def getArrayDescription(array, context, dataset):
if not array:
return None
if array.GetNumberOfComponents() == 9:
adapter = dsa.WrapDataObject(dataset)
name = array.GetName()
npArray = adapter.GetPointData().GetArray(name)
eigenvalues = algorithms.eigenvalue(npArray)
merge = np.column_stack((npArray, eigenvalues[:, np.newaxis, :]))
n = npArray.shape[0]
array.SetNumberOfComponents(12)
array.SetNumberOfTuples(n)
for i in range(n):
array.SetTypedTuple(i, merge[i].ravel())
pMd5 = digest(array)
context.cacheDataArray(pMd5, {
'array': array,
'mTime': array.GetMTime(),
'ts': time.time()
})
root = {}
root['hash'] = pMd5
root['vtkClass'] = 'vtkDataArray'
root['name'] = array.GetName()
root['dataType'] = getJSArrayType(array)
root['numberOfComponents'] = array.GetNumberOfComponents()
root['size'] = array.GetNumberOfComponents() * array.GetNumberOfTuples()
root['ranges'] = []
if root['numberOfComponents'] > 1:
for i in range(root['numberOfComponents']):
root['ranges'].append(getRangeInfo(array, i))
root['ranges'].append(getRangeInfo(array, -1))
else:
root['ranges'].append(getRangeInfo(array, 0))
return root
# Various numerical ops implemented in VTK g = algs.gradient(elev) assert algs.all(g[0] == (1, 0, 0)) v = algs.make_vector(elev, g[:,0], elev) assert algs.all(algs.gradient(v) == [[1, 0, 1], [0, 0, 0], [0, 0, 0]]) v = algs.make_vector(elev, g[:,0], elev2) assert algs.all(algs.curl(v) == [1, 0, 0]) v = algs.make_vector(elev, elev2, 2*elev3) g = algs.gradient(v) assert g.DataSet is v.DataSet assert algs.all(algs.det(g) == 2) assert algs.all(algs.eigenvalue(g) == [2, 1, 1]) assert algs.all(randomVec[:,0] == randomVec[:,0]) int_array1 = numpy.array([1, 0, 1], dtype=numpy.int) int_array2 = numpy.array([0, 1, 0], dtype=numpy.int) assert algs.all(algs.bitwise_or(int_array1, int_array2) == 1) assert algs.all(algs.bitwise_or(int_array1, dsa.NoneArray) == int_array1) assert algs.all(algs.bitwise_or(dsa.NoneArray, int_array1) == int_array1) comp_array1 = dsa.VTKCompositeDataArray([int_array1, int_array2]) comp_array2 = dsa.VTKCompositeDataArray([int_array2, int_array1]) comp_array3 = dsa.VTKCompositeDataArray([int_array2, dsa.NoneArray]) assert algs.all(algs.bitwise_or(comp_array1, comp_array2) == 1) assert algs.all(algs.bitwise_or(comp_array1, dsa.NoneArray) == comp_array1) assert algs.all(algs.bitwise_or(dsa.NoneArray, comp_array1) == comp_array1)
# Various numerical ops implemented in VTK
g = algs.gradient(elev)
assert algs.all(g[0] == (1, 0, 0))
v = algs.make_vector(elev, g[:,0], elev)
assert algs.all(algs.gradient(v) == [[1, 0, 0], [0, 0, 0], [1, 0, 0]])
v = algs.make_vector(elev, g[:,0], elev2)
assert algs.all(algs.curl(v) == [1, 0, 0])
v = algs.make_vector(elev, elev2, 2*elev3)
g = algs.gradient(v)
assert g.DataSet is v.DataSet
assert algs.all(algs.det(g) == 2)
assert algs.all(algs.eigenvalue(g) == [2, 1, 1])
assert algs.all(randomVec[:,0] == randomVec[:,0])
ssource = vtk.vtkSphereSource()
ssource.Update()
output = ssource.GetOutput()
fd = vtk.vtkFloatArray()
fd.SetNumberOfTuples(11)
fd.FillComponent(0, 5)
fd.SetName("field array")
output.GetFieldData().AddArray(fd)
def rans_les_uiuj_err(rans_vtk, les_vtk, thrsh, Ls=1, Us=1):
small = np.cbrt(np.finfo(float).tiny)
nnode = les_vtk.number_of_points
delij = np.zeros([nnode, 3, 3])
for i in range(0, 3):
delij[:, i, i] = 1.0
###############
# Get RANS uiuj
###############
rans_dsa = dsa.WrapDataObject(rans_vtk)
U = rans_dsa.PointData[
'U'] # NOTE - Getting variables from dsa obj not vtk obj as want to use algs etc later
# Velocity gradient tensor and its transpose
# J[:,i-1,j-1] is dUidxj
# Jt[:,i-1,j-1] is dUjdxi
Jt = algs.gradient(U) # Jt is this one as algs uses j,i ordering
J = algs.apply_dfunc(np.transpose, Jt, (0, 2, 1))
# Strain and vorticity tensors
Sij = 0.5 * (J + Jt)
nu_t = rans_dsa.PointData['mu_t'] / rans_dsa.PointData['ro']
tke = rans_dsa.PointData['k']
uiuj_rans = (2.0 / 3.0) * tke * delij - 2.0 * nu_t * Sij
# Find Caniso from linear EVM
aij = copy.deepcopy(Sij) * 0.0
for i in range(0, 3):
for j in range(0, 3):
aij[:, i,
j] = uiuj_rans[:, i, j] / (2.0 * tke + small) - delij[:, i,
j] / 3.0
# Get eigenvalues of aij
eig = algs.eigenvalue(aij)
eig1 = eig[:, 0]
eig2 = eig[:, 1]
eig3 = eig[:, 2]
# Get coords on barycentric triangle from eigenvalues
xc = [1.0, 0.0, 0.5] #x,y coords of corner of triangle
yc = [0.0, 0.0, np.cos(np.pi / 6.0)]
C1c = eig1 - eig2
C2c = 2 * (eig2 - eig3)
C3c = 3 * eig3 + 1
Caniso = 1.0 - C3c
rans_vtk.point_arrays['Caniso'] = Caniso
###############
# Get LES uiuj
###############
# Wrap vista object in dsa wrapper
les_dsa = dsa.WrapDataObject(les_vtk)
# Copy Reynolds stresses to tensor
# uiuj_les = np.zeros([nnode,3,3])
uiuj_les = copy.deepcopy(uiuj_rans) * 0.0
uiuj_les[:, 0, 0] = les_dsa.PointData['uu']
uiuj_les[:, 1, 1] = les_dsa.PointData['vv']
uiuj_les[:, 2, 2] = les_dsa.PointData['ww']
uiuj_les[:, 0, 1] = les_dsa.PointData['uv']
uiuj_les[:, 0, 2] = les_dsa.PointData['uw']
uiuj_les[:, 1, 2] = les_dsa.PointData['vw']
uiuj_les[:, 1, 0] = uiuj_les[:, 0, 1]
uiuj_les[:, 2, 0] = uiuj_les[:, 0, 2]
uiuj_les[:, 2, 1] = uiuj_les[:, 1, 2]
tke_les = 0.5 * (uiuj_les[:, 0, 0] + uiuj_les[:, 1, 1] + uiuj_les[:, 2, 2])
# Calc divergence of DNS and RANS uiuj
div_RANS = np.zeros([nnode, 3])
div_LES = np.zeros([nnode, 3])
for i in range(3):
for j in range(3):
div_RANS[:, i] += algs.gradient(uiuj_rans[:, i, j])[:, i]
div_LES[:, i] += algs.gradient(uiuj_les[:, i, j])[:, i]
div_err = 1.0 * div_LES - rans_dsa.PointData['ro'] * div_RANS
err = np.sqrt(div_err[:, 0]**2 + div_err[:, 1]**2 + div_err[:, 2]**2)
# err = err/(tke_les+small)
err = err * Ls / Us**2
# # Calc Frobenius norm distance between RANS and LES
# err = np.zeros(nnode)
# for i in range(3):
# for j in range(3):
# err[:] += ( (uiuj_rans[:,i,j]-uiuj_les[:,i,j]) )**2.
# err = np.sqrt(err)
# err = err/(tke_les+small)
index = algs.where(err > thrsh)
err_bool = np.zeros(nnode, dtype=int)
err_bool[index] = 1
y_raw = les_vtk.point_arrays['raw']
y_targ = les_vtk.point_arrays['target']
les_vtk.point_arrays['raw'] = np.append(y_raw, err.reshape(-1, 1), axis=1)
les_vtk.point_arrays['target'] = np.append(y_targ,
err_bool.reshape(-1, 1),
axis=1)
def make_targets(les_vtk, y_type, Ls=1, Us=1, ros=1):
from tqdm import tqdm
small = np.cbrt(np.finfo(float).tiny)
Ps = 0.5 * ros * Us**2
les_nnode = les_vtk.number_of_points
delij = np.zeros([les_nnode, 3, 3])
for i in range(0, 3):
delij[:, i, i] = 1.0
# Wrap vista object in dsa wrapper
les_dsa = dsa.WrapDataObject(les_vtk)
if (y_type == 'classification'):
ntarg = 5
y_targ = np.zeros([les_nnode, ntarg], dtype=int)
print('Classifier targets:')
elif (y_type == 'regression'):
ntarg = 2
y_targ = np.zeros([les_nnode, ntarg], dtype=float)
print('regressor targets:')
y_raw = np.zeros([les_nnode, ntarg])
target_labels = np.empty(ntarg, dtype='object')
targ = 0
# Copy Reynolds stresses to tensor
uiuj = np.zeros([les_nnode, 3, 3])
uiuj[:, 0, 0] = les_dsa.PointData['uu']
uiuj[:, 1, 1] = les_dsa.PointData['vv']
uiuj[:, 2, 2] = les_dsa.PointData['ww']
uiuj[:, 0, 1] = les_dsa.PointData['uv']
uiuj[:, 0, 2] = les_dsa.PointData['uw']
uiuj[:, 1, 2] = les_dsa.PointData['vw']
uiuj[:, 1, 0] = uiuj[:, 0, 1]
uiuj[:, 2, 0] = uiuj[:, 0, 2]
uiuj[:, 2, 1] = uiuj[:, 1, 2]
# resolved TKE
tke = 0.5 * (uiuj[:, 0, 0] + uiuj[:, 1, 1] + uiuj[:, 2, 2])
# Velocity vector
U = algs.make_vector(les_dsa.PointData['U'], les_dsa.PointData['V'],
les_dsa.PointData['W'])
# Velocity gradient tensor and its transpose
# J[:,i-1,j-1] is dUidxj
# Jt[:,i-1,j-1] is dUjdxi
Jt = algs.gradient(U) # Jt is this one as algs uses j,i ordering
J = algs.apply_dfunc(np.transpose, Jt, (0, 2, 1))
# Strain and vorticity tensors
Sij = 0.5 * (J + Jt)
Oij = 0.5 * (J - Jt)
# Anisotropy tensor and eigenvalues
aij = copy.deepcopy(Sij) * 0.0
inv2 = np.zeros(les_nnode)
inv3 = np.zeros(les_nnode)
for i in range(0, 3):
for j in range(0, 3):
aij[:, i,
j] = uiuj[:, i, j] / (2.0 * tke + small) - delij[:, i, j] / 3.0
# Get eigenvalues of aij
eig = algs.eigenvalue(aij)
eig1 = eig[:, 0]
eig2 = eig[:, 1]
eig3 = eig[:, 2]
# Get coords on barycentric triangle from eigenvalues
xc = [1.0, 0.0, 0.5] #x,y coords of corner of triangle
yc = [0.0, 0.0, np.cos(np.pi / 6.0)]
C1c = eig1 - eig2
C2c = 2 * (eig2 - eig3)
C3c = 3 * eig3 + 1
x0 = C1c * xc[0] + C2c * xc[1] + C3c * xc[2]
y0 = C1c * yc[0] + C2c * yc[1] + C3c * yc[2]
if (y_type == 'Classification'):
# Target 1: Negative eddy viscosity
#########################################
print('1: Negative eddy viscosity')
A = np.zeros(les_nnode)
B = np.zeros(les_nnode)
for i in range(0, 3):
for j in range(0, 3):
A += -uiuj[:, i, j] * Sij[:, i, j] + (
2.0 / 3.0) * tke * delij[:, i, j] * Sij[:, i, j]
B += 2.0 * Sij[:, i, j] * Sij[:, i, j]
Str = algs.sqrt(B) # magnitude of Sij strain tensor (used later)
nu_t = A / (B + small)
nu_t = nu_t / (Us * Ls)
y_raw[:, targ] = nu_t
index = algs.where(nu_t < 0.0)
y_targ[index, targ] = 1
target_labels[targ] = 'Negative eddy viscosity'
targ += 1
# Target 2: Deviation from plane shear
#################################################
print('2: Deviation from plane shear turbulence')
# Get distance from plane shear line
p1 = (1 / 3, 0)
p2 = (0.5, np.sqrt(3) / 2)
dist = abs((p2[1] - p1[1]) * x0 -
(p2[0] - p1[0]) * y0 + p2[0] * p1[1] -
p2[1] * p1[0]) / np.sqrt((p2[1] - p1[1])**2 +
(p2[0] - p1[0])**2)
y_raw[:, targ] = dist
index = algs.where(dist > 0.25)
y_targ[index, targ] = 1
target_labels[targ] = 'Deviation from plane shar turbulence'
targ += 1
# Target 3: Anisotropy of turbulence
##########################################
print('3: Anisotropy of turbulence')
Caniso = 1.0 - C3c
y_raw[:, targ] = Caniso
index = algs.where(Caniso > 0.5)
y_targ[index, targ] = 1
target_labels[targ] = 'Stress anisotropy'
targ += 1
# Target 4: Negative Pk
############################################
print('4: Negative Pk')
A = np.zeros(les_nnode)
for i in range(0, 3):
for j in range(0, 3):
A[:] += (-uiuj[:, i, j] * J[:, i, j])
A = A * Ls / Us**3
y_raw[:, targ] = A
index = algs.where(A < -0.0005)
y_targ[index, targ] = 1
target_labels[targ] = 'Negative Pk'
targ += 1
# Target 5: 2-eqn Cmu constant
############################################
print('5: 2-equation Cmu constant')
A = np.zeros(les_nnode)
for i in range(0, 3):
for j in range(0, 3):
A[:] += aij[:, i, j] * Sij[:, i, j]
Cmu = nu_t**2.0 * (Str / (tke + small))**2.0
y_raw[:, targ] = Cmu
allow_err = 0.25 #i.e. 10% err
Cmu_dist = algs.abs(Cmu - 0.09)
# index = algs.where(Cmu_dist>allow_err*0.09)
index = algs.where(Cmu > 1.1 * 0.09)
y_targ[index, targ] = 1
target_labels[targ] = 'Cmu != 0.09'
targ += 1
# ab = ((uiuj[:,1,1]-uiuj[:,0,0])*U[:,0]*U[:,1] + uiuj[:,0,1]*(U[:,0]**2-U[:,1]**2))/(U[:,0]**2+U[:,1]**2)
# y_raw[:,err] = ab
# # Target 3: Non-linearity
# ###############################
# print('3: Non-linearity')
#
# # Build cevm equation in form A*nut**3 + B*nut**2 + C*nut + D = 0
# B, A = build_cevm(Sij,Oij)
# B = B/(tke +1e-12)
# A = A/(tke**2.0 +1e-12)
#
# C = np.zeros_like(A)
# D = np.zeros_like(A)
# for i in range(0,3):
# for j in range(0,3):
# C += -2.0*Sij[:,i,j]*Sij[:,i,j]
# D += (2.0/3.0)*tke*Sij[:,i,j]*delij[:,i,j] - uiuj[:,i,j]*Sij[:,i,j]
#
# nu_t_cevm = np.empty_like(nu_t)
# for i in tqdm(range(0,les_nnode)):
# # Find the roots of the cubic equation (i.e. potential values for nu_t_cevm)
# roots = np.roots([A[i],B[i],C[i],D[i]])
# roots_orig = roots
#
# # Remove complex solutions (with imaginary part > a small number, to allow for numerical error)
# #roots = roots.real[abs(roots.imag)<1e-5] #NOTE - Matches nu_t much better without this?!
#
# # Out of remaining solutions(s), pick one that is closest to linear nu_t
# if(roots.size==0):
# nu_t_cevm[i] = nu_t[i]
# else:
# nu_t_cevm[i] = roots.real[np.argmin( np.abs(roots - np.full(roots.size,nu_t[i])) )]
#
# normdiff = algs.abs(nu_t_cevm - nu_t) / (algs.abs(nu_t_cevm) + algs.abs(nu_t) + 1e-12)
# y_raw[:,err] = nu_t_cevm
#
# index = algs.where(normdiff>0.15)
# y_targ[index,err] = 1
# error_labels[err] = 'Non-linearity'
# err += 1
elif (y_type == 'regression'):
# Target 3: Anisotropy of turbulence
##########################################
print('1: Anisotropy of turbulence')
Caniso = 1.0 - C3c
y_raw[:, targ] = Caniso
y_targ[:, targ] = Caniso
target_labels[targ] = 'Stress anisotropy'
targ += 1
return y_raw, y_targ, target_labels
