stefanproblem.c

Stefan Problem

Stefan Problem

A thin vapor layer is initialized close to a superheated wall. The high wall temperature heats up the vapor layer, which is in contact with a liquid phase that always remains at saturation temperature leading to the phase change. The simulation setup used here was adapted from Malan et al, 2021. The analytic solution to the problem describes the evolution of the vapor layer thickness: \displaystyle \delta(t) = 2\lambda\sqrt{\alpha_g t} where \alpha_g is the thermal diffusivity, defined as \alpha_g = \lambda_g/\rho_g/cp_g, while \lambda is a value, specific to the physical properties under investigation, which can be found from the solution of the transcendental equation: \displaystyle \lambda\exp(\lambda^2) \text{erf}(\lambda) = \dfrac{cp_g(T_{wall}-T_{sat})}{\Delta h_{ev}\sqrt{\pi}} The animation shows that, at the beginning of the simulation, the superheated wall heats up the vapor layer which becomes hotter than the liquid phase, leading to the evaporation which establishes the interface velocity jump.

Evolution of the interface and the temperature field

Phase Change Setup

We move the interface using the velocity uf, with the expansion term shifted toward the gas-phase. In this way uf is divergence-free at the interface. The double pressure velocity couping is used to obtain an extended velocity, used to transport the gas phase tracers.

#define INT_USE_UF
#define CONSISTENTPHASE2
#define SHIFT_TO_GAS

Simulation Setup

We use the centered solver with the divergence source term, and the extended velocity is obtained using the centered-doubled procedure. The evaporation model is used in combination with the temperature-gradient mechanism, which manages the solution of the temperature field.

#include "grid/multigrid.h"
#include "navier-stokes/centered-evaporation.h"
#include "navier-stokes/centered-doubled.h"
#include "two-phase.h"
#include "tension.h"
#include "evaporation.h"
#include "temperature-gradient.h"
#include "view.h"

We declare the variables required by the temperature-gradient model.

double lambda1, lambda2, cp1, cp2, dhev;
double TL0, TG0, TIntVal, Tsat, Twall;

Boundary conditions

Outflow boundary conditions for velocity and pressure are imposed on the left wall, while the temperature on the superheated wall is set to the superheated value.

u.n[left] = neumann (0);
u.t[left] = neumann (0);
p[left] = dirichlet (0);
uext.n[left] = neumann (0);
uext.t[left] = neumann (0);
pext[left] = dirichlet (0);

TG[left]  = dirichlet (Tsat);
TG[right] = dirichlet (Twall);
TL[left]  = dirichlet (Tsat);
TL[right] = dirichlet (Twall);

Problem Data

We declare the maximum and minimum levels of refinement, the \lambda parameter, and the initial thickness of the vapor layer.

int maxlevel, minlevel = 3;
double lambdaval = 0.06779249298045148;
double  delta0 = 322.5e-6;
double tshift, teff;


int main (void) {

We set the material properties of the fluids.

  rho1 = 958., rho2 = 0.6;
  mu1 = 2.82e-4, mu2 = 1.23e-5;
  lambda1 = 0.68, lambda2 = 0.025;
  cp1 = 4216., cp2 = 2080.;
  dhev = 2.256e6,

The initial temperature and the interface temperature are set to the same value.

  Tsat = 373., Twall = 383.;
  TL0 = Tsat, TG0 = Tsat; TIntVal = Tsat;

We change the dimension of the domain and the surface tension coefficient.

  L0 = 10e-3;
  f.sigma = 0.059;

We run the simulation for different maximum levels of refinement.

  for (maxlevel = 5; maxlevel <= 6; maxlevel++) {
    init_grid (1 << maxlevel);
    run();
  }
}

We initialize the volume fraction field and the temperature in the gas and in liquid phase.

event init (i = 0) {
  fraction (f, -(x - 0.0096775));

  foreach() {
    TL[] = f[]*TL0;
    TG[] = (1. - f[])*TG0;
    T[] = f[]*TL0 + (1. - f[])*TG0;
  }

At simulation time equal to zero, the thickness of the vapor layer is not zero. Therefore, we compute a time shifting factor (just for post-processing purposes).

  double effective_height;
  effective_height = (sq(L0) - statsf(f).sum)/L0;
  tshift = sq(effective_height/2./lambdaval)*rho2*cp2/lambda2;
}

We refine the interface and the region where the temperature field changes.

#if TREE
event adapt (i++) {
  adapt_wavelet_leave_interface ({T}, {f},
      (double[]){1.e-3}, maxlevel, minlevel, 1);
}
#endif

Post-Processing

The following lines of code are for post-processing purposes.

Exact Solution

We write a function that computes the exact solution to the thickness of the vapor layer, and the analytic temperature profile.

double exact (double time) {
  return 2.*lambdaval*sqrt(lambda2/rho2/cp2*time);
}

double tempsol (double time, double x) {
  return Twall + ((Tsat - Twall)/erf(lambdaval))*
    erf(x/2./sqrt(lambda2/rho2/cp2*time));
}

Output Files

We write the thickness of the vapor layer and the analytic solution on a file.

event output (t += 0.1) {
  double effective_height = 0.;
  foreach(reduction(+:effective_height))
    effective_height += (1. - f[])*dv();
  effective_height /= L0;

  double relerr = fabs (exact(t+tshift) - effective_height) / exact(t+tshift);

  char name[80];
  sprintf (name, "OutputData-%d", maxlevel);
  static FILE * fp = fopen (name, "w");

  fprintf (fp, "%g %g %g %g\n", t+tshift, effective_height, exact (t+tshift), relerr);
  fflush (fp);
}

Temperature Profile

We write on a file the temperature profile at the final time step.

event profiles (t = end) {
  char name[80];
  sprintf (name, "Temperature-%d", maxlevel);

We create an array with the temperature profile for each processor.

  Array * arrtemp = array_new();
  for (double x = 0.; x < L0; x += 0.5*L0/(1 << maxlevel)) {
    double val = interpolate (T, x, 0.);
    val = (val == nodata) ? 0. : val;
    array_append (arrtemp, &val, sizeof(double));
  }
  double * temps = (double *)arrtemp->p;

We sum each element of the arrays in every processor.

  @if _MPI
  int size = arrtemp->len/sizeof(double);
  MPI_Allreduce (MPI_IN_PLACE, temps, size, MPI_DOUBLE, MPI_SUM, MPI_COMM_WORLD);
  @endif

The master node writes the temperature profile.

  if (pid() == 0) {
    FILE * fpp = fopen (name, "w");
    int count = 0;
    for (double x = 0.; x < L0; x += 0.5*L0/(1 << maxlevel)) {
      double R = exact (t+tshift);
      double tempexact = (x >= L0-R) ? tempsol (t+tshift, L0-x) : Tsat;
      fprintf (fpp, "%g %g %g\n", x, temps[count], tempexact);
      count++;
    }
    fflush (fpp);
    fclose (fpp);
  }
  array_free (arrtemp);
}

Movie

We write the animation with the evolution of the temperature field and the gas-liquid interface.

event movie (t += 0.1; t <= 10) {
  if (maxlevel == 5) {
    clear();
    view (tx = -0.5, ty = -0.5);
    box();
    draw_vof ("f", lw = 1.5);
    squares ("T", min = Tsat, max = Twall, linear = true);
    vectors ("u", scale = 1.e-1, lc = {1.,1.,1.});
    save ("movie.mp4");
  }
}

Results

set xlabel "t[s]"
set ylabel "Vapor Layer Thickness [m]"
set key left top
set size square
set grid

plot "OutputData-5" every 10 u 1:3 w p ps 2 t "Analytic", \
     "OutputData-5" u 1:2 w l lw 2 t "LEVEL 5", \
     "OutputData-6" u 1:2 w l lw 2 t "LEVEL 6"

Evolution of the vapor layer thickness (script)

reset

stats "OutputData-4" using (last4=$4) nooutput
stats "OutputData-5" using (last5=$4) nooutput
stats "OutputData-6" using (last6=$4) nooutput
#stats "OutputData-7" using (last7=$4) nooutput

set print "Errors.csv"

print sprintf ("%d %.12f", 2**5, last5)
print sprintf ("%d %.12f", 2**6, last6)
#print sprintf ("%d %.12f", 2**7, last7)

unset print

reset
set xlabel "N"
set ylabel "Relative Error"

set logscale x 2
set logscale y

set xr[2**4:2**7]
set yr[1e-6:1e-2]

set size square
set grid

plot "Errors.csv" w p pt 8 ps 2 title "Results", \
  0.05*x**(-1) lw 2 title "1^{st} order", \
  0.5*x**(-2)  lw 2 title "2^{nd} order"

Relative Errors (script)

reset
set xlabel "Length [m]"
set ylabel "Temperature [K]"

set xr[0.0075:0.01]

set key bottom right
set size square
set grid

plot "Temperature-5" u 1:3 w p ps 2 t "Analytic", \
     "Temperature-5" u 1:2 w l lw 2 t "LEVEL 5", \
     "Temperature-6" u 1:2 w l lw 2 t "LEVEL 6"

Temperature Profile (script)

References

[malan2021geometric]

LC Malan, Arnaud G Malan, Stéphane Zaleski, and PG Rousseau. A geometric vof method for interface resolved phase change and conservative thermal energy advection. Journal of Computational Physics, 426:109920, 2021.