Prediction Of Material Properties Research Articles

Physical encoding improves OOD performance in deep learning materials property prediction

Physical encoding improves OOD performance in deep learning materials property prediction

Enhancing material property prediction via a dual stream model fusing spatial and topological information

Enhancing material property prediction via a dual stream model fusing spatial and topological information

Transformer-generated atomic embeddings to enhance prediction accuracy of crystal properties with machine learning

Accelerating the discovery of novel crystal materials by machine learning is crucial for advancing various technologies from clean energy to information processing. The machine-learning models for prediction of materials properties require embedding atomic information, while traditional methods have limited effectiveness in enhancing prediction accuracy. Here, we proposed an atomic embedding strategy called universal atomic embeddings (UAEs) for their broad applicability as atomic fingerprints, and generated the UAE tensors based on the proposed CrystalTransformer model. By performing experiments on widely-used materials database, our CrystalTransformer-based UAEs (ct-UAEs) are shown to accurately capture complex atomic features, leading to a 14% improvement in prediction accuracy on CGCNN and 18% on ALIGNN when using formation energies as the target, based on the Materials Project database. We also demonstrated the good transferability of ct-UAEs across various databases. Based on the clustering analysis for multi-task ct-UAEs, the elements in the periodic table can be categorized with reasonable connections between atomic features and targeted crystal properties. After applying ct-UAEs to predict formation energy in hybrid perovskites database, we realized an improvement in accuracy, with a 34% boost in MEGNET and 16% in CGCNN, showcasing their potential as atomic fingerprints to address the data scarcity challenges.

Recent Applications of Theoretical Calculations and Artificial Intelligence in Solid-State Electrolyte Research: A Review

Solid-state electrolytes (SSEs), as key materials for all-solid-state batteries (ASSBs), face challenges such as low ionic conductivity and poor interfacial stability. With the rapid advancement of computational science and artificial intelligence (AI) technologies, theoretical calculations and AI methods are emerging as efficient and important virtual tools for predicting and screening high-performance SSEs. To further promote the development of the SSEs, this review outlines recent applications of theoretical calculations and AI in this field. First, the current applications of theoretical calculation methods, such as density functional theory (DFT) and molecular dynamics (MD), in material structure optimization, electronic property analysis, and ionic transport dynamics are introduced, along with an analysis of their limitations. Second, innovative applications of AI methods, including machine learning (ML) and deep learning (DL), in predicting material properties, analyzing structural features, and simulating interfacial behaviors are elaborated. Subsequently, the synergistic application strategies combining high-throughput screening (HTS), theoretical calculations, and AI methods are highlighted, demonstrating the unique advantages of integrating multiple methodologies in material discovery and performance optimization. Finally, the current research progress is summarized, and future development trends are forecasted. The deep integration of theoretical calculations and AI methods is expected to significantly accelerate the development of high-performance SSE materials, thereby driving the industrial application of ASSBs.

Application of machine learning for environmentally friendly advancement: exploring biomass-derived materials in wastewater treatment and agricultural sector − a review

There are several uses for biomass-derived materials (BDMs) in the irrigation and farming industries. To solve problems with material, process, and supply chain design, BDM systems have started to use machine learning (ML), a new technique approach. This study examined articles published since 2015 to understand better the current status, future possibilities, and capabilities of ML in supporting environmentally friendly development and BDM applications. Previous ML applications were classified into three categories according to their objectives: material and process design, performance prediction and sustainability evaluation. ML helps optimize BDMs systems, predict material properties and performance, reverse engineering, and solve data difficulties in sustainability evaluations. Ensemble models and cutting-edge Neural Networks operate satisfactorily on these datasets and are easily generalized. Ensemble and neural network models have poor interpretability, and there have not been any studies in sustainability assessment that consider geo-temporal dynamics; thus, building ML methods for BDM systems is currently not practical. Future ML research for BDM systems should follow a workflow. Investigating the potential uses of ML in BDM system optimization, evaluation and sustainable development requires further investigation.

Accelerating materials property prediction via a hybrid Transformer Graph framework that leverages four body interactions

Machine learning has advanced the rapid prediction of inorganic materials properties, yet data scarcity for specific properties and capturing thermodynamic stability remains challenging. We propose a framework utilizing a Graph Neural Network with composition-based and crystal structure-based architectures, combined with a transfer learning scheme. This approach accurately predicts energy-related properties (e.g., total energy, energy above the convex hull, energy band gap) and data-scarce mechanical properties (e.g., bulk and shear modulus). Our model incorporates four-body interactions, capturing periodicity and structural characteristics. It outperforms state-of-the-art models in 8 materials property regression tasks. Also, this model predicts local atomic environments and global structural features better than several models. Transfer learning addresses mechanical property data scarcity, while separate architecture analysis allows application to materials lacking crystal structure information. Our framework’s interpretability aids in understanding elemental contributions, enhancing material design and discovery. Continuous advancements promise further performance improvements, driving efficient and accurate materials property prediction.

Assisted Energetic Material Property Prediction through Advanced Transfer Learning with Graph Neural Networks

Assisted Energetic Material Property Prediction through Advanced Transfer Learning with Graph Neural Networks

Benchmarking residue-resolution protein coarse-grained models for simulations of biomolecular condensates.

Intracellular liquid-liquid phase separation (LLPS) of proteins and nucleic acids is a fundamental mechanism by which cells compartmentalize their components and perform essential biological functions. Molecular simulations play a crucial role in providing microscopic insights into the physicochemical processes driving this phenomenon. In this study, we systematically compare six state-of-the-art sequence-dependent residue-resolution models to evaluate their performance in reproducing the phase behaviour and material properties of condensates formed by seven variants of the low-complexity domain (LCD) of the hnRNPA1 protein (A1-LCD)-a protein implicated in the pathological liquid-to-solid transition of stress granules. Specifically, we assess the HPS, HPS-cation-π, HPS-Urry, CALVADOS2, Mpipi, and Mpipi-Recharged models in their predictions of the condensate saturation concentration, critical solution temperature, and condensate viscosity of the A1-LCD variants. Our analyses demonstrate that, among the tested models, Mpipi, Mpipi-Recharged, and CALVADOS2 provide accurate descriptions of the critical solution temperatures and saturation concentrations for the multiple A1-LCD variants tested. Regarding the prediction of material properties for condensates of A1-LCD and its variants, Mpipi-Recharged stands out as the most reliable model. Overall, this study benchmarks a range of residue-resolution coarse-grained models for the study of the thermodynamic stability and material properties of condensates and establishes a direct link between their performance and the ranking of intermolecular interactions these models consider.

EOSnet: Embedded Overlap Structures for Graph Neural Networks in Predicting Material Properties.

Graph Neural Networks (GNNs) have emerged as powerful tools for predicting material properties, yet they often struggle to capture many-body interactions and require extensive manual feature engineering. Here, we present EOSnet (Embedded Overlap Structures for Graph Neural Networks), a novel approach that addresses these limitations by incorporating Gaussian Overlap Matrix (GOM) fingerprints as node features within the GNN architecture. Unlike models that rely on explicit angular terms or human-engineered features, EOSnet efficiently encodes many-body interactions through orbital overlap matrices, providing a rotationally invariant and transferable representation of atomic environments. The model demonstrates superior performance across various prediction tasks of materials' properties, achieving particularly notable results in properties sensitive to many-body interactions. For band gap prediction, EOSnet achieves a mean absolute error of 0.163 eV, surpassing previous state-of-the-art models. The model also excels in predicting mechanical properties and classifying materials, with 97.7% accuracy in metal/nonmetal classification. These results demonstrate that embedding GOM fingerprints into node features enhances the ability of GNNs to capture complex atomic interactions, making EOSnet a powerful tool for materials' discovery and property prediction.

Application of machine learning in polyimide structure design and property regulation

Polyimide (PI) is widely used in modern industry due to its excellent properties. Its synthesis methods and property research have significantly progressed. However, the design and regulation of PI structures through traditional technologies are slow and expensive, which make it difficult to meet the practical demand of modern materials. With the rapid development of high-throughput computing and data-driven technology, machine learning (ML) has become an important method for exploring new materials. Data-driven ML is envisaged as a decisive enabler for new PIs discovery. This paper first introduces the basic workflow and common algorithms of ML. Secondly, applications of ML in material properties prediction, assisting computational simulation technologies and inverse design for desired structures are reviewed. Finally, we discuss the main challenges and possible solutions of ML in PI research.

Predicting stacking fault energy in austenitic stainless steels via physical metallurgy-based machine learning approaches

Stacking fault energy (SFE) significantly influences plastic deformation, strength, and processing performance, making accurate assessment and prediction of SFE essential for material design and optimization. Traditional SFE calculations mainly rely on experimental measurements and thermodynamic theories, with the former usually being time-consuming and the latter limited in applicability at different compositions. To overcome these limitations, this study proposes a machine learning (ML) strategy introducing physical metallurgy (PM) parameters relevant to SFE, aiming to achieve robust predictions. Specifically, this study evaluates three methods for introducing PM information into ML (as an input, an intermediate parameter, and a transfer source), with transfer learning as the best strategy. Initially, various PM parameters were calculated based on alloy composition and temperature, and subsequently used as inputs to train a convolutional neural network (CNN). This source model was then transferred to the SFE prediction model. The results from the model transfer using different PM information show that incorporating phase-transformation driving force (DF) as a source model for SFE prediction provided the most accurate and reliable results. This approach of introducing PM parameters into ML significantly improves the predictive capability of SFE models, offering a new perspective and solution for the prediction of SFE. Furthermore, this method may also be applicable to the prediction of other material properties during material design and optimization.

Mesoscopic Modeling of Bio-Compatible PLGA Polymers with Coarse-Grained Molecular Dynamics Simulations.

A challenging topic in materials engineering is the development of numerical models that can accurately predict material properties with atomistic accuracy, matching the scale and level of detail achieved by experiments. In this regard, coarse-grained (CG) molecular dynamics (MD) simulations are a popular method for achieving this goal. Despite the efforts of the scientific community, a reliable CG model with quasi-atomistic accuracy has not yet been fully achieved for the design and prototyping of materials, especially polymers. In this paper, we describe a CG model for polymers, focusing on the biocompatible poly(lactic-co-glycolic acid) (PLGA), based on a general parametrization strategy with a potentially broader field of applications. In this model, polymers are represented with finite-size ellipsoids, short-range interactions are accounted for with the generalized Gay-Berne potential, while electrostatic and long-range interactions are accounted for with point charges within the ellipsoids. The model was validated against its atomistic counterpart, obtained through a back-mapping process, by comparing physical properties such as glass transition temperature, thermal conductivity, and elastic moduli. We observed quantitative agreement between the atomistic and CG representations, thus opening up the possibility of adopting the proposed model to expand the domain size of typical MD simulations to dimensions comparable to those of experimental setups.

Unlocking new horizons, challenges of integrating machine learning to energy conversion and storage research

ABSTRACT In recent times, artificial intelligence (AI) and machine learning (ML) have emerged as revolutionary technologies with wide-ranging applications across various fields, including energy conversion and storage (ECS) systems. These methods utilise large amounts of data and computational power to predict material properties, optimise energy storage systems, and develop control algorithms for energy conversion devices. This literature analysis focuses on the latest advancements and methodologies in AI/ML applications to ECS systems, encompassing material design and discovery, property prediction, and system optimisation. Furthermore, the study examines the main challenges of integrating ML into these systems. These problems include issues related to data availability and quality, model interpretability, transfer learning, experimental integration, and ethics. Despite the challenges, ML has the potential to revolutionise energy technologies and enhance performance. Advancements in ML-driven sustainable energy technologies are fostering interdisciplinary collaboration and research, offering promising solutions for sustainable energy.

ANÁLISIS DE MODELOS DE INTELIGENCIA ARTIFICIAL PARA PREDECIR LA TEMPERATURA VICAT DE DIVERSOS COMPUESTOS BASADO EN BIOHDPE COMPARANDO DIVERSAS TÉCNICAS DE MACHINE LEARNING

Materials science is an interdisciplinary field that studies the structure, properties, behavior, and performance of materials. This work focuses on polyethylene polymers, the most widely used in the industry, with a particular emphasis on bio-polyethylenes, which align with sustainability goals by being derived from renewable materials such as biomass and agricultural crops, reducing dependence on fossil resources and minimizing the carbon footprint. The VICAT temperature is a key measure for evaluating the thermal resistance of thermoplastic materials, indicating the temperature at which they begin to soften under load. A higher VICAT temperature suggests better resistance to thermal deformation. A BioHDPE matrix, known for its high impact and tensile strength, suitable for demanding applications, was used. To improve mechanical and thermal properties, hemp fibers, almond powder, argan powder, and Halloysite nanotubes (HNT) were incorporated into the matrix, using organic waste to reduce costs. These components, available in the Mediterranean region, affect the VICAT temperature differently: hemp fiber and HNT enhance mechanical strength, while almond and argan powders may reduce it due to their tendency to decompose. Machine learning (ML) techniques were applied to predict the VICAT temperature, aiming to reduce time and costs in the development of new materials. Regression models, including Multiple Linear Regression (MLR), Neural Networks (NN), Decision Trees (DT), and k-Nearest Neighbors (kNN), were compared, evaluated using the Mean Absolute Percentage Error (MAPE) and the coefficient of determination (R²). The best ML model was identified by a low MAPE and high R², ensuring high prediction accuracy. This study demonstrates the potential of ML in predicting material properties, contributing to the efficient and sustainable design of new polymers. Keywords: BioHDPE, Machine Learning, VICAT Temperature, Neural Networks, Prediction

Hybrid-LLM-GNN: integrating large language models and graph neural networks for enhanced materials property prediction

This study combines Graph Neural Networks (GNNs) and Large Language Models (LLMs) to improve material property predictions. By leveraging both embeddings, this hybrid approach achieves up to a 25% improvement over GNN-only model in accuracy.

A Machine Learning Feature Descriptor Approach: Revealing Potential Adsorption Mechanisms for SF6 Decomposition Product Gas-Sensitive Materials.

The man-made gas sulfur hexafluoride (SF6) is an excellent and stable insulating medium. However, some insulation defects can cause SF6 to decompose, threatening the safe operation of power grids. Based on this, it is of great significance to find and effectively control the decomposition products of SF6 in time. Gas sensors have proven to be an effective way to detect these decomposition gases (SO2, SOF2, SO2F2, H2S, and HF). Nanomaterials with gas-sensitive properties are at the heart of gas sensors. In recent years, data-driven machine learning (ML) has been widely used to predict material properties and discover new materials. However, it has become a major challenge to establish a common model between material properties derived from various types of calculations and intelligent algorithms. In order to make some progress in addressing this challenge. In this work, 250 data sets were extracted from 52 publications exploring the detection of SF6 decomposition products by nanocomposites based on relevant work over the past 10 years, and the adsorption behavior of SF6 decomposition products can be predictively analyzed. By comparing six different algorithmic models, the best model for predicting the adsorption distance (XGBoost: R2 = 91.94%) and adsorption energy (GBR: R2 = 78.63%) of SF6 decomposed gas was identified. Subsequently, the importance of each of the selected feature descriptors in predicting the gas adsorption effect was explained. This work combines first-principles computational results and machine-learning algorithms with each other to provide a new research idea for evaluating the gas sensing capability of nanocomposites.

Efficient prediction of potential energy surface and physical properties with Kolmogorov-Arnold Networks

The application of machine learning methods for predicting potential energy surface and physical properties within materials science has garnered significant attention. Among recent advancements, Kolmogorov-Arnold Networks (KANs) have emerged as a promising alternative to traditional Multi-Layer Perceptrons. This study evaluates the impact of substituting Multi-Layer Perceptrons with KANs within four established machine learning frameworks: Allegro, Neural Equivariant Interatomic Potentials, Higher Order Equivariant Message Passing Neural Network (MACE), and the Edge-Based Tensor Prediction Graph Neural Network. Our results demonstrate that the integration of KANs enhances prediction accuracies, especially for complex datasets such as the HfO2 structures. Notably, using KANs exclusively in the output block achieves the most significant improvements, improving prediction accuracy and computational efficiency. Furthermore, employing KANs exclusively in the output block facilitates faster inference and improved computational efficiency relative to utilizing KANs throughout the entire model. The selection of optimal basis functions for KANs depends on the specific problem. Our results demonstrate the strong potential of KANs in enhancing machine learning potentials and material property predictions. Additionally, the proposed methodology offers a generalizable framework that can be applied to other ML architectures.

Artificial Intelligence in the Educational Process of Students of Specialty 132 Material Science

Abstract. Problem. A current problem in recent years is an increase in interest in artificial intelligence, which is due to a significant increase in the power of computing equipment, as well as the availability of data through the global Internet network. The use of artificial intelligence is observed in many fields, in particular in materials science. Today, in the educational process, the issues of creating new materials, predicting properties, additional generation of design features of products are also being solved with the help of artificial intelligence. Goal. The goal of the research in the paper is to demonstrate the use of artificial intelligence in the educational process for students of higher education in the specialty 132 Materials science on the example of teaching the discipline of the selective block "Artificial intelligence in educational technologies". Methodology. Research methods for the application of the platform offer the opportunity to visualize the atomic structures of materials in a three-dimensional format. It helps students analyze material geometry, symmetry, and the spatial arrangement of atoms. Originality. The Materials Project platform is a powerful tool for analyzing and predicting material properties. Due to the fact that it is based on machine learning and provides open online access to the material base, this platform is promising for implementation in the practical activities of students within the discipline "Artificial Intelligence in Educational Technologies". Practical value. The practical training tasks described, which help to acquire the skills of materials scientists, may be used in the field of materials science and related disciplines in the future. The use of software products, in particular the Materials Project database, for the analysis of physico-chemical and mechanical properties of materials is demonstrated.

Optimal pre-train/fine-tune strategies for accurate material property predictions

A pathway to overcome limited data availability in materials science is to use the framework of transfer learning, where a pre-trained (PT) machine learning model (on a larger dataset) can be fine-tuned (FT) on a target (smaller) dataset. We systematically explore the effectiveness of various PT/FT strategies to learn and predict material properties and create generalizable models by PT on multiple properties (MPT) simultaneously. Specifically, we leverage graph neural networks (GNNs) to PT/FT on seven diverse curated materials datasets, with sizes ranging from 941 to 132,752. Besides identifying optimal PT/FT strategies and hyperparameters, we find our pair-wise PT-FT models to consistently outperform models trained from scratch on target datasets. Importantly, our MPT models outperform pair-wise models on several datasets and, more significantly, on a 2D material band gap dataset that is completely out-of-domain. Finally, we expect our PT/FT and MPT frameworks to accelerate materials design and discovery for various applications.

DenseGNN: universal and scalable deeper graph neural networks for high-performance property prediction in crystals and molecules

Modern generative models based on deep learning have made it possible to design millions of hypothetical materials. To screen these candidate materials and identify promising new materials, we need fast and accurate models to predict material properties. Graphical neural networks (GNNs) have become a current research focus due to their ability to directly act on the graphical representation of molecules and materials, enabling comprehensive capture of important information and showing excellent performance in predicting material properties. Nevertheless, GNNs still face several key problems in practical applications: First, although existing nested graph network strategies increase critical structural information such as bond angles, they significantly increase the number of trainable parameters in the model, resulting in a increase in training costs; Second, extending GNN models to broader domains such as molecules, crystalline materials, and catalysis, as well as adapting to small data sets, remains a challenge. Finally, the scalability of GNN models is limited by the over-smoothing problem. To address these issues, we propose the DenseGNN model, which combines Dense Connectivity Network (DCN), hierarchical node-edge-graph residual networks (HRN), and Local Structure Order Parameters Embedding (LOPE) strategies to create a universal, scalable, and efficient GNN model. We have achieved state-of-the-art performance (SOAT) on several datasets, including JARVIS-DFT, Materials Project, QM9, Lipop, FreeSolv, ESOL, and OC22, demonstrating the generality and scalability of our approach. By merging DCN and LOPE strategies into GNN models in computing, crystal materials, and molecules, we have improved the performance of models such as GIN, Schnet, and Hamnet on materials datasets such as Matbench. The LOPE strategy optimizes the embedding representation of atoms and allows our model to train efficiently with a minimal level of edge connections. This substantially reduces computational costs and shortens the time required to train large GNNs while maintaining accuracy. Our technique not only supports building deeper GNNs and avoids performance penalties experienced by other models, but is also applicable to a variety of applications that require large deep learning models. Furthermore, our study demonstrates that by using structural embeddings from pre-trained models, our model not only outperforms other GNNs in distinguishing crystal structures but also approaches the standard X-ray diffraction (XRD) method.