DAMASK_EICMD/code/CPFEM.f90

689 lines
33 KiB
Fortran
Raw Normal View History

! Copyright 2011 Max-Planck-Institut für Eisenforschung GmbH
!
! This file is part of DAMASK,
! the Düsseldorf Advanced MAterial Simulation Kit.
!
! DAMASK is free software: you can redistribute it and/or modify
! it under the terms of the GNU General Public License as published by
! the Free Software Foundation, either version 3 of the License, or
! (at your option) any later version.
!
! DAMASK is distributed in the hope that it will be useful,
! but WITHOUT ANY WARRANTY; without even the implied warranty of
! MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
! GNU General Public License for more details.
!
! You should have received a copy of the GNU General Public License
! along with DAMASK. If not, see <http://www.gnu.org/licenses/>.
!
!##############################################################
!* $Id$
!##############################################################
MODULE CPFEM
!##############################################################
! *** CPFEM engine ***
!
use prec, only: pReal, &
pInt
implicit none
real(pReal), parameter :: CPFEM_odd_stress = 1e15_pReal, &
CPFEM_odd_jacobian = 1e50_pReal
real(pReal), dimension (:,:,:), allocatable :: CPFEM_cs ! Cauchy stress
real(pReal), dimension (:,:,:,:), allocatable :: CPFEM_dcsdE ! Cauchy stress tangent
real(pReal), dimension (:,:,:,:), allocatable :: CPFEM_dcsdE_knownGood ! known good tangent
logical :: CPFEM_init_done = .false., & ! remember whether init has been done already
CPFEM_init_inProgress = .false., & ! remember whether first IP is currently performing init
CPFEM_calc_done = .false. ! remember whether first IP has already calced the results
CONTAINS
!*********************************************************
!*** call (thread safe) all module initializations ***
!*********************************************************
subroutine CPFEM_initAll(Temperature,element,IP)
use prec, only: pReal, &
prec_init
use numerics, only: numerics_init
use debug, only: debug_init
use FEsolving, only: FE_init
use math, only: math_init
use mesh, only: mesh_init
use lattice, only: lattice_init
use material, only: material_init
use constitutive, only: constitutive_init
use crystallite, only: crystallite_init
use homogenization, only: homogenization_init
use IO, only: IO_init
use DAMASK_interface
implicit none
integer(pInt), intent(in) :: element, & ! FE element number
IP ! FE integration point number
real(pReal), intent(in) :: Temperature ! temperature
real(pReal) rnd
integer(pInt) i,n
! initialization step (three dimensional stress state check missing?)
if (.not. CPFEM_init_done) then
call random_number(rnd)
do i=1,int(256.0*rnd)
n = n+1_pInt ! wasting random amount of time...
enddo ! ...to break potential race in multithreading
n = n+1_pInt
if (.not. CPFEM_init_inProgress) then ! yes my thread won!
CPFEM_init_inProgress = .true.
call prec_init()
call IO_init()
call numerics_init()
call debug_init()
call math_init()
call FE_init()
call mesh_init(IP, element) ! pass on coordinates to alter calcMode of first ip
call lattice_init()
call material_init()
call constitutive_init()
call crystallite_init(Temperature) ! (have to) use temperature of first IP for whole model
call homogenization_init(Temperature)
call CPFEM_init()
if (trim(FEsolver)/='Spectral') call DAMASK_interface_init() ! Spectral solver is doing initialization earlier
CPFEM_init_done = .true.
CPFEM_init_inProgress = .false.
else ! loser, loser...
do while (CPFEM_init_inProgress)
enddo
endif
endif
end subroutine
!*********************************************************
!*** allocate the arrays defined in module CPFEM ***
!*** and initialize them ***
!*********************************************************
subroutine CPFEM_init()
use prec, only: pInt
use debug, only: debug_verbosity
use IO, only: IO_read_jobBinaryFile
use FEsolving, only: parallelExecution, &
symmetricSolver, &
restartRead, &
FEmodelGeometry
use mesh, only: mesh_NcpElems, &
mesh_maxNips
use material, only: homogenization_maxNgrains, &
material_phase
use constitutive, only: constitutive_state0
use crystallite, only: crystallite_F0, &
crystallite_Fp0, &
crystallite_Lp0, &
crystallite_dPdF0, &
crystallite_Tstar0_v
use homogenization, only: homogenization_sizeState, &
homogenization_state0, &
materialpoint_F, &
materialpoint_F0
implicit none
integer(pInt) i,j,k,l,m
! initialize stress and jacobian to zero
allocate(CPFEM_cs(6,mesh_maxNips,mesh_NcpElems)) ; CPFEM_cs = 0.0_pReal
allocate(CPFEM_dcsdE(6,6,mesh_maxNips,mesh_NcpElems)) ; CPFEM_dcsdE = 0.0_pReal
allocate(CPFEM_dcsdE_knownGood(6,6,mesh_maxNips,mesh_NcpElems)) ; CPFEM_dcsdE_knownGood = 0.0_pReal
! *** restore the last converged values of each essential variable from the binary file
if (restartRead) then
if (debug_verbosity > 0) then
!$OMP CRITICAL (write2out)
write(6,'(a)') '<< CPFEM >> Restored state variables of last converged step from binary files'
!$OMP END CRITICAL (write2out)
endif
call IO_read_jobBinaryFile(777,'recordedPhase',FEmodelGeometry,size(material_phase))
read (777,rec=1) material_phase
close (777)
call IO_read_jobBinaryFile(777,'convergedF',FEmodelGeometry,size(crystallite_F0))
read (777,rec=1) crystallite_F0
close (777)
call IO_read_jobBinaryFile(777,'convergedFp',FEmodelGeometry,size(crystallite_Fp0))
read (777,rec=1) crystallite_Fp0
close (777)
call IO_read_jobBinaryFile(777,'convergedLp',FEmodelGeometry,size(crystallite_Lp0))
read (777,rec=1) crystallite_Lp0
close (777)
call IO_read_jobBinaryFile(777,'convergeddPdF',FEmodelGeometry,size(crystallite_dPdF0))
read (777,rec=1) crystallite_dPdF0
close (777)
call IO_read_jobBinaryFile(777,'convergedTstar',FEmodelGeometry,size(crystallite_Tstar0_v))
read (777,rec=1) crystallite_Tstar0_v
close (777)
call IO_read_jobBinaryFile(777,'convergedStateConst',FEmodelGeometry)
m = 0_pInt
do i = 1,homogenization_maxNgrains; do j = 1,mesh_maxNips; do k = 1,mesh_NcpElems
do l = 1,size(constitutive_state0(i,j,k)%p)
m = m+1_pInt
read(777,rec=m) constitutive_state0(i,j,k)%p(l)
enddo
enddo; enddo; enddo
close (777)
call IO_read_jobBinaryFile(777,'convergedStateHomog',FEmodelGeometry)
m = 0_pInt
do k = 1,mesh_NcpElems; do j = 1,mesh_maxNips
do l = 1,homogenization_sizeState(j,k)
m = m+1_pInt
read(777,rec=m) homogenization_state0(j,k)%p(l)
enddo
enddo; enddo
close (777)
call IO_read_jobBinaryFile(777,'convergeddcsdE',FEmodelGeometry,size(CPFEM_dcsdE))
read (777,rec=1) CPFEM_dcsdE
close (777)
restartRead = .false.
endif
! *** end of restoring
!$OMP CRITICAL (write2out)
write(6,*)
write(6,*) '<<<+- cpfem init -+>>>'
write(6,*) '$Id$'
#include "compilation_info.f90"
if (debug_verbosity > 0) then
write(6,'(a32,1x,6(i8,1x))') 'CPFEM_cs: ', shape(CPFEM_cs)
write(6,'(a32,1x,6(i8,1x))') 'CPFEM_dcsdE: ', shape(CPFEM_dcsdE)
write(6,'(a32,1x,6(i8,1x))') 'CPFEM_dcsdE_knownGood: ', shape(CPFEM_dcsdE_knownGood)
write(6,*)
write(6,*) 'parallelExecution: ', parallelExecution
write(6,*) 'symmetricSolver: ', symmetricSolver
endif
call flush(6)
!$OMP END CRITICAL (write2out)
endsubroutine
!***********************************************************************
!*** perform initialization at first call, update variables and ***
!*** call the actual material model ***
!***********************************************************************
subroutine CPFEM_general(mode, coords, ffn, ffn1, Temperature, dt, element, IP, cauchyStress,&
& jacobian, pstress, dPdF)
! note: cauchyStress = Cauchy stress cs(6) and jacobian = Consistent tangent dcs/dE
!*** variables and functions from other modules ***!
use prec, only: pReal, &
pInt
use numerics, only: relevantStrain, &
defgradTolerance, &
iJacoStiffness
use debug, only: debug_e, &
debug_i, &
debug_g, &
debug_selectiveDebugger, &
debug_verbosity, &
debug_stressMaxLocation, &
debug_stressMinLocation, &
debug_jacobianMaxLocation, &
debug_jacobianMinLocation, &
debug_stressMax, &
debug_stressMin, &
debug_jacobianMax, &
debug_jacobianMin
use FEsolving, only: parallelExecution, &
outdatedFFN1, &
terminallyIll, &
cycleCounter, &
theInc, &
theTime, &
theDelta, &
FEsolving_execElem, &
FEsolving_execIP, &
restartWrite
use math, only: math_identity2nd, &
math_mul33x33, &
math_det33, &
math_transpose33, &
math_I3, &
math_Mandel3333to66, &
math_Mandel66to3333, &
math_Mandel33to6, &
math_Mandel6to33
use mesh, only: mesh_FEasCP, &
mesh_NcpElems, &
mesh_maxNips, &
mesh_element, &
mesh_node0, &
mesh_node, &
mesh_ipCenterOfGravity, &
mesh_build_subNodeCoords, &
mesh_build_ipVolumes, &
mesh_build_ipCoordinates, &
FE_Nips, &
FE_Nnodes
use material, only: homogenization_maxNgrains, &
microstructure_elemhomo, &
material_phase
use constitutive, only: constitutive_state0,constitutive_state
use crystallite, only: crystallite_F0, &
crystallite_partionedF, &
crystallite_Fp0, &
crystallite_Fp, &
crystallite_Lp0, &
crystallite_Lp, &
crystallite_dPdF0, &
crystallite_dPdF, &
crystallite_Tstar0_v, &
crystallite_Tstar_v, &
crystallite_localConstitution
use homogenization, only: homogenization_sizeState, &
homogenization_state, &
homogenization_state0, &
materialpoint_F, &
materialpoint_F0, &
materialpoint_P, &
materialpoint_dPdF, &
materialpoint_results, &
materialpoint_sizeResults, &
materialpoint_Temperature, &
materialpoint_stressAndItsTangent, &
materialpoint_postResults
use IO, only: IO_write_jobBinaryFile, &
IO_warning
use DAMASK_interface
implicit none
!*** input variables ***!
integer(pInt), intent(in) :: element, & ! FE element number
IP ! FE integration point number
real(pReal), intent(inout) :: Temperature ! temperature
real(pReal), intent(in) :: dt ! time increment
real(pReal), dimension (3,*), intent(in) :: coords ! MARC: displacements for each node of the current element
! ABAQUS: coordinates of the current material point (IP)
! SPECTRAL: coordinates of the current material point (IP)
real(pReal), dimension (3,3), intent(in) :: ffn, & ! deformation gradient for t=t0
ffn1 ! deformation gradient for t=t1
integer(pInt), intent(in) :: mode ! computation mode 1: regular computation plus aging of results
! 2: regular computation
! 3: collection of FEM data
! 4: backup tangent from former converged inc
! 5: restore tangent from former converged inc
! 6: recycling of former results (MARC speciality)
!*** output variables ***!
real(pReal), dimension(6), intent(out) :: cauchyStress ! stress vector in Mandel notation
real(pReal), dimension(6,6), intent(out) :: jacobian ! jacobian in Mandel notation
real(pReal), dimension (3,3), intent(out) :: pstress ! Piola-Kirchhoff stress in Matrix notation
real(pReal), dimension (3,3,3,3), intent(out) :: dPdF !
!*** local variables ***!
real(pReal) J_inverse, & ! inverse of Jacobian
rnd
real(pReal), dimension (3,3) :: Kirchhoff, & ! Piola-Kirchhoff stress in Matrix notation
cauchyStress33 ! stress vector in Matrix notation
real(pReal), dimension (3,3,3,3) :: H_sym, &
H, &
jacobian3333 ! jacobian in Matrix notation
integer(pInt) cp_en, & ! crystal plasticity element number
i, &
j, &
k, &
l, &
m, &
n, &
e, &
node, &
FEnodeID
logical updateJaco ! flag indicating if JAcobian has to be updated
cp_en = mesh_FEasCP('elem',element)
if (debug_verbosity > 0 .and. cp_en == 1 .and. IP == 1) then
!$OMP CRITICAL (write2out)
write(6,*)
write(6,'(a)') '#############################################'
write(6,'(a1,a22,1x,f15.7,a6)') '#','theTime',theTime,'#'
write(6,'(a1,a22,1x,f15.7,a6)') '#','theDelta',theDelta,'#'
write(6,'(a1,a22,1x,i8,a13)') '#','theInc',theInc,'#'
write(6,'(a1,a22,1x,i8,a13)') '#','cycleCounter',cycleCounter,'#'
write(6,'(a1,a22,1x,i8,a13)') '#','computationMode',mode,'#'
write(6,'(a)') '#############################################'
write(6,*)
call flush (6)
!$OMP END CRITICAL (write2out)
endif
!*** according to our "mode" we decide what to do
select case (mode)
! --+>> REGULAR COMPUTATION (WITH AGING OF RESULTS IF MODE == 1) <<+--
case (1,2,8,9)
!*** age results
if (mode == 1 .or. mode == 8) then
crystallite_F0 = crystallite_partionedF ! crystallite deformation (_subF is perturbed...)
crystallite_Fp0 = crystallite_Fp ! crystallite plastic deformation
crystallite_Lp0 = crystallite_Lp ! crystallite plastic velocity
crystallite_dPdF0 = crystallite_dPdF ! crystallite stiffness
crystallite_Tstar0_v = crystallite_Tstar_v ! crystallite 2nd Piola Kirchhoff stress
forall ( i = 1:homogenization_maxNgrains, &
j = 1:mesh_maxNips, &
k = 1:mesh_NcpElems ) &
constitutive_state0(i,j,k)%p = constitutive_state(i,j,k)%p ! microstructure of crystallites
if (debug_verbosity > 0) then
!$OMP CRITICAL (write2out)
write(6,'(a)') '<< CPFEM >> Aging states'
if (debug_e == cp_en .and. debug_i == IP) then
write(6,'(a,1x,i8,1x,i2,1x,i4,/,(12x,6(e20.8,1x)))') '<< CPFEM >> AGED state of element ip grain',&
cp_en, IP, 1, constitutive_state(1,IP,cp_en)%p
write(6,*)
endif
!$OMP END CRITICAL (write2out)
endif
!$OMP PARALLEL DO
do k = 1,mesh_NcpElems
do j = 1,mesh_maxNips
if (homogenization_sizeState(j,k) > 0_pInt) &
homogenization_state0(j,k)%p = homogenization_state(j,k)%p ! internal state of homogenization scheme
enddo
enddo
!$OMP END PARALLEL DO
! * dump the last converged values of each essential variable to a binary file
if (restartWrite) then
if (debug_verbosity > 0) then
!$OMP CRITICAL (write2out)
write(6,'(a)') '<< CPFEM >> Writing state variables of last converged step to binary files'
!$OMP END CRITICAL (write2out)
endif
call IO_write_jobBinaryFile(777,'recordedPhase',size(material_phase))
write (777,rec=1) material_phase
close (777)
call IO_write_jobBinaryFile(777,'convergedF',size(crystallite_F0))
write (777,rec=1) crystallite_F0
close (777)
call IO_write_jobBinaryFile(777,'convergedFp',size(crystallite_Fp0))
write (777,rec=1) crystallite_Fp0
close (777)
call IO_write_jobBinaryFile(777,'convergedLp',size(crystallite_Lp0))
write (777,rec=1) crystallite_Lp0
close (777)
call IO_write_jobBinaryFile(777,'convergeddPdF',size(crystallite_dPdF0))
write (777,rec=1) crystallite_dPdF0
close (777)
call IO_write_jobBinaryFile(777,'convergedTstar',size(crystallite_Tstar0_v))
write (777,rec=1) crystallite_Tstar0_v
close (777)
call IO_write_jobBinaryFile(777,'convergedStateConst')
m = 0_pInt
do i = 1,homogenization_maxNgrains; do j = 1,mesh_maxNips; do k = 1,mesh_NcpElems
do l = 1,size(constitutive_state0(i,j,k)%p)
m = m+1_pInt
write(777,rec=m) constitutive_state0(i,j,k)%p(l)
enddo
enddo; enddo; enddo
close (777)
call IO_write_jobBinaryFile(777,'convergedStateHomog')
m = 0_pInt
do k = 1,mesh_NcpElems; do j = 1,mesh_maxNips
do l = 1,homogenization_sizeState(j,k)
m = m+1_pInt
write(777,rec=m) homogenization_state0(j,k)%p(l)
enddo
enddo; enddo
close (777)
call IO_write_jobBinaryFile(777,'convergeddcsdE',size(CPFEM_dcsdE))
write (777,rec=1) CPFEM_dcsdE
close (777)
endif
! * end of dumping
endif
if (mode == 8 .or. mode == 9) then ! Abaqus explicit skips collect
materialpoint_Temperature(IP,cp_en) = Temperature
materialpoint_F0(1:3,1:3,IP,cp_en) = ffn
materialpoint_F(1:3,1:3,IP,cp_en) = ffn1
endif
!*** deformation gradient outdated or any actual deformation gradient differs more than relevantStrain from the stored one
if (terminallyIll .or. outdatedFFN1 .or. any(abs(ffn1 - materialpoint_F(1:3,1:3,IP,cp_en)) > defgradTolerance)) then
if (.not. terminallyIll .and. .not. outdatedFFN1) then
if (debug_verbosity > 0) then
!$OMP CRITICAL (write2out)
write(6,'(a,1x,i8,1x,i2)') '<< CPFEM >> OUTDATED at element ip',cp_en,IP
write(6,'(a,/,3(12x,3(f10.6,1x),/))') '<< CPFEM >> FFN1 old:',math_transpose33(materialpoint_F(1:3,1:3,IP,cp_en))
write(6,'(a,/,3(12x,3(f10.6,1x),/))') '<< CPFEM >> FFN1 now:',math_transpose33(ffn1)
!$OMP END CRITICAL (write2out)
endif
outdatedFFN1 = .true.
endif
call random_number(rnd)
rnd = 2.0_pReal * rnd - 1.0_pReal
CPFEM_cs(1:6,IP,cp_en) = rnd*CPFEM_odd_stress
CPFEM_dcsde(1:6,1:6,IP,cp_en) = CPFEM_odd_jacobian*math_identity2nd(6)
!*** deformation gradient is not outdated
else
updateJaco = mod(cycleCounter,iJacoStiffness) == 0
!* no parallel computation, so we use just one single element and IP for computation
if (.not. parallelExecution) then
FEsolving_execElem(1) = cp_en
FEsolving_execElem(2) = cp_en
FEsolving_execIP(1,cp_en) = IP
FEsolving_execIP(2,cp_en) = IP
if (debug_verbosity > 0) then
!$OMP CRITICAL (write2out)
write(6,'(a,i8,1x,i2)') '<< CPFEM >> Calculation for element ip ',cp_en,IP
!$OMP END CRITICAL (write2out)
endif
call materialpoint_stressAndItsTangent(updateJaco, dt) ! calculate stress and its tangent
call materialpoint_postResults(dt) ! post results
!* parallel computation and calulation not yet done
elseif (.not. CPFEM_calc_done) then
if (debug_verbosity > 0) then
!$OMP CRITICAL (write2out)
write(6,'(a,i8,a,i8)') '<< CPFEM >> Calculation for elements ',FEsolving_execElem(1),' to ',FEsolving_execElem(2)
!$OMP END CRITICAL (write2out)
endif
if (FEsolver == 'Marc') then ! marc returns nodal coordinates, whereas Abaqus and spectral solver return ip coordinates. So for marc we have to calculate the ip coordinates from the nodal coordinates.
call mesh_build_subNodeCoords() ! update subnodal coordinates
call mesh_build_ipCoordinates() ! update ip coordinates
endif
if (debug_verbosity > 0) then
!$OMP CRITICAL (write2out)
2011-07-15 17:55:38 +05:30
write(6,'(a,i8,a,i8)') '<< CPFEM >> Start stress and tangent ',FEsolving_execElem(1),' to ',FEsolving_execElem(2)
!$OMP END CRITICAL (write2out)
endif
call materialpoint_stressAndItsTangent(updateJaco, dt) ! calculate stress and its tangent (parallel execution inside)
call materialpoint_postResults(dt) ! post results
!$OMP PARALLEL DO
do e = FEsolving_execElem(1),FEsolving_execElem(2) ! loop over all parallely processed elements
if (microstructure_elemhomo(mesh_element(4,e))) then ! dealing with homogeneous element?
forall (i = 2:FE_Nips(mesh_element(2,e))) ! copy results of first IP to all others
materialpoint_P(1:3,1:3,i,e) = materialpoint_P(1:3,1:3,1,e)
materialpoint_F(1:3,1:3,i,e) = materialpoint_F(1:3,1:3,1,e)
materialpoint_dPdF(1:3,1:3,1:3,1:3,i,e) = materialpoint_dPdF(1:3,1:3,1:3,1:3,1,e)
materialpoint_results(1:materialpoint_sizeResults,i,e) = materialpoint_results(1:materialpoint_sizeResults,1,e)
end forall
endif
enddo
!$OMP END PARALLEL DO
CPFEM_calc_done = .true.
endif
!*** map stress and stiffness (or return odd values if terminally ill)
if ( terminallyIll ) then
call random_number(rnd)
rnd = 2.0_pReal * rnd - 1.0_pReal
CPFEM_cs(1:6,IP,cp_en) = rnd * CPFEM_odd_stress
CPFEM_dcsde(1:6,1:6,IP,cp_en) = CPFEM_odd_jacobian * math_identity2nd(6)
else
! translate from P to CS
Kirchhoff = math_mul33x33(materialpoint_P(1:3,1:3,IP, cp_en), math_transpose33(materialpoint_F(1:3,1:3,IP,cp_en)))
J_inverse = 1.0_pReal / math_det33(materialpoint_F(1:3,1:3,IP,cp_en))
CPFEM_cs(1:6,IP,cp_en) = math_Mandel33to6(J_inverse * Kirchhoff)
! translate from dP/dF to dCS/dE
H = 0.0_pReal
do i=1,3; do j=1,3; do k=1,3; do l=1,3; do m=1,3; do n=1,3
H(i,j,k,l) = H(i,j,k,l) + &
materialpoint_F(j,m,IP,cp_en) * &
materialpoint_F(l,n,IP,cp_en) * &
materialpoint_dPdF(i,m,k,n,IP,cp_en) - &
math_I3(j,l) * materialpoint_F(i,m,IP,cp_en) * materialpoint_P(k,m,IP,cp_en) + &
0.5_pReal * (math_I3(i,k) * Kirchhoff(j,l) + math_I3(j,l) * Kirchhoff(i,k) + &
math_I3(i,l) * Kirchhoff(j,k) + math_I3(j,k) * Kirchhoff(i,l))
enddo; enddo; enddo; enddo; enddo; enddo
do i=1,3; do j=1,3; do k=1,3; do l=1,3
H_sym(i,j,k,l) = 0.25_pReal * (H(i,j,k,l) + H(j,i,k,l) + H(i,j,l,k) + H(j,i,l,k))
enddo; enddo; enddo; enddo
CPFEM_dcsde(1:6,1:6,IP,cp_en) = math_Mandel3333to66(J_inverse * H_sym)
endif
endif
! --+>> COLLECTION OF FEM INPUT WITH RETURNING OF RANDOMIZED ODD STRESS AND JACOBIAN <<+--
case (3,4,5)
if (mode == 4) then
CPFEM_dcsde_knownGood = CPFEM_dcsde ! --+>> BACKUP JACOBIAN FROM FORMER CONVERGED INC
else if (mode == 5) then
CPFEM_dcsde = CPFEM_dcsde_knownGood ! --+>> RESTORE CONSISTENT JACOBIAN FROM FORMER CONVERGED INC
end if
call random_number(rnd)
rnd = 2.0_pReal * rnd - 1.0_pReal
materialpoint_Temperature(IP,cp_en) = Temperature
materialpoint_F0(1:3,1:3,IP,cp_en) = ffn
materialpoint_F(1:3,1:3,IP,cp_en) = ffn1
CPFEM_cs(1:6,IP,cp_en) = rnd * CPFEM_odd_stress
CPFEM_dcsde(1:6,1:6,IP,cp_en) = CPFEM_odd_jacobian * math_identity2nd(6)
CPFEM_calc_done = .false.
select case (FEsolver)
case ('Abaqus','Spectral')
mesh_ipCenterOfGravity(1:3,IP,cp_en) = coords(1:3,1)
case ('Marc')
do node = 1,FE_Nnodes(mesh_element(2,cp_en))
FEnodeID = mesh_FEasCP('node',mesh_element(4+node,cp_en))
mesh_node(1:3,FEnodeID) = mesh_node0(1:3,FEnodeID) + coords(1:3,node)
enddo
end select
! --+>> RECYCLING OF FORMER RESULTS (MARC SPECIALTY) <<+--
case (6)
! do nothing
! --+>> RESTORE CONSISTENT JACOBIAN FROM FORMER CONVERGED INC
case (7)
CPFEM_dcsde = CPFEM_dcsde_knownGood
end select
!*** fill output variables with computed values
cauchyStress = CPFEM_cs(1:6,IP,cp_en)
jacobian = CPFEM_dcsdE(1:6,1:6,IP,cp_en)
pstress = materialpoint_P(1:3,1:3,IP,cp_en)
dPdF = materialpoint_dPdF(1:3,1:3,1:3,1:3,IP,cp_en)
if (theTime > 0.0_pReal) then
Temperature = materialpoint_Temperature(IP,cp_en) ! homogenized result except for potentially non-isothermal starting condition.
endif
if (mode < 3 .and. debug_verbosity > 0 .and. ((debug_e == cp_en .and. debug_i == IP) .or. .not. debug_selectiveDebugger)) then
!$OMP CRITICAL (write2out)
write(6,'(a,i8,1x,i2,/,12x,6(f10.3,1x)/)') '<< CPFEM >> stress/MPa at el ip ', cp_en, IP, cauchyStress/1.0e6_pReal
write(6,'(a,i8,1x,i2,/,6(12x,6(f10.3,1x)/))') '<< CPFEM >> jacobian/GPa at el ip ', cp_en, IP, transpose(jacobian)/1.0e9_pReal
call flush(6)
!$OMP END CRITICAL (write2out)
endif
!*** warn if stiffness close to zero
if (all(abs(jacobian) < 1e-10_pReal)) then
call IO_warning(601,cp_en,IP)
endif
!*** remember extreme values of stress and jacobian
if (mode < 3) then
cauchyStress33 = math_Mandel6to33(cauchyStress)
if (maxval(cauchyStress33) > debug_stressMax) then
debug_stressMaxLocation = (/cp_en, IP/)
debug_stressMax = maxval(cauchyStress33)
endif
if (minval(cauchyStress33) < debug_stressMin) then
debug_stressMinLocation = (/cp_en, IP/)
debug_stressMin = minval(cauchyStress33)
endif
jacobian3333 = math_Mandel66to3333(jacobian)
if (maxval(jacobian3333) > debug_jacobianMax) then
debug_jacobianMaxLocation = (/cp_en, IP/)
debug_jacobianMax = maxval(jacobian3333)
endif
if (minval(jacobian3333) < debug_jacobianMin) then
debug_jacobianMinLocation = (/cp_en, IP/)
debug_jacobianMin = minval(jacobian3333)
endif
endif
end subroutine
END MODULE CPFEM