- The optimal design of computer experiments algorithm (OPTDOE) (Jin, et al., JSPI, 2005) has made the just-in-time generation of large computer experiments possible. The code has been fully integrated into iSIGHT®, a multidisciplinary design optimization software, developed by Dassault Systèmes.
- The Sequential Optimization and Reliability Assessment (SORA) method (Du and Chen 2004) provides an effective strategy for decoupling probabilistic analyses from optimization loops to improve the efficiency of probabilistic optimization. The method has been integrated into Hyperworks®, an optimization software developed by Altair Engineering.
Material Reconstruction via Transfer Learning
Our transfer learning approach captures spatial correlations for microstructure reconstruction by leveraging the superior capabilities of deep convolutional networks. As described in the paper, “A Transfer Learning Approach for Microstructure Reconstruction and Structure-property Predictions“, our implementation incorporates an encoder-decoder process and a feature-matching optimization algorithm using the pre-trained VGG19 model. Unlike previous approaches that only apply to specific material systems or require a significant amount of prior knowledge for model selection and hyper-parameter tuning, our method provides an off-the-shelf solution to handle complex microstructures, and has the potential of expediting the discovery of new materials.
The code package may be requested for research purposes by clicking on the link below. You will be asked to fill in a form and we will email you following the form completion.
NanoMine: Material Informatics for Polymer Nanocomposites
Using the Material Data Curator System (MDCS) developed at the National Institute of Standards and Technology (NIST), and with the sponsorship of the National Science Foundation (NSF), we have developed a prototype system named “NanoMine” for nanocomposite material data curation, exploration, and analysis. NanoMine currently consists of a growing nanocomposite database; a collection of modular tools for statistical learning, and microstructure characterization and reconstruction (MCR); and simulation software to model bulk nano-polymer composite material response. The underlying principle of NanoMine is to create a living, open-source data resource for nanocomposites that provides data archiving and exchange, statistical analysis, and physics-based modeling for property prediction and materials design.
R Packages for Metamodeling
GPM: Gaussian Process Modeling of Multi-Response Datasets
A general and efficient package for modeling (possibly noisy) datasets via Gaussian processes. The modeling method is published in “Leveraging the nugget parameter for efficient Gaussian process modeling” by Ramin Bostanabad, Tucker Kearney, Siyu Tao, Daniel Apley, Wei Chen (2018) International Journal for Numerical Methods in Engineering, 114, no. 5 (2018): 501-516.
LVGP : Latent Variable Gaussian Process Modeling with Qualitative and Quantitative Input Variables
Fit response surfaces for datasets with latent-variable Gaussian process modeling, predict responses for new inputs, and plot latent variables locations in the latent space (only 1D or 2D). The input variables of the datasets can be quantitative, qualitative/categorical or mixed. The output variable of the datasets is a scalar (quantitative). The optimization of the likelihood function is done using a successive approximation/relaxation algorithm similar to another GP modeling package “GPM”. The modeling method is published in “A Latent Variable Approach to Gaussian Process Modeling with Qualitative and Quantitative Factors” by Yichi Zhang, Siyu Tao, Wei Chen, and Daniel W. Apley (2018) <arXiv:1806.07504>.
2D Orthotropic Metamaterial Dataset
- Data info (2D/3D, static/dynamic): 2D-Static
- Author(s): Northwestern University
- Description: This database contains 85595 orthotropic microstructures represented by 50x50 pixelated matrices as well as the associated independent components of the stiffness tensor calculator by energy-based homogenization, i.e.,C11 ,C12 , C22 and C66. The microstructures are composed of void (air) and solid (Young’s modulus=1(normalized), Poisson’s ratio=0.29). We first performed SIMP-based TO to find a corresponding pixelated microstructure design for each uniformly sampled target stiffness matrix. With 1400 microstructures generated by TO as initial seeds, an iterative stochastic shape perturbation algorithm is employed to perturb microstructure geometries that correspond to extreme and sparse properties.
- Reference publication: Wang, L., Chan, Y. C., Liu, Z., Zhu, P., & Chen, W. (2020). Data-driven metamaterial design with Laplace-Beltrami spectrum as “shape-DNA”. Structural and Multidisciplinary Optimization, 1-16.
- File format: *.mat files
Shape and Property Diverse 3D Isosurface Unit Cells for 3D metamaterials
- Data info (2D/3D, static/dynamic): 3D-Static
- Author(s): Northwestern University
- Description: This dataset contains 3,000 shape and property diverse 3D metamaterial unit cells, which were generated by taking 100 samples from 30 level-set functions (families) derived using cubic symmetry rules. The subsets of 30 families were autonomously selected by METASET, a methodology that uses similarity metrics to jointly measure the diversity of unit cells in both shape and property space through Determinantal Point Processes and an efficient, greedy selection algorithm. The properties provided are the 21 components of the effective 3D elastic tensors for each unit cell. Our diverse datasets can be directly used by metamaterials designers, especially in data-driven design or design of functionally graded structures.
- Reference publication: Chan, Y.-C., Ahmed, F., Wang, L., and Chen, W., 2020. “METASET: Exploring Shape and Property Spaces for Data-Driven Metamaterials Design.“arXiv preprint arXiv:2006.02142. (Submitted to Journal of Mechanical Design).
- File format: *.png and *.stl files