betaFeedback

WorksPhysics and Astronomy

Simplified models for unsteady three-dimensional flows in slowly varying microchannels

Abstract

We present a reduced order model for three dimensional unsteady pressure-driven flows in micro-channels of variable cross-section. This fast and accurate model is valid for long channels, but allows for large variations in the channel's cross-section along the axis. It is based on an asymptotic expansion of the governing equations in the aspect ratio of the channel. A finite Fourier transform in the plane normal to the flow direction is used to solve for the leading order axial velocity. The corresponding pressure and transverse velocity are obtained via a hybrid analytic-numerical scheme based on recursion. The channel geometry is such that one of the transverse velocity components is negligible, and the other component, in the plane of variation of channel height, is obtained from combination of the corresponding momentum equation and the continuity equations, assuming a low degree polynomial Ansatz of the pressure forcing. A key feature of the model is that it puts no restriction on the time dependence of the pressure forcing, in terms of shape and frequency, as long as the advective component of the inertia term is small. This is a major departure from many previous expositions which assume harmonic forcing. The model reveals to be accurate for a wide range of parameters and is two orders of magnitude faster than conventional three dimensional CFD simulations.

Introduction

Microfluidic devices allow the operation of fluids inside channels whose typical dimensions are on the order of microns. Since the micro scale matches the scale of some biological structures, one of the main advantages of these devices is that they make it possible to have multiple biological/chemical functions (a lab) in a small area (a chip). Applications that use these devices include the transport of biological materials such as proteins, DNA, cells, or transport of chemical samples and analytes. Other applications include sorting, mixing, drug delivery or mass spectroscopy. For a more detailed exposition of these applications, the reader is referred to the book of [1].

Of particular importance to these applications is the ability to accurately, yet efficiently predict unsteady flows in in three dimensional channels. Whenever we need to optimize the design of a microfluidic device, we need to swipe many dimensions and forcing signal parameters; this problem becomes prohibitively expensive to solve; specially if one resorts uniquely to using numerical solutions of the full system of the governing P.D.E. (typically CFD tools). The contribution of this work is to come up with an unsteady three dimensional reduced order model that is fast, yet accurate, to perform these simulations.

The alternative to fully simulate the governing equations numerically [2] is to resort to reduced order models. These can be based on simplifying the governing equations based on scaling. This is where our work fits.

Our model is based on approximating the Navier-Stokes equations using a regular perturbation expansion in the aspect ratio of the channel. It turns out that the resulting systems of equations can be partially solved for the leading order axial velocity using projections onto a space of appropriate eigenfunctions. The use of a recursive method to fully solve for the time dependence of the pressure distribution, and the fact that these expressions need only a few number of modes to converge render these models much faster than CFD. To solve for the transverse velocity components, vv and ww, we assume that the channel geometry is such that ww is negligible; The vv component is obtained from the corresponding momentum and continuity equations, assuming a low degree polynomial ansatz of the pressure forcing.

When the geometry is simple and the flow is steady, analytical solutions are possible. For example, this is the case for steady, laminar and fully developed flows in channels of constant cross section with regular geometry, such as rectangular or circular cross sections. For more complex geometries, solutions are often numerical or hybrid. In particular, some works focus on flows in channels of constant cross section with irregular geometry [3]. Flows in narrow and long geometries whose cross section is not constant are frequently investigated using the theory of lubrication. This theory was used to approximate the steady Navier-Stokes and continuity equations in the low Reynolds number regime, for slowly variable geometries [4] and later adapted to more general contexts. There is a vast literature on lubrication based models in the steady and two dimensional case, and those typically focus on obtaining to leading order, mass-flow relations. For a more detailed account on those, see e.g. the introduction in [5, 6]. As for flows in three dimensional rectangular channels, the work in [7] focuses on the steady case. The authors there use lubrication theory to show that if the cross-section is varying in the flow-wise direction, then the flow cannot be planar.

Concerning research on unsteady flows, the assumption is often that the pressure forcing is harmonic; for example this was done in the context of modeling bio-flows (see for e.g. [8, 9]). Before we proceed, we clarify here that what we mean by unsteady is the unsteadiness of the flow manifested in the inertia term, i.e., the time varying component of inertia, which we assume is non negligible. For instance the authors of [9] analyzed a pressure driven oscillatory flow in a circular channel of slowly varying cross-section, allowing small departure from symmetry. Their solution is developed for a harmonically oscillatory forcing with the oscillation frequency being too small (ReSt≪1\mathrm{ReSt}\ll1) or very large (ReSt≫1\mathrm{ReSt}\gg1).

In the context of lubrication theory, and up to our knowledge, this would be the first work to focus on three dimensional and unsteady flows in variable channels, with non-harmonic pressure forcing. The model we are proposing falls within the scope of the work done in [5, 6, 10, 11, 12]. In particular, it is an extension of the unsteady model developed for two dimensional rectangular channels in [5] to three dimensions, while relaxing the nearly parallel channel assumption. A key contribution of this work is that the model puts no restriction on the time dependence of the forcing, in terms of shape and frequency. The only condition on the forcing is such that the advective component of the inertia term is small. This is a major departure from these previous expositions that assume harmonic forcing. Another important feature of these models is that they allow us to extract ranges of validity in terms of the dependence of the error distributions of velocities and pressure on the parameters of interest.

This paper is organized as follows: in sections 2.1 and 2.2, we present the governing equations and the boundary conditions with the relevant scaling, and then, in section 2.3, we obtain the leading order equations from them. Next, in section 2.4.1, we present the semi-analytical solutions for uu and pp to these equations using a finite Fourier transform in the plane normal to the flow direction. Our approach to solving for the leading lateral and transverse velocities is presented in section 2.4.2. In the numerical results section 3, we show dependence of the error on key parameters for selected cases, in addition to comparisons of the pressure profiles and velocity contours obtained from our model, with those obtained from ANSYS. In section section:Conclusion, we conclude and propose future directions.

Model of an unsteady flow in three dimensional shallow channels

As illustrated in Figure 1, we consider a pressure driven flow along the x^\hat{x} direction, in a three dimensional micro-channel with constant height z=2H0z=2H_0, and a slowly varying cross-section y=D(x)y=D(x). The typical axial length scale over which the variations in the cross-section occur define the length scale LxL_x. We will assume in this work that the aspect ratio defined as α=H0Lx\alpha=\frac{H_0}{L_x} is small, i.e. α≪1\alpha\ll1. This will allow us later to formulate a solution based on an asymptotic expansion of our fields in the parameter α\alpha.

Schematic of channels Type I and Type II

Figure 1. Left: channel Type I. Surfaces y=D(x)y=D(x) and z=H0z=H_0 are solid impermeable walls. Planes y=0y=0 and z=0z=0 are planes of symmetry. Aspect ratio A=D0H0A=\frac{D_0}{H_0}. Right: channel Type II. Surfaces y=0y=0, y=D(x)y=D(x) and z=H0z=H_0 are solid impermeable walls. Plane z=0z=0 is a plane of symmetry. Aspect ratio A=D02H0A=\frac{D_0}{2H_0}.

Governing equations in non-dimensional form

The continuity equation and momentum conservation for an unsteady incompressible flow are

∂xu+∂yv+∂zw=0,\partial_x u+\partial_y v+\partial_z w=0,
ρ(∂tu+u⋅∇u)=−∂xp+μΔu,ρ(∂tv+u⋅∇v)=−∂yp+μΔv,ρ(∂tw+u⋅∇w)=−∂zp+μΔw.(1)\begin{aligned} \rho\left(\partial_t u+\mathbf{u}\cdot\nabla u\right)&=-\partial_xp+\mu\Delta u,\\ \rho\left(\partial_t v+\mathbf{u}\cdot\nabla v\right)&=-\partial_yp+\mu\Delta v,\\ \rho\left(\partial_t w+\mathbf{u}\cdot\nabla w\right)&=-\partial_zp+\mu\Delta w. \tag*{(1)} \end{aligned}

The gravitational body forces are neglected in the applications of micro-flows with dominant pressure variation p(x,y,z,t)p(x,y,z,t). The fluid is considered incompressible with a constant density ρ\rho, a velocity vector u=(u,v,w)\mathbf{u}=(u,v,w), and constant fluid dynamic viscosity μ\mu.

The equations in 1 are nondimensionalized in order to show the different characteristic parameters of the flow, which allows the proper scaling of the physical terms based on the geometric and physical assumptions of the problem. Hence, the following dimensionless parameters are used, with a star coupled to each

x∗=xLx, z∗=zH0, y∗=yD0, t∗=tf,x^*=\frac{x}{L_x},\ z^*=\frac{z}{H_0},\ y^*=\frac{y}{D_0},\ t^*=tf,
u∗=uU, v∗=vαU, w∗=wα2U, p∗=pΔP0u^*=\frac{u}{U},\ v^*=\frac{v}{\alpha U},\ w^*=\frac{w}{\alpha^2U},\ p^*=\frac{p}{\Delta P_0}

Since the flow in the channel is pressure-driven, the pressure force in x^\hat{x}-direction per unit volume can be scaled as ΔP0Lx\frac{\Delta P_0}{L_x}, where ΔP0\Delta P_0 is the pressure drop corresponding to a Poiseuille flow. The following Reynolds, Strouhal, and Poiseuille dimensionless numbers are also introduced into the governing equations

Re⁡≡ρUH0μ,St⁡≡fH0U,Po⁡≡ΔP0/LxμU/H02.\operatorname{Re} \equiv\frac{\rho U H_0}{\mu},\quad\operatorname{St} \equiv\frac{f H_0}{U},\quad\operatorname{Po} \equiv\frac{\Delta P_0/L_x}{\mu U/H_0^2}.

The continuity and momentum equations in 1 become in dimensionless form after dropping the stars and defining A=D0/H0A=D_0/H_0

∂xu+1A∂yv+∂zw=0,Re⁡(St⁡∂tu+α(u∂xu+vA∂yu+w∂zu))=−Po⁡∂xp+α2∂x2u+1A2∂y2u+∂z2u,Re⁡α2(St⁡∂tv+α(u∂xv+vA∂yv+w∂zv))=−Po⁡∂yp+α2(α2∂x2v+1A2∂y2v+∂z2v),Re⁡α3(St⁡∂tw+α(u∂xw+vA∂yw+w∂zw))=−Po⁡∂zp+α2(α2∂x2w+1A2∂y2w+∂z2w).(2)\begin{aligned} &\partial_x u+\frac{1}{A}\partial_y v+\partial_z w=0,\\ &\operatorname{Re}\left(\operatorname{St}\partial_tu+\alpha\left(u\partial_xu+\frac{v}{A}\partial_yu+w\partial_zu\right)\right)\\ &\qquad=-\operatorname{Po}\partial_xp+\alpha^2\partial_x^2u+\frac{1}{A^2}\partial_y^2u+\partial_z^2u,\\ &\operatorname{Re}\alpha^2\left(\operatorname{St}\partial_tv+\alpha\left(u\partial_xv+\frac{v}{A}\partial_yv+w\partial_zv\right)\right)\\ &\qquad=-\operatorname{Po}\partial_yp+\alpha^2\left(\alpha^2\partial_x^2v+\frac{1}{A^2}\partial_y^2v+\partial_z^2v\right),\\ &\operatorname{Re}\alpha^3\left(\operatorname{St}\partial_tw+\alpha\left(u\partial_xw+\frac{v}{A}\partial_yw+w\partial_zw\right)\right)\\ &\qquad=-\operatorname{Po}\partial_zp+\alpha^2\left(\alpha^2\partial_x^2w+\frac{1}{A^2}\partial_y^2w+\partial_z^2w\right). \tag*{(2)} \end{aligned}

Boundary Conditions

Two types of boundary conditions are considered

  1. Type I (the plane y=0y=0 is a symmetry condition)

∂yu(x,0,z,t)=0=∂yw(x,0,z,t)andv(x,0,z,t)=0,t>0\partial_yu(x,0,z,t)=0=\partial_yw(x,0,z,t)\quad\text{and}\quad v(x,0,z,t)=0,\quad t>0
∂zu(x,y,0,t)=0=∂zv(x,y,0,t)andw(x,y,0,t)=0,t>0\partial_zu(x,y,0,t)=0=\partial_zv(x,y,0,t)\quad\text{and}\quad w(x,y,0,t)=0,\quad t>0
u(x,D(x),z,t)=v(x,D(x),z,t)=w(x,D(x),z,t)=0,t>0u(x,D(x),z,t)=v(x,D(x),z,t)=w(x,D(x),z,t)=0,\quad t>0
u(x,y,1,t)=v(x,y,1,t)=w(x,y,1,t)=0,t>0u(x,y,1,t)=v(x,y,1,t)=w(x,y,1,t)=0,\quad t>0
  1. Type II (the plane y=0y=0 is a wall)

u(x,0,z,t)=w(x,0,z,t)=v(x,0,z,t)=0,t>0u(x,0,z,t)=w(x,0,z,t)=v(x,0,z,t)=0,\quad t>0
∂zu(x,y,0,t)=∂zv(x,y,0,t)=0,andw(x,y,0,t)=0,t>0\partial_zu(x,y,0,t)=\partial_zv(x,y,0,t)=0,\quad\text{and}\quad w(x,y,0,t)=0,\quad t>0

u(x,D(x),z,t)=v(x,D(x),z,t)=w(x,D(x),z,t)=0,t>0u(x,D(x),z,t)=v(x,D(x),z,t)=w(x,D(x),z,t)=0,\quad t>0

u(x,y,1,t)=v(x,y,1,t)=w(x,y,1,t)=0,t>0u(x,y,1,t)=v(x,y,1,t)=w(x,y,1,t)=0,\quad t>0

where D(x)D(x) is the nondimensional channel depth D∗(x)=D(x)/D0D^*(x)=D(x)/D_0 after removing the star, and initial conditions are

u(x,y,z,0)=v(x,y,z,0)=w(x,y,z,0)=0p(x,y,z,0)=pa(3)\begin{aligned} u(x,y,z,0)&=v(x,y,z,0)=w(x,y,z,0)=0\\ p(x,y,z,0)&=p_a \tag*{(3)} \end{aligned}

Leading Order Equations

The relevant fields are expanded asymptotically in the aspect ratio α\alpha which is assumed to be small (α≪1\alpha\ll1)

ϕ=ϕ0+αϕ1+⋯ ,(4)\phi=\phi_0+\alpha\phi_1+\cdots, \tag*{(4)}

where ϕ\phi could be uu, vv, ww or pp. Since the pressure boundary conditions at the entrance and exit of the channel are essential (p0p_0 is known there), the higher-orders of pressure vanish at these two locations. This implies that the initial condition for the zeroth order pressure distribution is p0(x,y,z,0)=pap_0(x,y,z,0)=p_a.

The asymptotic expansion yields the zeroth order equations (∼1\sim1), with velocities inheriting the same boundary conditions at the centerline and walls as in 2.2

∂xu0+1A∂yv0+∂zw0=0,Re⁡St⁡∂tu0=−Po⁡∂xp0+1A2∂y2u0+∂z2u0,0=−Po⁡∂yp0,0=−Po⁡∂zp0.(5)\begin{aligned} \partial_xu_0+\frac{1}{A}\partial_yv_0+\partial_zw_0&=0,\\ \operatorname{Re}\operatorname{St}\partial_tu_0&=-\operatorname{Po}\partial_xp_0+\frac{1}{A^2}\partial_y^2u_0+\partial_z^2u_0,\\ 0&=-\operatorname{Po}\partial_yp_0,\\ 0&=-\operatorname{Po}\partial_zp_0. \tag*{(5)} \end{aligned}

In what follows, and for the sake of simplicity of exposition, we only present the solutions for the boundary conditions of Type I. Type II conditions can be dealt with in a very similar way.

Leading order velocities and pressure for Type I channels

Axial velocity and pressure

The leading order axial velocity is obtained using a finite Fourier transform in the plane orthogonal to the flow direction on the momentum equation

u0(x,y,z,t)=−4PoReSt∑m,n=1∞(−1)m+nνmνncos⁡(νmyD(x))cos⁡(νnz)Im,n(∂p0∂x(x,t)),(6)u_0(x,y,z,t) = -\frac{4\mathrm{Po}}{\mathrm{ReSt}} \sum_{m,n=1}^{\infty} \frac{(-1)^{m+n}}{\nu_m\nu_n} \cos\left(\nu_m\frac{y}{D(x)}\right) \cos(\nu_n z) I_{m,n}\left(\frac{\partial p_0}{\partial x}(x,t)\right), \tag*{(6)}

where νn=π(n−1/2)\nu_n=\pi(n-1/2), n=1,2,⋯n=1,2,\cdots, and where we have introduced the following quantities for compactness

Im,n(f(x,t))=∫t′=0te−κm,n(t−t′)f(x,t′) dt′,I_{m,n}(f(x,t))=\int_{t'=0}^{t}e^{-\kappa_{m,n}(t-t')}f(x,t')\,\mathrm{d}t',
κm,n(x)=νm2A2D2(x)+νn2ReSt.\kappa_{m,n}(x)=\frac{\frac{\nu_m^2}{A^2D^2(x)}+\nu_n^2}{\mathrm{ReSt}}.

This expression depends on the unknown pressure p0p_0, which is obtained implicitly from the following methodology: we first integrate the continuity equation along the cross-sectional area normal to x^\hat{x}, and realize that the volumetric flow rate in the channel is independent of the flow direction, i.e.

Q0(t)=∫−11∫−D(x)D(x)u0 dx dy=−16PoReSt∑m,n=1∞D(x)νm2νn2Im,n(∂xp0(x,t)).(7)Q_0(t)=\int_{-1}^{1}\int_{-D(x)}^{D(x)}u_0\,\mathrm{d}x\,\mathrm{d}y =-\frac{16\mathrm{Po}}{\mathrm{ReSt}} \sum_{m,n=1}^{\infty} \frac{D(x)}{\nu_m^2\nu_n^2} I_{m,n}(\partial_xp_0(x,t)). \tag*{(7)}

Equation (7) is then solved recursively in time with

Q0(t+Δt)=−16PoReSt∑m,n=1∞D(x)νm2νn2Im,n(∂xp0(x,t+Δt))Q_0(t+\Delta t)=-\frac{16\mathrm{Po}}{\mathrm{ReSt}} \sum_{m,n=1}^{\infty} \frac{D(x)}{\nu_m^2\nu_n^2} I_{m,n}(\partial_xp_0(x,t+\Delta t))

where

Im,n(∂xp0(x,t+Δt))=e−κm,nΔt⋅Im,n(∂xp0(x,t))+∫tt+Δte−κm,n(t+Δt−t′)∂xp0(x,t′) dt′.(8)I_{m,n}(\partial_xp_0(x,t+\Delta t)) =e^{-\kappa_{m,n}\Delta t}\cdot I_{m,n}(\partial_xp_0(x,t)) +\int_{t}^{t+\Delta t}e^{-\kappa_{m,n}(t+\Delta t-t')}\partial_xp_0(x,t')\,\mathrm{d}t'. \tag*{(8)}

To resolve the last term in this equation, a quadrature with coefficients a1,2m,na_{1,2}^{m,n} is used, and the integral reduces to

∫tt+Δte−κm,n(t+Δt−t′)∂xp0(x,t′) dt′=a1m,n(x,Δt)∂xp0(x,t)+a2m,n(x,Δt)∂xp0(x,t+Δt).(9)\int_t^{t+\Delta t} e^{-\kappa_{m,n}(t+\Delta t-t')} \partial_x p_0(x,t')\,\mathrm{d}t' = a_1^{m,n}(x,\Delta t)\partial_x p_0(x,t) + a_2^{m,n}(x,\Delta t)\partial_x p_0(x,t+\Delta t). \tag*{(9)}

This yields the flow rate at t+Δtt+\Delta t

Q0(t+Δt)=−16P0ReSt∑m,n=1∞D(x)νm2νn2[e−κm,nΔtIm,n(∂xp0(x,t))+a1m,n(x,Δt)∂xp0(x,t)+a2m,n(x,Δt)∂xp0(x,t+Δt)]=−4C0D(x)(∑m,n=1∞1νm2νn2e−κm,nΔtIm,n(∂xp0(x,t))+∂xp0(x,t)S1(x)+∂xp0(x,t+Δt)S2(x)).(10)\begin{aligned} Q_0(t+\Delta t)={}&-\frac{16P_0}{\mathrm{ReSt}}\sum_{m,n=1}^{\infty}\frac{D(x)}{\nu_m^2\nu_n^2}\left[e^{-\kappa_{m,n}\Delta t}I_{m,n}\left(\partial_xp_0(x,t)\right)\right.\\ &\left.+a_1^{m,n}(x,\Delta t)\partial_xp_0(x,t)+a_2^{m,n}(x,\Delta t)\partial_xp_0(x,t+\Delta t)\right]\\ ={}&-4C_0D(x)\left(\sum_{m,n=1}^{\infty}\frac{1}{\nu_m^2\nu_n^2}e^{-\kappa_{m,n}\Delta t}I_{m,n}\left(\partial_xp_0(x,t)\right)\right.\\ &\left.+\partial_xp_0(x,t)S_1(x)+\partial_xp_0(x,t+\Delta t)S_2(x)\right). \tag*{(10)} \end{aligned}

Here, we have defined the following parameters for compactness

C0=4P0ReStC_0=\frac{4P_0}{\mathrm{ReSt}}
S1(x)=∑m,n=1∞a1m,n(x,Δt)νm2νn2S_1(x)=\sum_{m,n=1}^{\infty}\frac{a_1^{m,n}(x,\Delta t)}{\nu_m^2\nu_n^2}
S2(x)=∑m,n=1∞a2m,n(x,Δt)νm2νn2.S_2(x)=\sum_{m,n=1}^{\infty}\frac{a_2^{m,n}(x,\Delta t)}{\nu_m^2\nu_n^2}.

Upon integrating in xx from inlet to outlet, and after some algebraic manipulations, we obtain the volume flow rate at t+Δtt+\Delta t in terms of the pressure drop ΔP0\Delta P_0 as

Q0(t+Δt)=4C0∫0L/Lx1D(x)S2(x) dx[ΔP0(t+Δt)−∫0L/Lx1S2(x)∑m,n=1∞1νm2νn2e−κm,nΔtIm,n(∂xp0(x,t)) dx−∫0L/LxS1(x)S2(x)∂xp0(x,t) dx](11)Q_0(t+\Delta t)=\frac{4C_0}{\int_0^{L/L_x}\frac{1}{D(x)S_2(x)}\,dx}\left[\Delta P_0(t+\Delta t)-\int_0^{L/L_x}\frac{1}{S_2(x)}\sum_{m,n=1}^{\infty}\frac{1}{\nu_m^2\nu_n^2}e^{-\kappa_{m,n}\Delta t}I_{m,n}\left(\partial_xp_0(x,t)\right)\,dx-\int_0^{L/L_x}\frac{S_1(x)}{S_2(x)}\partial_xp_0(x,t)\,dx\right] \tag*{(11)}

This allows us to use the following relation to update the pressure drop

∂xp0(x,t+Δt)=−14C0D(x)S2(x)Q0(t+Δt)−1S2(x)∑m,n=1∞1νm2νn2e−κm,nΔtIm,n(∂xp0(x,t))−S1(x)S2(x)∂xp0(x,t)(12)\partial_xp_0(x,t+\Delta t)=-\frac{1}{4C_0D(x)S_2(x)}Q_0(t+\Delta t)-\frac{1}{S_2(x)}\sum_{m,n=1}^{\infty}\frac{1}{\nu_m^2\nu_n^2}e^{-\kappa_{m,n}\Delta t}I_{m,n}\left(\partial_xp_0(x,t)\right)-\frac{S_1(x)}{S_2(x)}\partial_xp_0(x,t) \tag*{(12)}

We summarize the procedure that allowed us to obtain the pressure in Algorithm 1.

Algorithm 1 Algorithm for computing p0p_0

  1. precompute:

    • number of modes NN

    • time independent quantities: S1(x)=∑m,n=1Na1m,nνm2νn2S_1(x)=\sum_{m,n=1}^{N}\frac{a_1^{m,n}}{\nu_m^2\nu_n^2} and S2(x)=∑m,n=1Na2m,nνm2νn2S_2(x)=\sum_{m,n=1}^{N}\frac{a_2^{m,n}}{\nu_m^2\nu_n^2}.

  2. initialize ∂p0∂x(x,t=0)=0\frac{\partial p_0}{\partial x}(x,t=0)=0 and Im,n(∂p0∂x(x,t=0))=0I_{m,n}\left(\frac{\partial p_0}{\partial x}(x,t=0)\right)=0

  3. loop over time: t→t+Δtt\to t+\Delta t, compute

    • Compute Q(t+Δt)Q(t+\Delta t) using the recurrence relation (11)

    • Update ∂p0∂x(x,t+Δt)\frac{\partial p_0}{\partial x}(x,t+\Delta t) using the recurrence relation (12)

Transverse and lateral velocities

To leading order, the ww component of velocity is zero according to our scaling. So one could use the conservation of mass to obtain v0v_0 from u0u_0. However, the trouble with this approach is that the solution obtained is not consistent with the boundary condition at the wall. In what follows we present a novel method to obtain a transverse velocity that satisfies the desired boundary conditions.

According to the O(α2)O(\alpha^{2}) equations, the zeroth order equation for vv satisfies

ReSt⁡∂tv0=−Po⁡∂yp2+1A2∂y2v0+∂z2v0,(13)\operatorname{ReSt}\partial_t v_0=-\operatorname{Po}\partial_y p_2+\frac{1}{A^2}\partial_y^2v_0+\partial_z^2v_0, \tag*{(13)}

and is forced by the unknown pressure term ∂yp2\partial_y p_2. Here, instead of running into the closure problem for the pressure term, we assume that it is for now an unknown function denoted by fv(x,y,z,t)f_v(x,y,z,t), thus 13 yields:

ReSt⁡∂tv0=fv(x,y,z,t)+1A2∂y2v0+∂z2v0,(14)\operatorname{ReSt}\partial_t v_0=f_v(x,y,z,t)+\frac{1}{A^2}\partial_y^2v_0+\partial_z^2v_0, \tag*{(14)}

subject to the boundary conditions presented in § 2.2. For the sake of compactness, we present this solution for Type I channel only, but a similar procedure can be reached for Type II channels.

To proceed with a solution methodology similar to that used for u0u_0 in 2.4.1, we assume a polynomial expansion of fv(x,y,z,t)f_v(x,y,z,t) in the plane orthogonal to the flow direction with coefficients dependent on time and flow direction

fv(x,y,z,t)≃∑i=0Ni∑j=0NjCij(x,t)ηizj,(15)f_v(x,y,z,t)\simeq\sum_{i=0}^{N_i}\sum_{j=0}^{N_j}C_{ij}(x,t)\eta^i z^j, \tag*{(15)}

where η=yAD(x)\eta=\frac{y}{AD(x)}.

The solution to (14) is then given by

v0(x,y,z,t)=4ReSt⁡∑m,n=1∞sin⁡(βmη)cos⁡(νnz)⋅∫0te−km,n(v)(t−t′)∫01∫01sin⁡(βmη′)cos⁡(νnz′)fv(x,η′,z′,t′) dη′dz′dt′,(16)v_0(x,y,z,t)=\frac{4}{\operatorname{ReSt}}\sum_{m,n=1}^{\infty}\sin(\beta_m\eta)\cos(\nu_n z)\cdot \int_0^t e^{-k_{m,n}^{(v)}(t-t')}\int_0^1\int_0^1\sin(\beta_m\eta')\cos(\nu_n z')f_v(x,\eta',z',t')\,\mathrm{d}\eta'\mathrm{d}z'\mathrm{d}t', \tag*{(16)}

where the following parameters are introduced

βm=mπ\beta_m=m\pi
km,n(v)=(βmAD(x))2+(νn)2ReSt⁡.k_{m,n}^{(v)}=\frac{\left(\frac{\beta_m}{AD(x)}\right)^2+(\nu_n)^2}{\operatorname{ReSt}}.

Substituting the form (15) in the expressions for v0v_0 allows to carry out the integrals in yy and zz

v0(x,y,z,t)=4ReSt∑m,n=1∞sin⁡(βmη)cos⁡(νnz)⋅∑i,jIm,n(v)(Cij(x,t))∫01∫01sin⁡(βmη′)cos⁡(νnz′)η′iz′j dη′dz′(17)v_0(x,y,z,t)=\frac{4}{\mathrm{ReSt}}\sum_{m,n=1}^{\infty}\sin(\beta_m\eta)\cos(\nu_n z)\cdot \sum_{i,j} I_{m,n}^{(v)}(C_{ij}(x,t))\int_0^1\int_0^1\sin(\beta_m\eta')\cos(\nu_n z'){\eta'}^i{z'}^j\,\mathrm{d}\eta'\mathrm{d}z' \tag*{(17)}

where

Im,n(v)(Cij(x,t))=∫0te−km,n(v)(t−t′)Cij(x,t′)dt′(18)I_{m,n}^{(v)}(C_{ij}(x,t))=\int_0^t e^{-k_{m,n}^{(v)}(t-t')}C_{ij}(x,t')\mathrm{d}t' \tag*{(18)}

From this, we obtain

v(x,y,z,t)=4ReSt∑m,n=1∞sin⁡(βmη)cos⁡(νnz)∑i,jIm,n(v)(Cij(x,t))MS(i,2m)MC(j,2n−1)(19)\begin{aligned} v(x,y,z,t)&=\\ &\frac{4}{\mathrm{ReSt}}\sum_{m,n=1}^{\infty}\sin(\beta_m\eta)\cos(\nu_n z)\sum_{i,j}I_{m,n}^{(v)}(C_{ij}(x,t))M_S(i,2m)M_C(j,2n-1) \tag*{(19)} \end{aligned}

where

MC(j,n)=∫01zjcos⁡(nπz2)dz(20)M_C(j,n)=\int_0^1 z^j\cos\left(\frac{n\pi z}{2}\right)\mathrm{d}z \tag*{(20)}
MS(i,m)=∫01yisin⁡(mπy2)dy(21)M_S(i,m)=\int_0^1 y^i\sin\left(\frac{m\pi y}{2}\right)\mathrm{d}y \tag*{(21)}

We now use the continuity equations to solve recursively for the unknown coefficients Ci,j(x,t)C_{i,j}(x,t) as we did for the pressure term in u0u_0. In particular, for Im,n(v)(Cij(x,t))I_{m,n}^{(v)}(C_{ij}(x,t)), we use similar expressions to that in equations (8) and (10), we get that the coefficients CijC_{ij} are updated in time according to

∑i,jCij(x,t+Δt)∑m,n=1∞bm,n1(v)(x,Δt)βmD(x)cos⁡(βmη)cos⁡(νnz)MS(i,2m)MC(j,2n−1)=Po⁡∑m,n=1∞(−1)m+nνnD′(x)D(x)ηsin⁡(νm1η)cos⁡(νnz)Im1,n1(u)(∂xp0(x,t+Δt))+Po⁡∑m,n=1∞(−1)m+nνm1νncos⁡(νm1η)cos⁡(νnz)∂∂x{rm1,n1(u)(∂xp0(x,t+Δt))}−∑m,n=1∞βmD(x)cos⁡(βmη)cos⁡(νnz)∑i,j[e−κm,n1(v)Δt⋅Im,n(v)(Cij(x,t))+am,n1(v)(x,Δt)Cij(x,t)]MS(i,2m)MC(j,2n−1)(22)\sum_{i,j} C_{ij}(x,t+\Delta t)\sum_{m,n=1}^{\infty} b^{(v)}_{m,n_1}(x,\Delta t)\frac{\beta_m}{D(x)}\cos(\beta_m\eta)\cos(\nu_n z)M_S(i,2m)M_C(j,2n-1) = \operatorname{Po}\sum_{m,n=1}^{\infty}\frac{(-1)^{m+n}}{\nu_n}\frac{D'(x)}{D(x)}\eta\sin(\nu_{m_1}\eta)\cos(\nu_n z)I^{(u)}_{m_1,n_1}\left(\partial_xp_0(x,t+\Delta t)\right) +\operatorname{Po}\sum_{m,n=1}^{\infty}\frac{(-1)^{m+n}}{\nu_{m_1}\nu_n}\cos(\nu_{m_1}\eta)\cos(\nu_n z)\frac{\partial}{\partial x}\left\{r^{(u)}_{m_1,n_1}\left(\partial_xp_0(x,t+\Delta t)\right)\right\} -\sum_{m,n=1}^{\infty}\frac{\beta_m}{D(x)}\cos(\beta_m\eta)\cos(\nu_n z)\sum_{i,j}\left[e^{-\kappa^{(v)}_{m,n_1}\Delta t}\cdot I^{(v)}_{m,n}\left(C_{ij}(x,t)\right) +a^{(v)}_{m,n_1}(x,\Delta t)C_{ij}(x,t)\right]M_S(i,2m)M_C(j,2n-1) \tag*{(22)}

We choose to solve for CijC_{ij} at a given xx and tt by satisfying equation (22) at uniformly spaced locations in the y−zy-z cross section, where the number of locations matches the number of unknowns, CijC_{ij}. The resulting square linear system is then solved for Cij(xk,tl)C_{ij}(x_k,t_l), corresponding to y−zy-z section at x=xkx=x_k and time t=tlt=t_l. We point out that the decoupling between x,tx,t and y,zy,z in the ansatz for ∂p/∂y\partial p/\partial y enables solving for vv at any y−zy-z cross section, and at any time tt, without the need to solve simultaneously for vv along all the coordinates, (x,y,z,t)(x,y,z,t), which saves on resources and make the method faster and more flexible.

Results

The accuracy of the model is compared to CFD simulations for the cases listed in Table 1. For all cases, a Type I channel geometry is considered (Figure 1) with

case nameforcing typeRe\mathrm{Re}St\mathrm{St}α\alphaγ\gamma
Case 1harmonic110.050.05
Case 2harmonic150.050.05
Case 3harmonic1100.050.05
Case 4harmonic510.050.05
Case 5harmonic2010.050.05
Case 6ramp510.050.05
Case 7harmonic110.050.25
Case 8harmonic110.050.5
Case 9harmonic110.051.0
Case 10harmonic110.10.05
Case 11harmonic110.30.05
Case 12harmonic110.50.05

Table 1. List of studied cases on Type I channel with their corresponding physical and geometrical parameters

D(x)=1+γ02(1−cos⁡(4πkxλ))(23)D(x)=1+\frac{\gamma_0}{2}\left(1-\cos\left(4\pi k\frac{x}{\lambda}\right)\right) \tag*{(23)}

where λ≡L=4\lambda\equiv L=4, k=2k=2, and γ0=0.05\gamma_0=0.05. Note that all quantities are normalized according to § 2.1 with the dimensions of the channel used as L=1000 μmL=1000\,\mu\mathrm{m}, α=0.05\alpha=0.05, D0=α2LD_0=\frac{\alpha}{2}L, and H0=2D0H_0=2D_0. In Table 1, Cases 1 is the reference case with Re⁡=1\operatorname{Re}=1, St⁡=1\operatorname{St}=1, α=0.05\alpha=0.05, and γ=0.05\gamma=0.05. Cases 1-3 aim to explore the performance of the proposed model for the range 1≤St⁡≤101\leq\operatorname{St}\leq10. Case 1,4, and 5 aim to explore the performance of the proposed model for the range 1≤Re⁡≤201\leq\operatorname{Re}\leq20. Case 1,7, 8 and 9 aim to explore the performance of the proposed model for the range 0.05≤γ≤10.05\leq\gamma\leq1. Case 1,10, 11 and 12 aim to explore the performance of the proposed model for the range 0.05≤α≤0.50.05 \le\alpha\le0.5. For all cases, except Case 6, the flow is driven by harmonic forcing of the inlet mass flux according to

m˙=ρUsin⁡(2πft)(24)\dot{m} = \rho U \sin(2\pi f t) \tag*{(24)}

For Case 6, the flow is driven by ramping up the inlet mass flux according to

m˙=ρU(1−e−ft)(25)\dot{m} = \rho U \left(1 - e^{-f t}\right) \tag*{(25)}

where the reference velocity is U=μReρH0U = \frac{\mu\mathrm{Re}}{\rho H_0} with μ=0.001003 N ⁣⋅ ⁣s/m2\mu= 0.001003\ \mathrm{N}\!\cdot\!\mathrm{s}/\mathrm{m}^2, ρ=998.2 kg/m3\rho= 998.2\ \mathrm{kg}/\mathrm{m}^3, and the forcing frequency is given by f=StUH0f = \mathrm{St}\frac{U}{H_0}. Figure 2-(a) plots the time variation of the pressure drop along the channel as predicted by Ansys simulation for the mass flux forcing of Equation (24), also plotted using the symbols in Figure 2-(b). The pressure drop history is then used an input to the reduced order model to predict the unsteady velocity field, the pressure variation along the channel and the mass flow rate. The mass flow rate history predicted by the model matches that of the mass flowrate used to drive the Ansys simulation, as shown in the plot of Figure 2-(b). In Figure 2-(c), plots of the deviation of the pressure distribution along

Normalized pressure drop history, volume flow rate, and departure from the linear profile

Figure 2. (a) Normalized pressure drop history along the channel for Case 9. (b) Resulting variation in volume flow rate. (c) Variation, over one cycle, along the channel of the departure from the linear profile at kT/8kT/8, k=1→8k = 1 \to8. Solid line: model. Symbols: Ansys.

the channel from the linear (Poiseuille) profile are presented at one eighth intervals of the forcing period, T=1/fT = 1/f. The model shows high similarity with the simulation results even at low deviations of order 10−310^{-3}.

Figure 3 shows the variations of the percent relative L2L_2 norm error of the two leading components of the velocity uu and vv with the dimensional parameters varied in Table 1. The %L2L_2 errors in uu and vv are computed respectively as follows

Variation of percent relative $L_2$ error in $u$ and $v$

Figure 3. Variation of % relative L2L_2 error in uu and vv with (a) St for cases {1,2,3}, (b) Re for cases {1,4,5}, (c) γ\gamma for cases {1,7,8,9} and (d) α\alpha for cases {1,10,11,12}.

%L2err(u)=100×∑i=18∑j=18∑k(um(ti,xj,yk,zk)−un(ti,xj,yk,zk))2∑i=18∑j=18∑kun(ti,xj,yk,zk)2(26)\%L_{2}err(u) = 100 \times\frac{\sum_{i=1}^{8}\sum_{j=1}^{8}\sum_{k}(u_m(t_i,x_j,y_k,z_k)-u_n(t_i,x_j,y_k,z_k))^2}{\sum_{i=1}^{8}\sum_{j=1}^{8}\sum_{k}u_n(t_i,x_j,y_k,z_k)^2} \tag*{(26)}
%L2err(v)=100×∑i=18∑j=18∑k(vm(ti,xj,yk,zk)−vn(ti,xj,yk,zk))2∑i=18∑j=18∑kvn(ti,xj,yk,zk)2(27)\%L_{2}err(v) = 100 \times\frac{\sum_{i=1}^{8}\sum_{j=1}^{8}\sum_{k}(v_m(t_i,x_j,y_k,z_k)-v_n(t_i,x_j,y_k,z_k))^2}{\sum_{i=1}^{8}\sum_{j=1}^{8}\sum_{k}v_n(t_i,x_j,y_k,z_k)^2} \tag*{(27)}

The error is evaluated as the sum of square differences between the model solution (subscript mm) and Ansys Fluent results (subscript nn) computed at y−zy-z cross sections that are evenly spaced over the wavelength (xj+1−xj=λ/8x_{j+1}-x_j=\lambda/8) and over evenly spaced intervals of the forcing period (ti+1−ti=T/8t_{i+1}-t_i=T/8). Figure 3-(a) shows that the model is accurate over the considered range of St\mathrm{St} in Cases 1-3, where the %L2L_2 error is less than 0.005% for uu and less than 0.03% for vv. This behavior is expected since the accuracy of the model depends on α\alpha and γ\gamma, both of which assume the small value of 0.05 for Cases 1-3.

Figure 3-(b) presents the variation of the %L2L_2 errors with Re\mathrm{Re} for Cases 1, 4 and 5. The error in uu remains less than 0.005% while the error in vv is less than 0.045%. We notice that the difference between the relative errors of the axial and transverse components of velocity mimics the difference in their scales as the error in vv is about one order of magnitude greater than that of uu. Note that the L2L_2 error in vv is normalized by the L2L_2 norm of vv which scales as αU\alpha U. The modest dependence of %L2L_2 error on Re\mathrm{Re} and St\mathrm{St} numbers is clearly visualized in Figure 4, Figure 5, and Figure 6. The model solution contours, represented in red color, show pronounced agreement with the simulation results, with a subtle decline of accuracy for the transverse component of the velocity.

Contours of $u_0$ and $v_0$ at $T/8$ time intervals of one cycle for Case 1

Figure 4. Contours of u0u_0 and v0v_0 at T/8T/8 time intervals of one cycle for Case 1. Top-left: u0u_0, xyxy-plane of symmetry (z=0z=0), Top-right: u0u_0, xzxz-plane of symmetry (y=0y=0), Bottom-left: u0u_0, yzyz-plane at x=1.875x=1.875, Bottom-right: v0v_0, yzyz-plane at x=1.875x=1.875.

Contours of $u_0$ and $v_0$ at $T/8$ time intervals of one cycle for Case 2

Figure 5. Contours of u0u_0 and v0v_0 at T/8T/8 time intervals of one cycle for Case 2. Top-left: u0u_0, xyxy-plane of symmetry (z=0z = 0), Top-right: u0u_0, xzxz-plane of symmetry (y=0y = 0), Bottom-left: u0u_0, yzyz-plane at x=1.875x = 1.875, Bottom-right: v0v_0, yzyz-plane at x=1.875x = 1.875.

Contours of $u_0$ and $v_0$ at $T/8$ time intervals of one cycle for Case 4

Figure 6. Contours of u0u_0 and v0v_0 at T/8T/8 time intervals of one cycle for Case 4. Top-left: u0u_0, xyxy-plane of symmetry (z=0z = 0), Top-right: u0u_0, xzxz-plane of symmetry (y=0y = 0), Bottom-left: u0u_0, yzyz-plane at x=1.875x = 1.875, Bottom-right: v0v_0, yzyz-plane at x=1.875x = 1.875.

On the other hand, Figure 3-(c) and (d) show a stronger dependence of the %L2L_2 error on the dimensionless geometric parameters γ\gamma and α\alpha. For the range 0.05≤γ≤10.05 \le\gamma\le1 in Cases 1, 7, 8, and 9 of Table 1, the %L2L_2 error, plotted in Figure-(c), increases fro 0.002% to 0.05% for the axial velocity and from 0.03% to 0.24% for the transverse velocity. For Case 7, corresponding to Re=1\mathrm{Re}=1, St=1\mathrm{St}=1, α=0.05\alpha=0.05, and γ=0.25\gamma=0.25, This accuracy of the model can also be assessed by inspecting Figure 7, which shows the model-predicted contours (solid lines) on top of the Ansys-predicted color distributions of uu and vv at selected x−yx-y, y−zy-z and x−zx-z planes at T/8T/8 time intervals of one cycle.

Contours of $u_0$ and $v_0$ at $T/8$ time intervals

Figure 7. Contours of u0u_0 and v0v_0 at T/8T/8 time intervals of one cycle for Case 7. Top-left: u0u_0, xyxy-plane of symmetry (z=0z = 0), Top-right: u0u_0, xzxz-plane of symmetry (y=0y = 0), Bottom-left: u0u_0, yzyz-plane at x=3.125x = 3.125, Bottom-right: v0v_0, yzyz-plane at x=3.125x = 3.125.

For the range 0.05≤α≤0.50.05 \le\alpha\le0.5 in Cases 1, 10, 11 and 12 of Table 1, the %L2L_2 error in vv, plotted in Figure-(d), increases abruptly beyond α=0.1\alpha=0.1 to reach 60% for α=0.5\alpha=0.5. Presenting the error in this way exaggerates the discrepancy. The error in vv increases to reach 60% because the average vv decreases for larger α\alpha, which means that the denominator in Equation (27) decreases as α\alpha increases while keeping Re\mathrm{Re}, St\mathrm{St} and γ\gamma constant. Actually for a fully developed inertia-free flow, and for small values of γ\gamma, the streamlines are nearly parallel to the channel axis for y/D(x)≪1y/D(x) \ll1, and as y/D(x)y/D(x) approaches 1, the streamlines become nearly parallel to the channel wall. As α\alpha increases for a given small γ\gamma, the effect of the curvature of the wall on the streamlines gets smaller. For α/γ≫1\alpha/\gamma\gg1, most of the streamlines are parallel to the channel axis, which implies that v≃0v \simeq0 in most of the domain, except near the wall. As such, a smaller error in vv gets significantly amplified when normalized by the L2L_2 norm of vv. Upon inspecting Figure 8, one can observe that the model-predicted contours (solid lines) of uu are in good agreement with the color distributions predicted by Ansys-Fluent simulations at selected x−yx-y, y−zy-z and x−zx-z planes at T/8T/8 time intervals of one cycle. This is not the case for contours of vv, where the model-predicted contours diverge considerably from the Ansys solution, as shown in the bottom right subfigure. As pointed out in the discussion above, the values of vv in this cases are very small.

Contours of $u_0$ and $v_0$ at $T/8$ time intervals of one cycle for Case 11

Figure 8. Contours of u0u_0 and v0v_0 at T/8T/8 time intervals of one cycle for Case 11. Top-left: u0u_0, xyxy-plane of symmetry (z=0z = 0), Top-right: u0u_0, xzxz-plane of symmetry (y=0y = 0), Bottom-left: u0u_0, yzyz-plane at x=3.125x = 3.125, Bottom-right: v0v_0, yzyz-plane at x=3.125x = 3.125.

For Case 6 of Table 1, the flow is driven by an inlet mass flux obeying equation (25). The corresponding time variation of the pressure drop along the channel as predicted by Ansys simulation experiences a sharp rise to it steady value, as plotted in Figure 9-(a). The pressure drop history is then used an input to the reduced order model to predict the unsteady velocity field, the pressure variation along the channel and the mass flow rate. The mass flow rate history predicted by the model matches that of the mass flowrate used to drive the Ansys simulation, as shown in the plot of Figure 9-(b). The accuracy of the model is also apparent in Figure 10, where the model-predicted contours of uu and vv match those predicted by the Ansys simulation with a slight deviation on the transverse velocity component vv.

Normalized pressure drop history and resulting volume flow rate

Figure 9. (a) Normalized pressure drop history along the channel for Case 6. (b) Resulting variation in volume flow rate. Solid line: model. Symbols: Ansys.

Contours of $u_0$ and $v_0$ at $T/8$ time intervals of one cycle for Case 6

Figure 10. Contours of u0u_0 and v0v_0 at T/8T/8 time intervals of one cycle for Case 6. Top-left: u0u_0, xyxy-plane of symmetry (z=0z = 0), Top-right: u0u_0, xzxz-plane of symmetry (y=0y = 0), Bottom-left: u0u_0, yzyz-plane at x=1.75x = 1.75, Bottom-right: v0v_0, yzyz-plane at x=1.75x = 1.75.

Conclusion

In this paper, we presented a simplified model for predicting the dynamic behavior of low αRe\alpha Re flows in three-dimensional channels of slowly varying cross section. The model is based on an asymptotic expansion of the governing equations in the aspect ratio of the channel α≪1\alpha\ll1.

The solution of these leading order equations is a hybrid analytical-numerical solution. The leading order axial velocity uu is obtained using a finite Fourier transform in the plane orthogonal to the flow direction on the momentum equation; the pressure is obtained via a recursive numerical scheme that uses the mass equation and the expression of the solution to the axial velocity. In order to solve the yy-component of the momentum equation for vv, we use and ansatz for ∂p/∂y\partial p/\partial y in the form of a polynomial in y,zy,z, where the coefficients are dependent on xx and tt. We then use the continuity equation

tion while assuming that w≃0w \simeq0, which is consistent with our assumption that the channel height only depends on xx. This allows us to solve for the unknown coefficients. In this fashion, we obtain vv at any yy–zz cross-section, at any instant, without having to solve for the entire v(x,y,z,t)v(x,y,z,t), thus saving on resources and speed.

In contrast to previous works, the proposed approach allows for a general time dependence of the driving pressure force, provided that the advective inertia term remains much smaller than the driving force. As such, the model accurately predicts the dynamic flow behavior in response to fast temporal changes in Δp\Delta p, which are characterized by large values of the Strouhal number, associated with the high frequency components comprising the pressure drop.

The accuracy of the model is assessed by inspecting the behavior of the relative L2L_2 norm of the error in the uu and vv components, where the true solution is obtained from converged Ansys Fluent simulations. In addition, contours of the velocity components at selected cross sections, resulting from the simplified model are compared with those of Ansys. The results demonstrate that the proposed model is capable of accurately predicting the velocity field for most cases considered. In fact, in the ranges (St<10St < 10, Re<20Re < 20, γ<1\gamma< 1, α<0.5\alpha< 0.5), the relative percent L2L_2 error in uu (and pp) is less than 1%. As for the error in vv, the relative percent L2L_2 error in vv is less than 10% for all the cases considered in the ranges (St<10St < 10, Re<20Re < 20, γ<1\gamma< 1, α<0.3\alpha< 0.3), except when α=0.5\alpha= 0.5 and for small γ\gamma. As discussed above, the relative error in vv increases with α\alpha, for small γ\gamma, and this is because vv itself is very small for this case. In fact, the streamlines are nearly parallel to the xx-axis over most of the domain. We note that the way we presented the error dependence in Figure 3 can also be used to gain information about the error sensitivity to variations in ReRe, StSt, α\alpha, γ\gamma.

The speed of the simplified model is around two orders of magnitude larger than that of the detailed 3D CFD simulations of Ansys Fluent.

We envision that future directions for this work can proceed along two tracks: a methodology track and an applications track. (i) In the methodology track, directions that are worth exploring include the possibility of relaxing the condition on ww by allowing for variations of the channel cross section along zz. Another direction is to further investigate whether we can use a better alternative to the low degree polynomial ansatz for ∂p/∂y\partial p/\partial y, and explore alternative ways to solve for the unknown polynomial coefficients. (ii) in the application track, a potentially impactful future direction is to couple this simplified model to an advection diffusion solver. The solver can then be embedded within a framework to quickly assess proposed mixing/heat transfer mechanisms in micro-fluidic applications and arrive at conditions that optimize the efficiency.

Acknowledgments

This work was supported by LAU SRDC grant SRDC-r-2017-21

References

  1. [1]G. Karniadakis, A. Beskok, N. Aluru, Microflows and nanoflows: fundamentals and simulation, Vol. 29, Springer, 2005.
  2. [2]S. Bhattacharyya, S. Bera, Combined electroosmosis-pressure driven flow and mixing in a microchannel with surface heterogeneity, Applied Mathematical Modelling 39 (15) (2015) 4337–4350.DOI
  3. [3]I. Pinheiro, G. Puccetti, G. Morini, L. Sphaier, Integral transform analysis of microchannel fluid flow: Irregular geometry estimation using velocimetry data, Applied Mathematical Modelling 90 943–954.
  4. [4]G. K. Batchelor, An Introduction to Fluid Dynamics, Cambridge University Press, 2000.
  5. [5]L. Issa, Perturbation-based simplified models for unsteady incompressible microchannel flows, Applied Mathematical Modelling 40 (9) (2016) 5301–5323.DOI
  6. [6]L. Issa, A simplified model for unsteady pressure driven flows in circular microchannels of variable cross-section, Applied Mathematical Modelling 59 (2018) 410 – 426. doi:https://doi.org/10.1016/j.apm.2018.01.037. URL http://www.sciencedirect.com/science/article/pii/S0307904X18300490
  7. [7]E. Lauga, A. D. Stroock, H. A. Stone, Three-dimensional flows in slowly varying planar geometries, Physics of Fluids (1994-present) 16 (8) (2004) 3051–3062.DOI
  8. [8]A. R. Rao, R. Devanathan, Pulsatile flow in tubes of varying cross-sections, Zeitschrift für angewandte Mathematik und Physik ZAMP 24 (2) (1973) 203–213.DOI
  9. [9]P. Hall, Unsteady viscous flow in a pipe of slowly varying cross-section, Journal of Fluid Mechanics 64 (02) (1974) 209–226.DOI
  10. [10]L. Issa, I. Lakkis, Reduced order models of low mach number isothermal flows in microchannels, in: ASME 2013 11th International Conference on Nanochannels, Microchannels, and Minichannels, American Society of Mechanical Engineers, 2013, pp. V001T12A012–V001T12A012.DOI
  11. [11]L. Issa, I. Lakkis, Reduced-order modeling of low mach number unsteady microchannel flows, Journal of Fluids Engineering 136 (5) (2014) 051201.DOI
  12. [12]I. Lakkis, System-level modeling of microflows in circular and rectangular channels, ICNMM2008, ASME, 2008.DOI

Paper details

vlnc:W_x94ss75y
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

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

  • Physics and Astronomy

Not recorded for this paper

View statement
Looking for related papers…
Contents