Neural Co-Optimization of Structural Topology, Manufacturable Layers, and Path Orientations for Fiber​-Reinforced Composites

SIGGRAPH 2025 (ToG)

Authors:
Tao Liu*, Tianyu Zhang*, Yongxue Chen, Weiming Wang,  Yu Jiang, Yuming Huang, and Charlie C.L. Wang†

---

📖 Overview

We propose a neural network-based computational framework for the simultaneous optimization of:

• Structural topology
• Curved manufacturing layers
• Path orientations

Our method targets fiber-reinforced thermoplastic composites, aiming to maximize anisotropic strength while preserving manufacturability for filament-based multi-axis 3D printing.

Key features:

• Three implicit neural fields to represent geometry, layer sequence, and fiber orientation.
• Integrated and differentiable objectives:Anisotropic strength, Structural volume,Machine motion control, Layer curvature and thickness
• Physical validation: up to 33.1% improvement in failure loads over sequential optimization.

Figure: Overview of our slicing framework for co-optimization of design and manufacturing constraints in multi-axis 3D printing.

🚀 Installation

Environment

• OS: Ubuntu 20.04 LTS
• GPU: Nvidia RTX 4080 / 4090 (≥16 GB VRAM)
• CPU: 13th Gen Intel(R) Core(TM) i7-13700K
• Memory: 32 GB

Performance

• The estimated runtime of TipCantilever (30×20×20) is about 694 seconds on the tested hardware.

Git download the code

git clone https://github.com/RyanTaoLiu/NeuralTOMO
cd NeuralTOMO
mkdir data

Create and activate the Python environment

conda create -n NeuralTO python=3.10
conda activate NeuralTO

Install dependencies

conda install -y -c conda-forge gxx_linux-64=11 binutils_linux-64 sysroot_linux-64
conda install pytorch==2.3.0 torchvision==0.18.0 torchaudio==2.3.0 pytorch-cuda=11.8 -c pytorch -c nvidia
#conda install nvidia/label/cuda-11.8.0::cuda-toolkit
conda install -y -c "nvidia/label/cuda-11.8.0" cuda-nvcc=11.8.89 cuda-cudart=11.8.89 cuda-toolkit=11.8.0

conda install -c conda-forge scikit-sparse

pip install comet_ml paramiko matplotlib ninja pyvista meshio
pip install pymkl
pip install diso==0.1.2

🧪Example of TipCantilever

python main.py \
  --problem=TipCantilever_30_20_20_midLoad \
  --material=PLAPlus \
  --no-isotropic \
  --wSF 0 \
  --wSR 1 \
  --wStress 0 \
  --desireVolumeFraction 0.15 \
  --wHarmonic 1 \
  --numLayers 5 \
  --numNeuronsPerLayer 128 \
  --fourierMap \
  --learningRate 0.001

Results will be saved in ./data.

python main.py --help

Will show the help documents for args.

📂Data structure & Notes

• Boundary condition, See './settings/problems/testproblem.py'. Based on the paper, An efficient 3D topology optimization code written in Matlab[1]. Add any new boundary condition as the new py file follow the testproblem.py and can be called by the arg '--problem'.

• Voigt tensor order: [xx, yy, zz, yz, xz, xy]

• Two rotation matrix quasi-isotropy [2](https://doi.org/10.1007/s00158-023-03586-w) and anisotropy[3](https://doi.org/10.1007/s00158-019-02461-x),

• For stress only based optimization, a small 1e-3regulartion weight should be used for wSR(rigid).

• The result saved every 50 iterations, includes a '.obj' file for the topology optimization marching cubes result(via diso lib), and '.vtk’ file for the voxel-based density('density'), fiber direction ('fiber'), and local printing direction('lpd').

Reference

• [1] K. Liu and A. Tovar. An efficient 3D topology optimization code written in Matlab. Structural and Multidisciplinary Optimization, 50(6):1175–1196, 2014. https://doi.org/10.1007/s00158-014-1107-x

• [2] Y.R. Luo, R. Hewson, and M. Santer. Spatially optimised fibre-reinforced composites with isosurface-controlled additive manufacturing constraints. Structural and Multidisciplinary Optimization, 66, 130, 2023. https://doi.org/10.1007/s00158-023-03586-w

• [3] D.R. Jantos, K. Hackl, and P. Junker. Topology optimization with anisotropic materials, including a filter to smooth fiber pathways. Structural and Multidisciplinary Optimization, 61:2135–2154, 2020. https://doi.org/10.1007/s00158-019-02461-x

• [4] T. Liu, T. Zhang, Y. Chen, Y. Huang, and C.C.L. Wang. Neural Slicer for Multi-Axis 3D Printing. ACM Transactions on Graphics, 43(4), Article 85, 15 pages, July 2024. https://doi.org/10.1145/3658212

• [5] T. Liu, T. Zhang, Y. Chen, W. Wang, Y. Jiang, Y. Huang, and C.C.L. Wang. Neural Co-Optimization of Structural Topology, Manufacturable Layers, and Path Orientations for Fiber-Reinforced Composites. ACM Transactions on Graphics, 44(4), Article 128, 17 pages, August 2025. https://doi.org/10.1145/3730922

Contact information:

• Tao Liu (tao.liu@manchester.ac.uk)
• Tianyu Zhang (tianyu.zhang@manchester.ac.uk)
• Charlie C.L. Wang (changling.wang@manchester.ac.uk)