Data-driven Materials Research Articles

Coupling physics in machine learning to predict properties of high-temperatures alloys

High-temperature alloy design requires a concurrent consideration of multiple mechanisms at different length scales. We propose a workflow that couples highly relevant physics into machine learning (ML) to predict properties of complex high-temperature alloys with an example of the 9–12 wt% Cr steels yield strength. We have incorporated synthetic alloy features that capture microstructure and phase transformations into the dataset. Identified high impact features that affect yield strength of 9Cr from correlation analysis agree well with the generally accepted strengthening mechanism. As a part of the verification process, the consistency of sub-datasets has been extensively evaluated with respect to temperature and then refined for the boundary conditions of trained ML models. The predicted yield strength of 9Cr steels using the ML models is in excellent agreement with experiments. The current approach introduces physically meaningful constraints in interrogating the trained ML models to predict properties of hypothetical alloys when applied to data-driven materials.

First-principles based machine learning study of oxygen evolution reactions of perovskite oxides using a surface center-environment feature model

Perovskite-type oxide ABO3 catalysts can be used to prepare inexpensive, efficient, durable, and environmentally friendly electrochemical energy conversion devices. The numerous combinatorial substitutions of perovskite oxides provide rich choices of candidates but pose a great challenge to examine all the possibilities both experimentally and computationally. Inspired by the data-driven materials design approach, we proposed a surface center-environment (SCE) feature model and developed first-principles based machine learning (ML) methods to predict the adsorption free energies of intermediate species (HO*, O*, and HOO*) on the surfaces of perovskite oxides and their overpotentials, critical to evaluating the catalytic performance of oxygen evolution reaction (OER). The SCE feature models contain the elementary properties of substitution elements and the structure information of the perovskite surfaces, demonstrated as an effective description of catalytic reactions on surface in ML models. The ML models predicted that the perovskite oxides adsorb the intermediate species most easily when B = Nb, Mo, Ta, W, and Os, and their overpotentials are low when B = Mn, Fe, Co, Ru, Rh, and Ir. This work demonstrated that the machine learning methods based on SCE features can be applied to describe surface chemical reactions and accelerate the screening for potential perovskite catalysts with target properties.

Machine-learning-assisted metasurface design for high-efficiency thermal emitter optimization

Nanophotonic devices can provide solutions to challenges in energy conversion, information technologies, chemical or biological sensing, quantum computing, and secure communications. The realization of practical optical structures and devices is a complex problem due to the multitude of constraints on their optical performance, materials, scalability, and experimental tolerances, all of which are requirements implying large optimization spaces. However, despite the complexity of the process, to date, almost all nanophotonic structures are designed either intuitively or based on a priori selected topologies, and by adjusting a limited number of parameters. These intuition-based models are limited to ad hoc needs and have narrow applicability and predictive power, with the exhaustive parameter searches often performed manually. Since the comprehensive search in hyper-dimensional design space is highly resource-heavy, multi-objective optimization has so far been almost impossible. Humans' restrained capacity to think hyper-dimensionally also limits the perception of multivariate optimization models, and, therefore, advanced machinery is needed to manage the multi-domain, hyper-dimensional design parameter space. In this work, we merge the topology optimization method with deep learning algorithms, such as adversarial autoencoders, and show substantial improvement of the optimization process in terms of computational time (4900 times faster) and final devices efficiencies (∼98%) by providing unparalleled control of the compact design space representations. By enabling efficient, global optimization searches within complex landscapes, the proposed compact hyperparametric representations could become crucial for multi-constrained problems. The proposed approach could enable a much broader scope of the optimal designs and data-driven materials synthesis that goes beyond photonic and optoelectronic applications.

Development of database and information management system for data-driven materials design

With the progress of data-driven materials design, the scale and quality of datasets for intelligence algorithms such as machine learning and data mining have actually become the bottleneck for the practical benefit. The massive experimental data reported in the literatures and those unpublished are significant treasure trove for producing high-quality datasets. However, there has been a lack of specific database or information management system for storing and managing data as well as generating high-quality datasets. Using the Sm-Co alloy system as an example, which is a typical representative of the high-temperature permanent magnetic materials, we established an intelligent database and the corresponding information management system, where the data acquisition, data annotation, data extraction and data format conversion are flexibly integrated. We set up correlation models for the elemental composition, phase constitution, crystal structure, processing, measurement, and properties of the materials. From these models, the collected data are structured and the logical associations of related data are explicit. Thus, high-quality datasets can be produced from the developed system according to certain requirements for materials design. In the developed information management system, the online data annotation by multi-user has been realized. Moreover, the system is advanced in high-efficiency data entry and unifying data format, and precise search for redundant data. It has been confirmed that the present specific database and information management system play an important role in the data-driven materials design, and are promising in the applications in the fields of materials genome initiative and materials informatics.

A Design-to-Device Pipeline for Data-Driven Materials Discovery

The world needs new materials to stimulate the chemical industry in key sectors of our economy: environment and sustainability, information storage, optical telecommunications, and catalysis. Yet, nearly all functional materials are still discovered by "trial-and-error", of which the lack of predictability affords a major materials bottleneck to technological innovation. The average "molecule-to-market" lead time for materials discovery is currently 20 years. This is far too long for industrial needs, as highlighted by the Materials Genome Initiative, which has ambitious targets of up to 4-fold reductions in average molecule-to-market lead times. Such a large step change in progress can only be realistically achieved if one adopts an entirely new approach to materials discovery. Fortunately, a fundamentally new approach to materials discovery has been emerging, whereby data science with artificial intelligence offers a prospective solution to speed up these average molecule-to-market lead times.This approach is known as data-driven materials discovery. Its broad prospects have only recently become a reality, given the timely and major advances in "big data", artificial intelligence, and high-performance computing (HPC). Access to massive data sets has been stimulated by government-regulated open-access requirements for data and literature. Natural-language processing (NLP) and machine-learning (ML) tools that can mine data and find patterns therein are becoming mainstream. Exascale HPC capabilities that can aid data mining and pattern recognition and also generate their own data from calculations are now within our grasp. These timely advances present an ideal opportunity to develop data-driven materials-discovery strategies to systematically design and predict new chemicals for a given device application.This Account shows how data science can afford materials discovery via a four-step "design-to-device" pipeline that entails (1) data extraction, (2) data enrichment, (3) material prediction, and (4) experimental validation. Massive databases of cognate chemical and property information are first forged from "chemistry-aware" natural-language-processing tools, such as ChemDataExtractor, and enriched using machine-learning methods and high-throughput quantum-chemical calculations. New materials for a bespoke application can then be predicted by mining these databases with algorithmic encodings of relationships between chemical structures and physical properties that are known to deliver functional materials. These may take the form of classification, enumeration, or machine-learning algorithms. A data-mining workflow short-lists these predictions to a handful of lead candidate materials that go forward to experimental validation. This design-to-device approach is being developed to offer a roadmap for the accelerated discovery of new chemicals for functional applications. Case studies presented demonstrate its utility for photovoltaic, optical, and catalytic applications. While this Account is focused on applications in the physical sciences, the generic pipeline discussed is readily transferable to other scientific disciplines such as biology and medicine.

Advanced data science toolkit for non-data scientists – A user guide

Emerging modern data analytics attracts much attention in materials research and shows great potential for enabling data-driven design. Data populated from the high-throughput CALPHAD approach enables researchers to better understand underlying mechanisms and to facilitate novel hypotheses generation, but the increasing volume of data makes the analysis extremely challenging. Herein, we introduce an easy-to-use, versatile, and open-source data analytics frontend, ASCENDS (Advanced data SCiENce toolkit for Non-Data Scientists), designed with the intent of accelerating data-driven materials research and development. The toolkit is also of value beyond materials science as it can analyze the correlation between input features and target values, train machine learning models, and make predictions from the trained surrogate models of any scientific dataset. Various algorithms implemented in ASCENDS allow users performing quantified correlation analyses and supervised machine learning to explore any datasets of interest without extensive computing and data science background. The detailed usage of ASCENDS is introduced with an example of experimental high-temperature alloy data.

Mixing enthalpies of alloys with dynamical instability: bcc Ti-V system

Enthalpy of mixing is among the key materials parameters to determine phase stability and phase transformations in solid solutions. The possibility to predict it from first principles in the framework of the density functional theory is one of the corner stones of the modern materials modeling and the future data-driven materials design. Here we have considered body-centered cubic (bcc) Ti-V alloys, a system with high potential for aerospace, automotive biomedical and energy applications, which is known to exhibit the dynamical instability of the crystal lattice for Ti-rich alloys at low temperature. We have calculated the mixing enthalpies ΔH of bcc Ti-V alloys in the whole interval of concentration at high temperature using ab initio molecular dynamics (AIMD) simulations. A comparison with state-of-the-art static calculations at temperature 0 K shows drastic difference between the two methods: while AIMD predicts positive values of ΔH in the whole range of concentrations, the static zero-temperature simulations result in negative values of ΔH for Ti-rich alloys. We have measured the mixing enthalpy of bcc Ti-V alloys experimentally at 1073 K using an isoperibol high temperature Tian-Calvet calorimeter and found that the enthalpies are positive, in agreement with our finite temperature AIMD calculations. We attribute the failure of the standard static calculations of ΔH to lattice distortions associated with the dynamical instability of bcc Ti-V alloys at zero temperature and argue that the effect should be generally important in theoretical predictions of thermodynamic properties, especially for systems with dynamical instability.

The Materials Simulation Toolkit for Machine learning (MAST-ML): An automated open source toolkit to accelerate data-driven materials research

As data science and machine learning methods are taking on an increasingly important role in the materials research community, there is a need for the development of machine learning software tools that are easy to use (even for nonexperts with no programming ability), provide flexible access to the most important algorithms, and codify best practices of machine learning model development and evaluation. Here, we introduce the Materials Simulation Toolkit for Machine Learning (MAST-ML), an open source Python-based software package designed to broaden and accelerate the use of machine learning in materials science research. MAST-ML provides predefined routines for many input setup, model fitting, and post-analysis tasks, as well as a simple structure for executing a multi-step machine learning model workflow. In this paper, we describe how MAST-ML is used to streamline and accelerate the execution of machine learning problems. We walk through how to acquire and run MAST-ML, demonstrate how to execute different components of a supervised machine learning workflow via a customized input file, and showcase a number of features and analyses conducted automatically during a MAST-ML run. Further, we demonstrate the utility of MAST-ML by showcasing examples of recent materials informatics studies which used MAST-ML to formulate and evaluate various machine learning models for an array of materials applications. Finally, we lay out a vision of how MAST-ML, together with complementary software packages and emerging cyberinfrastructure, can advance the rapidly growing field of materials informatics, with a focus on producing machine learning models easily, reproducibly, and in a manner that facilitates model evolution and improvement in the future.

Machine-enabled inverse design of inorganic solid materials: promises and challenges.

Developing high-performance advanced materials requires a deeper insight and search into the chemical space. Until recently, exploration of materials space using chemical intuitions built upon existing materials has been the general strategy, but this direct design approach is often time and resource consuming and poses a significant bottleneck to solve the materials challenges of future sustainability in a timely manner. To accelerate this conventional design process, inverse design, which outputs materials with pre-defined target properties, has emerged as a significant materials informatics platform in recent years by leveraging hidden knowledge obtained from materials data. Here, we summarize the latest progress in machine-enabled inverse materials design categorized into three strategies: high-throughput virtual screening, global optimization, and generative models. We analyze challenges for each approach and discuss gaps to be bridged for further accelerated and rational data-driven materials design.

Lithium adsorption on 2D transition metal dichalcogenides: towards a descriptor for machine learned materials design

Adsorption energy: an efficient descriptor for prediction of single-atom adsorption on 2D TMDs.

Comparative dataset of experimental and computational attributes of UV/vis absorption spectra

The ability to auto-generate databases of optical properties holds great prospects in data-driven materials discovery for optoelectronic applications. We present a cognate set of experimental and computational data that describes key features of optical absorption spectra. This includes an auto-generated database of 18,309 records of experimentally determined UV/vis absorption maxima, λmax, and associated extinction coefficients, ϵ, where present. This database was produced using the text-mining toolkit, ChemDataExtractor, on 402,034 scientific documents. High-throughput electronic-structure calculations using fast (simplified Tamm-Dancoff approach) and traditional (time-dependent) density functional theory were executed to predict λmax and oscillation strengths, f (related to ϵ) for a subset of validated compounds. Paired quantities of these computational and experimental data show strong correlations in λmax, f and ϵ, laying the path for reliable in silico calculations of additional optical properties. The total dataset of 8,488 unique compounds and a subset of 5,380 compounds with experimental and computational data, are available in MongoDB, CSV and JSON formats. These can be queried using Python, R, Java, and MATLAB, for data-driven optoelectronic materials discovery.

ImageDataExtractor: A Tool To Extract and Quantify Data from Microscopy Images.

The rise of data science is leading to new paradigms in data-driven materials discovery. This carries an essential notion that large data sources containing chemical structure and property information can be mined in a fashion that detects and exploits structure-property relationships, such that chemicals can be predicted to suit a given material application. The success of material predictions is predicated on these large data sources of chemical structure and property information being suited to a target application. Microscopy is commonly used to characterize chemical structure, especially in fields such as nanotechnology where material properties are highly dependent on the size and shape of nanoparticles. Large data sources of nanoparticle information stemming from microscopy images would thus be highly beneficial. Millions of microscopy images exist, but they lie fragmented across the literature, typically presented individually within a paper article and usually in a qualitative fashion therein, even though they harbor a wealth of numeric information. We present the ImageDataExtractor toolkit that autoidentifies and autoextracts microscopy images from scientific documents, whereupon it autonomously analyzes each image to produce quantitative particle size and shape information about its subject material. Each image is quantified by decoding its scale bar information using optical character recognition, with help from super-resolution convolutional neural networks where required. Individual particles are detected and profiled using various thresholding, segmentation, polygon fitting, and edge correction routines. The high-throughput operational capability of ImageDataExtractor means that it can be used to generate large-data sources of particle information for data-driven materials discovery. Evaluation metrics, precision and recall, are greater than 80% for the majority of the image processing steps, and precision is above 80% for all critical steps. The ImageDataExtractor tool is released under the MIT license and is available to download from http://www.imagedataextractor.org.

Introducing self-organized maps (SOM) as a visualization tool for materials research and education

Materials informatics is an emerging discipline that opens a new paradigm of science: data-driven materials discovery. The visualization of high dimensional material properties data is challenging due to a lack of appropriate tools. This article describes a workflow for using Self-Organizing Maps (SOM) as a time- and cost-efficient method of visualizing and validating property relationships between materials. SOM is a machine learning technique that uses dimensionality reduction based on similarities between properties to visualize the relationships between materials in a high-dimensional dataset. While the conventionally used visualization method, Ashby plot, displays a pair of material’s properties, SOM is a technique that can display multiple properties that are effective for high-dimensional materials correlation studies. In this paper, SOM was used to validate property correlations among 21 material properties in a dataset of over 398 commercial materials. We employed U-matrix map, clustering map and heatmaps to visualize the SOM we trained. The clustering and heat maps produced from SOM were used to identify unique materials and infer correlations between material properties and fundamental material structure. We have also shown that SOM method can provide multiple layers of interpretation through visualizing not only numerical properties but also categorical information of materials. Lastly, we have used SOM to provide new insight into the quantity of Grüneisen parameter across metals and ceramics. Through these examples, it is demonstrated that SOM can be used for different types of materials analyses for various applications. This paper advocates SOM as an integrated approach to study material property relationships that can be used for both materials education and discovery.

First-Principles Design and Simulations Promote the Development of Nonlinear Optical Crystals.

A hot topic in materials science is to search for nonlinear optical (NLO) crystals, which are indispensable in current laser technology, future optical information, and precision measurements. In the period of the 1980s and 1990s, the anionic group theory proposed by Prof. Chuangtian Chen has greatly promoted the inventions of BaB2O4 (BBO), LiB3O5 (LBO), and KBe2BO3F2 (KBBF) which are widely applied in the ultraviolet (UV) spectral region today. From the beginning of this century, the rapid development of laser science and technology urgently demands new NLO crystals for wider application ranges. However, commercial NLO crystals in deep-UV and mid-infrared (mid-IR) regions are scarce. The challenge arises from the stringent criteria at various wavelengths and inefficient exploration strategy. As such, more comprehensive and quantitative theoretical guidance is necessary to improve and supplement the NLO structure-property understandings. Benefiting from high-performance computing resources, first-principles design and simulations came into being, which is more applicable to the understanding of mid-IR NLO mechanism and suitable for the efficient design of new NLO structures for current needs. In the past decade, a complete set of computational research programs based on first-principles simulations have been developed, which have promoted the development of NLO crystals in the deep-UV and mid-IR regions, and guided the subsequent and further experimental explorations. Based on our developed first-principles materials design system, the discoveries of NLO materials have ranged from basic theoretical design to rapid-prototyping and final experimental synthesis. In this Account, we will concisely summarize our ab initio guided and forward-looking studies on NLO crystals, which are our original contributions to this field and can be consulted by other material fields. First, we will review the development of NLO crystals and the important features of NLO materials. Second, we will summarize the important role of computer-aided design in advancing the NLO material field and our developed NLO material design system based on the first-principles simulations. Third, we will introduce the first-principles design for new deep-UV NLO crystals using two novel design proposals, i.e., interlayer cationic replacement and intralayer anionic substitution. Meanwhile, we will illustrate the hierarchical molecular engineering optimizations for mid-IR NLO crystals by illustrating an extended mid-IR NLO family pedigree, from which many promising mid-IR NLO systems were predicted theoretically and confirmed experimentally. Finally, we will give an outlook to explore new functional NLO crystals guided by our first-principles design and simulations. We believe that the computer-assisted exploration for new functional NLO materials is useful for understanding structure-property relationships and can provide researchers with a new approach to cost-effective and data-driven materials design.

Simultaneous learning of several materials properties from incomplete databases with multi-task SISSO

The identification of descriptors of materials properties and functions that capture the underlying physical mechanisms is a critical goal in data-driven materials science. Only such descriptors will enable a trustful and efficient scanning of materials spaces and possibly the discovery of new materials. Recently, the sure-independence screening and sparsifying operator (SISSO) has been introduced and was successfully applied to a number of materials-science problems. SISSO is a compressed sensing based methodology yielding predictive models that are expressed in form of analytical formulas, built from simple physical properties. These formulas are systematically selected from an immense number (billions or more) of candidates. In this work, we describe a powerful extension of the methodology to a ‘multi-task learning’ approach, which identifies a single descriptor capturing multiple target materials properties at the same time. This approach is specifically suited for a heterogeneous materials database with scarce or partial data, e.g. in which not all properties are reported for all materials in the training set. As showcase examples, we address the construction of materials properties maps for the relative stability of octet-binary compounds, considering several crystal phases simultaneously, and the metal/insulator classification of binary materials distributed over many crystal prototypes.

Data-Driven Learning: A Scaffolding Methodology for CLIL and LSP Teaching and Learning

When dealing with Language for Specific Purposes (LSP), teachers always have to confront with issues which are strictly linked to the specificities of the language of a given field. This is not only the case of language teachers, but it is also and particularly true for CLIL teachers in Italy, who are subject teachers sharing with language teachers some aspects of pupils’ language education. Corpus-based methodology in CLIL classes (the LSP learning environment par excellence) means to empower both teachers and students to develop competences in moving away from mere surface features of text to selecting and understanding meanings and structures. In doing so, they use texts with specific intentions, becoming familiar with tools such as corpora to compensate, for instance, the deficiencies of general dictionaries. Over 25 years ago Tim Johns advocated the learning-centered value of Data-driven Learning (DDL), calling “every student a Sherlock Holmes”. In fact, DDL good practices perfectly align with current theories and practices of SLA, namely the constructivist and learner-centered approaches to language acquisition. It underpins the mandate in contemporary communicative language instruction for the use of authentic language materials and for the development of metalinguistic knowledge and learner autonomy. The uses and benefits of corpora for language learning are widely reported in the literature, although there is still little field practice, in Italy at least. This paper is grounded in the discussion of DDL as a scaffolding method to support the language aspects of CLIL, and the role of data-driven materials in enhancing learning in general. A case study where DDL was successfully used in a vocational context will be presented. The reference corpus was based on oral and written productions by English native speakers, elicited from a picture specifically sketched for the activity (a hairdresser’s salon with multiple actions and objects). Both lexical items and syntactic structures were extracted by students who were confronted with the data and had to deal with different tasks for their analysis. Results were encouraging and students who were exposed to DDL engaged in an involving activity that considerably improved their language skills in their actual working practice.

Robust and synthesizable photocatalysts for CO2 reduction: a data-driven materials discovery

The photocatalytic conversion of the greenhouse gas CO2 to chemical fuels such as hydrocarbons and alcohols continues to be a promising technology for renewable generation of energy. Major advancements have been made in improving the efficiencies and product selectiveness of currently known CO2 reduction electrocatalysts, nonetheless, materials discovery is needed to enable economically viable, industrial-scale CO2 reduction. We report here the largest CO2 photocathode search to date, starting with 68860 candidate materials, using a rational first-principles computation-based screening strategy to evaluate synthesizability, corrosion resistance, visible-light absorption, and compatibility of the electronic structure with fuel synthesis. The results confirm the observation of the literature that few materials meet the stringent CO2 photocathode requirements, with only 52 materials meeting all requirements. The results are well validated with respect to the literature, with 9 of these materials having been studied for CO2 reduction, and the remaining 43 materials are discoveries from our pipeline that merit further investigation.

Data-Driven Materials Modeling with XGBoost Algorithm and Statistical Inference Analysis for Prediction of Fatigue Strength of Steels

With the rapid development of the industry, the demand for new materials is increasing. However, new material development is time-consuming and costly. In this study, we proposed a workflow that uses data to create a materials model that accurately reflects the properties of materials. Six different numerical models for predicting the fatigue strength of steels were constructed with an empirical dataset extracted from a certified database (NIMS MatNavi material database). Because it is very difficult to understand the structure and patterns of large amounts of datasets and develop good predictive models at once, we have sought reliable models through statistical inference analysis, which has not been done in previous studies. We also chose the highest performance model with the accuracy (R2 = 0.9850) by applying the latest XGBoost algorithm. Through further study, we believe that this workflow can be used to develop predictive models on various properties of materials.

Property prediction and properties-to-microstructure inverse analysis of steels by a machine-learning approach

The design of new materials with useful properties is becoming increasingly important. Machine-learning tools Materials Genome Integration System Phase and Property Analysis (MIPHA) and rMIPHA (based on the R programming environment) have been independently developed to accelerate the process of materials discovery via a data-driven materials research approach. In the present work, MIPHA and rMIPHA are applied to steel, where machine-learning-based 2D/3D microstructural analysis, direct analysis of property predictions, and properties-to-microstructure inverse analysis were conducted. The results demonstrate that the prediction models deliver satisfactory performance. The inverse exploration of microstructures related to desired target properties (e.g., stress–strain curve, tensile strength, and total elongation) was realized. MIPHA and rMIPHA are still under improvement. The microstructure-to-processing inverse analysis is expected to be realized in the future.

NOMAD: The FAIR concept for big data-driven materials science

Abstract