Development of a Robust CNN Model for Capturing Microstructure-Property Linkages and Building Property Closures Supporting Material Design

Abstract

Recent works have demonstrated the viability of convolutional neural networks (CNN) for capturing the highly non-linear microstructure-property linkages in high contrast composite material systems. In this work, we develop a new CNN architecture that utilizes a drastically reduced number of trainable parameters for building these linkages, compared to the benchmarks in current literature. This is accomplished by creating CNN architectures that completely avoid the use of fully connected layers, while using the 2-point spatial correlations of the microstructure as the input to the CNN. In addition to increased robustness (because of the much smaller number of trainable parameters), the CNN models developed in this work facilitate the construction of property closures at very low computational cost. This is because it allows for easy exploration of the space of valid 2-point spatial correlations, which is known to be a convex hull. Consequently, one can generate new sets of valid 2-point spatial correlations from previously available valid sets of 2-point spatial correlations, simply as convex combinations. This work demonstrates the significant benefits of utilizing 2-point spatial correlations as the input to the CNN, in place of the voxelated discrete microstructures used in current benchmarks.

An interpretable deep learning approach for designing nanoporous silicon nitride membranes with tunable mechanical properties

“…Machine learning (ML) models are alternative promising tools that can explore the design space in a significantly faster pace in comparison with performing massive number of MD simulations to conduct the forward design approach. Different ML modes such as: support vector machine (SVM) 19 , random forest 20,21 , convolutional neural network (CNN) [22][23][24][25][26] , multi-layer perceptron (MLP) neural network 27 , attention-based transformer neural network 28 and graph-based neural networks 29 are adopted as surrogate forward models to relate the microstructures or microstructural features into mechanical properties in many applications. For instance, Yang et al 22 combined principal component analysis (PCA) and CNN to predict the stress-strain behavior of binary composites up to the failure point.…”

A generalised deep learning-based surrogate model for homogenisation utilising material property encoding and physics-based bounds

“…Dense connections of the last stage increase the number of learnable parameters drastically, thus leading to heavier computation costs and longer training times. Hence, Mann and Kaidindi 20 have developed a CNN model wherein the output of the first stage is directly mapped to outputs. Also, at the end of the first stage, using globally averaged pooling instead of simple flattening was proved to reduce the number of parameters and over-fitting in the model 18 , 23 .…”

“…We have used the working principle of this simple and standard architecture because the primary focus of the present work is to develop data sets that are aware of the material information and to evaluate its influence on the model performance. Though CNN models are free of feature engineering, some models have demonstrated that by supplying modified input instead of simple raw images, model learning capability can be enhanced 17 , 20 , 25 . For example, Mann and Kalidindi 20 used two-point spatial correlations of the micro-structure; Cheng and Wagner 17 have developed RVE-net which uses loading conditions and parameterised geometry (by level set fields) as the input.…”

“…Though CNN models are free of feature engineering, some models have demonstrated that by supplying modified input instead of simple raw images, model learning capability can be enhanced 17 , 20 , 25 . For example, Mann and Kalidindi 20 used two-point spatial correlations of the micro-structure; Cheng and Wagner 17 have developed RVE-net which uses loading conditions and parameterised geometry (by level set fields) as the input. As the preparation of labels is computationally intensive, some CNN models have developed using physics information to learn labels implicitly 17 , 26 .…”

Learning from 2D: Machine learning of 3D effective properties of heterogeneous materials based on 2D microstructure sections

“…In MKS, 3D microstructures are quantified using the n-point spatial statistics with subsequent establishment of quantitative microstructure-property relationships in the form of polynomial functions fitted to data from numerical (e.g., FE) simulations. Other forms of microstructure-property relationships besides polynomial functions are also seen in literature, including statistical learning models [e.g., Gaussian process regression (Marshall and Kalidindi, 2021)], or neural networks, including convolutional neural networks, CNN (Cecen et al, 2018;Yang et al, 2018;Ibragimova et al, 2022;Mann and Kalidindi, 2022) and graph neural networks (Dai et al, 2021;Hestroffer et al, 2023). Owing to the computational efficiency and account for 3D microstructure, surrogate models offer a promising pathway towards practical implementation and industrial adoption of microstructure-sensitive multiscale models of full-scale metal forming processes.…”