Abstract:Effective thermal conductivity (ETC) is an important property of heterogeneous materials in many thermal management applications. Recently, there is increasing interest to establish the structure-property linkage through machine learning methods. One limitation is that their prediction accuracy is highly dependent on the extremely large space of descriptors of heterogeneous materials due to complex microstructure. Besides, they are also based on a large quantity of data samples, especially for the high contrast heterogeneous materials due to the strongly nonlinear relationship between ETC and volume fraction ratios. In this study, a novel two-stage reduced-order deep learning approach assisted multiscale analysis is proposed to predict the ETC of arbitrary contrast heterogeneous materials with complex microstructure. The key point of the approach is that a two-stage reduced-order strategy is proposed to extract the sample features from any data sample and the feature samples from the database by deep learning approach. The high-fidelity dataset relating the microstructural images of heterogeneous materials with different volume fractions to their corresponding ETC is generated by the asymptotic homogenization method assisted with finite element algorithm. The multiscalebased database and the two-stage reduced-order learning strategy ease the training of neural networks, and enable us to efficiently build more simple networks for approximating mappings of very high complexity. 2D and 3D interpenetrating phase composites were employed as the examples to illustrate the method. The trained network is validated to be able to establish the structure-property linkage accurately and efficiently. Not limited to interpenetrating phase composites, the proposed method can be further extended to predict ETC of other heterogeneous materials.

Transfer Learning for Deep Learning-based Prediction of Lattice Thermal Conductivity

Predicting the Effective Thermal Conductivities of Composite Materials and Porous Media by Machine Learning Methods

Exploring lattice thermal conductivity models via interpretable deep learning to accelerate the discovery of novel materials

Prediction of lattice thermal conductivity with two-stage interpretable machine learning

Predicting lattice thermal conductivity from fundamental material properties using machine learning techniques

Machine Learning Models for the Lattice Thermal Conductivity Prediction of Inorganic Materials

Exploring diamond-like lattice thermal conductivity crystals via feature-based transfer learning

Improving neural network predictions of material properties with limited data using transfer learning

End-to-end Material Thermal Conductivity Prediction through Machine Learning

Machine Learning for Accelerated Prediction of Lattice Thermal Conductivity at Arbitrary Temperature

Predicting lattice thermal conductivity of semiconductors from atomic-information-enhanced CGCNN combined with transfer learning

Lattice Thermal Conductivity: An Accelerated Discovery Guided by Machine Learning

Fast and accurate machine learning prediction of phonon scattering rates and lattice thermal conductivity

Lattice Thermal Conductivity Prediction using Symbolic Regression and Machine Learning

High-throughput computational discovery of 3218 ultralow thermal conductivity and dynamically stable materials by dual machine learning models

A Multiscale Analysis-Assisted Two-Stage Reduced-Order Deep Learning Approach for Effective Thermal Conductivity of Arbitrary Contrast Heterogeneous Materials

Enhancing materials property prediction by leveraging computational and experimental data using deep transfer learning

An interpretable deep learning strategy for effective thermal conductivity prediction of porous materials

CrysGraphFormer: an equivariant graph transformer for prediction of lattice thermal conductivity with interpretability

Predicting lattice thermal conductivity via machine learning: a mini review

On-the-fly training of polynomial machine learning potentials in computing lattice thermal conductivity