Large-Scale Worst-Case Topology Optimization

Compilation

The source can be compiled using CMake, while most of the dependcies has be packed into a conda environment. You can activate it by :

conda env create -f environment.yml
conda activate robtop

Then configure and build the project in conda environment:

mkdir build
cd buld
cmake .. -DUSE_CONDA=ON 
make -j4

Usage

Option

• -meshfile: The input mesh model.

• -jsonfile: The input boundary condition in Json format.

• -gridreso: default=200, set the grid resolution along the longest axis (must<1024)

• -volume_ratio: The goal volume ratio of optimized model

• -outdir: The output directory of the results.

• -workmode: 4 alternative mode (wscf/wsff/nscf/nsff), ws/ns means with/no support(fixed) boundary, cf/ff means constrain force direction to surface normal or not.

• -filter_radius: default=2, the sensitivity filter radius in the unit of the voxel length.

• -damp_ratio: default=0.5, the damp ratio of the Optimality Criteria method

• -design_step: default=0.03, the change limit (maximal step length) when updating the density.

• -shell_width: default=3, the shell width to be enforced during the optimization, given in the unit of the voxel length.

• -poisson_ratio:default=0.4, Poisson ratio of the solid material.

• -youngs_modulus: default=1e6, Young's Modulus of the solid material.

• -min_density: default=1e-3, minimal density value to avoid numerical problem.

• -power_penalty: default=3, the power exponent of density penalty.

• -[no]logdensity: output density field in each iteration or not.

• -testname: additional test suits. For example, -testname=testordtop will do general topology optimization, not worst-case optimization.

example

1. Copy the compiled executable to ./bench

2. Copy the dependent shared library to ./bench .

3. Run the following command in CMD or shell

   • Bridge Example

     distribute force Optimization

     ./robtop -jsonfile=./mirbridge/config2.json -meshfile=./mirbridge/mirbridge.obj -outdir=./result/mirbridge/distri/ -power_penalty=3 -volume_ratio=0.4 -filter_radius=2 -gridreso=511 -damp_ratio=0.5 -shell_width=0 -workmode=wscf -poisson_ratio=0.4 -design_step=0.06 -vol_reduction=0.05 -min_density=1e-3 -logdensity -nologcompliance -testname=testdistributeforce
     

     Worst-Case Optimization

     ./robtop -jsonfile=./mirbridge/config2.json -meshfile=./mirbridge/mirbridge.obj -outdir=./result/mirbridge/rob/ -power_penalty=3 -volume_ratio=0.4 -filter_radius=2 -gridreso=511 -damp_ratio=0.5 -shell_width=0 -workmode=wscf -design_step=0.06 -poisson_ratio=0.4 -vol_reduction=0.05 -min_density=1e-3 -logdensity -nologcompliance -testname=None
     
     

Dependency

• CGAL
• Trimesh2 (https://github.com/Forceflow/trimesh2.git)
• CUDA 11
• OpenVDB
• OpenMesh
• glm (https://glm.g-truc.net/0.9.9/)
• Boost
• RapidJson
• gflags
• Spectra (https://github.com/yixuan/spectra.git)

Docker Image

• We provide a docker environment for compiling the code and reproducing the results easily, the docker image can be generated by building the Dockerfile:

  docker build -f ./Dockerfile -t robtop_dev .

  entering the image with GPU enabled:

  docker run -it --gpus=all robtop_dev /bin/bash

  the source code is located in /home/src/robtop。

• There is also a prebuilt version which can be obtained by

  docker pull dockzd/robtop:2.0
  

  This image only contains the executable file and runtime libraries.

Note

To see our original source of paper "Large-Scale Worst-Cast Topology Optimization", Please checkout tag CGF-Paper-Version . The source tree has been refactored.

Reference

• This repository incorporates a voxelization routine that credits the original source, cuda_voxelizer.