Vous êtes sur la page 1sur 8

Proceedings of the 1st GASMEMS Workshop- Eindhoven, September 7-8, 2009

GASMEMS09-10

STUDY OF RAREFACTION AND NON-EQUILIBRIUM


EFFECTS USING MODEL KINETIC EQUATIONS
A.A. Shershnev1, A.N. Kudryavtsev2, I.A. Graur3
1,2
Khristianovich Institute of Theoretical and Applied Mechanics,
630090 Novosibirsk, Russia
3
Université de Provence – Ecole Polytechnique Universitaire de Marseille,
13453 Marseille cedex 13, France

Abstract
In the present paper a numerical method based on direct solving of relaxation-type kinetic equations
with high-order finite-difference schemes and the discrete ordinate method is used for simulation of
rarefied non-equilibrium gas flows. The BGK, Shakhov and ellipsoidal statistical models are
employed for 1D computations of the shock wave structure and 2D computations of super- and
hypersonic flows over a finite-length flat plate. The results obtained by solving the kinetic equations
with the present numerical method are validated and compared with Navier-Stokes and DSMC
solutions as well as with experimental data.

1 Introduction
Gas flows in the transitional regime between continuum and free-molecular flow are widely present
in nature and can be often encountered in various practical applications. Noticeable attention has
been recently paid to modelling that sort of flows owing to intensive development of micro- and
nanotechnologies and design of miniature technical devices. The Navier-Stokes equations, even
with the velocity slip and temperature jump boundary conditions, are known to be inadequate for
the description of transitional flows. At present time, the most common approach is the direct
statistical simulation based on solving the Boltzmann equation with the Monte Carlo method (Bird
(1994)). Owing to the stochastic character of the DSMC (direct simulation Monte Carlo) method,
however, there arise a number of difficulties in modeling unsteady and low-speed flows with this
method. The approach based on the direct deterministic solution of the distribution-function kinetic
equations is free of that kind of difficulties. On other hand, when solving the Boltzmann equation,
an accurate calculation of multidimensional integrals in the collision term is very resource-
consuming. A more efficient approach is to solve relaxation-type kinetic equations, which are
approximations of the Boltzmann equation, such as the Bhatnagar-Gross-Krook (BGK) equation,
the Shakhov model, or the ellipsoidal statistical BGK model. Many features of transitional flows are
properly described by that sort of approximating equations (Bhatnagar et al. (1954)).
The goal of the present paper is to develop an efficient method for numerical simulation of
transitional flows in microdevices based on solving the relaxation-type kinetic equation and to use it
for numerical simulation of rarefied gas flows, especially characterized by significant non-
equilibrium effects.
Proceedings of the 1st GASMEMS Workshop- Eindhoven, September 7-8, 2009

2 Kinetic Approach and Numerical Method


2.1 Model Kinetic Equations
We consider a class of model kinetic equations, which can be deduced from the Boltzmann equation
by replacement of the collision integral by a simple relaxation-type term:

f  f
    ( f N  f ) .
t x
  
Here, f ( x,  , t ) is the velocity distribution function, x is the spatial coordinate vector,  is the
molecular velocity vector,  is the collision frequency, and f N is an appropriate equilibrium
distribution function depending on the model used.
Macroscopic quantities, such as the number density n , the flow velocity u , the temperature T , the

heat flux vector q , can be determined using appropriate moments of the distribution function:

       
n( x , t )   f ( x ,  , t ) d  x d  y d  z , n u ( x , t )    f ( x ,  , t )d  x d  y d  z ,

      c2   
3nRT ( x , t )   c 2 f ( x ,  , t ) d  x d  y d  z , q ( x , t )  m  c f ( x ,  , t ) d  x d  y d  z ,
2

  
where c    u is the thermal velocity.
In present paper, three different models, the BGK model, the model of Shakhov, and the ellipsoid
statistical model are used.
In the BGK model, f N is simply a local Maxwellian function,

n  1 3 
fN  fM  exp    (i  ui )(i  ui )  .
(2 RT )3 / 2  2 RT i , j 1 

The BGK model is the simplest and most widely used kinetic equation. However, when employing
the Chapman-Enskog procedure it leads to a wrong value of the Prandtl number in the resulting
continuum equations. This disadvantage can be eliminated by adjustment of the equilibrium
distribution function. Two commonly used generalizations of the BGK equation are the model of
Shakhov (1974) and the ellipsoidal statistical model proposed by Holway (1966).
In the Shakhov model, f N is defined as follows:

 1     c2 
f N  f M 1  (1  Pr)(  u )  q   5  .
 RT 
 5 pRT   


Here q is the heat flux vector, p is the pressure, c denotes the speed of sound and Pr is the Prandtl
number, which equals 2/3 for monatomic gases.
In the ellipsoidal statistical model, the equilibrium distribution function f N is an anisotropic
Gaussian:

n  3 
fN  fE  det A  exp    Aij (i  ui )( j  u j )  ,
 3/ 2  i , j 1 

1
 2RT 2(1  Pr) pij   
where Aij    ij   , pij   (i  ui )( j  u j ) f ( x ,  , t ) d x d y d z .
 Pr n Pr 
Proceedings of the 1st GASMEMS Workshop- Eindhoven, September 7-8, 2009

It is worth noting that while the equilibrium distribution functions of both the BGK and ES models
include only the first and second moments of f , the Shakhov model utilizes also the third-order
moments.

2.2 Non-dimensional variables

Introducing a characteristic velocity C  2RT we define non-dimensional variables as

     
t  t ND /( L / C ) , x  xND  L , u  u ND  C ,    ND  C , n  nND  n ,
1 C   1 n
p  pND  mn C2 , T  TND  T ,    ND  , q  qND  mn C3 , f  f ND 3 .
2 L 2 C

Here the subscript ND denotes non-dimensional variables. In the following considerations we drop
this subscript for the sake of simplicity. Non-dimensional macroscopic quantities can also be found
as moments of the distribution function.

2.3 Reduced Distribution Functions


Generally speaking, the velocity space is three-dimensional even for 1D and 2D problems.
However, the number of independent variables and, as a consequence, numerical efforts in these
cases can be substantially reduced if the so-called reduced distribution functions are introduced. In
one-dimensional case they can be written as
   
g ( x,  x , t )   f ( x ,  , t ) d  y d  z , h( x,  x , t )   ( y2   z2 ) f ( x ,  , t ) d  y d  z .

Knowledge of the reduced distribution function is sufficient for calculation of macroscopic flow
variables. Thus, we get a system of two equations for functions depending on 3 variables instead of
5. In two-dimensional case reduced functions are defined in a similar way by integrating over one
of the molecular velocity components.

2.4 Discrete Ordinate Method

Figure 1. Velocity grid construction. 1D Case.

To solve kinetic equations with a finite-difference scheme, the discrete ordinate method is
commonly used. Its basic concept is to replace exact integration over the whole velocity space with
an approximate numerical integration over a finite domain with a discrete set of points. The Gauss-
Hermite quadrature or the Simpson integration rule are used in numerical computations presented
below.
The position of maximum of the distribution function depends on the mean flow velocity and the
distribution width depends on the temperature. Thus, we can majorize the limits of integration and
determine the required number of points in the velocity space using some a priori information. Test
Proceedings of the 1st GASMEMS Workshop- Eindhoven, September 7-8, 2009

computations show that the following choice of the limits and the increment of the grid in the
velocity space allows one to integrate with decent accuracy:

Tmin
U min  3.5 Tmax ; U min  3.5 Tmax  , dv  ,
  3

where U min , U max are the minimum and maximum flow velocity, Tmax and Tmin are the maximum
and minimum temperature, respectively. Figure 1 gives scheme of velocity construction in one-
dimensional case.

2.5 Spatial and Temporal Approximation


Since numerical integration over the velocity space requires considerable computational efforts
(especially at high flow Mach numbers), it is preferable to use for spatial approximation an effective
high-order numerical scheme. The high resolution shock-capturing WENO schemes have proven to
be a robust instrument suitable for these purposes. In present paper, the 5th order WENO scheme of
Jiang and Shu (1988) was used for spatial approximation and the 2nd order Runge-Kutta scheme
was employed for time stepping.

2.6 Wall Boundary Conditions


Boundary conditions for the distribution function on the solid wall depend on details of gas-surface
interaction. In many cases it can be described with good accuracy by assuming the diffuse
  
reflection of molecules. Namely, for all particles coming on the surface (i.e. n   < 0 , where n is
the interior surface normal) the functions g  and h are calculated as extrapolations of the
corresponding distribution functions from within the computational domain. For all particles
coming off the surface it is assumed that molecules are emitted with the Maxwellian distribution
functions corresponding the zero mean flow velocity and the temperature equal to the wall
temperature Tw Their density nw is calculated from the condition that the fluxes of particles coming
on and off the wall are equal.

3 Results of Numerical Simulations


The numerical method described is used to perform numerical simulations of the shock wave
structure problem and to calculate the supersonic flow over a flat plate at different Mach numbers.
The chosen test cases are fundamental problems, good understanding of them is essential for a
variety of modern applications, such as design of micro-engines (powered by shock-induced
combustion) and re-entry capsules and perspective hypersonic planes construction.

3.1 Shock Wave Structure


The investigation of the shock wave structure is a classical problem including strong non-
equilibrium phenomena. It has been treated both numerically and experimentally by many
researchers – Gilbarg and Paolucci(1954), Alsmeyer (1976), Robben and Talbot (1966), Segal
(1971), Chu (1966), which makes it a suitable test case.
The simulations are performed for shock waves in argon. In the coordinate space 400 grid points are
used. Shock waves with different Mach numbers (up to 10) are considered. The distribution
functions far ahead and behind the shock wave are specified as Maxwellian distributions with
parameters obtained from the Rankine-Hugoniot equations. The results are compared with Navier-
Stokes calculations and experimental measurements of Alsmeyer (1976).
Proceedings of the 1st GASMEMS Workshop- Eindhoven, September 7-8, 2009

The inverse shock-wave thicknesses obtained with three different kinetic models and the Navier-
Stokes equations are depicted in figure 2 as functions of the shock-wave Mach number.
As expected, the solution of the Navier-Stokes equations highly underestimates the stationary
shockwave thickness and gives the thinnest profile of all considered models. Although there is a
significant scatter in experimental data, it can be seen that the ellipsoidal statistical model shows the
best agreement with the experimental results. The BGK model slightly underestimates the shock-
wave thickness, and the Shakhov model exhibits rather a large deviation from experiments, with the
shock-wave thickness being overestimated, especially at high Mach numbers.

Figure 2. Inverse shock wave thickness as function of shock wave Mach number.

Also, data on macroscopic quantities distribution have been obtained. As against the density, which
varies monotonously, the temperature in the shock wave has a more complex behavior. It was
shown by Yen (1966) that the longitudinal temperature Tx can be expressed as a function of density
n and possesses a maximum within a shock wave provided that the free-stream Mach number is
greater than 9 / 5 (for monatomic gases).

Figure 4. Longitudinal temperature in a shock wave, M   3.8 .

Since this relationship is deduced only from the conservation laws, one can use it as a criterion of
the accuracy of the numerical method. As it can be seen in Fig. 4, all models give different
longitudinal temperature distributions. Regardless of this fact, the kinetic models show the same
Proceedings of the 1st GASMEMS Workshop- Eindhoven, September 7-8, 2009

value of the temperature overshoot and are in full agreement with the analytical relation between
the longitudinal temperature and the density.
As against model equations, the Mott-Smith approximation does not have an overshoot in
longitudinal temperature. Comparison of the distribution functions in the density median point
reveals double-peak distributions, which are qualitatively similar to the Mott-Smith bimodal
approximation.

n1  n2
Figure 5. Reduced distribution functions g in the point x* , where n*  , M   9.8
2

3.2 Flat Plate


In this section we consider a rarefied gas flow past finite flat plate at zero angle of attack. It is
another well studied fundamental problem (see, e.g. Pullin and Harvey (1976)).
All computations are performed for the hard sphere gas at the free-stream Mach numbers M   2
and M   10 and the Knudsen number Kn  0.01 . The free-stream flow temperature is equal to the
temperature of the plate, i.e. Tw  T . All computations are performed on the grid with 100  150
points. The Obtained data are validated by comparison with solution of the Navier-Stokes equations
and results of the DSMC simulations.
In figure 6 the distributions of density and pressure at the cross-section x  0.8 are shown. As it can
be seen, in case M   2 the model of Shakhov and the ellipsoid statistical model are in perfect
agreement with each other and in very good agreement with the DSMC results.

Figure 6. Density and pressure profiles. Section x=0.8


Also, the profiles obtained from the Navier-Stokes equations match those from DSMC method in
general quite well, however, significant discrepancies can be observed close to the plate surface.
Proceedings of the 1st GASMEMS Workshop- Eindhoven, September 7-8, 2009

The BGK model demonstrates a big deviation from other profiles, although this result is predictable
due to a wrong value of the Prandtl number in this model.

Figure 7. Density and pressure profiles. Section x=0.8

The distinction between different approaches used is much more pronounced for the M   10 flow.
The BGK model is not used in calculations because of poor results on lower Mach number flows.
The S-model and the ellipsoid statistical model show relatively good agreement with the DSMC
method whereas the Navier-Stokes results differ drastically. Figure 7 shows that in same section
x  0.8 (which is quite far from sharp leading edge) the Navier-Stokes profiles do not coincide
even qualitatively with all other approaches.
This phenomenon can be explained by a strong violation of equilibrium in the hypersonic rarefied
flow. In figure 8 the flowfields of ratio of the longitudinal temperature Tx to the total temperature
T computed with the ellipsoidal statistical model kinetic equations are presented. This parameter
characterizes the non-equilibrium between different translational degrees of freedom of molecules.
As it can be seen in case of the relatively low free-stream Mach number the flow is non-equilibrium
only in a small vicinity of the plate leading edge. However, for the hypersonic Mach number, the
non-equilibrium is observed in an extensive region around the plate.

Figure 8. Nonequilibrium of temperature, Tx / T . M   2 (plot above) and M   10 (plot below).


Proceedings of the 1st GASMEMS Workshop- Eindhoven, September 7-8, 2009

Acknowledgements

This work has been supported by the Russian Foundation for Basic Research (Joint Russian-French
Project No. 06-01-22000).

References

Alsmeyer, H., 1976, Density profiles in argon and nitrogen shock waves measured by the
absorbtion of an electron beam. J. Fluid Mech. 74, pp. 497–513.
Bhatnagar, P.L., Gross, E.P., Krook, M., 1954, A model for collision processes in gases. I. Small
amplitude processes in charged and neutral one-component systems. Phys. Rev. 94, 511–525.
Bird, G.A., 1994, Molecular Gas Dynamics and Direct Simulation of Gas Flows. Clarendon Press:
Oxford.
Chu, C.K., 1966, Kinetic-Theoretic Description of shock Wave Formation. II. Phys.Fluids 8,
1450–1455.
Gilbarg, D. and Paolucci, D., 1953, The structure of shock waves in the continuum theory of fluids,
J. Rat. Mech. Anal. 2, 617–642.
Holway, L.H., 1966, New Statistical Models for Kinetic Theory: Methods of Construction. Phys.
Fluids, 9, 1658–1673.
Pullin, D.I., Harvey, J.K., 1976, Anumerical simulation of the rarefied hypersonic flat-plate
problem. J. Fluid Mech. 78, 689–707.
Robben, F., Talbot, L., 1966, Measurement of Shock Wave Thickness by Electron Beam
Fluorescence Method. Phys.Fluids 9, 633–643.
Segal, Ben M., 1971, Shock wave structure using nonlinear model Boltzmann equations. Stanford
University.
Shakhov, E.M., 1974, in russian, Method for Investigation of Rarefied Gas Motion. Nauka:
Moscow.
Shu, C.-W., Osher, S., 1988, J. Comp. Phys. 77, 439–471.
Shu, C.-W., Osher, S., 1989, J. Comp. Phys. 83, 32–78 .
Yen , S.-M., 1966, Physics of Fluids 9, 1417–1418.

Vous aimerez peut-être aussi