Abstract:Determining, understanding, and predicting the so-called structure-property relation is an important task in many scientific disciplines, such as chemistry, biology, meteorology, physics, engineering, and materials science. Structure refers to the spatial distribution of, e.g., substances, material, or matter in general, while property is a resulting characteristic that usually depends in a non-trivial way on spatial details of the structure. Traditionally, forward simulations models have been used for such tasks. Recently, several machine learning algorithms have been applied in these scientific fields to enhance and accelerate simulation models or as surrogate models. In this work, we develop and investigate the applications of six machine learning techniques based on two different datasets from the domain of materials science: data from a two-dimensional Ising model for predicting the formation of magnetic domains and data representing the evolution of dual-phase microstructures from the Cahn-Hilliard model. We analyze the accuracy and robustness of all models and elucidate the reasons for the differences in their performances. The impact of including domain knowledge through tailored features is studied, and general recommendations based on the availability and quality of training data are derived from this.

What problem does this paper attempt to address?

### Problems the Paper Attempts to Solve This paper aims to investigate how machine learning methods can be used to predict structure-property relationships in materials science. Specifically, the authors seek to develop and evaluate six different machine learning techniques based on two datasets from materials science (2D Ising model for predicting magnetic domain formation and Cahn-Hilliard model representing the evolution of biphasic microstructures) to analyze the accuracy and robustness of these models and explore the impact of incorporating domain knowledge through feature engineering on model performance. ### Summary of Main Content 1. **Background and Motivation** - Identifying, understanding, and predicting the so-called structure-property relationships is an important task in many scientific fields such as chemistry, biology, meteorology, physics, engineering, and materials science. - Traditionally, forward simulation models have been used for such tasks, but in recent years, machine learning algorithms have been applied in these scientific fields to enhance and accelerate simulation models or serve as surrogate models. 2. **Datasets and Methods** - Two datasets from materials science were used: one based on the 2D Ising model for predicting magnetic domain formation, and the other based on the Cahn-Hilliard model for representing the evolution of biphasic microstructures. - Four different machine learning methods were studied: regression models based on simple features (grad-PCR), support vector regression based on power spectral density features (physFeat1-SVR), highly specialized physical feature support vector regression (physFeat2-SVR), convolutional neural networks (CNN), and multi-channel convolutional neural networks (MuCha-CNN). 3. **Experimental Results and Analysis** - The accuracy and robustness of all models were analyzed, and the reasons for their performance differences were explained. - The impact of incorporating domain knowledge through feature engineering was studied, and general recommendations were made based on the availability and quality of training data. ### Conclusion By developing and evaluating various machine learning techniques, this paper demonstrates how to effectively predict structure-property relationships in materials science and explores the impact of different feature engineering methods on model performance. These methods not only help overcome the limitations of traditional simulation methods in terms of computational cost and robustness but also provide new ideas and technical means for further research in the field of materials science.

Efficient Surrogate Models for Materials Science Simulations: Machine Learning-based Prediction of Microstructure Properties

Machine-learned multi-system surrogate models for materials prediction

Self-supervised Learning and Prediction of Microstructure Evolution with Convolutional Recurrent Neural Networks

Accelerating phase-field-based microstructure evolution predictions via surrogate models trained by machine learning methods

What Information is Necessary and Sufficient to Predict Materials Properties using Machine Learning?

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

Improving direct physical properties prediction of heterogeneous materials from imaging data via convolutional neural network and a morphology-aware generative model

What can machine learning help with microstructure-informed materials modeling and design?

Machine Learning in Materials Design: Algorithm and Application*

Towards High-Throughput Microstructure Simulation in Compositionally Complex Alloys Via Machine Learning

Advancing materials science through next-generation machine learning

Recent advances and applications of machine learning in solid-state materials science

Machine Learning for Structural Materials

Materials Discovery and Design Using Machine Learning

Learning to Predict Material Structure from Neutron Scattering Data

Application of machine learning for advanced material prediction and design

Time series forecasting of multiphase microstructure evolution using deep learning

Synergy of semiempirical models and machine learning in computational chemistry

Application of Machine Learning in Material Synthesis and Property Prediction

Accelerating microstructure modelling via machine learning: a new method combining Autoencoder and ConvLSTM