Removed unused variables and renamed some variables for TRIP model
This commit is contained in:
parent
e243632eb5
commit
0d36a2d3e2
|
@ -1021,9 +1021,8 @@ subroutine lattice_init
|
|||
integer(pInt) :: section = 0_pInt,i
|
||||
real(pReal), dimension(:), allocatable :: &
|
||||
CoverA, & !< c/a ratio for hex type lattice
|
||||
aA, & !< lattice parameter a for fcc austenite
|
||||
aM, & !< lattice paramater a for bcc martensite
|
||||
cM !< lattice parameter c for bcc martensite
|
||||
a_fcc, & !< lattice parameter a for fcc austenite
|
||||
a_bcc !< lattice paramater a for bcc martensite
|
||||
|
||||
mainProcess: if (worldrank == 0) then
|
||||
write(6,'(/,a)') ' <<<+- lattice init -+>>>'
|
||||
|
@ -1164,9 +1163,8 @@ subroutine lattice_init
|
|||
allocate(lattice_interactionTwinTwin(lattice_maxNtwin,lattice_maxNtwin,Nphases),source=0_pInt) ! other:me
|
||||
|
||||
allocate(CoverA(Nphases),source=0.0_pReal)
|
||||
allocate(aA(Nphases),source=0.0_pReal)
|
||||
allocate(aM(Nphases),source=0.0_pReal)
|
||||
allocate(cM(Nphases),source=0.0_pReal)
|
||||
allocate(a_fcc(Nphases),source=0.0_pReal)
|
||||
allocate(a_bcc(Nphases),source=0.0_pReal)
|
||||
|
||||
rewind(fileUnit)
|
||||
line = '' ! to have it initialized
|
||||
|
@ -1224,12 +1222,10 @@ subroutine lattice_init
|
|||
lattice_C66(6,6,section) = IO_floatValue(line,positions,2_pInt)
|
||||
case ('covera_ratio','c/a_ratio','c/a')
|
||||
CoverA(section) = IO_floatValue(line,positions,2_pInt)
|
||||
case ('aa', 'a_a', 'a_austenite', 'a_fcc')
|
||||
aA(section) = IO_floatValue(line,positions,2_pInt)
|
||||
case ('am', 'a_m', 'a_martensite', 'a_bcc')
|
||||
aM(section) = IO_floatValue(line,positions,2_pInt)
|
||||
case ('cm', 'c_m', 'c_martensite')
|
||||
cM(section) = IO_floatValue(line,positions,2_pInt)
|
||||
case ('a_fcc')
|
||||
a_fcc(section) = IO_floatValue(line,positions,2_pInt)
|
||||
case ('a_bcc')
|
||||
a_bcc(section) = IO_floatValue(line,positions,2_pInt)
|
||||
case ('thermal_conductivity11')
|
||||
lattice_thermalConductivity33(1,1,section) = IO_floatValue(line,positions,2_pInt)
|
||||
case ('thermal_conductivity22')
|
||||
|
@ -1301,10 +1297,10 @@ subroutine lattice_init
|
|||
do i = 1_pInt,Nphases
|
||||
if ((CoverA(i) < 1.0_pReal .or. CoverA(i) > 2.0_pReal) &
|
||||
.and. lattice_structure(i) == LATTICE_hex_ID) call IO_error(131_pInt,el=i) ! checking physical significance of c/a
|
||||
call lattice_initializeStructure(i, CoverA(i), aA(i), aM(i), cM(i))
|
||||
call lattice_initializeStructure(i, CoverA(i), a_fcc(i), a_bcc(i))
|
||||
enddo
|
||||
|
||||
deallocate(CoverA,aA,aM,cM)
|
||||
deallocate(CoverA,a_fcc,a_bcc)
|
||||
|
||||
end subroutine lattice_init
|
||||
|
||||
|
@ -1312,7 +1308,7 @@ end subroutine lattice_init
|
|||
!--------------------------------------------------------------------------------------------------
|
||||
!> @brief Calculation of Schmid matrices, etc.
|
||||
!--------------------------------------------------------------------------------------------------
|
||||
subroutine lattice_initializeStructure(myPhase,CoverA,aA,aM,cM)
|
||||
subroutine lattice_initializeStructure(myPhase,CoverA,a_fcc,a_bcc)
|
||||
use prec, only: &
|
||||
tol_math_check
|
||||
use math, only: &
|
||||
|
@ -1337,9 +1333,8 @@ subroutine lattice_initializeStructure(myPhase,CoverA,aA,aM,cM)
|
|||
integer(pInt), intent(in) :: myPhase
|
||||
real(pReal), intent(in) :: &
|
||||
CoverA, &
|
||||
aA, &
|
||||
aM, &
|
||||
cM
|
||||
a_fcc, &
|
||||
a_bcc
|
||||
|
||||
real(pReal), dimension(3) :: &
|
||||
sdU, snU, &
|
||||
|
@ -1426,10 +1421,10 @@ subroutine lattice_initializeStructure(myPhase,CoverA,aA,aM,cM)
|
|||
zb(1:3,i) = real(LATTICE_fcc_bainVariant(7:9,i),pReal)
|
||||
|
||||
ub(1:3,1:3,i) = 0.0_pReal
|
||||
if ((aA > 0.0_pReal) .and. (aM > 0.0_pReal) .and. (abs(cM) <= tiny(0.0_pReal))) then
|
||||
ub(1:3,1:3,i) = (aM/aA)*math_tensorproduct(xb(1:3,i), xb(1:3,i)) + &
|
||||
sqrt(2.0_pReal)*(aM/aA)*math_tensorproduct(yb(1:3,i), yb(1:3,i)) + &
|
||||
sqrt(2.0_pReal)*(aM/aA)*math_tensorproduct(zb(1:3,i), zb(1:3,i))
|
||||
if ((a_fcc > 0.0_pReal) .and. (a_bcc > 0.0_pReal)) then
|
||||
ub(1:3,1:3,i) = (a_bcc/a_fcc)*math_tensorproduct(xb(1:3,i), xb(1:3,i)) + &
|
||||
sqrt(2.0_pReal)*(a_bcc/a_fcc)*math_tensorproduct(yb(1:3,i), yb(1:3,i)) + &
|
||||
sqrt(2.0_pReal)*(a_bcc/a_fcc)*math_tensorproduct(zb(1:3,i), zb(1:3,i))
|
||||
endif
|
||||
enddo
|
||||
do i = 1_pInt, myNcleavage ! assign cleavage system vectors
|
||||
|
|
Loading…
Reference in New Issue