You cannot select more than 25 topics
Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
1214 lines
52 KiB
Python
1214 lines
52 KiB
Python
2 years ago
|
'''
|
||
|
@ The script is developed by Qingbin Liu under the supervision of Jie Liu
|
||
|
E-mail to Qingbin Liu: liuqingb@mail2.sysu.edu.cn; liuqingb@pku.edu.cn
|
||
|
to discuss technical details.
|
||
|
--------------------------------------------------------------------------------
|
||
|
The script is to convert the data from Abaqus output database format
|
||
|
to vtk file format for parallel visualization
|
||
|
Python script performs the following three major steps:
|
||
|
1) reading Abaqus output file according to the architecture of ODB;
|
||
|
2) data decomposition for parallel visualization;
|
||
|
3) writing VTK for Paraview.
|
||
|
'''
|
||
|
|
||
|
#import necessary modules to handle Abaqus output database, files and string
|
||
|
from odbAccess import *
|
||
|
from textRepr import *
|
||
|
from string import *
|
||
|
from time import *
|
||
|
|
||
|
#Function 1
|
||
|
def ConvertOdb2Vtk(filename = 'C:\Temp\odb2vtk.txt'): #Modify the default value of filename here to specify the default configuration file
|
||
|
|
||
|
starttime = time()
|
||
|
# Check if filename points to existing file
|
||
|
if not os.path.isfile(filename):
|
||
|
print 'Parameter file "%s" not found'%(filename)
|
||
|
sys.exit(2)
|
||
|
|
||
|
#read odb2vtk file to get parameter setting
|
||
|
odb2vtk = open(filename,'rt')
|
||
|
read = odb2vtk.read()
|
||
|
input = read.split("'")
|
||
|
#get odb file's path
|
||
|
odb_path = input[1]
|
||
|
#get odb file name
|
||
|
odbname = input[3]
|
||
|
#get the output files' path
|
||
|
vtk_path = input[5]
|
||
|
#get the mesh type
|
||
|
mesh_type = int(input[7])
|
||
|
mesh_conner = 0
|
||
|
if (mesh_type == 12):
|
||
|
mesh_conner = 8
|
||
|
mesh_name = "Hexahedron"
|
||
|
if (mesh_type == 10):
|
||
|
mesh_conner = 4
|
||
|
mesh_name = "Tetra"
|
||
|
if (mesh_conner == 0):
|
||
|
print "Mesh type error or unidentified"
|
||
|
os._exit(0)
|
||
|
#get the quantity of pieces to partition
|
||
|
piecenum = int(input[9])
|
||
|
#get the frame
|
||
|
input_frame = input[11].split("-")
|
||
|
input_frame = range(int(input_frame[0]),int(input_frame[1])+1)
|
||
|
#get the step
|
||
|
input_step = input[13].split(",")
|
||
|
#get the instance
|
||
|
input_instance = input[15].split(",")
|
||
|
#end reding and close odb2vtk file
|
||
|
odb2vtk.close()
|
||
|
#display the reading result of odb2vtk file
|
||
|
print "odb2vtk reading finished, time elapsed: ", time()-starttime
|
||
|
print "Basic Information:"
|
||
|
print "Model:",odbname,"; Mesh type:",mesh_name,"; Number of blocks:",piecenum
|
||
|
print "Convert frames: ",input_frame[0]," to ",input_frame[-1]
|
||
|
print "Step & Instance : ",str(input_step),", ",str(input_instance)
|
||
|
|
||
|
#open an ODB ( Abaqus output database )
|
||
|
odb = openOdb(os.path.join(odb_path,odbname)+'.odb',readOnly=True)
|
||
|
print "ODB opened"
|
||
|
|
||
|
#access geometry and topology information ( odb->rootAssembly->instances->(nodes, elements) )
|
||
|
rootassembly = odb.rootAssembly
|
||
|
instance = rootassembly.instances
|
||
|
#access attribute information
|
||
|
step = odb.steps
|
||
|
#get instance & step information : Quantity and all names
|
||
|
allinstancestr = str(instance)
|
||
|
autoins = allinstancestr.split("'")
|
||
|
inslen = len(autoins)/4
|
||
|
instance_N = range(0,inslen)
|
||
|
allstepstr = str(step)
|
||
|
autostep = allstepstr.split("'")
|
||
|
steplen = len(autostep)/4
|
||
|
step_N = range(0,steplen)
|
||
|
|
||
|
for i in input_step:
|
||
|
if(steplen < int(i)):
|
||
|
print "input step exceeds the range of steps"
|
||
|
os._exit(0)
|
||
|
for i in input_instance:
|
||
|
if(inslen < int(i)):
|
||
|
print "input instance exceeds the range of instances"
|
||
|
os._exit(0)
|
||
|
|
||
|
|
||
|
#step cycle
|
||
|
for step_i in input_step:
|
||
|
n = int(step_i)*4+1
|
||
|
stepname = autostep[n]
|
||
|
print "Step: ",stepname
|
||
|
#instance cycle
|
||
|
for ins_i in input_instance:
|
||
|
n = int(ins_i)*4+1
|
||
|
instancename = autoins[n]
|
||
|
print "Instance: ",instancename
|
||
|
|
||
|
#access nodes & elements
|
||
|
node = instance[instancename].nodes
|
||
|
element = instance[instancename].elements
|
||
|
n_nodes = len(node)
|
||
|
n_elements = len(element)
|
||
|
#access attribute(fieldOutputs) information
|
||
|
frame = step[stepname].frames
|
||
|
|
||
|
|
||
|
#compute the number of element of each block
|
||
|
p_elements = n_elements/piecenum + 1
|
||
|
lp_elements = n_elements - (p_elements*(piecenum-1)) #last block
|
||
|
|
||
|
|
||
|
#match nodes' label and its order in sequence (for empty nodes in tetra mesh)
|
||
|
MLN = node[n_nodes-1].label
|
||
|
TOTAL=[]
|
||
|
#read node in sequence, and get the largest label of node(non-empty)
|
||
|
#MLN is the max label of nodeset
|
||
|
for i in node:
|
||
|
TOTAL.append(i.label)
|
||
|
if(i.label > MLN):
|
||
|
MLN = i.label
|
||
|
#match (the key)
|
||
|
L=[]
|
||
|
n = 0
|
||
|
for i in range(MLN):
|
||
|
L.append(0)
|
||
|
for i in TOTAL:
|
||
|
L[i-1] = n
|
||
|
n += 1
|
||
|
|
||
|
|
||
|
#frame cycle
|
||
|
for i_frame in input_frame:
|
||
|
|
||
|
#Detect whether the input frame is out of range
|
||
|
try:
|
||
|
TRY = odb.steps[stepname].frames[int(i_frame)]
|
||
|
except:
|
||
|
print "input frame exceeds the range of frames"
|
||
|
os._exit(0)
|
||
|
break
|
||
|
|
||
|
#Access a frame
|
||
|
N_Frame = odb.steps[stepname].frames[int(i_frame)]
|
||
|
print "Frame:",i_frame
|
||
|
|
||
|
#create array for store result data temporarily
|
||
|
# Vector-U,A,V,RF
|
||
|
L0=[]
|
||
|
# Tensors-S
|
||
|
L1=[]
|
||
|
# Tensors-LE
|
||
|
L2=[]
|
||
|
# Tensors-PE
|
||
|
L3=[]
|
||
|
# Scalars-PEEQ
|
||
|
L4=[]
|
||
|
for i in range(MLN):
|
||
|
L0.append([0,0,0,0,0,0,0,0,0,0,0,0])
|
||
|
L1.append([0,0,0,0,0,0,0,0,0,0,0,0,0,0])
|
||
|
L2.append([0,0,0,0,0,0,0,0,0])
|
||
|
L3.append([0,0,0,0,0,0,0,0,0])
|
||
|
L4.append([0,0])
|
||
|
|
||
|
|
||
|
print "Reading U, A, V, RF ......"
|
||
|
time1 = time()
|
||
|
#Access Spatial displacement
|
||
|
displacement = N_Frame.fieldOutputs['U']
|
||
|
fieldValues = displacement.values
|
||
|
for valueX in fieldValues :
|
||
|
i = valueX.nodeLabel
|
||
|
L0[i-1][0] = valueX.data[0]
|
||
|
L0[i-1][1] = valueX.data[1]
|
||
|
L0[i-1][2] = valueX.data[2]
|
||
|
#Access Spatial acceleration
|
||
|
acceleration = N_Frame.fieldOutputs['A']
|
||
|
fieldValues = acceleration.values
|
||
|
for valueX in fieldValues :
|
||
|
i = valueX.nodeLabel
|
||
|
L0[i-1][3] = valueX.data[0]
|
||
|
L0[i-1][4] = valueX.data[1]
|
||
|
L0[i-1][5] = valueX.data[2]
|
||
|
#Access Spatial velocity
|
||
|
velocity = N_Frame.fieldOutputs['V']
|
||
|
fieldValues = velocity.values
|
||
|
for valueX in fieldValues :
|
||
|
i = valueX.nodeLabel
|
||
|
L0[i-1][6] = valueX.data[0]
|
||
|
L0[i-1][7] = valueX.data[1]
|
||
|
L0[i-1][8] = valueX.data[2]
|
||
|
#Access Reaction force
|
||
|
Reaction_force = N_Frame.fieldOutputs['RF']
|
||
|
fieldValues = Reaction_force.values
|
||
|
for valueX in fieldValues :
|
||
|
i = valueX.nodeLabel
|
||
|
L0[i-1][9] = valueX.data[0]
|
||
|
L0[i-1][10] = valueX.data[1]
|
||
|
L0[i-1][11] = valueX.data[2]
|
||
|
print "Time elapsed: ", time() - time1, "s"
|
||
|
print "Reading Stress ......"
|
||
|
time1 = time()
|
||
|
#access Stress components
|
||
|
Stress = N_Frame.fieldOutputs['S']
|
||
|
node_Stress = Stress.getSubset(position=ELEMENT_NODAL)
|
||
|
fieldValues = node_Stress.values
|
||
|
for valueX in fieldValues :
|
||
|
L1[valueX.nodeLabel-1][0] += 1
|
||
|
L1[valueX.nodeLabel-1][1] += valueX.data[0]
|
||
|
L1[valueX.nodeLabel-1][2] += valueX.data[1]
|
||
|
L1[valueX.nodeLabel-1][3] += valueX.data[2]
|
||
|
L1[valueX.nodeLabel-1][4] += valueX.data[3]
|
||
|
L1[valueX.nodeLabel-1][5] += valueX.data[4]
|
||
|
L1[valueX.nodeLabel-1][6] += valueX.data[5]
|
||
|
L1[valueX.nodeLabel-1][7] += valueX.mises
|
||
|
L1[valueX.nodeLabel-1][8] += valueX.maxPrincipal
|
||
|
L1[valueX.nodeLabel-1][9] += valueX.midPrincipal
|
||
|
L1[valueX.nodeLabel-1][10] += valueX.minPrincipal
|
||
|
L1[valueX.nodeLabel-1][11] += valueX.press
|
||
|
L1[valueX.nodeLabel-1][12] += valueX.tresca
|
||
|
L1[valueX.nodeLabel-1][13] += valueX.inv3
|
||
|
# can first ave
|
||
|
print "Time elapsed: ", time() - time1, "s"
|
||
|
print "Reading Logarithmic strain ......"
|
||
|
time1 = time()
|
||
|
#Logarithmic strain components
|
||
|
Logarithmic_strain = N_Frame.fieldOutputs['LE']
|
||
|
node_Logarithmic_strain = Logarithmic_strain.getSubset(position=ELEMENT_NODAL)
|
||
|
fieldValues = node_Logarithmic_strain.values
|
||
|
for valueX in fieldValues :
|
||
|
L2[valueX.nodeLabel-1][0] += 1
|
||
|
L2[valueX.nodeLabel-1][1] += valueX.data[0]
|
||
|
L2[valueX.nodeLabel-1][2] += valueX.data[1]
|
||
|
L2[valueX.nodeLabel-1][3] += valueX.data[2]
|
||
|
L2[valueX.nodeLabel-1][4] += valueX.data[3]
|
||
|
L2[valueX.nodeLabel-1][5] += valueX.data[4]
|
||
|
L2[valueX.nodeLabel-1][6] += valueX.data[5]
|
||
|
L2[valueX.nodeLabel-1][7] += valueX.maxPrincipal
|
||
|
L2[valueX.nodeLabel-1][8] += valueX.minPrincipal
|
||
|
print "Time elapsed: ", time() - time1, "s"
|
||
|
print "Reading Plastic strain ......"
|
||
|
time1 = time()
|
||
|
#Plastic strain components
|
||
|
Plastic_strain = N_Frame.fieldOutputs['PE']
|
||
|
node_Plastic_strain = Plastic_strain.getSubset(position=ELEMENT_NODAL)
|
||
|
fieldValues = node_Plastic_strain.values
|
||
|
for valueX in fieldValues :
|
||
|
L3[valueX.nodeLabel-1][0] += 1
|
||
|
L3[valueX.nodeLabel-1][1] += valueX.data[0]
|
||
|
L3[valueX.nodeLabel-1][2] += valueX.data[1]
|
||
|
L3[valueX.nodeLabel-1][3] += valueX.data[2]
|
||
|
L3[valueX.nodeLabel-1][4] += valueX.data[3]
|
||
|
L3[valueX.nodeLabel-1][5] += valueX.data[4]
|
||
|
L3[valueX.nodeLabel-1][6] += valueX.data[5]
|
||
|
L3[valueX.nodeLabel-1][7] += valueX.maxPrincipal
|
||
|
L3[valueX.nodeLabel-1][8] += valueX.minPrincipal
|
||
|
print "Time elapsed: ", time() - time1, "s"
|
||
|
print "Reading Equivalent plastic strain ......"
|
||
|
time1 = time()
|
||
|
#Equivalent plastic strain
|
||
|
Equivalent_plastic_strain = N_Frame.fieldOutputs['PEEQ']
|
||
|
node_Equivalent_plastic_strain = Equivalent_plastic_strain.getSubset(position=ELEMENT_NODAL)
|
||
|
fieldValues = node_Equivalent_plastic_strain.values
|
||
|
for valueX in fieldValues :
|
||
|
L4[valueX.nodeLabel-1][0] += 1
|
||
|
L4[valueX.nodeLabel-1][1] += valueX.data
|
||
|
print "Time elapsed: ", time() - time1, "s"
|
||
|
|
||
|
|
||
|
'''============================================================'''
|
||
|
|
||
|
print "Partitionning model and writing vtk files ......"
|
||
|
#piece cycle, to partion the model and create each piece for vtk files
|
||
|
for pn in range(piecenum):
|
||
|
time1 = time()
|
||
|
print "frame:",i_frame,"; block:",pn
|
||
|
#Reorganization
|
||
|
#Control&Storage
|
||
|
#estimate whether the node has already existed
|
||
|
stg_p = []
|
||
|
#store the reorganized node for element
|
||
|
stg_e = []
|
||
|
#store the reorganized node for node
|
||
|
stg_n = []
|
||
|
for i in range(MLN):
|
||
|
stg_p.append(-1)
|
||
|
nodecount = 0
|
||
|
#reorganize the node and element (reconstruct the mesh)
|
||
|
if(pn == piecenum-1):
|
||
|
M = range(pn*p_elements,n_elements)
|
||
|
else:
|
||
|
M = range(pn*p_elements,(pn+1)*p_elements)
|
||
|
for i in M:
|
||
|
for j in range(mesh_conner):
|
||
|
k = element[i].connectivity[j] - 1
|
||
|
if(stg_p[k] < 0):
|
||
|
stg_p[k] = nodecount
|
||
|
stg_n.append(L[k])
|
||
|
stg_e.append(nodecount)
|
||
|
nodecount += 1
|
||
|
else:
|
||
|
stg_e.append(stg_p[k])
|
||
|
#compute point quantity
|
||
|
n_reop = len(stg_n)
|
||
|
reop_N = range(0,len(stg_n))
|
||
|
|
||
|
|
||
|
#create and open a VTK(.vtu) files
|
||
|
if(piecenum > 1):
|
||
|
outfile = open (os.path.join(vtk_path,odbname)+'_'+stepname+'_'+instancename+'f%03d'%int(i_frame)+' '+'p'+str(pn)+'.vtu','w')
|
||
|
if(piecenum == 1):
|
||
|
outfile = open (os.path.join(vtk_path,odbname)+'_'+stepname+'_'+instancename+'f%03d'%int(i_frame)+'.vtu','w')
|
||
|
|
||
|
#<VTKFile>, including the type of mesh, version, and byte_order
|
||
|
outfile.write('<VTKFile type="UnstructuredGrid" version="0.1" byte_order="LittleEndian">'+'\n')
|
||
|
#<UnstructuredGrid>
|
||
|
outfile.write('<UnstructuredGrid>'+'\n')
|
||
|
#<Piece>, including the number of points and cells
|
||
|
if(pn == piecenum-1):
|
||
|
outfile.write('<Piece NumberOfPoints="'+str(n_reop)+'"'+' '+'NumberOfCells="'+str(lp_elements)+'">'+'\n')
|
||
|
else:
|
||
|
outfile.write('<Piece NumberOfPoints="'+str(n_reop)+'"'+' '+'NumberOfCells="'+str(p_elements)+'">'+'\n')
|
||
|
|
||
|
|
||
|
print "Writing Nodes ......"
|
||
|
#<Points> Write nodes into vtk files
|
||
|
displacement = N_Frame.fieldOutputs['U']
|
||
|
fieldValues = displacement.values
|
||
|
outfile.write('<Points>'+'\n')
|
||
|
outfile.write('<DataArray type="Float64" NumberOfComponents="3" format="ascii">'+'\n')
|
||
|
for i in reop_N:
|
||
|
nt = stg_n[i]
|
||
|
k = node[stg_n[i]].label-1
|
||
|
X,Y,Z = node[nt].coordinates[0]+L0[k][0],node[nt].coordinates[1]+L0[k][1],node[nt].coordinates[2]+L0[k][2]
|
||
|
outfile.write(' '+'%11.8e'%X+' '+'%11.8e'%Y+' '+'%11.8e'%Z+'\n')
|
||
|
outfile.write('</DataArray>'+'\n')
|
||
|
outfile.write('</Points>'+'\n')
|
||
|
#</Points>
|
||
|
|
||
|
|
||
|
print "Writing Results data ......"
|
||
|
#<PointData> Write results data into vtk files
|
||
|
outfile.write("<"+"PointData"+" "+"Tensors="+'"'+"Stress_Components,Plastic_strain_components"+'"'\
|
||
|
+" "+"Vevtors="+'"'+"Spatial_displacement,Reaction_force"+'"'\
|
||
|
+" "+"Scalars="+'"'+"Equivalent_plastic_strain,Stress_Mises,Stress_Max_Principal,Stress_Mid_Principal,Stress_Min_Principal,Stress_Pressure,Stress_Tresca,Stress_Third_Invariant,Plastic_strain_Max_Principal,Plastic_strain_Min_Principal"+'"'+">"+'\n')
|
||
|
|
||
|
#Stress components, <DataArray>
|
||
|
outfile.write("<"+"DataArray"+" "+"type="+'"'+"Float32"+'"'+" "+"Name="+'"'+"Stress_Components"+'"'+" "+"NumberOfComponents="+'"'+"9"+'"'+" "+"format="+'"'+"ascii"+'"'+">"+'\n')
|
||
|
for i in reop_N:
|
||
|
k = node[stg_n[i]].label-1
|
||
|
XX,XY,XZ,YX,YY,YZ,ZX,ZY,ZZ = L1[k][1]/L1[k][0],L1[k][4]/L1[k][0],L1[k][6]/L1[k][0],L1[k][4]/L1[k][0],L1[k][2]/L1[k][0],L1[k][5]/L1[k][0],L1[k][6]/L1[k][0],L1[k][5]/L1[k][0],L1[k][3]/L1[k][0]
|
||
|
outfile.write('%11.8e'%XX+' '+'%11.8e'%XY+' '+'%11.8e'%XZ+' '+'%11.8e'%YX+' '+'%11.8e'%YY+' '+'%11.8e'%YZ+' '+'%11.8e'%ZX+' '+'%11.8e'%ZY+' '+'%11.8e'%ZZ+'\n')
|
||
|
outfile.write("</DataArray>"+'\n')
|
||
|
#</DataArray>
|
||
|
|
||
|
#Logarithmic strain components, <DataArray>
|
||
|
outfile.write("<"+"DataArray"+" "+"type="+'"'+"Float32"+'"'+" "+"Name="+'"'+"Logarithmic_strain_components"+'"'+" "+"NumberOfComponents="+'"'+"9"+'"'+" "+"format="+'"'+"ascii"+'"'+">"+'\n')
|
||
|
for i in reop_N:
|
||
|
k = node[stg_n[i]].label-1
|
||
|
XX,XY,XZ,YX,YY,YZ,ZX,ZY,ZZ = L2[k][1]/L2[k][0],L2[k][4]/L2[k][0],L2[k][6]/L2[k][0],L2[k][4]/L2[k][0],L2[k][2]/L2[k][0],L2[k][5]/L2[k][0],L2[k][6]/L2[k][0],L2[k][5]/L2[k][0],L2[k][3]/L2[k][0]
|
||
|
outfile.write('%11.8e'%XX+' '+'%11.8e'%XY+' '+'%11.8e'%XZ+' '+'%11.8e'%YX+' '+'%11.8e'%YY+' '+'%11.8e'%YZ+' '+'%11.8e'%ZX+' '+'%11.8e'%ZY+' '+'%11.8e'%ZZ+'\n')
|
||
|
outfile.write("</DataArray>"+'\n')
|
||
|
#</DataArray>
|
||
|
|
||
|
#Plastic strain components, <DataArray>
|
||
|
outfile.write("<"+"DataArray"+" "+"type="+'"'+"Float32"+'"'+" "+"Name="+'"'+"Plastic_strain_components"+'"'+" "+"NumberOfComponents="+'"'+"9"+'"'+" "+"format="+'"'+"ascii"+'"'+">"+'\n')
|
||
|
for i in reop_N:
|
||
|
k = node[stg_n[i]].label-1
|
||
|
XX,XY,XZ,YX,YY,YZ,ZX,ZY,ZZ = L3[k][1]/L3[k][0],L3[k][4]/L3[k][0],L3[k][6]/L3[k][0],L3[k][4]/L3[k][0],L3[k][2]/L3[k][0],L3[k][5]/L3[k][0],L3[k][6]/L3[k][0],L3[k][5]/L3[k][0],L3[k][3]/L3[k][0]
|
||
|
outfile.write('%11.8e'%XX+' '+'%11.8e'%XY+' '+'%11.8e'%XZ+' '+'%11.8e'%YX+' '+'%11.8e'%YY+' '+'%11.8e'%YZ+' '+'%11.8e'%ZX+' '+'%11.8e'%ZY+' '+'%11.8e'%ZZ+'\n')
|
||
|
outfile.write("</DataArray>"+'\n')
|
||
|
#</DataArray>
|
||
|
|
||
|
#Spatial displacement, <DataArray>
|
||
|
outfile.write("<"+"DataArray"+" "+"type="+'"'+"Float32"+'"'+" "+"Name="+'"'+"Spatial_displacement"+'"'+" "+"NumberOfComponents="+'"'+"3"+'"'+" "+"format="+'"'+"ascii"+'"'+">"+'\n')
|
||
|
for i in reop_N:
|
||
|
k = node[stg_n[i]].label-1
|
||
|
X,Y,Z = L0[k][0],L0[k][1],L0[k][2]
|
||
|
outfile.write('%11.8e'%X+' '+'%11.8e'%Y+' '+'%11.8e'%Z+'\n')
|
||
|
outfile.write("</DataArray>"+'\n')
|
||
|
#</DataArray>
|
||
|
|
||
|
#Spatial acceleration, <DataArray>
|
||
|
outfile.write("<"+"DataArray"+" "+"type="+'"'+"Float32"+'"'+" "+"Name="+'"'+"Spatial_acceleration"+'"'+" "+"NumberOfComponents="+'"'+"3"+'"'+" "+"format="+'"'+"ascii"+'"'+">"+'\n')
|
||
|
for i in reop_N:
|
||
|
k = node[stg_n[i]].label-1
|
||
|
X,Y,Z = L0[k][3],L0[k][4],L0[k][5]
|
||
|
outfile.write('%11.8e'%X+' '+'%11.8e'%Y+' '+'%11.8e'%Z+'\n')
|
||
|
outfile.write("</DataArray>"+'\n')
|
||
|
#</DataArray>
|
||
|
|
||
|
#Spatial velocity, <DataArray>
|
||
|
outfile.write("<"+"DataArray"+" "+"type="+'"'+"Float32"+'"'+" "+"Name="+'"'+"Spatial_velocity"+'"'+" "+"NumberOfComponents="+'"'+"3"+'"'+" "+"format="+'"'+"ascii"+'"'+">"+'\n')
|
||
|
for i in reop_N:
|
||
|
k = node[stg_n[i]].label-1
|
||
|
X,Y,Z = L0[k][6],L0[k][7],L0[k][8]
|
||
|
outfile.write('%11.8e'%X+' '+'%11.8e'%Y+' '+'%11.8e'%Z+'\n')
|
||
|
outfile.write("</DataArray>"+'\n')
|
||
|
#</DataArray>
|
||
|
|
||
|
#Reaction force
|
||
|
outfile.write("<"+"DataArray"+" "+"type="+'"'+"Float32"+'"'+" "+"Name="+'"'+"Reaction_force"+'"'+" "+"NumberOfComponents="+'"'+"3"+'"'+" "+"format="+'"'+"ascii"+'"'+">"+'\n')
|
||
|
for i in reop_N:
|
||
|
k = node[stg_n[i]].label-1
|
||
|
X,Y,Z = L0[k][9],L0[k][10],L0[k][11]
|
||
|
outfile.write('%11.8e'%X+' '+'%11.8e'%Y+' '+'%11.8e'%Z+'\n')
|
||
|
outfile.write("</DataArray>"+'\n')
|
||
|
#</DataArray>
|
||
|
|
||
|
#Equivalent plastic strain, <DataArray>
|
||
|
outfile.write("<"+"DataArray"+" "+"type="+'"'+"Float32"+'"'+" "+"Name="+'"'+"Equivalent_plastic_strain"+'"'+" "+"format="+'"'+"ascii"+'"'+">"+'\n')
|
||
|
for i in reop_N:
|
||
|
k = node[stg_n[i]].label-1
|
||
|
X = L4[k][1]/L4[k][0]
|
||
|
outfile.write('%11.8e'%X+'\n')
|
||
|
outfile.write('</DataArray>'+'\n')
|
||
|
#</DataArray>
|
||
|
|
||
|
#Stress Mises, <DataArray>
|
||
|
outfile.write("<"+"DataArray"+" "+"type="+'"'+"Float32"+'"'+" "+"Name="+'"'+"Stress_Mises"+'"'+" "+"format="+'"'+"ascii"+'"'+">"+'\n')
|
||
|
for i in reop_N:
|
||
|
k = node[stg_n[i]].label-1
|
||
|
X = L1[k][7]/L1[k][0]
|
||
|
outfile.write('%11.8e'%X+'\n')
|
||
|
outfile.write('</DataArray>'+'\n')
|
||
|
#</DataArray>
|
||
|
|
||
|
#Stress Max.Principal, <DataArray>
|
||
|
outfile.write("<"+"DataArray"+" "+"type="+'"'+"Float32"+'"'+" "+"Name="+'"'+"Stress_Max_Principal"+'"'+" "+"format="+'"'+"ascii"+'"'+">"+'\n')
|
||
|
for i in reop_N:
|
||
|
k = node[stg_n[i]].label-1
|
||
|
X = L1[k][8]/L1[k][0]
|
||
|
outfile.write('%11.8e'%X+'\n')
|
||
|
outfile.write('</DataArray>'+'\n')
|
||
|
#</DataArray>
|
||
|
|
||
|
#Stress Mid.Principal, <DataArray>
|
||
|
outfile.write("<"+"DataArray"+" "+"type="+'"'+"Float32"+'"'+" "+"Name="+'"'+"Stress_Mid_Principal"+'"'+" "+"format="+'"'+"ascii"+'"'+">"+'\n')
|
||
|
for i in reop_N:
|
||
|
k = node[stg_n[i]].label-1
|
||
|
X = L1[k][9]/L1[k][0]
|
||
|
outfile.write('%11.8e'%X+'\n')
|
||
|
outfile.write('</DataArray>'+'\n')
|
||
|
#</DataArray>
|
||
|
|
||
|
#Stress Min.Principal, <DataArray>
|
||
|
outfile.write("<"+"DataArray"+" "+"type="+'"'+"Float32"+'"'+" "+"Name="+'"'+"Stress_Min_Principal"+'"'+" "+"format="+'"'+"ascii"+'"'+">"+'\n')
|
||
|
for i in reop_N:
|
||
|
k = node[stg_n[i]].label-1
|
||
|
X = L1[k][10]/L1[k][0]
|
||
|
outfile.write('%11.8e'%X+'\n')
|
||
|
outfile.write('</DataArray>'+'\n')
|
||
|
#</DataArray>
|
||
|
|
||
|
#Stress Pressure, <DataArray>
|
||
|
outfile.write("<"+"DataArray"+" "+"type="+'"'+"Float32"+'"'+" "+"Name="+'"'+"Stress_Pressure"+'"'+" "+"format="+'"'+"ascii"+'"'+">"+'\n')
|
||
|
for i in reop_N:
|
||
|
k = node[stg_n[i]].label-1
|
||
|
X = L1[k][11]/L1[k][0]
|
||
|
outfile.write('%11.8e'%X+'\n')
|
||
|
outfile.write('</DataArray>'+'\n')
|
||
|
#</DataArray>
|
||
|
|
||
|
#Stress Tresca, <DataArray>
|
||
|
outfile.write("<"+"DataArray"+" "+"type="+'"'+"Float32"+'"'+" "+"Name="+'"'+"Stress_Tresca"+'"'+" "+"format="+'"'+"ascii"+'"'+">"+'\n')
|
||
|
for i in reop_N:
|
||
|
k = node[stg_n[i]].label-1
|
||
|
X = L1[k][12]/L1[k][0]
|
||
|
outfile.write('%11.8e'%X+'\n')
|
||
|
outfile.write('</DataArray>'+'\n')
|
||
|
#</DataArray>
|
||
|
|
||
|
#Stress Third_Invariant, <DataArray>
|
||
|
outfile.write("<"+"DataArray"+" "+"type="+'"'+"Float32"+'"'+" "+"Name="+'"'+"Stress_Third_Invariant"+'"'+" "+"format="+'"'+"ascii"+'"'+">"+'\n')
|
||
|
for i in reop_N:
|
||
|
k = node[stg_n[i]].label-1
|
||
|
X = L1[k][13]/L1[k][0]
|
||
|
outfile.write('%11.8e'%X+'\n')
|
||
|
outfile.write('</DataArray>'+'\n')
|
||
|
#</DataArray>
|
||
|
|
||
|
#Logarithmic_strain_Max_Principal, <DataArray>
|
||
|
outfile.write("<"+"DataArray"+" "+"type="+'"'+"Float32"+'"'+" "+"Name="+'"'+"Logarithmic_strain_Max_Principal"+'"'+" "+"format="+'"'+"ascii"+'"'+">"+'\n')
|
||
|
for i in reop_N:
|
||
|
k = node[stg_n[i]].label-1
|
||
|
X = L2[k][7]/L2[k][0]
|
||
|
outfile.write('%11.8e'%X+'\n')
|
||
|
outfile.write('</DataArray>'+'\n')
|
||
|
#</DataArray>
|
||
|
|
||
|
#Logarithmic strain Min.Principal, <DataArray>
|
||
|
outfile.write("<"+"DataArray"+" "+"type="+'"'+"Float32"+'"'+" "+"Name="+'"'+"Logarithmic_strain_Min_Principal"+'"'+" "+"format="+'"'+"ascii"+'"'+">"+'\n')
|
||
|
for i in reop_N:
|
||
|
k = node[stg_n[i]].label-1
|
||
|
X = L2[k][8]/L2[k][0]
|
||
|
outfile.write('%11.8e'%X+'\n')
|
||
|
outfile.write('</DataArray>'+'\n')
|
||
|
#</DataArray>'''
|
||
|
|
||
|
#Plastic strain Max.Principal, <DataArray>
|
||
|
outfile.write("<"+"DataArray"+" "+"type="+'"'+"Float32"+'"'+" "+"Name="+'"'+"Plastic_strain_Max_Principal"+'"'+" "+"format="+'"'+"ascii"+'"'+">"+'\n')
|
||
|
for i in reop_N:
|
||
|
k = node[stg_n[i]].label-1
|
||
|
X = L3[k][7]/L3[k][0]
|
||
|
outfile.write('%11.8e'%X+'\n')
|
||
|
outfile.write('</DataArray>'+'\n')
|
||
|
#</DataArray>
|
||
|
|
||
|
#Plastic strain Min.Principal, <DataArray>
|
||
|
outfile.write("<"+"DataArray"+" "+"type="+'"'+"Float32"+'"'+" "+"Name="+'"'+"Plastic_strain_Min_Principal"+'"'+" "+"format="+'"'+"ascii"+'"'+">"+'\n')
|
||
|
for i in reop_N:
|
||
|
k = node[stg_n[i]].label-1
|
||
|
X = L3[k][8]/L3[k][0]
|
||
|
outfile.write('%11.8e'%X+'\n')
|
||
|
outfile.write('</DataArray>'+'\n')
|
||
|
#</DataArray>
|
||
|
|
||
|
outfile.write("</PointData>"+'\n')
|
||
|
#</PointData>
|
||
|
|
||
|
|
||
|
print "Writing Cells ......"
|
||
|
#<Cells> Write cells into vtk files
|
||
|
outfile.write('<Cells>'+'\n')
|
||
|
#Connectivity
|
||
|
outfile.write('<DataArray type="Int32" Name="connectivity" format="ascii">'+'\n')
|
||
|
if (mesh_type == 12):
|
||
|
for i in range(len(stg_e)/8):
|
||
|
outfile.write(str(stg_e[i*8])+' '+str(stg_e[i*8+1])+' '+str(stg_e[i*8+2])+' '+str(stg_e[i*8+3])+' '+str(stg_e[i*8+4])+' '+str(stg_e[i*8+5])+' '+str(stg_e[i*8+6])+' '+str(stg_e[i*8+7])+'\n')
|
||
|
if (mesh_type == 10):
|
||
|
for i in range(len(stg_e)/4):
|
||
|
outfile.write(str(stg_e[i*4])+' '+str(stg_e[i*4+1])+' '+str(stg_e[i*4+2])+' '+str(stg_e[i*4+3])+'\n')
|
||
|
outfile.write('</DataArray>'+'\n')
|
||
|
#Offsets
|
||
|
outfile.write('<DataArray type="Int32" Name="offsets" format="ascii">'+'\n')
|
||
|
for i in range(len(stg_e)/mesh_conner):
|
||
|
outfile.write(str(i*mesh_conner+mesh_conner)+'\n')
|
||
|
outfile.write('</DataArray>'+'\n')
|
||
|
#Type
|
||
|
outfile.write('<DataArray type="UInt8" Name="types" format="ascii">'+'\n')
|
||
|
for i in range(len(stg_e)/mesh_conner):
|
||
|
outfile.write(str(mesh_type)+'\n')
|
||
|
outfile.write('</DataArray>'+'\n')
|
||
|
outfile.write('</Cells>'+'\n')
|
||
|
#</Cells>
|
||
|
|
||
|
|
||
|
#</Piece>
|
||
|
outfile.write('</Piece>'+'\n')
|
||
|
#</UnstructuredGrid>
|
||
|
outfile.write('</UnstructuredGrid>'+'\n')
|
||
|
#</VTKFile>
|
||
|
outfile.write('</VTKFile>'+'\n')
|
||
|
|
||
|
outfile.close()
|
||
|
print "Time elapsed: ", time() - time1, "s"
|
||
|
|
||
|
'''====================================================================='''
|
||
|
print "Creating .pvtu file for frame ", i_frame," ......"
|
||
|
#create .pvtu files for parallel visualization
|
||
|
if ( piecenum > 1 ):
|
||
|
outfile = open (os.path.join(vtk_path,odbname)+'_'+stepname+'_'+'f%03d'%int(i_frame)+'.pvtu','w')
|
||
|
|
||
|
#write the basic information for .pvtu files
|
||
|
outfile.write('<?xml version="1.0"?>'+'\n')
|
||
|
outfile.write('<VTKFile type="PUnstructuredGrid" version="0.1" byte_order="LittleEndian">'+'\n')
|
||
|
outfile.write("<PUnstructuredGrid GhostLevel="+'"'+str(piecenum)+'"'+">"+'\n')
|
||
|
#pointdata
|
||
|
outfile.write("<"+"PPointData"+" "+"Tensors="+'"'+"Stress_Components,Logarithmic_strain_components,Plastic_strain_components"+'"'\
|
||
|
+" "+"Vevtors="+'"'+"Spatial_displacement,Spatial_acceleration,Spatial_velocity,Reaction_force"+'"'\
|
||
|
+" "+"Scalars="+'"'+"Equivalent_plastic_strain,Stress_Mises,Stress_Max_Principal,Stress_Mid_Principal,Stress_Min_Principal,Stress_Pressure,Stress_Tresca,Stress_Third_Invariant,Logarithmic_strain_Max_Principal,Logarithmic_strain_Min_Principal,Plastic_strain_Max_Principal,Plastic_strain_Min_Principal"+'"'+">"+'\n')
|
||
|
|
||
|
outfile.write("<"+"PDataArray"+" "+"type="+'"'+"Float32"+'"'+" "+"Name="+'"'+"Stress_Components"+'"'+" "+"NumberOfComponents="+'"'+"9"+'"'+" "+"/>"+'\n')
|
||
|
outfile.write("<"+"PDataArray"+" "+"type="+'"'+"Float32"+'"'+" "+"Name="+'"'+"Logarithmic_strain_components"+'"'+" "+"NumberOfComponents="+'"'+"9"+'"'+" "+"/>"+'\n')
|
||
|
outfile.write("<"+"PDataArray"+" "+"type="+'"'+"Float32"+'"'+" "+"Name="+'"'+"Plastic_strain_components"+'"'+" "+"NumberOfComponents="+'"'+"9"+'"'+" "+"/>"+'\n')
|
||
|
outfile.write("<"+"PDataArray"+" "+"type="+'"'+"Float32"+'"'+" "+"Name="+'"'+"Spatial_displacement"+'"'+" "+"NumberOfComponents="+'"'+"3"+'"'+" "+"/>"+'\n')
|
||
|
outfile.write("<"+"PDataArray"+" "+"type="+'"'+"Float32"+'"'+" "+"Name="+'"'+"Spatial_acceleration"+'"'+" "+"NumberOfComponents="+'"'+"3"+'"'+" "+"/>"+'\n')
|
||
|
outfile.write("<"+"PDataArray"+" "+"type="+'"'+"Float32"+'"'+" "+"Name="+'"'+"Spatial_velocity"+'"'+" "+"NumberOfComponents="+'"'+"3"+'"'+" "+"/>"+'\n')
|
||
|
outfile.write("<"+"PDataArray"+" "+"type="+'"'+"Float32"+'"'+" "+"Name="+'"'+"Reaction_force"+'"'+" "+"NumberOfComponents="+'"'+"3"+'"'+" "+"/>"+'\n')
|
||
|
outfile.write("<"+"PDataArray"+" "+"type="+'"'+"Float32"+'"'+" "+"Name="+'"'+"Equivalent_plastic_strain"+'"'+" "+"/>"+'\n')
|
||
|
outfile.write("<"+"PDataArray"+" "+"type="+'"'+"Float32"+'"'+" "+"Name="+'"'+"Stress_Mises"+'"'+" "+"/>"+'\n')
|
||
|
outfile.write("<"+"PDataArray"+" "+"type="+'"'+"Float32"+'"'+" "+"Name="+'"'+"Stress_Max_Principal"+'"'+" "+"/>"+'\n')
|
||
|
outfile.write("<"+"PDataArray"+" "+"type="+'"'+"Float32"+'"'+" "+"Name="+'"'+"Stress_Mid_Principal"+'"'+" "+"/>"+'\n')
|
||
|
outfile.write("<"+"PDataArray"+" "+"type="+'"'+"Float32"+'"'+" "+"Name="+'"'+"Stress_Min_Principal"+'"'+" "+"/>"+'\n')
|
||
|
outfile.write("<"+"PDataArray"+" "+"type="+'"'+"Float32"+'"'+" "+"Name="+'"'+"Stress_Pressure"+'"'+" "+"/>"+'\n')
|
||
|
outfile.write("<"+"PDataArray"+" "+"type="+'"'+"Float32"+'"'+" "+"Name="+'"'+"Stress_Tresca"+'"'+" "+"/>"+'\n')
|
||
|
outfile.write("<"+"PDataArray"+" "+"type="+'"'+"Float32"+'"'+" "+"Name="+'"'+"Stress_Third_Invariant"+'"'+" "+"/>"+'\n')
|
||
|
outfile.write("<"+"PDataArray"+" "+"type="+'"'+"Float32"+'"'+" "+"Name="+'"'+"Logarithmic_strain_Max_Principal"+'"'+" "+"/>"+'\n')
|
||
|
outfile.write("<"+"PDataArray"+" "+"type="+'"'+"Float32"+'"'+" "+"Name="+'"'+"Logarithmic_strain_Min_Principal"+'"'+" "+"/>"+'\n')
|
||
|
outfile.write("<"+"PDataArray"+" "+"type="+'"'+"Float32"+'"'+" "+"Name="+'"'+"Plastic_strain_Max_Principal"+'"'+" "+"/>"+'\n')
|
||
|
outfile.write("<"+"PDataArray"+" "+"type="+'"'+"Float32"+'"'+" "+"Name="+'"'+"Plastic_strain_Min_Principal"+'"'+" "+"/>"+'\n')
|
||
|
outfile.write("</PPointData>"+'\n')
|
||
|
|
||
|
#points
|
||
|
outfile.write("<PPoints>"+'\n')
|
||
|
outfile.write("<PDataArray type="+'"'+"Float64"+'"'+" "+"NumberOfComponents="+'"'+"3"+'"'+"/>"+'\n')
|
||
|
outfile.write("</PPoints>"+'\n')
|
||
|
|
||
|
#write the path of each piece for reading it through the .pvtu file
|
||
|
for pn in range(piecenum):
|
||
|
outfile.write("<Piece Source="+'"'+odbname+'_'+stepname+'_'+instancename+'f%03d'%int(i_frame)+' '+'p'+str(pn)+'.vtu'+'"'+"/>"+'\n')
|
||
|
|
||
|
outfile.write("</PUnstructuredGrid>"+'\n')
|
||
|
outfile.write("</VTKFile>")
|
||
|
|
||
|
outfile.close()
|
||
|
|
||
|
odb.close()
|
||
|
|
||
|
print "Total time elapsed: ", time() - starttime, "s"
|
||
|
|
||
|
#Function 2
|
||
|
def ConvertOdb2VtkP(Odbpath = ' ',Odbname = ' ',vtkpath = ' ',MeshType = ' ',Piecenum = ' ', BeginFrame = ' ', EndFrame =' ', Steps =' ', Instances = ' '):
|
||
|
|
||
|
starttime = time()
|
||
|
|
||
|
#get odb file's path
|
||
|
odb_path = Odbpath
|
||
|
#get odb file name
|
||
|
odbname = Odbname
|
||
|
#get the output files' path
|
||
|
vtk_path = vtkpath
|
||
|
#get the mesh type
|
||
|
mesh_type = int(MeshType)
|
||
|
mesh_conner = 0
|
||
|
if (mesh_type == 12):
|
||
|
mesh_conner = 8
|
||
|
mesh_name = "Hexahedron"
|
||
|
if (mesh_type == 10):
|
||
|
mesh_conner = 4
|
||
|
mesh_name = "Tetra"
|
||
|
if (mesh_conner == 0):
|
||
|
print "Mesh type error or unidentified"
|
||
|
os._exit(0)
|
||
|
#get the quantity of pieces to partition
|
||
|
piecenum = int(Piecenum)
|
||
|
#get the frame
|
||
|
input_frame = range(int(BeginFrame),int(EndFrame)+1)
|
||
|
#get the step
|
||
|
input_step = Steps.split(",")
|
||
|
#get the instance
|
||
|
input_instance = Instances.split(",")
|
||
|
|
||
|
#display the read parameter
|
||
|
print "Basic Information:"
|
||
|
print "Model:",odbname,"; Mesh type:",mesh_name,"; Number of blocks:",piecenum
|
||
|
print "Convert frames: ",input_frame[0]," to ",input_frame[-1]
|
||
|
print "Step & Instance : ",str(input_step),", ",str(input_instance)
|
||
|
|
||
|
#open an ODB ( Abaqus output database )
|
||
|
odb = openOdb(os.path.join(odb_path,odbname)+'.odb',readOnly=True)
|
||
|
print "ODB opened"
|
||
|
|
||
|
#access geometry and topology information ( odb->rootAssembly->instances->(nodes, elements) )
|
||
|
rootassembly = odb.rootAssembly
|
||
|
instance = rootassembly.instances
|
||
|
#access attribute information
|
||
|
step = odb.steps
|
||
|
#get instance & step information : Quantity and all names
|
||
|
allinstancestr = str(instance)
|
||
|
autoins = allinstancestr.split("'")
|
||
|
inslen = len(autoins)/4
|
||
|
instance_N = range(0,inslen)
|
||
|
allstepstr = str(step)
|
||
|
autostep = allstepstr.split("'")
|
||
|
steplen = len(autostep)/4
|
||
|
step_N = range(0,steplen)
|
||
|
|
||
|
for i in input_step:
|
||
|
if(steplen < int(i)):
|
||
|
print "input step exceeds the range of steps"
|
||
|
os._exit(0)
|
||
|
for i in input_instance:
|
||
|
if(inslen < int(i)):
|
||
|
print "input instance exceeds the range of instances"
|
||
|
os._exit(0)
|
||
|
|
||
|
|
||
|
#step cycle
|
||
|
for step_i in input_step:
|
||
|
n = int(step_i)*4+1
|
||
|
stepname = autostep[n]
|
||
|
print "Step: ",stepname
|
||
|
#instance cycle
|
||
|
for ins_i in input_instance:
|
||
|
n = int(ins_i)*4+1
|
||
|
instancename = autoins[n]
|
||
|
print "Instance: ",instancename
|
||
|
|
||
|
#access nodes & elements
|
||
|
node = instance[instancename].nodes
|
||
|
element = instance[instancename].elements
|
||
|
n_nodes = len(node)
|
||
|
n_elements = len(element)
|
||
|
#access attribute(fieldOutputs) information
|
||
|
frame = step[stepname].frames
|
||
|
|
||
|
|
||
|
#compute the number of element of each block
|
||
|
p_elements = n_elements/piecenum + 1
|
||
|
lp_elements = n_elements - (p_elements*(piecenum-1)) #last block
|
||
|
|
||
|
|
||
|
#match nodes' label and its order in sequence (for empty nodes in tetra mesh)
|
||
|
MLN = node[n_nodes-1].label
|
||
|
TOTAL=[]
|
||
|
#read node in sequence, and get the largest label of node(non-empty)
|
||
|
#MLN is the max label of nodeset
|
||
|
for i in node:
|
||
|
TOTAL.append(i.label)
|
||
|
if(i.label > MLN):
|
||
|
MLN = i.label
|
||
|
#match (the key)
|
||
|
L=[]
|
||
|
n = 0
|
||
|
for i in range(MLN):
|
||
|
L.append(0)
|
||
|
for i in TOTAL:
|
||
|
L[i-1] = n
|
||
|
n += 1
|
||
|
|
||
|
|
||
|
#frame cycle
|
||
|
for i_frame in input_frame:
|
||
|
|
||
|
#Detect whether the input frame is out of range
|
||
|
try:
|
||
|
TRY = odb.steps[stepname].frames[int(i_frame)]
|
||
|
except:
|
||
|
print "input frame exceeds the range of frames"
|
||
|
os._exit(0)
|
||
|
break
|
||
|
|
||
|
#Access a frame
|
||
|
N_Frame = odb.steps[stepname].frames[int(i_frame)]
|
||
|
print "Frame:",i_frame
|
||
|
|
||
|
#create array for store result data temporarily
|
||
|
# Vector-U,A,V,RF
|
||
|
L0=[]
|
||
|
# Tensors-S
|
||
|
L1=[]
|
||
|
# Tensors-LE
|
||
|
L2=[]
|
||
|
# Tensors-PE
|
||
|
L3=[]
|
||
|
# Scalars-PEEQ
|
||
|
L4=[]
|
||
|
for i in range(MLN):
|
||
|
L0.append([0,0,0,0,0,0,0,0,0,0,0,0])
|
||
|
L1.append([0,0,0,0,0,0,0,0,0,0,0,0,0,0])
|
||
|
L2.append([0,0,0,0,0,0,0,0,0])
|
||
|
L3.append([0,0,0,0,0,0,0,0,0])
|
||
|
L4.append([0,0])
|
||
|
|
||
|
|
||
|
print "Reading U, A, V, RF ......"
|
||
|
time1 = time()
|
||
|
#Access Spatial displacement
|
||
|
displacement = N_Frame.fieldOutputs['U']
|
||
|
fieldValues = displacement.values
|
||
|
for valueX in fieldValues :
|
||
|
i = valueX.nodeLabel
|
||
|
L0[i-1][0] = valueX.data[0]
|
||
|
L0[i-1][1] = valueX.data[1]
|
||
|
L0[i-1][2] = valueX.data[2]
|
||
|
#Access Spatial acceleration
|
||
|
acceleration = N_Frame.fieldOutputs['A']
|
||
|
fieldValues = acceleration.values
|
||
|
for valueX in fieldValues :
|
||
|
i = valueX.nodeLabel
|
||
|
L0[i-1][3] = valueX.data[0]
|
||
|
L0[i-1][4] = valueX.data[1]
|
||
|
L0[i-1][5] = valueX.data[2]
|
||
|
#Access Spatial velocity
|
||
|
velocity = N_Frame.fieldOutputs['V']
|
||
|
fieldValues = velocity.values
|
||
|
for valueX in fieldValues :
|
||
|
i = valueX.nodeLabel
|
||
|
L0[i-1][6] = valueX.data[0]
|
||
|
L0[i-1][7] = valueX.data[1]
|
||
|
L0[i-1][8] = valueX.data[2]
|
||
|
#Access Reaction force
|
||
|
Reaction_force = N_Frame.fieldOutputs['RF']
|
||
|
fieldValues = Reaction_force.values
|
||
|
for valueX in fieldValues :
|
||
|
i = valueX.nodeLabel
|
||
|
L0[i-1][9] = valueX.data[0]
|
||
|
L0[i-1][10] = valueX.data[1]
|
||
|
L0[i-1][11] = valueX.data[2]
|
||
|
print "Time elapsed: ", time() - time1, "s"
|
||
|
print "Reading Stress ......"
|
||
|
time1 = time()
|
||
|
#access Stress components
|
||
|
Stress = N_Frame.fieldOutputs['S']
|
||
|
node_Stress = Stress.getSubset(position=ELEMENT_NODAL)
|
||
|
fieldValues = node_Stress.values
|
||
|
for valueX in fieldValues :
|
||
|
L1[valueX.nodeLabel-1][0] += 1
|
||
|
L1[valueX.nodeLabel-1][1] += valueX.data[0]
|
||
|
L1[valueX.nodeLabel-1][2] += valueX.data[1]
|
||
|
L1[valueX.nodeLabel-1][3] += valueX.data[2]
|
||
|
L1[valueX.nodeLabel-1][4] += valueX.data[3]
|
||
|
L1[valueX.nodeLabel-1][5] += valueX.data[4]
|
||
|
L1[valueX.nodeLabel-1][6] += valueX.data[5]
|
||
|
L1[valueX.nodeLabel-1][7] += valueX.mises
|
||
|
L1[valueX.nodeLabel-1][8] += valueX.maxPrincipal
|
||
|
L1[valueX.nodeLabel-1][9] += valueX.midPrincipal
|
||
|
L1[valueX.nodeLabel-1][10] += valueX.minPrincipal
|
||
|
L1[valueX.nodeLabel-1][11] += valueX.press
|
||
|
L1[valueX.nodeLabel-1][12] += valueX.tresca
|
||
|
L1[valueX.nodeLabel-1][13] += valueX.inv3
|
||
|
# can first ave
|
||
|
print "Time elapsed: ", time() - time1, "s"
|
||
|
print "Reading Logarithmic strain ......"
|
||
|
time1 = time()
|
||
|
#Logarithmic strain components
|
||
|
Logarithmic_strain = N_Frame.fieldOutputs['LE']
|
||
|
node_Logarithmic_strain = Logarithmic_strain.getSubset(position=ELEMENT_NODAL)
|
||
|
fieldValues = node_Logarithmic_strain.values
|
||
|
for valueX in fieldValues :
|
||
|
L2[valueX.nodeLabel-1][0] += 1
|
||
|
L2[valueX.nodeLabel-1][1] += valueX.data[0]
|
||
|
L2[valueX.nodeLabel-1][2] += valueX.data[1]
|
||
|
L2[valueX.nodeLabel-1][3] += valueX.data[2]
|
||
|
L2[valueX.nodeLabel-1][4] += valueX.data[3]
|
||
|
L2[valueX.nodeLabel-1][5] += valueX.data[4]
|
||
|
L2[valueX.nodeLabel-1][6] += valueX.data[5]
|
||
|
L2[valueX.nodeLabel-1][7] += valueX.maxPrincipal
|
||
|
L2[valueX.nodeLabel-1][8] += valueX.minPrincipal
|
||
|
print "Time elapsed: ", time() - time1, "s"
|
||
|
print "Reading Plastic strain ......"
|
||
|
time1 = time()
|
||
|
#Plastic strain components
|
||
|
Plastic_strain = N_Frame.fieldOutputs['PE']
|
||
|
node_Plastic_strain = Plastic_strain.getSubset(position=ELEMENT_NODAL)
|
||
|
fieldValues = node_Plastic_strain.values
|
||
|
for valueX in fieldValues :
|
||
|
L3[valueX.nodeLabel-1][0] += 1
|
||
|
L3[valueX.nodeLabel-1][1] += valueX.data[0]
|
||
|
L3[valueX.nodeLabel-1][2] += valueX.data[1]
|
||
|
L3[valueX.nodeLabel-1][3] += valueX.data[2]
|
||
|
L3[valueX.nodeLabel-1][4] += valueX.data[3]
|
||
|
L3[valueX.nodeLabel-1][5] += valueX.data[4]
|
||
|
L3[valueX.nodeLabel-1][6] += valueX.data[5]
|
||
|
L3[valueX.nodeLabel-1][7] += valueX.maxPrincipal
|
||
|
L3[valueX.nodeLabel-1][8] += valueX.minPrincipal
|
||
|
print "Time elapsed: ", time() - time1, "s"
|
||
|
print "Reading Equivalent plastic strain ......"
|
||
|
time1 = time()
|
||
|
#Equivalent plastic strain
|
||
|
Equivalent_plastic_strain = N_Frame.fieldOutputs['PEEQ']
|
||
|
node_Equivalent_plastic_strain = Equivalent_plastic_strain.getSubset(position=ELEMENT_NODAL)
|
||
|
fieldValues = node_Equivalent_plastic_strain.values
|
||
|
for valueX in fieldValues :
|
||
|
L4[valueX.nodeLabel-1][0] += 1
|
||
|
L4[valueX.nodeLabel-1][1] += valueX.data
|
||
|
print "Time elapsed: ", time() - time1, "s"
|
||
|
|
||
|
|
||
|
'''============================================================'''
|
||
|
|
||
|
print "Partitionning model and writing vtk files ......"
|
||
|
#piece cycle, to partion the model and create each piece for vtk files
|
||
|
for pn in range(piecenum):
|
||
|
time1 = time()
|
||
|
print "frame:",i_frame,"; block:",pn
|
||
|
#Reorganization
|
||
|
#Control&Storage
|
||
|
#estimate whether the node has already existed
|
||
|
stg_p = []
|
||
|
#store the reorganized node for element
|
||
|
stg_e = []
|
||
|
#store the reorganized node for node
|
||
|
stg_n = []
|
||
|
for i in range(MLN):
|
||
|
stg_p.append(-1)
|
||
|
nodecount = 0
|
||
|
#reorganize the node and element (reconstruct the mesh)
|
||
|
if(pn == piecenum-1):
|
||
|
M = range(pn*p_elements,n_elements)
|
||
|
else:
|
||
|
M = range(pn*p_elements,(pn+1)*p_elements)
|
||
|
for i in M:
|
||
|
for j in range(mesh_conner):
|
||
|
k = element[i].connectivity[j] - 1
|
||
|
if(stg_p[k] < 0):
|
||
|
stg_p[k] = nodecount
|
||
|
stg_n.append(L[k])
|
||
|
stg_e.append(nodecount)
|
||
|
nodecount += 1
|
||
|
else:
|
||
|
stg_e.append(stg_p[k])
|
||
|
#compute point quantity
|
||
|
n_reop = len(stg_n)
|
||
|
reop_N = range(0,len(stg_n))
|
||
|
|
||
|
|
||
|
#create and open a VTK(.vtu) files
|
||
|
if(piecenum > 1):
|
||
|
outfile = open (os.path.join(vtk_path,odbname)+'_'+stepname+'_'+instancename+'f%03d'%int(i_frame)+' '+'p'+str(pn)+'.vtu','w')
|
||
|
if(piecenum == 1):
|
||
|
outfile = open (os.path.join(vtk_path,odbname)+'_'+stepname+'_'+instancename+'f%03d'%int(i_frame)+'.vtu','w')
|
||
|
|
||
|
#<VTKFile>, including the type of mesh, version, and byte_order
|
||
|
outfile.write('<VTKFile type="UnstructuredGrid" version="0.1" byte_order="LittleEndian">'+'\n')
|
||
|
#<UnstructuredGrid>
|
||
|
outfile.write('<UnstructuredGrid>'+'\n')
|
||
|
#<Piece>, including the number of points and cells
|
||
|
if(pn == piecenum-1):
|
||
|
outfile.write('<Piece NumberOfPoints="'+str(n_reop)+'"'+' '+'NumberOfCells="'+str(lp_elements)+'">'+'\n')
|
||
|
else:
|
||
|
outfile.write('<Piece NumberOfPoints="'+str(n_reop)+'"'+' '+'NumberOfCells="'+str(p_elements)+'">'+'\n')
|
||
|
|
||
|
|
||
|
print "Writing Nodes ......"
|
||
|
#<Points> Write nodes into vtk files
|
||
|
displacement = N_Frame.fieldOutputs['U']
|
||
|
fieldValues = displacement.values
|
||
|
outfile.write('<Points>'+'\n')
|
||
|
outfile.write('<DataArray type="Float64" NumberOfComponents="3" format="ascii">'+'\n')
|
||
|
for i in reop_N:
|
||
|
nt = stg_n[i]
|
||
|
k = node[stg_n[i]].label-1
|
||
|
X,Y,Z = node[nt].coordinates[0]+L0[k][0],node[nt].coordinates[1]+L0[k][1],node[nt].coordinates[2]+L0[k][2]
|
||
|
outfile.write(' '+'%11.8e'%X+' '+'%11.8e'%Y+' '+'%11.8e'%Z+'\n')
|
||
|
outfile.write('</DataArray>'+'\n')
|
||
|
outfile.write('</Points>'+'\n')
|
||
|
#</Points>
|
||
|
|
||
|
|
||
|
print "Writing Results data ......"
|
||
|
#<PointData> Write results data into vtk files
|
||
|
outfile.write("<"+"PointData"+" "+"Tensors="+'"'+"Stress_Components,Plastic_strain_components"+'"'\
|
||
|
+" "+"Vevtors="+'"'+"Spatial_displacement,Reaction_force"+'"'\
|
||
|
+" "+"Scalars="+'"'+"Equivalent_plastic_strain,Stress_Mises,Stress_Max_Principal,Stress_Mid_Principal,Stress_Min_Principal,Stress_Pressure,Stress_Tresca,Stress_Third_Invariant,Plastic_strain_Max_Principal,Plastic_strain_Min_Principal"+'"'+">"+'\n')
|
||
|
|
||
|
#Stress components, <DataArray>
|
||
|
outfile.write("<"+"DataArray"+" "+"type="+'"'+"Float32"+'"'+" "+"Name="+'"'+"Stress_Components"+'"'+" "+"NumberOfComponents="+'"'+"9"+'"'+" "+"format="+'"'+"ascii"+'"'+">"+'\n')
|
||
|
for i in reop_N:
|
||
|
k = node[stg_n[i]].label-1
|
||
|
XX,XY,XZ,YX,YY,YZ,ZX,ZY,ZZ = L1[k][1]/L1[k][0],L1[k][4]/L1[k][0],L1[k][6]/L1[k][0],L1[k][4]/L1[k][0],L1[k][2]/L1[k][0],L1[k][5]/L1[k][0],L1[k][6]/L1[k][0],L1[k][5]/L1[k][0],L1[k][3]/L1[k][0]
|
||
|
outfile.write('%11.8e'%XX+' '+'%11.8e'%XY+' '+'%11.8e'%XZ+' '+'%11.8e'%YX+' '+'%11.8e'%YY+' '+'%11.8e'%YZ+' '+'%11.8e'%ZX+' '+'%11.8e'%ZY+' '+'%11.8e'%ZZ+'\n')
|
||
|
outfile.write("</DataArray>"+'\n')
|
||
|
#</DataArray>
|
||
|
|
||
|
#Logarithmic strain components, <DataArray>
|
||
|
outfile.write("<"+"DataArray"+" "+"type="+'"'+"Float32"+'"'+" "+"Name="+'"'+"Logarithmic_strain_components"+'"'+" "+"NumberOfComponents="+'"'+"9"+'"'+" "+"format="+'"'+"ascii"+'"'+">"+'\n')
|
||
|
for i in reop_N:
|
||
|
k = node[stg_n[i]].label-1
|
||
|
XX,XY,XZ,YX,YY,YZ,ZX,ZY,ZZ = L2[k][1]/L2[k][0],L2[k][4]/L2[k][0],L2[k][6]/L2[k][0],L2[k][4]/L2[k][0],L2[k][2]/L2[k][0],L2[k][5]/L2[k][0],L2[k][6]/L2[k][0],L2[k][5]/L2[k][0],L2[k][3]/L2[k][0]
|
||
|
outfile.write('%11.8e'%XX+' '+'%11.8e'%XY+' '+'%11.8e'%XZ+' '+'%11.8e'%YX+' '+'%11.8e'%YY+' '+'%11.8e'%YZ+' '+'%11.8e'%ZX+' '+'%11.8e'%ZY+' '+'%11.8e'%ZZ+'\n')
|
||
|
outfile.write("</DataArray>"+'\n')
|
||
|
#</DataArray>
|
||
|
|
||
|
#Plastic strain components, <DataArray>
|
||
|
outfile.write("<"+"DataArray"+" "+"type="+'"'+"Float32"+'"'+" "+"Name="+'"'+"Plastic_strain_components"+'"'+" "+"NumberOfComponents="+'"'+"9"+'"'+" "+"format="+'"'+"ascii"+'"'+">"+'\n')
|
||
|
for i in reop_N:
|
||
|
k = node[stg_n[i]].label-1
|
||
|
XX,XY,XZ,YX,YY,YZ,ZX,ZY,ZZ = L3[k][1]/L3[k][0],L3[k][4]/L3[k][0],L3[k][6]/L3[k][0],L3[k][4]/L3[k][0],L3[k][2]/L3[k][0],L3[k][5]/L3[k][0],L3[k][6]/L3[k][0],L3[k][5]/L3[k][0],L3[k][3]/L3[k][0]
|
||
|
outfile.write('%11.8e'%XX+' '+'%11.8e'%XY+' '+'%11.8e'%XZ+' '+'%11.8e'%YX+' '+'%11.8e'%YY+' '+'%11.8e'%YZ+' '+'%11.8e'%ZX+' '+'%11.8e'%ZY+' '+'%11.8e'%ZZ+'\n')
|
||
|
outfile.write("</DataArray>"+'\n')
|
||
|
#</DataArray>
|
||
|
|
||
|
#Spatial displacement, <DataArray>
|
||
|
outfile.write("<"+"DataArray"+" "+"type="+'"'+"Float32"+'"'+" "+"Name="+'"'+"Spatial_displacement"+'"'+" "+"NumberOfComponents="+'"'+"3"+'"'+" "+"format="+'"'+"ascii"+'"'+">"+'\n')
|
||
|
for i in reop_N:
|
||
|
k = node[stg_n[i]].label-1
|
||
|
X,Y,Z = L0[k][0],L0[k][1],L0[k][2]
|
||
|
outfile.write('%11.8e'%X+' '+'%11.8e'%Y+' '+'%11.8e'%Z+'\n')
|
||
|
outfile.write("</DataArray>"+'\n')
|
||
|
#</DataArray>
|
||
|
|
||
|
#Spatial acceleration, <DataArray>
|
||
|
outfile.write("<"+"DataArray"+" "+"type="+'"'+"Float32"+'"'+" "+"Name="+'"'+"Spatial_acceleration"+'"'+" "+"NumberOfComponents="+'"'+"3"+'"'+" "+"format="+'"'+"ascii"+'"'+">"+'\n')
|
||
|
for i in reop_N:
|
||
|
k = node[stg_n[i]].label-1
|
||
|
X,Y,Z = L0[k][3],L0[k][4],L0[k][5]
|
||
|
outfile.write('%11.8e'%X+' '+'%11.8e'%Y+' '+'%11.8e'%Z+'\n')
|
||
|
outfile.write("</DataArray>"+'\n')
|
||
|
#</DataArray>
|
||
|
|
||
|
#Spatial velocity, <DataArray>
|
||
|
outfile.write("<"+"DataArray"+" "+"type="+'"'+"Float32"+'"'+" "+"Name="+'"'+"Spatial_velocity"+'"'+" "+"NumberOfComponents="+'"'+"3"+'"'+" "+"format="+'"'+"ascii"+'"'+">"+'\n')
|
||
|
for i in reop_N:
|
||
|
k = node[stg_n[i]].label-1
|
||
|
X,Y,Z = L0[k][6],L0[k][7],L0[k][8]
|
||
|
outfile.write('%11.8e'%X+' '+'%11.8e'%Y+' '+'%11.8e'%Z+'\n')
|
||
|
outfile.write("</DataArray>"+'\n')
|
||
|
#</DataArray>
|
||
|
|
||
|
#Reaction force
|
||
|
outfile.write("<"+"DataArray"+" "+"type="+'"'+"Float32"+'"'+" "+"Name="+'"'+"Reaction_force"+'"'+" "+"NumberOfComponents="+'"'+"3"+'"'+" "+"format="+'"'+"ascii"+'"'+">"+'\n')
|
||
|
for i in reop_N:
|
||
|
k = node[stg_n[i]].label-1
|
||
|
X,Y,Z = L0[k][9],L0[k][10],L0[k][11]
|
||
|
outfile.write('%11.8e'%X+' '+'%11.8e'%Y+' '+'%11.8e'%Z+'\n')
|
||
|
outfile.write("</DataArray>"+'\n')
|
||
|
#</DataArray>
|
||
|
|
||
|
#Equivalent plastic strain, <DataArray>
|
||
|
outfile.write("<"+"DataArray"+" "+"type="+'"'+"Float32"+'"'+" "+"Name="+'"'+"Equivalent_plastic_strain"+'"'+" "+"format="+'"'+"ascii"+'"'+">"+'\n')
|
||
|
for i in reop_N:
|
||
|
k = node[stg_n[i]].label-1
|
||
|
X = L4[k][1]/L4[k][0]
|
||
|
outfile.write('%11.8e'%X+'\n')
|
||
|
outfile.write('</DataArray>'+'\n')
|
||
|
#</DataArray>
|
||
|
|
||
|
#Stress Mises, <DataArray>
|
||
|
outfile.write("<"+"DataArray"+" "+"type="+'"'+"Float32"+'"'+" "+"Name="+'"'+"Stress_Mises"+'"'+" "+"format="+'"'+"ascii"+'"'+">"+'\n')
|
||
|
for i in reop_N:
|
||
|
k = node[stg_n[i]].label-1
|
||
|
X = L1[k][7]/L1[k][0]
|
||
|
outfile.write('%11.8e'%X+'\n')
|
||
|
outfile.write('</DataArray>'+'\n')
|
||
|
#</DataArray>
|
||
|
|
||
|
#Stress Max.Principal, <DataArray>
|
||
|
outfile.write("<"+"DataArray"+" "+"type="+'"'+"Float32"+'"'+" "+"Name="+'"'+"Stress_Max_Principal"+'"'+" "+"format="+'"'+"ascii"+'"'+">"+'\n')
|
||
|
for i in reop_N:
|
||
|
k = node[stg_n[i]].label-1
|
||
|
X = L1[k][8]/L1[k][0]
|
||
|
outfile.write('%11.8e'%X+'\n')
|
||
|
outfile.write('</DataArray>'+'\n')
|
||
|
#</DataArray>
|
||
|
|
||
|
#Stress Mid.Principal, <DataArray>
|
||
|
outfile.write("<"+"DataArray"+" "+"type="+'"'+"Float32"+'"'+" "+"Name="+'"'+"Stress_Mid_Principal"+'"'+" "+"format="+'"'+"ascii"+'"'+">"+'\n')
|
||
|
for i in reop_N:
|
||
|
k = node[stg_n[i]].label-1
|
||
|
X = L1[k][9]/L1[k][0]
|
||
|
outfile.write('%11.8e'%X+'\n')
|
||
|
outfile.write('</DataArray>'+'\n')
|
||
|
#</DataArray>
|
||
|
|
||
|
#Stress Min.Principal, <DataArray>
|
||
|
outfile.write("<"+"DataArray"+" "+"type="+'"'+"Float32"+'"'+" "+"Name="+'"'+"Stress_Min_Principal"+'"'+" "+"format="+'"'+"ascii"+'"'+">"+'\n')
|
||
|
for i in reop_N:
|
||
|
k = node[stg_n[i]].label-1
|
||
|
X = L1[k][10]/L1[k][0]
|
||
|
outfile.write('%11.8e'%X+'\n')
|
||
|
outfile.write('</DataArray>'+'\n')
|
||
|
#</DataArray>
|
||
|
|
||
|
#Stress Pressure, <DataArray>
|
||
|
outfile.write("<"+"DataArray"+" "+"type="+'"'+"Float32"+'"'+" "+"Name="+'"'+"Stress_Pressure"+'"'+" "+"format="+'"'+"ascii"+'"'+">"+'\n')
|
||
|
for i in reop_N:
|
||
|
k = node[stg_n[i]].label-1
|
||
|
X = L1[k][11]/L1[k][0]
|
||
|
outfile.write('%11.8e'%X+'\n')
|
||
|
outfile.write('</DataArray>'+'\n')
|
||
|
#</DataArray>
|
||
|
|
||
|
#Stress Tresca, <DataArray>
|
||
|
outfile.write("<"+"DataArray"+" "+"type="+'"'+"Float32"+'"'+" "+"Name="+'"'+"Stress_Tresca"+'"'+" "+"format="+'"'+"ascii"+'"'+">"+'\n')
|
||
|
for i in reop_N:
|
||
|
k = node[stg_n[i]].label-1
|
||
|
X = L1[k][12]/L1[k][0]
|
||
|
outfile.write('%11.8e'%X+'\n')
|
||
|
outfile.write('</DataArray>'+'\n')
|
||
|
#</DataArray>
|
||
|
|
||
|
#Stress Third_Invariant, <DataArray>
|
||
|
outfile.write("<"+"DataArray"+" "+"type="+'"'+"Float32"+'"'+" "+"Name="+'"'+"Stress_Third_Invariant"+'"'+" "+"format="+'"'+"ascii"+'"'+">"+'\n')
|
||
|
for i in reop_N:
|
||
|
k = node[stg_n[i]].label-1
|
||
|
X = L1[k][13]/L1[k][0]
|
||
|
outfile.write('%11.8e'%X+'\n')
|
||
|
outfile.write('</DataArray>'+'\n')
|
||
|
#</DataArray>
|
||
|
|
||
|
#Logarithmic_strain_Max_Principal, <DataArray>
|
||
|
outfile.write("<"+"DataArray"+" "+"type="+'"'+"Float32"+'"'+" "+"Name="+'"'+"Logarithmic_strain_Max_Principal"+'"'+" "+"format="+'"'+"ascii"+'"'+">"+'\n')
|
||
|
for i in reop_N:
|
||
|
k = node[stg_n[i]].label-1
|
||
|
X = L2[k][7]/L2[k][0]
|
||
|
outfile.write('%11.8e'%X+'\n')
|
||
|
outfile.write('</DataArray>'+'\n')
|
||
|
#</DataArray>
|
||
|
|
||
|
#Logarithmic strain Min.Principal, <DataArray>
|
||
|
outfile.write("<"+"DataArray"+" "+"type="+'"'+"Float32"+'"'+" "+"Name="+'"'+"Logarithmic_strain_Min_Principal"+'"'+" "+"format="+'"'+"ascii"+'"'+">"+'\n')
|
||
|
for i in reop_N:
|
||
|
k = node[stg_n[i]].label-1
|
||
|
X = L2[k][8]/L2[k][0]
|
||
|
outfile.write('%11.8e'%X+'\n')
|
||
|
outfile.write('</DataArray>'+'\n')
|
||
|
#</DataArray>'''
|
||
|
|
||
|
#Plastic strain Max.Principal, <DataArray>
|
||
|
outfile.write("<"+"DataArray"+" "+"type="+'"'+"Float32"+'"'+" "+"Name="+'"'+"Plastic_strain_Max_Principal"+'"'+" "+"format="+'"'+"ascii"+'"'+">"+'\n')
|
||
|
for i in reop_N:
|
||
|
k = node[stg_n[i]].label-1
|
||
|
X = L3[k][7]/L3[k][0]
|
||
|
outfile.write('%11.8e'%X+'\n')
|
||
|
outfile.write('</DataArray>'+'\n')
|
||
|
#</DataArray>
|
||
|
|
||
|
#Plastic strain Min.Principal, <DataArray>
|
||
|
outfile.write("<"+"DataArray"+" "+"type="+'"'+"Float32"+'"'+" "+"Name="+'"'+"Plastic_strain_Min_Principal"+'"'+" "+"format="+'"'+"ascii"+'"'+">"+'\n')
|
||
|
for i in reop_N:
|
||
|
k = node[stg_n[i]].label-1
|
||
|
X = L3[k][8]/L3[k][0]
|
||
|
outfile.write('%11.8e'%X+'\n')
|
||
|
outfile.write('</DataArray>'+'\n')
|
||
|
#</DataArray>
|
||
|
|
||
|
outfile.write("</PointData>"+'\n')
|
||
|
#</PointData>
|
||
|
|
||
|
|
||
|
print "Writing Cells ......"
|
||
|
#<Cells> Write cells into vtk files
|
||
|
outfile.write('<Cells>'+'\n')
|
||
|
#Connectivity
|
||
|
outfile.write('<DataArray type="Int32" Name="connectivity" format="ascii">'+'\n')
|
||
|
if (mesh_type == 12):
|
||
|
for i in range(len(stg_e)/8):
|
||
|
outfile.write(str(stg_e[i*8])+' '+str(stg_e[i*8+1])+' '+str(stg_e[i*8+2])+' '+str(stg_e[i*8+3])+' '+str(stg_e[i*8+4])+' '+str(stg_e[i*8+5])+' '+str(stg_e[i*8+6])+' '+str(stg_e[i*8+7])+'\n')
|
||
|
if (mesh_type == 10):
|
||
|
for i in range(len(stg_e)/4):
|
||
|
outfile.write(str(stg_e[i*4])+' '+str(stg_e[i*4+1])+' '+str(stg_e[i*4+2])+' '+str(stg_e[i*4+3])+'\n')
|
||
|
outfile.write('</DataArray>'+'\n')
|
||
|
#Offsets
|
||
|
outfile.write('<DataArray type="Int32" Name="offsets" format="ascii">'+'\n')
|
||
|
for i in range(len(stg_e)/mesh_conner):
|
||
|
outfile.write(str(i*mesh_conner+mesh_conner)+'\n')
|
||
|
outfile.write('</DataArray>'+'\n')
|
||
|
#Type
|
||
|
outfile.write('<DataArray type="UInt8" Name="types" format="ascii">'+'\n')
|
||
|
for i in range(len(stg_e)/mesh_conner):
|
||
|
outfile.write(str(mesh_type)+'\n')
|
||
|
outfile.write('</DataArray>'+'\n')
|
||
|
outfile.write('</Cells>'+'\n')
|
||
|
#</Cells>
|
||
|
|
||
|
|
||
|
#</Piece>
|
||
|
outfile.write('</Piece>'+'\n')
|
||
|
#</UnstructuredGrid>
|
||
|
outfile.write('</UnstructuredGrid>'+'\n')
|
||
|
#</VTKFile>
|
||
|
outfile.write('</VTKFile>'+'\n')
|
||
|
|
||
|
outfile.close()
|
||
|
print "Time elapsed: ", time() - time1, "s"
|
||
|
|
||
|
'''====================================================================='''
|
||
|
print "Creating .pvtu file for frame ", i_frame," ......"
|
||
|
#create .pvtu files for parallel visualization
|
||
|
if ( piecenum > 1 ):
|
||
|
outfile = open (os.path.join(vtk_path,odbname)+'_'+stepname+'_'+'f%03d'%int(i_frame)+'.pvtu','w')
|
||
|
|
||
|
#write the basic information for .pvtu files
|
||
|
outfile.write('<?xml version="1.0"?>'+'\n')
|
||
|
outfile.write('<VTKFile type="PUnstructuredGrid" version="0.1" byte_order="LittleEndian">'+'\n')
|
||
|
outfile.write("<PUnstructuredGrid GhostLevel="+'"'+str(piecenum)+'"'+">"+'\n')
|
||
|
#pointdata
|
||
|
outfile.write("<"+"PPointData"+" "+"Tensors="+'"'+"Stress_Components,Logarithmic_strain_components,Plastic_strain_components"+'"'\
|
||
|
+" "+"Vevtors="+'"'+"Spatial_displacement,Spatial_acceleration,Spatial_velocity,Reaction_force"+'"'\
|
||
|
+" "+"Scalars="+'"'+"Equivalent_plastic_strain,Stress_Mises,Stress_Max_Principal,Stress_Mid_Principal,Stress_Min_Principal,Stress_Pressure,Stress_Tresca,Stress_Third_Invariant,Logarithmic_strain_Max_Principal,Logarithmic_strain_Min_Principal,Plastic_strain_Max_Principal,Plastic_strain_Min_Principal"+'"'+">"+'\n')
|
||
|
|
||
|
outfile.write("<"+"PDataArray"+" "+"type="+'"'+"Float32"+'"'+" "+"Name="+'"'+"Stress_Components"+'"'+" "+"NumberOfComponents="+'"'+"9"+'"'+" "+"/>"+'\n')
|
||
|
outfile.write("<"+"PDataArray"+" "+"type="+'"'+"Float32"+'"'+" "+"Name="+'"'+"Logarithmic_strain_components"+'"'+" "+"NumberOfComponents="+'"'+"9"+'"'+" "+"/>"+'\n')
|
||
|
outfile.write("<"+"PDataArray"+" "+"type="+'"'+"Float32"+'"'+" "+"Name="+'"'+"Plastic_strain_components"+'"'+" "+"NumberOfComponents="+'"'+"9"+'"'+" "+"/>"+'\n')
|
||
|
outfile.write("<"+"PDataArray"+" "+"type="+'"'+"Float32"+'"'+" "+"Name="+'"'+"Spatial_displacement"+'"'+" "+"NumberOfComponents="+'"'+"3"+'"'+" "+"/>"+'\n')
|
||
|
outfile.write("<"+"PDataArray"+" "+"type="+'"'+"Float32"+'"'+" "+"Name="+'"'+"Spatial_acceleration"+'"'+" "+"NumberOfComponents="+'"'+"3"+'"'+" "+"/>"+'\n')
|
||
|
outfile.write("<"+"PDataArray"+" "+"type="+'"'+"Float32"+'"'+" "+"Name="+'"'+"Spatial_velocity"+'"'+" "+"NumberOfComponents="+'"'+"3"+'"'+" "+"/>"+'\n')
|
||
|
outfile.write("<"+"PDataArray"+" "+"type="+'"'+"Float32"+'"'+" "+"Name="+'"'+"Reaction_force"+'"'+" "+"NumberOfComponents="+'"'+"3"+'"'+" "+"/>"+'\n')
|
||
|
outfile.write("<"+"PDataArray"+" "+"type="+'"'+"Float32"+'"'+" "+"Name="+'"'+"Equivalent_plastic_strain"+'"'+" "+"/>"+'\n')
|
||
|
outfile.write("<"+"PDataArray"+" "+"type="+'"'+"Float32"+'"'+" "+"Name="+'"'+"Stress_Mises"+'"'+" "+"/>"+'\n')
|
||
|
outfile.write("<"+"PDataArray"+" "+"type="+'"'+"Float32"+'"'+" "+"Name="+'"'+"Stress_Max_Principal"+'"'+" "+"/>"+'\n')
|
||
|
outfile.write("<"+"PDataArray"+" "+"type="+'"'+"Float32"+'"'+" "+"Name="+'"'+"Stress_Mid_Principal"+'"'+" "+"/>"+'\n')
|
||
|
outfile.write("<"+"PDataArray"+" "+"type="+'"'+"Float32"+'"'+" "+"Name="+'"'+"Stress_Min_Principal"+'"'+" "+"/>"+'\n')
|
||
|
outfile.write("<"+"PDataArray"+" "+"type="+'"'+"Float32"+'"'+" "+"Name="+'"'+"Stress_Pressure"+'"'+" "+"/>"+'\n')
|
||
|
outfile.write("<"+"PDataArray"+" "+"type="+'"'+"Float32"+'"'+" "+"Name="+'"'+"Stress_Tresca"+'"'+" "+"/>"+'\n')
|
||
|
outfile.write("<"+"PDataArray"+" "+"type="+'"'+"Float32"+'"'+" "+"Name="+'"'+"Stress_Third_Invariant"+'"'+" "+"/>"+'\n')
|
||
|
outfile.write("<"+"PDataArray"+" "+"type="+'"'+"Float32"+'"'+" "+"Name="+'"'+"Logarithmic_strain_Max_Principal"+'"'+" "+"/>"+'\n')
|
||
|
outfile.write("<"+"PDataArray"+" "+"type="+'"'+"Float32"+'"'+" "+"Name="+'"'+"Logarithmic_strain_Min_Principal"+'"'+" "+"/>"+'\n')
|
||
|
outfile.write("<"+"PDataArray"+" "+"type="+'"'+"Float32"+'"'+" "+"Name="+'"'+"Plastic_strain_Max_Principal"+'"'+" "+"/>"+'\n')
|
||
|
outfile.write("<"+"PDataArray"+" "+"type="+'"'+"Float32"+'"'+" "+"Name="+'"'+"Plastic_strain_Min_Principal"+'"'+" "+"/>"+'\n')
|
||
|
outfile.write("</PPointData>"+'\n')
|
||
|
|
||
|
#points
|
||
|
outfile.write("<PPoints>"+'\n')
|
||
|
outfile.write("<PDataArray type="+'"'+"Float64"+'"'+" "+"NumberOfComponents="+'"'+"3"+'"'+"/>"+'\n')
|
||
|
outfile.write("</PPoints>"+'\n')
|
||
|
|
||
|
#write the path of each piece for reading it through the .pvtu file
|
||
|
for pn in range(piecenum):
|
||
|
outfile.write("<Piece Source="+'"'+odbname+'_'+stepname+'_'+instancename+'f%03d'%int(i_frame)+' '+'p'+str(pn)+'.vtu'+'"'+"/>"+'\n')
|
||
|
|
||
|
outfile.write("</PUnstructuredGrid>"+'\n')
|
||
|
outfile.write("</VTKFile>")
|
||
|
|
||
|
outfile.close()
|
||
|
|
||
|
odb.close()
|
||
|
|
||
|
print "Total time elapsed: ", time() - starttime, "s"
|
||
|
|
||
|
|
||
|
|