Thermo-elastic Topology Optimization

(Optional) temperature limits, optimization of voxel mesh.

This is the implementation of the paper Thermo-elastic topology optimization with stress and temperature constraints.

Files

• 3rd/: third-party library
• assets/: user-defined assets
• examples/: several teaching examples
• output/: output directory
• ref/: reference material
• src/: source code
• cmake/: CMake files

Dependencies

A inside library:

• mma: constrained optimization algorithm
• Eigen: linear algebra
• libigl: basic geometry functions

The following dependencies require user installation:

• json: parsing input JSON scenes
• spdlog: logging information
• OpenMP: CPU parallel processing.

sudo apt install libomp-dev

• SuiteSparse: Linear solver. Optional, NOTE: Use of the Intel MKL BLAS is strongly recommended.
• boost: Use filesystem
• AMGCL: Linear solver. Optional.
• CUDA Toolkit: CUDA support. Optional.

Select a Linear solver

If your matrix has less than 50w of freedom, then it is recommended to choose a direct solver (e.g. SuiteSparse):

1. install SuiteSparse.
2. Set ENABLE_AMGCL to OFF and set ENABLE_SUITESPARSE to ON in CMakeLists.txt.

Otherwise, it is recommended to choose an iterative solver (e.g. AMGCL),in CPU:

1. install OpenMP and AMGCL.
2. Set ENABLE_AMGCL to ON, ENABLE_AMGCL_CUDA to OFF and ENABLE_SUITESPARSE to OFF in CMakeLists.txt.

Further, CUDA can be used to speed up the iterative solver:

1. install OpenMP, CUDA Toolkit and AMGCL.
2. Set ENABLE_AMGCL to ON, ENABLE_AMGCL_CUDA to ON and ENABLE_SUITESPARSE to OFF in CMakeLists.txt.

Finally, if all options are set to OFF, then the Eigen build-in iterative solver will be chosen.(not recommended).

Build

1. set path in CMakeLists.txt.

set(CMAKE_CUDA_COMPILER "/path/to/nvcc") # set path to nvcc

mkdir build
cd build
cmake -DCMAKE_BUILD_TYPE=Release ..
make -j 16

Usage

3/28 update

0. Git switch to multiple_top branch.
1. Use example/top-thermolastic-compare-3d to run mechanical(Me)/mechanical thermal(MeTh) topology optimization(Top) and simulation(Sim). The procedure run in following order:
   1. Me Top & MeTh Top -> density(*_MeTop_rho.vtk & _MethTop_rho.vtk) and compliance/volume each iteration(_MeTop_compliance.txt *_MeTop_volume.txt & ...)
   2. clamp density by different threshold(.XX) -> 0/1 density(*_MeSim_threshXX_rho.vtk & *_MeThSim_threshXX_rho.vtk)
   3. MeTh Sim -> temperature(_T.vtk), displacement(_U.vtk), Von Mise Stress(*_von_stress.vtk). Note: open .vtk via Paraview software.
2. Input
   1. Set parameters in *.json. see comments in example/top-thermoelastic-*.json(see comments in examples/top-thermoelastic-BiclampedStructure/config.json and ref paper for MeTh parameters; see comments in examples/top-thermoelastic-compare-3d/config_beam.json)
   2. Redirect in main.cpp or main.cu if ENABLE_AMGCL_CUDA is ON:

   top::fs_path config_file(
   CMAKE_SOURCE_DIR "/examples/top-thermoelastic-compare-3d/${your_config_file}.json");
   

   3. For irregular voxel model(e.g. Lshape), you can define the initial density in main.cpp or main.cu:

    // NOTE: USER DEFINE GRID HERE!!! 
    std::shared_ptr<top::Mesh> sp_mech_mesh;
    std::shared_ptr<top::HeatMesh> sp_thermal_mesh;
    if (ex_name=="Lshape") {
        // L-shape condition
        spdlog::critical("Using User Defined density!");
        top::Tensor3d L_shape_model(len_x, len_y, len_z);
        L_shape_model.setConstant(1);
        // set the initial voxel model
        for (int k = 0; k < len_z; ++k) {
            for (int j = 0; j < len_y; ++j) {
                for (int i = 0; i < len_x; ++i) {
                    if (j > len_y* 0.6 & k >len_z * 0.5) {
                        L_shape_model(i, j, k) = 0;
                    }
                }
            }
        }
   //...
   }
   

3. Output see output/txt/${example_name}/${example_name}_* and output/vtk/${example_name}/${example_name}_*.

---

1. See example/top-thermoelastic-BiclampedStructure or examples/top-thermoelastic-Lshape-condition.

NOTE:

1. "//*" in examples/*/config.json file mean comments.
2. you can modify CONFIG_FILE, OUTPUT_DIR and ASSETS_DIR in examples/*/CMAKEList.txt.
3. you can modify the linear solver arguments prm.solver.tol and prm.solver.maxiter in src/LinearSolver/Amgcl.h or src/LinearSolver/AmgclCuda.h.
4. you should modify example content in *.cpp rather than *.cu, the latter is copied from the former by cmake.