Thermo-elasticTopologyOptimization/README.md

# Thermo-elastic Topology Optimization
![Screenshot from 2023-07-02 13-57-29.png](docs%2Fimgs%2FScreenshot%20from%202023-07-02%2013-57-29.png)
<b> (Optional) temperature limits, optimization of voxel mesh. </b>

This is the implementation of the paper [Thermo-elastic topology optimization with stress
and temperature constraints.](ref%2FThermo-elastic%20topology%20optimization%20with%20stress%0Aand%20temperature%20constraints.pdf)

## 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](3rd%2Fmma): constrained optimization algorithm
* [Eigen](https://eigen.tuxfamily.org/): linear algebra
* [libigl](https://github.com/libigl/libigl): basic geometry functions
* 
**The following dependencies require user installation**:
* [json](https://github.com/nlohmann/json): parsing input JSON scenes
* [spdlog](https://github.com/gabime/spdlog): logging information
* OpenMP: CPU parallel processing. 
```bash
sudo apt install libomp-dev
```
* [SuiteSparse](https://github.com/DrTimothyAldenDavis/SuiteSparse): Linear solver. Optional, NOTE: Use of the Intel MKL BLAS is strongly recommended.
* [boost](https://github.com/boostorg/boost): Use filesystem
* [AMGCL](https://github.com/ddemidov/amgcl): Linear solver. Optional. 
* [CUDA Toolkit](https://developer.nvidia.com/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](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](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](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](CMakeLists.txt).
```cmake
set(CMAKE_CUDA_COMPILER "/path/to/nvcc") # set path to nvcc
```
2. 
```bash
mkdir build
cd build
cmake -DCMAKE_BUILD_TYPE=Release ..
make -j 16
```

## Usage
1. See `example/top-thermoelastic-BiclampedStructure` or `examples/top-thermoelastic-Lshape-condition`.
2. If CUDA version is selected, see `examples/top-thermoelastic-BiclampedStructure-cuda` as a template.

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`.