ResInsight/Fwk/AppFwk/cafTensor/cafTensor3.inl
2018-09-07 10:49:00 +02:00

420 lines
14 KiB
C++

/////////////////////////////////////////////////////////////////////////////////
//
// Copyright (C) 2015- Statoil ASA
// Copyright (C) 2015- Ceetron Solutions AS
//
// ResInsight 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.
//
// ResInsight 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 at <http://www.gnu.org/licenses/gpl.html>
// for more details.
//
/////////////////////////////////////////////////////////////////////////////////
#include "cvfAssert.h"
#include "cvfMath.h"
#include "cvfSystem.h"
#include <algorithm>
#include "cvfMatrix3.h"
#include <math.h>
namespace caf {
//--------------------------------------------------------------------------------------------------
///
//--------------------------------------------------------------------------------------------------
template< typename S>
caf::Tensor3<S>::Tensor3()
{
m_tensor[0] = (S) 0.0;
m_tensor[1] = (S) 0.0;
m_tensor[2] = (S) 0.0;
m_tensor[3] = (S) 0.0;
m_tensor[4] = (S) 0.0;
m_tensor[5] = (S) 0.0;
}
//----------------------------------------------------------------------------------------------------
/// Copy constructor
//----------------------------------------------------------------------------------------------------
template <typename S>
inline Tensor3<S>::Tensor3(const Tensor3& other)
{
cvf::System::memcpy(m_tensor, sizeof(m_tensor), other.m_tensor, sizeof(other.m_tensor));
}
//----------------------------------------------------------------------------------------------------
/// Explicit Cast constructor
//----------------------------------------------------------------------------------------------------
template <typename S>
template <typename T>
Tensor3<S>::Tensor3(const Tensor3<T>& other)
{
m_tensor[SXX] = other[Tensor3<T>::SXX];
m_tensor[SYY] = other[Tensor3<T>::SYY];
m_tensor[SZZ] = other[Tensor3<T>::SZZ];
m_tensor[SXY] = other[Tensor3<T>::SXY];
m_tensor[SYZ] = other[Tensor3<T>::SYZ];
m_tensor[SZX] = other[Tensor3<T>::SZX];
}
//----------------------------------------------------------------------------------------------------
/// Constructor with explicit initialization of all tensor elements.
///
//----------------------------------------------------------------------------------------------------
template <typename S>
Tensor3<S>::Tensor3(S sxx, S syy, S szz, S sxy, S syz, S szx)
{
m_tensor[0] = sxx;
m_tensor[1] = syy;
m_tensor[2] = szz;
m_tensor[3] = sxy;
m_tensor[4] = syz;
m_tensor[5] = szx;
}
//--------------------------------------------------------------------------------------------------
///
//--------------------------------------------------------------------------------------------------
template <typename S>
Tensor3<S> caf::Tensor3<S>::invalid()
{
return caf::Tensor3<S>(std::numeric_limits<S>::infinity(), std::numeric_limits<S>::infinity(),
std::numeric_limits<S>::infinity(), std::numeric_limits<S>::infinity(),
std::numeric_limits<S>::infinity(), std::numeric_limits<S>::infinity());
}
//----------------------------------------------------------------------------------------------------
/// Assignment operator
//----------------------------------------------------------------------------------------------------
template <typename S>
inline Tensor3<S>& Tensor3<S>::operator=(const Tensor3& obj)
{
cvf::System::memcpy(m_tensor, sizeof(m_tensor), obj.m_tensor, sizeof(obj.m_tensor));
return *this;
}
//--------------------------------------------------------------------------------------------------
/// Component-wise addition
//--------------------------------------------------------------------------------------------------
template< typename S>
Tensor3<S> caf::Tensor3<S>::operator+(const Tensor3& rhs) const
{
Tensor3<S> result(*this);
result.m_tensor[0] += rhs.m_tensor[0];
result.m_tensor[1] += rhs.m_tensor[1];
result.m_tensor[2] += rhs.m_tensor[2];
result.m_tensor[3] += rhs.m_tensor[3];
result.m_tensor[4] += rhs.m_tensor[4];
result.m_tensor[5] += rhs.m_tensor[5];
return result;
}
//--------------------------------------------------------------------------------------------------
///
//--------------------------------------------------------------------------------------------------
template< typename S>
Tensor3<S> caf::Tensor3<S>::operator*(S scale) const
{
Tensor3<S> result(*this);
result.m_tensor[0] *= scale;
result.m_tensor[1] *= scale;
result.m_tensor[2] *= scale;
result.m_tensor[3] *= scale;
result.m_tensor[4] *= scale;
result.m_tensor[5] *= scale;
return result;
}
//----------------------------------------------------------------------------------------------------
/// Check if matrices are equal using exact comparisons.
//----------------------------------------------------------------------------------------------------
template<typename S>
bool Tensor3<S>::equals(const Tensor3& ten) const
{
for (int i = 0; i < 6; i++)
{
if (m_tensor[i] != ten.m_tensor[i]) return false;
}
return true;
}
//----------------------------------------------------------------------------------------------------
/// Comparison operator. Checks for equality using exact comparisons.
//----------------------------------------------------------------------------------------------------
template <typename S>
bool Tensor3<S>::operator==(const Tensor3& rhs) const
{
return this->equals(rhs);
}
//----------------------------------------------------------------------------------------------------
/// Comparison operator. Checks for not equal using exact comparisons.
//----------------------------------------------------------------------------------------------------
template <typename S>
bool Tensor3<S>::operator!=(const Tensor3& rhs) const
{
int i;
for (i = 0; i < 6; i++)
{
if (m_tensor[i] != rhs.m_tensor[i]) return true;
}
return false;
}
//--------------------------------------------------------------------------------------------------
/// Get modifiable component 0,1,2. E.g. x = v[0];
//--------------------------------------------------------------------------------------------------
template<typename S>
inline S Tensor3<S>::operator[](TensorComponentEnum index) const
{
CVF_TIGHT_ASSERT(index >= 0);
CVF_TIGHT_ASSERT(index < 6);
return m_tensor[index];
}
//--------------------------------------------------------------------------------------------------
/// Get const component 0,1,2. E.g. x = v[0];
//--------------------------------------------------------------------------------------------------
template<typename S>
inline S& Tensor3<S>::operator[](TensorComponentEnum index)
{
CVF_TIGHT_ASSERT(index >= 0);
CVF_TIGHT_ASSERT(index < 6);
return m_tensor[index];
}
//--------------------------------------------------------------------------------------------------
///
//--------------------------------------------------------------------------------------------------
template< typename S>
void Tensor3<S>::setFromInternalLayout(S* tensorData)
{
m_tensor[0] = tensorData[0];
m_tensor[1] = tensorData[1];
m_tensor[2] = tensorData[2];
m_tensor[3] = tensorData[3];
m_tensor[4] = tensorData[4];
m_tensor[5] = tensorData[5];
}
//--------------------------------------------------------------------------------------------------
///
//--------------------------------------------------------------------------------------------------
template< typename S>
void Tensor3<S>::setFromAbaqusLayout(S* tensorData)
{
m_tensor[0] = tensorData[0];
m_tensor[1] = tensorData[1];
m_tensor[2] = tensorData[2];
m_tensor[3] = tensorData[3];
m_tensor[4] = tensorData[5];
m_tensor[5] = tensorData[4];
}
cvf::Mat3d cofactor3(const cvf::Mat3d& mx);
cvf::Vec3d eigenVector3(const cvf::Mat3d& mx, double eigenValue, bool* computedOk);
//--------------------------------------------------------------------------------------------------
/// Compute principal values and optionally the principal directions
/// The tensor must be laid out as follows: SXX, SYY, SZZ, SXY, SYZ, SZX
//--------------------------------------------------------------------------------------------------
template<typename S>
cvf::Vec3f Tensor3<S>::calculatePrincipals( cvf::Vec3f principalDirections[3]) const
{
CVF_TIGHT_ASSERT(m_tensor);
const float floatThreshold = 1.0e-30f;
const double doubleThreshold = 1.0e-60;
cvf::Vec3f principalValues;
// Init return arrays to invalid
principalValues[0] = std::numeric_limits<float>::infinity();
principalValues[1] = std::numeric_limits<float>::infinity();
principalValues[2] = std::numeric_limits<float>::infinity();
if (principalDirections)
{
principalDirections[0] = cvf::Vec3f::ZERO;
principalDirections[1] = cvf::Vec3f::ZERO;
principalDirections[2] = cvf::Vec3f::ZERO;
}
// Return if we have an undefined component
int i;
for (i = 0; i < 6; i++)
{
if (m_tensor[i] == std::numeric_limits<S>::infinity())
{
return principalValues;
}
}
// Return 0, 0, 0 if all components are zero
bool isAllTensCompsZero = true;
for (i = 0; i < 6; i++)
{
if (!(fabs(m_tensor[i]) < floatThreshold))
{
isAllTensCompsZero = false;
break;
}
}
if (isAllTensCompsZero)
{
return cvf::Vec3f::ZERO;
}
double SXX = m_tensor[0], SYY = m_tensor[1], SZZ = m_tensor[2];
double SXY = m_tensor[3], SYZ = m_tensor[4], SZX = m_tensor[5];
double pressure = -(SXX + SYY + SZZ)/3.0;
// Normally we would solve the eigenvalues by solving the 3'rd degree equation:
// -sigma^3 + A*sigma^2 - B*sigma + C = 0
// in which A, B, and C are the invariants of the stress tensor.
// http://www.engapplets.vt.edu/Mohr/java/nsfapplets/MohrCircles2-3D/Theory/theory.htm
// But the roots(eigenvalues) are calculated by transforming the above equation into
// s**3 + aa*s + b = 0 and using the trignometric solution.
// See crc standard mathematical tables 19th edition pp. 103-104.
SXX += pressure;
SYY += pressure;
SZZ += pressure;
double S1, S2, S3;
double AA, BB, CC, DD, angleP;
AA = SXY*SXY + SYZ*SYZ + SZX*SZX - SXX*SYY - SYY*SZZ - SXX*SZZ;
BB = SXX * SYZ * SYZ
+ SYY * SZX * SZX
+ SZZ * SXY * SXY
- SXX * SYY * SZZ
- 2.0 * SXY * SYZ * SZX;
if (fabs(AA) < doubleThreshold)
{
S1 = 0.0;
S2 = 0.0;
S3 = 0.0;
}
else
{
CC = -sqrt(27.0/AA) * BB * 0.5 / AA;
if (CC > 1.0) CC = 1.0;
else if (CC < -1.0) CC = -1.0;
angleP = acos(CC)/3.0;
DD = 2.0*sqrt(AA/3.0);
S1 = DD*cos(angleP);
S2 = DD*cos(angleP + 4.0*cvf::PI_D/3.0);
S3 = DD*cos(angleP + 2.0*cvf::PI_D/3.0);
}
int idxPMin = 2;
int idxPMid = 1;
int idxPMax = 0;
double principalsd[3];
principalsd[idxPMax] = (S1 - pressure);
principalsd[idxPMid] = (S2 - pressure);
principalsd[idxPMin] = (S3 - pressure);
// Sort the principals if we have no Z component in the tensor at all
if ((m_tensor[2] == 0.0f) && (m_tensor[4] == 0.0f) && (m_tensor[5] == 0.0f))
{
if (fabs(principalsd[idxPMin]) > fabs(principalsd[idxPMid])) std::swap(idxPMin, idxPMid);
if (fabs(principalsd[idxPMin]) > fabs(principalsd[idxPMax])) std::swap(idxPMin, idxPMax);
if (principalsd[idxPMax] < principalsd[idxPMid]) std::swap(idxPMax, idxPMid);
principalsd[idxPMin] = 0;
}
// Calculate the principal directions if needed
if (principalDirections)
{
cvf::Mat3d T;
T(0,0) = m_tensor[0]; T(0,1) = m_tensor[3]; T(0,2) = m_tensor[5];
T(1,0) = m_tensor[3]; T(1,1) = m_tensor[1]; T(1,2) = m_tensor[4];
T(2,0) = m_tensor[5]; T(2,1) = m_tensor[4]; T(2,2) = m_tensor[2];
principalDirections[0] = cvf::Vec3f(eigenVector3(T, principalsd[idxPMax], NULL));
principalDirections[0].normalize();
principalDirections[1] = cvf::Vec3f(eigenVector3(T, principalsd[idxPMid], NULL));
principalDirections[1].normalize();
principalDirections[2] = cvf::Vec3f(eigenVector3(T, principalsd[idxPMin], NULL));
principalDirections[2].normalize();
}
principalValues[0] = (float)principalsd[idxPMax];
principalValues[1] = (float)principalsd[idxPMid];
principalValues[2] = (float)principalsd[idxPMin];
return principalValues;
}
//--------------------------------------------------------------------------------------------------
///
//--------------------------------------------------------------------------------------------------
template< typename S>
float caf::Tensor3<S>::calculateVonMises() const
{
return (float) sqrt( ( (m_tensor[0]*m_tensor[0] + m_tensor[1]*m_tensor[1] + m_tensor[2]*m_tensor[2]) ) +
( -(m_tensor[0]*m_tensor[1] + m_tensor[1]*m_tensor[2] + m_tensor[0]*m_tensor[2]) ) +
( 3*(m_tensor[3]*m_tensor[3] + m_tensor[4]*m_tensor[4] + m_tensor[5]*m_tensor[5]) ) );
}
//--------------------------------------------------------------------------------------------------
/// Calculates Trot = rotMx*T*transpose(rotMx)
//--------------------------------------------------------------------------------------------------
template< typename S>
Tensor3<S> caf::Tensor3<S>::rotated(const cvf::Matrix3<S>& rotMx) const
{
cvf::Matrix3<S> tensor(m_tensor[SXX], m_tensor[SXY], m_tensor[SZX],
m_tensor[SXY], m_tensor[SYY], m_tensor[SYZ],
m_tensor[SZX], m_tensor[SYZ], m_tensor[SZZ]);
cvf::Matrix3<S> transposedRotMx = rotMx;
transposedRotMx.transpose();
cvf::Matrix3<S> rotatedTensor = rotMx * tensor * transposedRotMx;
return Tensor3(rotatedTensor(0,0), rotatedTensor(1,1), rotatedTensor(2,2),
rotatedTensor(1,0), rotatedTensor(1,2), rotatedTensor(0,2));
}
}