematics, 56(9):1179–1195, 2006.
  18. [18]M. S. Dodd and A. Ferrante. A fast pressure-correction method for incompressible two-fluid flows. Journal of Computational Physics, 273:416–434, 2014.
  19. [19]T. Dong et al. Simulation of flow and mixing for highly viscous fluid in a twin screw extruder with a conveying element using parallelized smoothed particle hydrodynamics. Chemical Engineering Science, 212:115311, 2020.
  20. [20]T. Dong, H. Liu, S. Jiang, L. Gu, Q. Xiao, Z. Yu, and X. Liu. Simulation of free surface flow with a revolving moving boundary for screw extrusion using smoothed particle hydrodynamics. CMES-Computer Modeling in Engineering & Sciences, 95(5):339–360, 2013.
  21. [21]T. Dong and J. Wu. Simulation of non-Newtonian flows in a partially filled twin-screw extruder by smoothed particle hydrodynamics. Polymer Engineering & Science, 62(3):802–814, 2022.
  22. [22]T. Dong, J. Wu, Y. Ruan, et al. Simulation of multiphase flow and mixing in a conveying element of a co-rotating twin-screw extruder using SPH. International Journal of Chemical Engineering, 2023:8383763, 2023.
  23. [23]C. Dopazo. On conditioned averages for intermittent turbulent flows. Journal of Fluid Mechanics, 81(3):433–438, 1977.
  24. [24]K. Edwards. A designers’ guide to engineering polymer technology. Materials & Design, 19(1):57–67, 1998.
  25. [25]M. El Ouafa, S. Vincent, and V. Le Chenadec. Monolithic solvers for incompressible two-phase flows at large density and viscosity ratios. Fluids, 6(1):23, 2021.
  26. [26]S. Elgeti and H. Sauerland. Deforming fluid domains within the finite element method: five mesh-based tracking methods in comparison. Archives of Computational Methods in Engineering, 23(2):323–361, 2016.
  27. [27]J. L. Favero, A. R. Secchi, N. S. M. Cardozo, and H. Jasak. Viscoelastic fluid analysis in internal and in free surface flows using the software OpenFOAM. Computers & Chemical Engineering, 34(12):1984–1993, 2010.
  28. [28]J. H. Ferziger, M. Perić, and R. L. Street. Computational Methods for Fluid Dynamics. Springer, 4 edition, 2020.
  29. [29]C. Galusinski and P. Vigneaux. On stability condition for bifluid flows with surface tension: Application to microfluidics. Journal of Computational Physics, 227(12):6140–6164, 2008.
  30. [30]M. Garcia-Villalba, T. Colonius, O. Desjardins, D. Lucas, A. Mani, D. Marchisio, O. K. Matar, F. Picano, and S. Zaleski. Numerical methods for multiphase flows. International Journal of Multiphase Flow, 191:105285, 2025.
  31. [31]A. Georgoulas, P. Koukouvinis, M. Gavaises, and M. Marengo. Numerical treatment of the interface in two-phase flows using a compressible framework in OpenFOAM: demonstration on a high velocity droplet impact case. Fluids, 6(2):78, 2021.
  32. [32]E. Gutiérrez, F. Favre, N. Balcázar, A. Amani, and J. Rigola. Numerical approach to study bubbles and drops evolving through complex geometries by using a level set–Moving mesh–Immersed boundary method. Chemical Engineering Journal, 349:662–682, 2018.
  33. [33]F. Habla, H. Marschall, O. Hinrichsen, L. Dietsche, H. Jasak, and J. L. Favero. Numerical simulation of viscoelastic two-phase flows using OpenFOAM®. Chemical Engineering Science, 66(22):5487–5496, 2011.
  34. [34]W. H. Herschel and R. Bullkley. Konsistenzmessungen von gummi-benzollösungen. Kolloid-Zeitschrift, 39:291–300, 1926.
  35. [35]D. P. Hill. The computer simulation of dispersed two-phase flow. PhD thesis, Citeseer, 1998.
  36. [36]C. Hirt, A. Amsden, and J. Cook. An arbitrary Lagrangian-Eulerian computing method for all flow speeds. Journal of Computational Physics, 14(3):227–253, 1974.
  37. [37]C. W. Hirt and B. D. Nichols. Volume of fluid (VOF) method for the dynamics of free boundaries. Journal of Computational Physics, 39(1):202–225, 1981.
  38. [38]P. Hold. Mixing of polymers— an overview part i. Advances in Polymer Technology, 2(2):141–151, 1982.
  39. [39]P. Horgue, M. Prat, and M. Quintard. A penalization technique applied to the Volume-Of-Fluid method: Wettability condition on immersed boundaries. Computers & Fluids, 100:255–266, 2014.
  40. [40]T. Ikeno and T. Kajishima. Finite-difference immersed boundary method consistent with wall conditions for incompressible turbulent flow simulations. Journal of Computational Physics, 226(2):1485–1508, Oct 2007.
  41. [41]R. I. Issa. Solution of the implicitly discretised fluid flow equations by operator-splitting. Journal of computational physics, 62(1):40–65, 1986.
  42. [42]D. Jacqmin. Calculation of two-phase Navier-Stokes flows using phase-field modeling. Journal of Computational Physics, 155(1):96–127, 1999.
  43. [43]H. Jasak. Error Analysis and Estimation for the Finite Volume Method with Applications to Fluid Flows. PhD thesis, Imperial College of Science, Technology and Medicine, University of London, London, United Kingdom, 1996.
  44. [44]H. Jasak, D. Rigler, and Ž. Tuković. Design and implementation of immersed boundary method with discrete forcing approach for boundary conditions. In 6th European Conference on Computational Fluid Dynamics, ECFD 2014, pages 5319–5332. International Center for Numerical Methods in Engineering (CIMNE), 2014.
  45. [45]N.-S. Kim, H.-B. Kim, and J.-W. Lee. Numerical analysis of internal flow and mixing performance in polymer extruder I: single screw element. Korea-Australia rheology journal, 18(3):143–151, 2006.
  46. [46]T. M. Kousemaker, P. D. Druetta, F. Picchioni, and A. I. Vakis. Modelling supercritical CO₂ flow in a co-rotating twin screw extruder using the level-set method. Chemical Engineering Research and Design, 205:569–577, 2024.
  47. [47]C.-Y. Ku. A novel method for solving ill-conditioned systems of linear equations with extreme physical property contrasts. CMES, 96(6):409–434, 2013.
  48. [48]M. R. Larsen, E. T. Holmen Olofsson, J. H. Hattel, and J. Spangenberg. Numerical investigation of continuous mixing of highly viscous, non-Newtonian polymer suspensions by a starved-fed single-screw extruder. In AIP Conference Proceedings, volume 3158, page 070004. AIP Publishing, 2024.
  49. [49]M. R. Larsen, T. Ottsen, E. T. Holmen Olofsson, and J. Spangenberg. Numerical modeling of the mixing of highly viscous polymer suspensions in partially filled sigma blade mixers. Polymers, 15(8):1938, 2023.
  50. [50]L.-p. Ma, J.-j. Lian, and D.-m. Liu. A GPU-accelerated two-phase flow model for fluid-solid interaction using the sharp interface immersed boundary method. Journal of Hydrodynamics, 36:883–897, 2024.
  51. [51]C. W. Macosko. Rheology: Principles, Measurements and Applications. Wiley, 1994.
  52. [52]V. Maronnier, M. Picasso, and J. Rappaz. Numerical simulation of free surface flows. Journal of Computational Physics, 155(2):439–455, 1999.
  53. [53]J. A. D. Marquez, Y. Quan, X. Zhu, H.-J. Sue, and Q. Wang. Predicting the processability of polymers in a twin-screw extruder: CFD simulation and experimental validation. Industrial & Engineering Chemistry Research, 63(22):9823–9832, 2024.
  54. [54]H. Marschall. Towards the numerical simulation of multi-scale two-phase flows. PhD thesis, Technische Universität München, 2011.
  55. [55]T. Matzerath and C. Bonten. Automated optimization of mixing elements for single-screw extrusion using CFD simulations. Polymers, 17(4):438, 2025.
  56. [56]S. Middleman. Fundamentals of Polymer Processing. McGraw-Hill, New York, 1977.
  57. [57]S. Mirjalili, S. Jain, and M. Dodd. Interface-capturing methods for two-phase flows: An overview and recent developments. Center for Turbulence Research - Annual research brief, pages 117–135, 12 2017.
  58. [58]R. Mittal and G. Iaccarino. Immersed boundary methods. Annual Review of Fluid Mechanics, 37:239–261, 2005.
  59. [59]F. Moukalled, L. Mangani, and M. Darwish. The Finite Volume Method in Computational Fluid Dynamics: An Advanced Introduction with OpenFOAM® and Matlab, volume 113 of Fluid Mechanics and Its Applications. Springer International Publishing, Cham, Switzerland, 2016.
  60. [60]G. Negrini. Non-conforming methods for the simulation of industrial polymer mixing processes. PhD thesis, Politecnico di Milano, 2023.
  61. [61]G. Negrini, N. Parolini, and M. Verani. On the convergence of the Rhie–Chow stabilized box method for the Stokes problem. International Journal for Numerical Methods in Fluids, 96(8):1489–1516, 2024.
  62. [62]G. Negrini, N. Parolini, and M. Verani. An immersed boundary method for polymeric continuous mixing. In Emerging Technologies in Computational Sciences for Industry, Sustainability and Innovation, pages 163–178. Springer Nature Switzerland, 2025.
  63. [63]E. H. Olofsson, M. Roland, J. Spangenberg, N. H. Jokil, and J. H. Hattel. A CFD model with free surface tracking: predicting fill level and residence time in a starve-fed single-screw extruder. The International Journal of Advanced Manufacturing Technology, 126(7):3579–3591, 2023.
  64. [64]E. Olsson and G. Kreiss. A conservative level set method for two phase flow. Journal of computational physics, 210(1):225–246, 2005.
  65. [65]E. Olsson, G. Kreiss, and S. Zahedi. A conservative level set method for two phase flow ii. Journal of Computational Physics, 225(1):785–807, 2007.
  66. [66]S. Osher and J. A. Sethian. Fronts propagating with curvature-dependent speed: algorithms based on Hamilton–Jacobi formulations. Journal of Computational Physics, 79(1):12–49, 1988.
  67. [67]W. Ostwald. Concerning the function rate of the viscosity of dispersion systems. Kolloid-Zeitschrift, 36:99–117, 1925.
  68. [68]N. Parolini and E. Burman. A finite element level set method for viscous free-surface flows. In Applied and industrial mathematics in Italy, pages 416–427. World Scientific, 2005.
  69. [69]S. V. Patankar. Numerical Heat Transfer and Fluid Flow. Series in Computational Methods in Mechanics and Thermal Sciences. Hemisphere Publishing Corporation, 1980.
  70. [70]S. V. Patankar and D. B. Spalding. A calculation procedure for heat, mass and momentum transfer in three-dimensional parabolic flows. International Journal of Heat and Mass Transfer, 15(10):1787–1806, 1972.
  71. [71]H. V. Patel et al. A coupled Volume of Fluid and Immersed Boundary Method for simulating 3D multiphase flows with contact line dynamics in complex geometries. Chemical Engineering Science, 166:28–41, 2017.
  72. [72]F. Pimenta and M. A. Alves. Stabilization of an open-source finite-volume solver for viscoelastic fluid flows. Journal of Non-Newtonian Fluid Mechanics, 239:85–104, 2017.
  73. [73]H. Poudyal, I. Ahmed, and A. J. Chandy. Three-dimensional, non-isothermal simulations of the effect of speed ratio in partially-filled rubber mixing. International Polymer Processing, 34(2):219–230, 2019.
  74. [74]S. Rajendran, R. M. Manglik, and M. A. Jog. New property averaging scheme for volume of fluid method for two-phase flows with large viscosity ratios. Journal of Fluids Engineering, 144(6):061101, 2022.
  75. [75]C. Rauwendaal. Polymer extrusion. Carl Hanser Verlag GmbH Co KG, 2014.
  76. [76]H. Rusche. Computational Fluid Dynamics of Dispersed Two-Phase Flows at High Phase Fractions. PhD thesis, Imperial College of Science, Technology and Medicine, 01 2002.
  77. [77]R. Scardovelli and S. Zaleski. Direct numerical simulation of free-surface and interfacial flow. Annu. Rev. Fluid Mech., 31, 01 1999.
  78. [78]X. Sun and M. Sakai. Numerical simulation of two-phase flows in complex geometries by using the volume-of-fluid/immersed-boundary method. Chemical Engineering Science, 139:221–240, 2016.
  79. [79]M. Sussman, P. Smereka, and S. Osher. A level set approach for computing solutions to incompressible two-phase flow. Journal of Computational Physics, 114(1):146–159, 1994.
  80. [80]Z. Tadmor and C. G. Gogos. Principles of polymer processing. John Wiley & Sons, 2013.
  81. [81]M. Tamellini, N. Parolini, and M. Verani. An optimal control problem for two-phase compressible–incompressible flows. Computers & Fluids, 172:538–548, 2018.
  82. [82]The OpenFOAM Foundation. OpenFOAM, Free CFD Software. https://openfoam.org/, 2023.
  83. [83]T. Tolle, D. Bothe, and T. Marić. Saample: A segregated accuracy-driven algorithm for multiphase pressure-linked equations. Computers & Fluids, 200:104450, 2020.
  84. [84]G. Tryggvason, B. Bunner, A. Esmaeeli, D. Juric, N. Al-Rawahi, W. Tauber, J. Han, S. Nas, and Y.-J. Jan. A front-tracking method for the computations of multiphase flow. Journal of Computational Physics, 169(2):708–759, 2001.
  85. [85]R. Urraca, A. Pernía-Espinoza, I. Díaz, and A. Sanz-Garcia. Practical methodology for validating constitutive models for the simulation of rubber compounds in extrusion processes. International Journal of Advanced Manufacturing Technology, 90(5-8):2377–2387, 2017.
  86. [86]S. Vakilpour, M. Mohammadi, and S. Ormiston. A fully coupled ALE interface tracking method for a pressure-based finite volume solver. Journal of Computational Physics, 427:110054, 2021.
  87. [87]B. Van Leer. Towards the ultimate conservative difference scheme. II. Monotonicity and conservation combined in a second-order scheme. Journal of Computational Physics, 14(4):361–370, 1974.
  88. [88]H. Wang et al. Simulation method for rubber compounding under isothermal partial filling conditions. Advances in Polymer Technology, 2023:6062543, 2023.
  89. [89]Z. Wang, J. Yang, and F. Stern. A new volume-of-fluid method with a constructed distance function on general structured grids. Journal of Computational Physics, 231:3703–3722, 2012.
  90. [90]H. Weller. Non-orthogonal version of the arbitrary polygonal c-grid and a new diamond grid. Geoscientific Model Development, 7(3):779–797, 2014.
  91. [91]H. Weller and A. Shahrokhi. Curl-free pressure gradients over orography in a solution of the fully compressible euler equations with implicit treatment of acoustic and gravity waves. Monthly Weather Review, 142(12):4439–4457, 2014.
  92. [92]H. G. Weller. The development of a new flame area combustion model using conditional averaging. In Technical report, Thermo-Fluids Section Report TF 9307. Imperial College of Science, Technology and Medicine, 1993.
  93. [93]H. G. Weller. A new approach to VOF-based interface capturing methods for incompressible and compressible flow. In Technical report. OpenCFD Limited, 1993.
  94. [94]H. G. Weller, G. Tabor, H. Jasak, and C. Fureby. A tensorial approach to computational continuum mechanics using object-oriented techniques. Computers in Physics, 12(6):620–631, 1998.
  95. [95]M. Wörner. Numerical modeling of multiphase flows in microfluidics and micro process engineering: a review of methods and applications. Microfluidics and nanofluidics, 12(6):841–886, 2012.
  96. [96]K. Yang and T. Aoki. A momentum-conserving weakly compressible Navier–Stokes solver for simulation of violent two-phase flows with high density ratio. International Journal of Computational Fluid Dynamics, 37(9):776–796, 2023.
  97. [97]H. Yu, P. Dong, B. Xu, W. Xu, J. Huang, L. Tan, Q. Wang, and L.-S. Tung. Simulation of flow and mixing of partially-filled fluid in an asymmetric twin-screw extruder with an internal baffle using moving-particle semi-implicit method. Chemical Engineering Science, 318:122198, 2025.
  98. [98]S. Zhang, X. Zhao, and S. Bayyuk. Generalized formulations for the Rhie–Chow interpolation. Journal of Computational Physics, 258:880–914, 2014.
  99. [99]X. Zheng and G. Karniadakis. A phase-field/ALE method for simulating fluid-structure interactions in two-phase flow. Computer Methods in Applied Mechanics and Engineering, 309:19–40, 2016.

Paper details

vlnc:W_czu49bey
Version
v1 · latest
Published
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