Data-driven Materials Research Articles

From Small Data Modeling to Large Language Model Screening: A Dual-Strategy Framework for Materials Intelligent Design.

Small data in materials present significant challenges to constructing highly accurate machine learning models, severely hindering the widespread implementation of data-driven materials intelligent design. In this study, the Dual-Strategy Materials Intelligent Design Framework (DSMID) is introduced, which integrates two innovative methods. The Adversarial domain Adaptive Embedding Generative network (AAEG) transfers data between related property datasets, even with only 90 data points, enhancing material composition characterization and improving property prediction. Additionally, to address the challenge of screening and evaluating numerous alloy designs, the Automated Material Screening and Evaluation Pipeline (AMSEP) is implemented. This pipeline utilizes large language models with extensive domain knowledge to efficiently identify promising experimental candidates through self-retrieval and self-summarization. Experimental findings demonstrate that this approach effectively identifies and prepares new eutectic High Entropy Alloy (EHEA), notably Al14(CoCrFe)19Ni28, achieving an ultimate tensile strength of 1085 MPa and 24% elongation without heat treatment or extra processing. This demonstrates significantly greater plasticity and equivalent strength compared to the typical as-cast eutectic HEA AlCoCrFeNi2.1. The DSMID framework, combining AAEG and AMSEP, addresses the challenges of small data modeling and extensive candidate screening, contributing to cost reduction and enhanced efficiency of material design. This framework offers a promising avenue for intelligent material design, particularly in scenarios constrained by limited data availability.

Predictive Modeling of High-Entropy Alloys and Amorphous Metallic Alloys Using Machine Learning.

High entropy alloys and amorphous metallic alloys represent two distinct classes of advanced alloy materials, each with unique structural characteristics. Their emergence has garnered considerable interest across the materials science and engineering communities, driven by their promising properties, including exceptional strength. However, their extensive compositional diversity poses substantial challenges for systematic exploration, as traditional experimental approaches and high-throughput calculations struggle to efficiently navigate this vast space. While the recent development in data-driven materials discovery could potentially help, such efforts are hindered by the scarcity of comprehensive data and the lack of robust predictive tools that can effectively link alloy composition with specific properties. To address these challenges, we have deployed a machine-learning-based workflow for feature selection and statistical analysis to afford predictive models that accelerate the data-driven discovery and optimization of these advanced materials. Our methodology is validated through two case studies: (i) a regression analysis of the bulk modulus, and (ii) a classification analysis based on glass-forming ability. The Bayesian-optimized regression model trained for the prediction of bulk modulus achieved an R2 of 0.969, an mean absolute error (MAE) of 3.958 GPa, and an root mean square error (RMSE) of 5.411 GPa, while our classification model for predicting glass-forming ability achieved an F1-score of 0.91, an area-under-the-curve of the receiver-operating-characteristic curve of 0.98, and an accuracy of 0.91. Furthermore, by leveraging a wide array of chemical data from diverse literature sources, we have successfully predicted a broad range of properties. This success underscores the efficacy of our modeling approach and emphasizes the importance of a comprehensive feature analysis and judicious feature selection strategy over a mere reliance on complex modeling techniques.

Data-Driven Materials Research and Development for Functional Coatings.

Functional coatings, including organic and inorganic coatings, play a vital role in various industries by providing a protective layer and introducing unique functionalities. However, its design often involves time-consuming experimentation with multiple materials and processing parameters. To overcome these limitations, data-driven approaches are gaining traction in materials science. In this paper, recent advances in data-driven materials research and development (R&D) for functional coatings, highlighting the importance, data sources, working processes, and applications of this paradigm are summarized. It is begun by discussing the challenges of traditional methods, then introduce typical data-driven processes. It is demonstrated how data-driven approaches enable the identification of correlations between input parameters and coating performance, thus allowing for efficient prediction and design. Furthermore, carefully selected case studies are presented across diverse industries that exemplify the effectiveness of data-driven methods in accelerating the discovery of new functional coatings with tailored properties. Finally, the emerging research directions, involving integrating advanced techniques and data from different sources, are addressed. Overall, this review provides an overview of data-driven materials R&D for functional coatings, shedding light on its potential and future developments.

Predicting effective thermal conductivity of HGM composite using ML

Data-driven materials research can be used as an effective tool to supplement conventional approaches in designing new materials. Hollow glass microsphere (HGM) composites are widely used for high-temperature thermal insulation applications, and their synthesis is traditionally done by varying parameters of the constituents on a trial-and-error basis. In this work, a prediction tool based on a supervised machine learning (ML) model is developed to predict the effective thermal conductivity (ETC) of an HGM composite by tailoring the composition and parameters of the constituents to reduce the time and cost involved. A comprehensive database containing the various input features is generated from previous experimental investigations conducted on various HGM composites. ML models, namely random forest regression (RF), K-nearest neighbor (KNN), support vector regression (SVR), and artificial neural network (ANN), are used to predict the ETC of HGM composites. Feature importance analysis showed that the matrix material’s thermal conductivity and the composite’s bulk density have the highest impact on the ETC of the HGM composite, followed by porosity and average microsphere size. ANN emerged as the best model to predict HGM composite ETC with the lowest root mean square error and highest R2 value for predictions. Moreover, a systematic approach for key feature selection using ANN shows that adding or omitting additional features beyond an optimal combination degrades the model’s predictive accuracy.

NJmat: Data-Driven Machine Learning Interface to Accelerate Material Design.

Machine learning techniques have significantly transformed the way materials scientists conduct research. However, the widespread deployment of machine learning software in daily experimental and simulation research for materials and chemical design has been limited. This is partly due to the substantial time investment and learning curve associated with mastering the necessary codes and computational environments. In this paper, we introduce a user-friendly, data-driven machine learning interface featuring multiple "button-clicking" functionalities to streamline the design of materials and chemicals. This interface automates the processes of transforming materials and molecules, performing feature selection, constructing machine learning models, making virtual predictions, and visualizing results. Such automation accelerates materials prediction and analysis in the inverse design process, aligning with the time criteria outlined by the Materials Genome Initiative. With simple button clicks, researchers can build machine learning models and predict new materials once they have gathered experimental or simulation data. Beyond the ease of use, NJmat offers three additional features for data-driven materials design: (1) automatic feature generation for both inorganic materials (from chemical formulas) and organic molecules (from SMILES), (2) automatic generation of Shapley plots, and (3) automatic construction of "white-box" genetic models and decision trees to provide scientific insights. We present case studies on surface design for halide perovskite materials encompassing both inorganic and organic species. These case studies illustrate general machine learning models for virtual predictions as well as the automatic featurization and Shapley/genetic model construction capabilities. We anticipate that this software tool will expedite materials and molecular design within the scope of the Materials Genome Initiative, particularly benefiting experimentalists.

Prevention of Leakage in Machine Learning Prediction for Polymer Composite Properties.

Machine learning (ML) has facilitated property prediction for intricate materials by integrating materials and experimental features such as processing and measurement conditions. However, ML models designed for material properties have often disregarded a common issue of "leakage," resulting in an overestimation of model performance and a decrease in model transferability. This issue can arise from biases inherent in multiple data points obtained from the same experimental group. We provide a critical examination and prevention method of leakage in property prediction for polymer composites. Our proposed method utilizes data partitioning based on the experimental group to ensure that data from the same group are not mixed in both the training and test sets. Evaluation results highlight that the conventional random partitioning unintentionally inflates ML performance through the misuse of experimental features for leaking data bias within the same experimental group rather than explaining the physical causality. In contrast, the proposed method enables the leakage-free utilization of experimental features to improve prediction accuracy while ensuring model transferability. Specifically, when integrating experimental features with polymer and filler features, the conventional method overestimates the prediction performance of electrical conductivity in reducing RMSE by 26% depending on leakage, whereas the proposed method achieves a reduction in RMSE by 5% without leakage. These findings offer valuable guidance for the effective utilization of experimental features in data-driven materials science.

AFLOW-CCE for the thermodynamics of ionic materials.

Accurate thermodynamic stability predictions enable data-driven computational materials design. Standard density functional theory (DFT) approximations have limited accuracy with average errors of a few hundred meV/atom for ionic materials, such as oxides and nitrides. Thus, insightful correction schemes as given by the coordination corrected enthalpies (CCE) method, based on an intuitive parametrization of DFT errors with respect to coordination numbers and cation oxidation states, present a simple, yet accurate solution to enable materials stability assessments. Here, we illustrate the computational capabilities of our AFLOW-CCE software by utilizing our previous results for oxides and introducing new results for nitrides. The implementation reduces the deviations between theory and experiment to the order of the room temperature thermal energy scale, i.e., ∼25 meV/atom. The automated corrections for both materials classes are freely available within the AFLOW ecosystem via the AFLOW-CCE module, requiring only structural inputs.

Extrapolative prediction of small-data molecular property using quantum mechanics-assisted machine learning

Data-driven materials science has realized a new paradigm by integrating materials domain knowledge and machine-learning (ML) techniques. However, ML-based research has often overlooked the inherent limitation in predicting unknown data: extrapolative performance, especially when dealing with small-scale experimental datasets. Here, we present a comprehensive benchmark for assessing extrapolative performance across 12 organic molecular properties. Our large-scale benchmark reveals that conventional ML models exhibit remarkable performance degradation beyond the training distribution of property range and molecular structures, particularly for small-data properties. To address this challenge, we introduce a quantum-mechanical (QM) descriptor dataset, called QMex, and an interactive linear regression (ILR), which incorporates interaction terms between QM descriptors and categorical information pertaining to molecular structures. The QMex-based ILR achieved state-of-the-art extrapolative performance while preserving its interpretability. Our benchmark results, QMex dataset, and proposed model serve as valuable assets for improving extrapolative predictions with small experimental datasets and for the discovery of novel materials/molecules that surpass existing candidates.

A reproducibility study of atomistic line graph neural networks for materials property prediction

ALIGNN performance on 29 regression tasks can be generally well reproduced with minor disparity due to stochasticity.

Future Aspects of Data-Driven Materials Design Using both Simulation and AI

Future Aspects of Data-Driven Materials Design Using both Simulation and AI

Streamlining the automated discovery of porous organic cages.

Self-assembly through dynamic covalent chemistry (DCC) can yield a range of multi-component organic assemblies. The reversibility and dynamic nature of DCC has made prediction of reaction outcome particularly difficult and thus slows the discovery rate of new organic materials. In addition, traditional experimental processes are time-consuming and often rely on serendipity. Here, we present a streamlined hybrid workflow that combines automated high-throughput experimentation, automated data analysis, and computational modelling, to accelerate the discovery process of one particular subclass of molecular organic materials, porous organic cages. We demonstrate how the design and implementation of this workflow aids in the identification of organic cages with desirable properties. The curation of a precursor library of 55 tri- and di-topic aldehyde and amine precursors enabled the experimental screening of 366 imine condensation reactions experimentally, and 1464 hypothetical organic cage outcomes to be computationally modelled. From the screen, 225 cages were identified experimentally using mass spectrometry, 54 of which were cleanly formed as a single topology as determined by both turbidity measurements and 1H NMR spectroscopy. Integration of these characterisation methods into a fully automated Python pipeline, named cagey, led to over a 350-fold decrease in the time required for data analysis. This work highlights the advantages of combining automated synthesis, characterisation, and analysis, for large-scale data curation towards an accessible data-driven materials discovery approach.

(Invited) Data-Driven Design for Energy Materials

Fueled by our abilities to compute materials properties and characteristics orders of magnitude faster than they can be measured and recent advancements in harnessing literature data, we are entering the era of the fourth paradigm of science: data-driven materials design. The Materials Project uses supercomputing together with state-of-the-art quantum mechanical theory to compute the properties of inorganic materials, design novel materials and offer the data for free to the community together with online analysis and design algorithms. Specifically relevant to the Energy Materials Network effort, the Project produces and disseminates materials properties relevant for a broad range of energy materials, interfacial reaction predictions, and Pourbaix diagrams to assess corrosion. We will in this talk survey the existing capabilities, current and ongoing developments and highlight design of novel materials for photocatalysis; from ab initio prediction to synthesis and characterization. Specifically, we identify bottlenecks in the design process and our efforts to accelerate it. Finally, we recognize that the highest impact from the Materials Project comes from its rapidly growing user community, currently counting over 400,000 registered users, and their use of our resources.

Recent progress in the JARVIS infrastructure for next-generation data-driven materials design

The joint automated repository for various integrated simulations (JARVIS) infrastructure at the National Institute of Standards and Technology is a large-scale collection of curated datasets and tools with more than 80 000 materials and millions of properties. JARVIS uses a combination of electronic structure, artificial intelligence, advanced computation, and experimental methods to accelerate materials design. Here, we report some of the new features that were recently included in the infrastructure, such as (1) doubling the number of materials in the database since its first release, (2) including more accurate electronic structure methods such as quantum Monte Carlo, (3) including graph neural network-based materials design, (4) development of unified force-field, (5) development of a universal tight-binding model, (6) addition of computer-vision tools for advanced microscopy applications, (7) development of a natural language processing tool for text-generation and analysis, (8) debuting a large-scale benchmarking endeavor, (9) including quantum computing algorithms for solids, (10) integrating several experimental datasets, and (11) staging several community engagement and outreach events. New classes of materials, properties, and workflows added to the database include superconductors, two-dimensional (2D) magnets, magnetic topological materials, metal-organic frameworks, defects, and interface systems. The rich and reliable datasets, tools, documentation, and tutorials make JARVIS a unique platform for modern materials design. JARVIS ensures the openness of data and tools to enhance reproducibility and transparency and to promote a healthy and collaborative scientific environment.

High-throughput solubility determination for data-driven materials design and discovery in redox flow battery research

High-throughput solubility determination for data-driven materials design and discovery in redox flow battery research

A scalable crystal representation for reverse engineering of novel inorganic materials using deep generative models

The efficient search for crystals with targeted properties is a significant challenge in materials discovery. The rapidly growing field of materials informatics has so far primarily focused on the application of AI/ML models to predict the properties of known crystals from their fundamental and derived properties as descriptors. In the last few years, deep learning-based approaches have spawned a slew of innovative data-driven materials research applications. Materials scientists have used these techniques for the reverse engineering of crystal structures for target applications. However, one of the challenges has been the representation of the crystal structures in the machine readable format. Proposed representations in the literature lack in generality and scalability. In this paper, we train a conditional variational autoencoder with a scalable and invertible representation along with the elemental properties of the constituents as descriptors to inverse-design new crystal structures with specified attributes. When targeting formation energy, we show that our model predicts structures that are not in the complete OQMD database. Finally, we use first-principles density functional theory calculations to validate our findings and show that the developed model is able to generate novel crystal structures for targeted property, i.e. formation energy in this case.

Rapid design of top-performing metal-organic frameworks with qualitative representations of building blocks

Data-driven materials design often encounters challenges where systems possess qualitative (categorical) information. Specifically, representing Metal-organic frameworks (MOFs) through different building blocks poses a challenge for designers to incorporate qualitative information into design optimization, and leads to a combinatorial challenge, with large number of MOFs that could be explored. In this work, we integrated Latent Variable Gaussian Process (LVGP) and Multi-Objective Batch-Bayesian Optimization (MOBBO) to identify top-performing MOFs adaptively, autonomously, and efficiently. We showcased that our method (i) requires no specific physical descriptors and only uses building blocks that construct the MOFs for global optimization through qualitative representations, (ii) is application and property independent, and (iii) provides an interpretable model of building blocks with physical justification. By searching only ~1% of the design space, LVGP-MOBBO identified all MOFs on the Pareto front and 97% of the 50 top-performing designs for the CO2 working capacity and CO2/N2 selectivity properties.

FFMDFPA: A FAIRification Framework for Materials Data with No-Code Flexible Semi-Structured Parser and Application Programming Interfaces.

The FAIR Data Principles are guidelines to ensure Findability, Accessibility, Interoperability, and Reusability of digital resources, which are essential to accelerate data-driven materials science. Despite the development and growing adoption of the FAIR principles, appropriate implementation solutions and software to make data FAIR are still sparse, particularly in standardization of heterogeneous data and subsequent data access. Here, we introduce a FAIRification Framework for Materials Data with No-Code Flexible Semi-Structured Parser and API (FFMDFPA) (API, application programming interface) for raw data processing. Using a template-based parser, FFMDFPA can extract and transform semistructured data in various text formats, providing the flexibility to extend data manipulation without coding. Additionally, FFMDFPA provides a standardized API with efficient query syntax that facilitates seamless data sharing. Taking various text files generated by computational software as examples, we demonstrate the potential utility of FFMDFPA. This work offers important insights toward efficient utilization and reuse of materials data, and the data semantic manipulation implemented in the parser and API can be extended to textual data, which has implications for future data FAIRification.

Data driven design of alkali-activated concrete using sequential learning

This paper presents a novel approach for developing sustainable building materials through Sequential Learning. Data sets with a total of 1367 formulations of different types of alkali-activated building materials, including fly ash and blast furnace slag-based concrete and their respective compressive strength and CO2-footprint, were compiled from the literature to develop and evaluate this approach. Utilizing this data, a comprehensive computational study was undertaken to evaluate the efficacy of the proposed material design methodologies, simulating laboratory conditions reflective of real-world scenarios. The results indicate a significant reduction in development time and lower research costs enabled through predictions with machine learning. This work challenges common practices in data-driven materials development for building materials. Our results show, training data required for data-driven design may be much less than commonly suggested. Further, it is more important to establish a practical design framework than to choose more accurate models. This approach can be immediately implemented into practical applications and can be translated into significant advances in sustainable building materials development.

A Bi-Gradient Dielectric Polymer/High-Κ Nanoparticle/Molecular Semiconductor Ternary Composite for High-Temperature Capacitive Energy Storage.

Attaining compact energy storage under extreme temperature conditions is of paramount importance in the development of advanced dielectric materials. The polymer composite approach has proved effective towards this goal, and addressing the correlation between filler distribution and electrical properties is foremost in designing composite dielectrics, especially in multifiller systems. Here, the design of a bi-gradient polymer composite dielectric using an integrated framework based on the phase field model is reported. This framework can predict the charge-inhibiting behavior of composite dielectrics, which is a key factor impacting the high-temperature capacitive performance but unfortunately is ignored in conventional phase field models. It is found that due to the traps provided by the functional organic fillers, more carriers are trapped near the electrodes and weaken the electric field, thus significantly suppressing the breakdown initialization process. An interpenetrating gradient structure is designed rationally and synthesized experimentally, which exhibits concurrent high energy density (5.51 J cm-3 ) and high charge-discharge efficiency (90%) up to 200 °C. This work provides a strategy to predict the high-temperature energy storage performance of polymer composites containing charge-inhibiting components and helps broaden the scope of data-driven materials design based on phase-field modeling.

EwaldSolidSolution: A High-Throughput Application to Quickly Sample Stable Site Arrangements for Ionic Solid Solutions.

Data-driven materials design of ionic solid solutions often requires sampling (meta)stable site arrangements among the massive number of possibilities, which has been hampered by the lack of relevant methods. Herein, we develop a quick high-throughput sampling application for site arrangements of ionic solid solutions. Given the Ewald Coulombic energies for an initial site arrangement, EwaldSolidSolution updates the modified parts of the energy with varying sites only, which can be exhaustively estimated by using massively parallel processing. Given two representative examples of solid electrolytes, Li10GeP2S12 and Na3Zr2Si2PO12, EwaldSolidSolution successfully calculates the Ewald Coulombic energies of 211,266,225 (235,702,467) site arrangements for Li10GeP2S12 (Na3Zr2Si2PO12) with 216 (160) ion sites per unit cell in 1223.2 (1187.9) seconds: 0.0057898 (0.0050397) milliseconds per site arrangement. The computational cost is enormously saved in comparison with an existing application, which estimates the energy of a site arrangement on the second timescale. The positive correlations between the Ewald Coulombic energies and those estimated by density functional theory calculations show that (meta)stable samples are easily revealed by our computationally inexpensive algorithm. We also reveal that the different-valence nearest-neighbor pairs are distinctively formed in the low-energy site arrangements. EwaldSolidSolution will boost the materials design of ionic solid solutions by attracting broad interest.