Demo and data of the multiclass shape blending method

The multiclass blending code and data from "Remixing Functionally Graded Structures: Data-Driven Topology Optimization with Multiclass Shape Blending" by Chan et al. Using this blending scheme, functionally graded structures can be optimized under a data-driven framework. Check out our paper and extensions of our work at the end of this page!

Blending Scripts and Demo

The code of the proposed multiclass blending scheme is in .\src\blending\shapeBlending.m.

To run a demo of blending in MATLAB or Octave, navigate to .\src\ and run:

demo_blending_2d.m

By interpolating between sets of blending coefficients (not included in the demo), one can achieve smooth and connected changes between different microstructures, such as below.

Data

We also provide the datasets described in Sec. 3.1 of our paper.

Basis Classes

.\src\data_basis_classes contains the basis classes (signed distance functions and other relevant parameters) in *.mat format. See the demo for more information on how to use them for blending.

Training Data for Property Prediction Neural Networks

.\src\data_training contains the datasets created by sampling the blending coefficients. They are used to train the property prediction models for data-driven topology optimization.

• *_coeffs.csv: Coefficients of the shape blending scheme, or the predictors (X) in the models
• *_props.csv: Linear elastic stiffness tensor components and volume fractions, or the responses (Y) in the models

Morphology Types

Each folder above contains the datasets for two morphology types:

• dpp_2d_sp20: Shape and property diverse freeform classes (20 total; only the first 5 are used in the paper)
• truss_2d_red5: Truss-type classes (5 total)

Citation

If our data and/or code has been useful in your research, please cite our work:

Chan, Y.-C., Da, D., Wang, L. et al. (2022). Remixing functionally graded structures: data-driven topology optimization with multiclass shape blending. Structural and Multidisciplinary Optimization, 65(5), 135.

• arXiv version available

@article{Chan2022Remix,
	doi = {10.1007/s00158-022-03224-x},
	year = 2022,
	month = {apr},
	publisher = {Springer Science and Business Media {LLC}},
	volume = {65},
	number = {5},
	author = {Yu-Chin Chan and Daicong Da and Liwei Wang and Wei Chen},
	title = {Remixing functionally graded structures: data-driven topology optimization with multiclass shape blending},
	journal = {Structural and Multidisciplinary Optimization}
}

Papers using the Multiclass Shape Blending Method

• Lee, D., Chan, Y. C., Chen, W., Wang, L., van Beek, A., & Chen, W. (2023). t-METASET: Task-Aware Acquisition of Metamaterial Datasets Through Diversity-Based Active Learning. Journal of Mechanical Design, 145(3), 031704.