A Transfer Learning Approach for Microstructure Reconstruction and Structure-property Predictions

Abstract

Stochastic microstructure reconstruction has become an indispensable part of computational materials science, but ongoing developments are specific to particular material systems. In this paper, we address this generality problem by presenting a transfer learning-based approach for microstructure reconstruction and structure-property predictions that is applicable to a wide range of material systems. The proposed approach incorporates an encoder-decoder process and feature-matching optimization using a deep convolutional network. For microstructure reconstruction, model pruning is implemented in order to study the correlation between the microstructural features and hierarchical layers within the deep convolutional network. Knowledge obtained in model pruning is then leveraged in the development of a structure-property predictive model to determine the network architecture and initialization conditions. The generality of the approach is demonstrated numerically for a wide range of material microstructures with geometrical characteristics of varying complexity. Unlike previous approaches that only apply to specific material systems or require a significant amount of prior knowledge in model selection and hyper-parameter tuning, the present approach provides an off-the-shelf solution to handle complex microstructures, and has the potential of expediting the discovery of new materials.

Deep materials informatics: Applications of deep learning in materials science

“…Reconstruction of the structure of a disordered heterogeneous material using limited structural information about the original system remains an important problem in modeling of heterogeneous materials. [83] Li et al [35] developed a deep transfer learning based approach for reconstructing statistically equivalent microstructures from arbitrary material systems based on a single given microstructure. In their approach, the input microstructure with k labeled material phases is first encoded to a three-channel (RGB) representation to make it amenable to be used as an input to a pruned version of a pretrained CNN called VGG19.…”

“…In this paper, we discuss some of the recent advances in deep materials informatics for exploring PSPP linkages in materials, after a brief introduction to the basics of deep learning, and its challenges and opportunities. Illustrative examples of deep materials informatics that we review in this paper include learning the chemistry of materials using only elemental composition, [24] structure-aware property prediction, [25,26] crystal structure prediction, [27] learning multiscale homogenization [28,29] and localization [30] linkages in high-contrast composites, structure characterization [31,32] and quantification, [33,34] and microstructure reconstruction [35] and design. [36] We also discuss the future outlook and envisioned impact of deep learning in materials science before summarizing and concluding the paper.…”

Microstructural Materials Design Via Deep Adversarial Learning Methodology

“…In the field of material science, Cang et al [31] included the style transfer loss into the total loss function as a penalty term when training a Variational Auto-Encoder network [29]. In our early work, Li et al [23] takes the style transfer loss as an optimization objective and uses its gradients with respect to each entry in the microstructure image to reconstruct statistically equivalent microstructures. They also discover an interesting intrinsic relationship between the layers included in the calculation of style transfer loss and the reconstructed microstructure: higher level convolutional layers could be dropped to reduce the computational cost while preserving the reconstruction accuracy.…”

“…Spectral Density Function (SDF)-based methods [15], and 7. Transfer Learning-based methods [23,24] However, not all existing MCR techniques are applicable for microstructural materials design. Two major limitations exist: 1) Some MCR methods (methods 3, 4, 5 and 7) are not applicable for microstructural materials design, because no parameters are available to serve as design variables for generating new microstructure designs.…”

iQSPR in XenonPy: A Bayesian Molecular Design Algorithm

“…The sampling efficiency of iQSPR-X is highly influenced by the reliability of the evaluator that predicts the material properties for any given chemical structure. [18,[34][35][36][37][38][39][40][41] In this study, we applied a specific type of transfer learning using pre-trained neural networks. XenonPy currently provides 140,000 pretrained neural networks for the prediction of physical, chemical, electronic, thermodynamic, and mechanical properties of small organic molecules, polymers, and inorganic crystalline materials, with models for 15, 18, and 12 properties of these material types, respectively.…”