You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
 
 
 
 
 
 

302 lines
14 KiB

//
// Created by cflin on 4/20/23.
//
#include "Top3d.h"
#include <spdlog/spdlog.h>
#include <cassert>
#include "Eigen/src/Core/Matrix.h"
#include "Util.h"
#include "LinearSolver/Amgcl.h"
#include "LinearSolver/AmgclCuda.h"
namespace da::sha {
namespace top {
Tensor3d Top3d::TopOptMainLoop() {
Eigen::VectorXd xPhys_col(sp_mesh_->GetNumEles());
Eigen::VectorXi chosen_ele_id(sp_mesh_->GetChosenEleIdx());
bool flg_chosen = chosen_ele_id.size() != 0;
if (!flg_chosen) {
// no part chosen
xPhys_col.setConstant(sp_para_->volfrac);
} else {
// pick chosen part to sim
xPhys_col = sp_mesh_->GetInitEleRho();
xPhys_col(chosen_ele_id).setConstant(sp_para_->volfrac);
}
int loop = 0;
double change = 1.0;
double E0 = sp_fea_->sp_material_->E;
double Emin = E0 * sp_para_->E_factor;
// precompute
Eigen::VectorXd dv(sp_mesh_->GetNumEles());
dv.setOnes();
dv = H_ * (dv.array() / Hs_.array()).matrix().eval();
Eigen::VectorXd ele_to_write =
Eigen::VectorXd::Zero(
sp_mesh_->GetLx() * sp_mesh_->GetLy() * sp_mesh_->GetLz());
Eigen::VectorXi pixel_idx = sp_mesh_->GetGlobalIdxOfEleInUse();
spdlog::info("end precompute");
// // clear output dir
// clear_dir(CMAKE_SOURCE_DIR "/output");
LOG_SOLVER
// start iteration
while (change > sp_para_->tol_x && loop < sp_para_->max_loop) {
++loop;
// filter
xPhys_col = H_ * (xPhys_col.array() / Hs_.array()).matrix().eval();
Eigen::VectorXd sK =
(sKe_ * (Emin + xPhys_col.array().pow(sp_para_->penal) *
(E0 - Emin)).matrix().transpose())
.reshaped();
auto v_tri = Vec2Triplet(iK_, jK_, sK);
K_.setFromTriplets(v_tri.begin(), v_tri.end());
IntroduceFixedDofs(K_, F_);
INIT_SOLVER(solver,K_);
U_ = solver.solve(F_);
// compliance
Eigen::VectorXd ce(sp_mesh_->GetNumEles());
for (int i = 0; i < sp_mesh_->GetNumEles(); ++i) {
Eigen::VectorXi dofs_in_ele_i = sp_mesh_->MapEleId2Dofs(i);
Eigen::VectorXd Ue = U_(dofs_in_ele_i);
ce(i) = Ue.transpose() * Ke_ * Ue;
}
double c =
ce.transpose() *
(Emin + xPhys_col.array().pow(sp_para_->penal) * (E0 - Emin)).matrix();
double v = flg_chosen ? xPhys_col(chosen_ele_id).sum() : xPhys_col.sum();
// sensitivity
Eigen::VectorXd dc_drho =
-sp_para_->penal * (E0 - Emin) *
xPhys_col.array().pow(sp_para_->penal - 1.0) * ce.array();
Eigen::VectorXd dc_dx = drho_dx_ * dc_drho;
// mma solver
size_t num_constrants = 1;
size_t num_variables = flg_chosen ? chosen_ele_id.size() : sp_mesh_->GetNumEles();
auto mma = std::make_shared<MMASolver>(num_variables, num_constrants);
Eigen::VectorXd variables_tmp = flg_chosen ? xPhys_col(chosen_ele_id) : xPhys_col;
double f0val = c;
Eigen::VectorXd df0dx = flg_chosen
? dc_dx(chosen_ele_id).eval() /
dc_dx(chosen_ele_id).cwiseAbs().maxCoeff()
: dc_dx / dc_dx.cwiseAbs().maxCoeff();
double fval = v - num_variables * sp_para_->volfrac;
Eigen::VectorXd dfdx = flg_chosen ? dv(chosen_ele_id) : dv;
static Eigen::VectorXd low_bounds = Eigen::VectorXd::Zero(num_variables);
static Eigen::VectorXd up_bounds = Eigen::VectorXd::Ones(num_variables);
// spdlog::info("mma update");
mma->Update(variables_tmp.data(), df0dx.data(), &fval, dfdx.data(),
low_bounds.data(),
up_bounds.data());
if (flg_chosen) {
change = (variables_tmp - xPhys_col(chosen_ele_id)).cwiseAbs().maxCoeff();
xPhys_col(chosen_ele_id) = variables_tmp;
} else {
change = (variables_tmp - xPhys_col).cwiseAbs().maxCoeff();
xPhys_col = variables_tmp;
}
spdlog::critical("Iter: {:3d}, Comp: {:.3e}, Vol: {:.2f}, Change: {:f}", loop, c, v,
change);
#ifdef WRITE_TENSOR_IN_LOOP
// extract vtk
ele_to_write(pixel_idx) = xPhys_col;
Tensor3d ten_xPhys_to_write(sp_mesh_->GetLx() * sp_mesh_->GetLy() * sp_mesh_->GetLz(), 1, 1);
for (int i = 0; i < ele_to_write.size(); ++i) {
ten_xPhys_to_write(i, 0, 0) = ele_to_write(i);
}
ten_xPhys_to_write = ten_xPhys_to_write.reshape(Eigen::array<Eigen::DenseIndex, 3>{
sp_mesh_->GetLx(), sp_mesh_->GetLy(), sp_mesh_->GetLz()});
top::WriteTensorToVtk(
da::WorkingResultDirectoryPath() / ("field_matrix" + std::to_string(loop) + ".vtk"),
ten_xPhys_to_write, sp_mesh_);
#endif
}
// result
rho_ = xPhys_col;
// set 0 to rho of unchosen part
assert(xPhys_col.size());
Eigen::VectorXi continue_idx =
Eigen::VectorXi::LinSpaced(xPhys_col.size(), 0, xPhys_col.size() - 1);
Eigen::VectorXi unchosen_idx = flg_chosen ? SetDifference(continue_idx, chosen_ele_id)
: Eigen::VectorXi();
{
xPhys_col(unchosen_idx).setZero();
ele_to_write(pixel_idx) = xPhys_col;
Tensor3d ten_xPhys_to_write(
sp_mesh_->GetLx() * sp_mesh_->GetLy() * sp_mesh_->GetLz(), 1, 1);
for (int i = 0; i < ele_to_write.size(); ++i) {
ten_xPhys_to_write(i, 0, 0) = ele_to_write(i);
}
ten_xPhys_to_write = ten_xPhys_to_write.reshape(Eigen::array<Eigen::DenseIndex, 3>{
sp_mesh_->GetLx(), sp_mesh_->GetLy(), sp_mesh_->GetLz()});
rho_field_zero_filled_ = ten_xPhys_to_write;
}
{
xPhys_col(unchosen_idx).setOnes();
ele_to_write(pixel_idx) = xPhys_col;
Tensor3d ten_xPhys_to_write(
sp_mesh_->GetLx() * sp_mesh_->GetLy() * sp_mesh_->GetLz(), 1, 1);
for (int i = 0; i < ele_to_write.size(); ++i) {
ten_xPhys_to_write(i, 0, 0) = ele_to_write(i);
}
ten_xPhys_to_write = ten_xPhys_to_write.reshape(Eigen::array<Eigen::DenseIndex, 3>{
sp_mesh_->GetLx(), sp_mesh_->GetLy(), sp_mesh_->GetLz()});
rho_field_one_filled_ = ten_xPhys_to_write;
}
return rho_field_zero_filled_;
}
std::vector<Tensor3d> Top3d::GetTensorOfStress(const Eigen::VectorXd &which_col_of_stress) {
Eigen::VectorXd ele_to_write =
Eigen::VectorXd::Zero(
sp_mesh_->GetLx() * sp_mesh_->GetLy() * sp_mesh_->GetLz());
Eigen::VectorXi pixel_idx = sp_mesh_->GetGlobalIdxOfEleInUse();
// stress
Eigen::MatrixXd mat_stress(sp_mesh_->GetNumEles(), 6);
Eigen::MatrixXd B = sp_fea_->computeBe({0, 0, 0});
for (int i = 0; i < sp_mesh_->GetNumEles(); ++i) {
Eigen::VectorXi dofs_in_ele_i = sp_mesh_->MapEleId2Dofs(i);
Eigen::VectorXd Ue = U_(dofs_in_ele_i);
mat_stress.row(i) = rho_(i) * sp_fea_->computeD() * B * Ue;
}
// fill
std::vector<Tensor3d> vt;
for (int i = 0; i < which_col_of_stress.size(); ++i) {
ele_to_write(pixel_idx) = mat_stress.col(which_col_of_stress(i));
vt.push_back(GetTensorFromCol(ele_to_write));
}
return vt;
}
Tensor3d Top3d::GetVonStress() {
Eigen::VectorXd ele_to_write =
Eigen::VectorXd::Zero(
sp_mesh_->GetLx() * sp_mesh_->GetLy() * sp_mesh_->GetLz());
Eigen::VectorXi pixel_idx = sp_mesh_->GetGlobalIdxOfEleInUse();
// stress
Eigen::MatrixXd mat_stress(sp_mesh_->GetNumEles(), 6);
Eigen::MatrixXd B = sp_fea_->computeBe({0, 0, 0});
for (int i = 0; i < sp_mesh_->GetNumEles(); ++i) {
Eigen::VectorXi dofs_in_ele_i = sp_mesh_->MapEleId2Dofs(i);
Eigen::VectorXd Ue = U_(dofs_in_ele_i);
mat_stress.row(i) = rho_(i) * sp_fea_->computeD() * B * Ue;
}
// fill
Tensor3d vt;
auto computeVonStress = [&](Eigen::VectorXd stress) -> double {
double x = stress(0);
double y = stress(1);
double z = stress(2);
double xy = stress(3);
double yz = stress(4);
double zx = stress(5);
return sqrt(0.5 * (x * x + y * y + z * z) + 3 * (xy * xy + yz * yz + zx * zx));
};
Eigen::VectorXd vonStress = Eigen::VectorXd::Zero(mat_stress.rows());
for (int i = 0; i < sp_mesh_->GetNumEles(); ++i) {
vonStress(i) = computeVonStress(mat_stress.row(i));
}
ele_to_write(pixel_idx) = vonStress;
vt = GetTensorFromCol(ele_to_write);
return vt;
}
Tensor3d Top3d::GetTensorFromCol(const Eigen::VectorXd &proprty_col) {
Tensor3d ten_prop_to_write(sp_mesh_->GetLx() * sp_mesh_->GetLy() * sp_mesh_->GetLz(), 1,
1);
assert(proprty_col.size() == ten_prop_to_write.size());
for (int i = 0; i < proprty_col.size(); ++i) {
ten_prop_to_write(i, 0, 0) = proprty_col(i);
}
ten_prop_to_write = ten_prop_to_write.reshape(
Eigen::array<Eigen::DenseIndex, 3>{sp_mesh_->GetLx(), sp_mesh_->GetLy(),
sp_mesh_->GetLz()});
return ten_prop_to_write;
}
void Top3d::Precompute() {
Eigen::MatrixXi mat_ele2dofs = sp_mesh_->GetEleId2DofsMap();
int dofs_each_ele = sp_mesh_->Get_DOFS_EACH_ELE();// 24 for mathe; 8 for heat
iK_ = Eigen::KroneckerProduct(mat_ele2dofs, Eigen::VectorXi::Ones(dofs_each_ele))
.transpose()
.reshaped();
jK_ = Eigen::KroneckerProduct(mat_ele2dofs, Eigen::RowVectorXi::Ones(dofs_each_ele))
.transpose()
.reshaped();
Ke_ = sp_fea_->computeKe(1.0);
sKe_ = Ke_.reshaped();
// precompute filter
Eigen::VectorXi iH = Eigen::VectorXi::Ones(
sp_mesh_->GetNumEles() *
std::pow(2.0 * (std::ceil(sp_para_->r_min) - 1.0) + 1, 3));
Eigen::VectorXi jH = iH;
Eigen::VectorXd sH(iH.size());
sH.setZero();
int cnt = 0;
Hs_ = Eigen::VectorXd::Zero(sp_mesh_->GetNumEles());
int delta = std::ceil(sp_para_->r_min) - 1;
for (int k = 0; k < sp_mesh_->GetLz(); ++k) {
for (int j = 0; j < sp_mesh_->GetLy(); ++j) {
for (int i = 0; i < sp_mesh_->GetLx(); ++i) {
int ele_id0 = sp_mesh_->MapEleCoord2Id(
(Eigen::MatrixXi(1, 3) << i, j, k).finished())(0);
if (ele_id0 == -1) {
continue;
}
for (int k2 = std::max(k - delta, 0);
k2 <= std::min(k + delta, sp_mesh_->GetLz() - 1);
++k2) {
for (int j2 = std::max(j - delta, 0);
j2 <= std::min(j + delta, sp_mesh_->GetLy() - 1);
++j2) {
for (int i2 = std::max(i - delta, 0);
i2 <= std::min(i + delta, sp_mesh_->GetLx() - 1);
++i2) {
int ele_id1 =
sp_mesh_->MapEleCoord2Id((Eigen::MatrixXi(1, 3)
<< i2, j2, k2).finished())(
0);
if (ele_id1 == -1) {
continue;
}
iH(cnt) = ele_id0;
jH(cnt) = ele_id1;
sH(cnt) =
std::max(0.0,
sp_para_->r_min -
Eigen::Vector3d(i - i2, j - j2,
k - k2).norm());
Hs_(ele_id0) += sH(cnt);
++cnt;
}
}
}
}
}
}
std::vector<Eigen::Triplet<double>> v_tri = Vec2Triplet(iH, jH, sH);
H_ = SpMat(sp_mesh_->GetNumEles(), sp_mesh_->GetNumEles());
H_.setFromTriplets(v_tri.begin(), v_tri.end());
Eigen::VectorXi i_Hs = Eigen::VectorXi::LinSpaced(Hs_.size(), 0,
Hs_.size());
Eigen::SparseMatrix<double> sp_inv_Hs(i_Hs.size(), i_Hs.size());
auto v_inv_Hs_tri = Vec2Triplet(i_Hs, i_Hs, Eigen::VectorXd(1.0 / Hs_.array()));
sp_inv_Hs.setFromTriplets(v_inv_Hs_tri.begin(), v_inv_Hs_tri.end());
drho_dx_ = sp_inv_Hs * H_.transpose();
}
} // namespace top
} // namespace da::sha