Mechanical MNIST

Mechanical MNIST 集合中的每个数据集都包含 70,000 个（60,000 个训练示例 + 10,000 个测试示例）对受大变形影响的异质材料的有限元模拟结果。通过首先将 MNIST 位图图像 (http://www.pymvpa.org/datadb/mnist.html) 转换为 2D 异构材料块来生成机械 MNIST。与 MNIST 位图（28 美元 × 28 美元像素）一致，材质域是一个 28 美元 × 28 美元的单位正方形。所有模拟均使用 FEniCS 计算平台 (https://fenicsproject.org) 进行。重现这些模拟的代码托管在 GitHub (https://github.com/elejeune11/Mechanical-MNIST/tree/master/generate_dataset) 上。论文“Mechanical MNIST：机械元模型的基准数据集”可在 https://doi.org/10.1016/j.eml.2020.100659 找到。复制手稿中演示的元模型所需的所有代码都可以在 GitHub (https://github.com/elejeune11/Mechanical-MNIST) 上找到。如有问题，请联系 Emma Lejeune ([e-mail])。

Each dataset in the Mechanical MNIST collection contains 70,000 examples (60,000 training samples + 10,000 test samples) of finite element simulation results for heterogeneous materials subjected to large deformations. Mechanical MNIST is generated by first converting MNIST bitmap images (http://www.pymvpa.org/datadb/mnist.html) into 2D heterogeneous material blocks. Consistent with the MNIST bitmaps (28 × 28 pixels), the material domain is a 28 × 28 unit square. All simulations were conducted using the FEniCS computational platform (https://fenicsproject.org). Code for reproducing these simulations is hosted on GitHub at https://github.com/elejeune11/Mechanical-MNIST/tree/master/generate_dataset. The paper "Mechanical MNIST: A Benchmark Dataset for Mechanical Meta-Modeling" is available at https://doi.org/10.1016/j.eml.2020.100659. All code required to replicate the meta-models demonstrated in the manuscript can be found on GitHub at https://github.com/elejeune11/Mechanical-MNIST. For inquiries, please contact Emma Lejeune ([e-mail]).