☆ 4.5 Article

Capabilities of Auto-encoders and Principal Component Analysis of the reduction of microstructural images; Application on the acceleration of Phase-Field simulations

COMPUTATIONAL MATERIALS SCIENCE (2023)

期刊

COMPUTATIONAL MATERIALS SCIENCE

卷 216, 期 -, 页码 -

出版社

ELSEVIER

DOI: 10.1016/j.commatsci.2022.111820

关键词

Phase field; Spinodal decomposition; LSTM; GRU; Auto-encoders; PCA; HPC

类别

Materials Science, Multidisciplinary

向作者/读者索取更多资源

Protocol

社区支持

Reagent

社区支持

智能总结 New
摘要

In this study, a data-driven framework based on Phase-Field simulations is proposed, demonstrating the capabilities of neural networks in accurate low dimensionality reduction of simulated microstructural images and time-series analysis. The dataset used was constructed from high-fidelity Phase-Field simulations. Analysis showed that combining auto-encoder neural networks and principal component analysis ensured efficient and significant dimensionality reduction, achieving a reduction ratio of 1/196 with over 80% accuracy. The application of Long Short Term Memory (LSTM) neural networks allowed for next frame predictions, enabling the acceleration of Phase-Field simulation without high computing resources. Various areas of research can benefit from this framework, with different methods proposed for dimensionality reduction (auto-encoders, principal component analysis, Artificial Neural Networks) and time-series analysis (LSTM, Gated Recurrent Unit (GRU)).

In this work, a data-driven framework based on Phase-Field simulations data is proposed to highlight the capabilities of neural networks to ensure accurate low dimensionality reduction of simulated microstructural images and to provide time-series analysis. The dataset was indeed constructed from high-fidelity Phase -Field simulations. Analyses demonstrated that the association of auto-encoder neural networks and principal component analyses leads to ensure efficient and significant dimensionality reduction: 1/196 of reduction ratio with more than 80% of accuracy. These findings give insight to apply analyses on data from the latent dimension. Application of Long Short Term Memory (LSTM) neural networks showed the possibility of making next frame predictions; that makes possible the acceleration of Phase-Field simulation without the need of high computing resources. We discussed the application of such a framework on various areas of research. Different methods are proposed from the conducted analyses, in order to ensure dimensionality reduction (auto-encoders, principal component analysis, Artificial Neural Networks) and time-series analysis (LSTM, Gated Recurrent Unit (GRU)).

作者

我是这篇论文的作者

点击您的名字以认领此论文并将其添加到您的个人资料中。

主要评分

4.5

评分不足

次要评分

新颖性

重要性

科学严谨性

评价这篇论文

Atomistic simulation model on a diffusive timescale based on the extension of the cluster-activation method to continuous space

Ryo Yamada, Munekazu Ohno

Summary: Recently, the phase-field crystal method has gained attention for its ability to simulate the behavior of atoms in a system over diffusive timescales. In this study, a new atomistic simulation model called the continuous cluster-activation method (CAM) was proposed, which extends the previous discrete model to continuous space. This continuous CAM can effectively simulate various physical phenomena on diffusive timescales and utilizes well-defined atomistic properties as input parameters, such as interatomic interaction energies. Its versatility was demonstrated through simulations of crystal growth, homogeneous nucleation, and grain boundary formation.

PHYSICAL REVIEW E (2023)