class Obstacle:
"""
Flow class to simulate the flow around an object (mask).
It consists of one inflow (equilibrium boundary)
and one outflow (anti-bounce-back-boundary), leading to a flow in positive x direction.
Parameters
----------
shape : Tuple[int]:
Grid resolution.
domain_length_x : float
Length of the domain in physical units.
Attributes
----------
mask : np.array with dtype = np.bool
Boolean mask to define the obstacle. The shape of this object is the shape of the grid.
Initially set to zero (no obstacle).
Examples
--------
Initialization of flow around a cylinder:
>>> from lettuce import Lattice, D2Q9
>>> flow = Obstacle(
>>> shape=(101, 51),
>>> reynolds_number=100,
>>> mach_number=0.1,
>>> lattice=lattice,
>>> domain_length_x=10.1
>>> )
>>> x, y = flow.grid
>>> condition = np.sqrt((x-2.5)**2+(y-2.5)**2) < 1.
>>> flow.mask[np.where(condition)] = 1
"""
def __init__(self,
shape,
reynolds_number,
mach_number,
lattice,
domain_length_x,
char_length=1,
char_velocity=1):
if len(shape) != lattice.D:
raise ValueError(
f"{lattice.D}-dimensional lattice requires {lattice.D}-dimensional `shape`"
)
self.shape = shape
char_length_lu = shape[0] / domain_length_x * char_length
self.units = UnitConversion(lattice,
reynolds_number=reynolds_number,
mach_number=mach_number,
characteristic_length_lu=char_length_lu,
characteristic_length_pu=char_length,
characteristic_velocity_pu=char_velocity)
self._mask = np.zeros(shape=self.shape, dtype=np.bool)
@property
def mask(self):
return self._mask
@mask.setter
def mask(self, m):
assert isinstance(m, np.ndarray) and m.shape == self.shape
self._mask = m.astype(np.bool)
def initial_solution(self, x):
p = np.zeros_like(x[0], dtype=float)[None, ...]
u_char = self.units.characteristic_velocity_pu * self._unit_vector()
u_char = append_axes(u_char, self.units.lattice.D)
u = (1 - self.mask) * u_char
return p, u
@property
def grid(self):
xyz = tuple(
self.units.convert_length_to_pu(np.arange(n)) for n in self.shape)
return np.meshgrid(*xyz, indexing='ij')
@property
def boundaries(self):
x = self.grid[0]
return [
EquilibriumBoundaryPU(
np.abs(x) < 1e-6, self.units.lattice, self.units,
self.units.characteristic_velocity_pu * self._unit_vector()),
AntiBounceBackOutlet(self.units.lattice,
self._unit_vector().tolist()),
BounceBackBoundary(self.mask, self.units.lattice)
]
def _unit_vector(self, i=0):
return np.eye(self.units.lattice.D)[i]
class DecayingTurbulence:
def __init__(self, resolution, reynolds_number, mach_number, lattice, k0=20, ic_energy=0.5):
self.k0 = k0
self.ic_energy = ic_energy
self.resolution = resolution
self.units = UnitConversion(
lattice,
reynolds_number=reynolds_number, mach_number=mach_number,
characteristic_length_lu=resolution, characteristic_length_pu=2 * np.pi,
characteristic_velocity_pu=None
)
self.wavenumbers = []
self.spectrum = []
def analytic_solution(self, x, t=0):
return
def _generate_wavenumbers(self):
self.dimensions = self.grid[0].shape
frequencies = [np.fft.fftfreq(dim, d=1 / dim) for dim in self.dimensions]
wavenumber = np.meshgrid(*frequencies)
wavenorms = np.linalg.norm(wavenumber, axis=0)
self.wavenumbers = np.arange(int(np.max(wavenorms)))
wavemask = (wavenorms[..., None] > self.wavenumbers - 0.5) & (wavenorms[..., None] <= self.wavenumbers + 0.5)
return wavenorms, wavenumber, wavemask
def _generate_spectrum(self):
wavenorms, wavenumber, wavemask = self._generate_wavenumbers()
ek = (wavenorms) ** 4 * np.exp(-2 * (wavenorms / self.k0) ** 2)
ek /= np.sum(ek)
ek *= self.ic_energy
self.spectrum = ek[..., None] * wavemask
self.spectrum = np.sum(self.spectrum, axis=tuple((np.arange(self.units.lattice.D))))
return ek, wavenumber
def _generate_initial_velocity(self, ek, wavenumber):
dx = self.units.convert_length_to_pu(1.0)
u = np.random.random(np.array(wavenumber).shape) * 2 * np.pi + 0j
u = [np.fft.fftn(u[dim], axes=tuple((np.arange(self.units.lattice.D)))) for dim in range(self.units.lattice.D)]
u_real = [u[dim].real for dim in range(self.units.lattice.D)]
u_imag = [u[dim].imag for dim in range(self.units.lattice.D)]
for dim in range(self.units.lattice.D):
u_real[dim].ravel()[0] = 0
u_imag[dim].ravel()[0] = 0
u_real_h = [np.sqrt(2 / self.units.lattice.D * ek / (u_imag[dim] ** 2 + u_real[dim] ** 2 + 1.e-15))
* u_real[dim] for dim in range(self.units.lattice.D)]
u_imag_h = [np.sqrt(2 / self.units.lattice.D * ek / (u_imag[dim] ** 2 + u_real[dim] ** 2 + 1.e-15))
* u_imag[dim] for dim in range(self.units.lattice.D)]
for dim in range(self.units.lattice.D):
u_real_h[dim].ravel()[0] = 0
u_imag_h[dim].ravel()[0] = 0
### Remove divergence
# modified wave number sin(k*dx) is used, as the gradient below uses second order cental differences
# Modify if other schemes are used or use kx, ky if you don't know the modified wavenumber !!!
wavenumber_modified = [np.sin(wavenumber[dim] * dx) / dx for dim in range(self.units.lattice.D)]
wavenorm_modified = np.linalg.norm(wavenumber_modified, axis=0) + 1e-16
divergence_real = np.zeros(self.dimensions)
divergence_imag = np.zeros(self.dimensions)
for dim in range(self.units.lattice.D):
divergence_real += wavenumber_modified[dim] * u_real_h[dim]
divergence_imag += wavenumber_modified[dim] * u_imag_h[dim]
u_real = [u_real_h[dim] - divergence_real * wavenumber_modified[dim]
/ wavenorm_modified ** 2 for dim in range(self.units.lattice.D)]
u_imag = [u_imag_h[dim] - divergence_imag * wavenumber_modified[dim]
/ wavenorm_modified ** 2 for dim in range(self.units.lattice.D)]
for dim in range(self.units.lattice.D):
u_real[dim].ravel()[0] = 0
u_imag[dim].ravel()[0] = 0
### Scale velocity field to achieve the desired inicial energy
e_kin = [np.sum(u_real[dim] ** 2 + u_imag[dim] ** 2) for dim in range(self.units.lattice.D)]
e_kin = np.sum(e_kin) * .5
factor = np.sqrt(self.ic_energy / e_kin)
u_real = [u_real[dim] * factor for dim in range(self.units.lattice.D)]
u_imag = [u_imag[dim] * factor for dim in range(self.units.lattice.D)]
### Backtransformation to physical space
norm = ((self.resolution * dx ** (1 - self.units.lattice.D) * np.sqrt(self.units.characteristic_length_pu))
if self.units.lattice.D == 3 else (self.resolution / dx))
u = np.asarray([
(np.fft.ifftn(u_real[dim] + u_imag[dim] * 1.0j, axes=tuple((np.arange(self.units.lattice.D)))) * norm).real
for dim in range(self.units.lattice.D)])
return u
def _compute_initial_pressure(self):
return np.zeros(self.dimensions)[None, ...]
def initial_solution(self, x):
"""Return initial solution. Note: this function sets the characteristic velocity in phyiscal units."""
ek, wavenumber = self._generate_spectrum()
u = self._generate_initial_velocity(ek, wavenumber)
p = self._compute_initial_pressure()
self.units.characteristic_velocity_pu = np.linalg.norm(u, axis=0).max()
return p, u
@property
def energy_spectrum(self):
return self.spectrum, self.wavenumbers
@property
def grid(self):
grid = [np.linspace(0, 2 * np.pi, num=self.resolution, endpoint=False) for _ in range(self.units.lattice.D)]
return np.meshgrid(*grid)
@property
def boundaries(self):
return []
