DeepXRD, a Deep Learning Model for Predicting XRD spectrum from Material Composition

ACS Applied Materials & Interfaces, Journal Year: 2022, Volume and Issue: 14(35), P. 40102 - 40115

Published: Aug. 26, 2022

One of the long-standing problems in materials science is how to predict a material's structure and then its properties given only composition. Experimental characterization crystal structures has been widely used for determination, which is, however, too expensive high-throughput screening. At same time, directly predicting from compositions remains challenging unsolved problem. Herein we propose deep learning algorithm XRD spectrum composition material, can be infer key structural features downstream analysis such as system or space group classification lattice parameter determination property prediction. Benchmark studies on two data sets show that our DeepXRD achieve good performance prediction evaluated over test sets. It thus screening huge discovery.

Language: Английский

---

Predicting the Crystal Structure and Lattice Parameters of the Perovskite Materials via Different Machine Learning Models Based on Basic Atom Properties

Crystals, Journal Year: 2022, Volume and Issue: 12(11), P. 1570 - 1570

Published: Nov. 3, 2022

Perovskite materials have high potential for the renewable energy sources such as solar PV cells, fuel etc. Different structural distortions crystal structure and lattice parameters a critical impact on determination of perovskite’s strength, stability, overall performance in applications. To improve perovskite accelerate prediction different distortions, few ML models been established to predict type structures their using basic atom characteristics materials. In this work, random forest (RF), support vector machine (SVM), neural network (NN), genetic algorithm (GA) supported (GA-NN) established, whereas regression (SVR) algorithm-supported (GA-SVR) assessed parameters. The model accuracy classification is almost 88% average GA-NN constants GA-SVR gives ~95% which can be further improved by accumulating more robust datasets into database. These used an alternative process development finding out new material providing valuable insight behaviours

Language: Английский

---

Enhancing deep learning predictive models with HAPPY (Hierarchically Abstracted rePeat unit of PolYmers) representation

npj Computational Materials, Journal Year: 2024, Volume and Issue: 10(1)

Published: May 24, 2024

Abstract We introduce HAPPY (Hierarchically Abstracted rePeat unit of PolYmers), a string representation for polymers, designed to efficiently encapsulate essential polymer structure features property prediction. assigns single constituent elements groups sub-structures and employs grammatically complete independent connectors between chemical linkages. Using limited number datapoints, we trained neural networks utilizing both conventional SMILES encoding repeated structures compared their performance in predicting five properties: dielectric constant, glass transition temperature, thermal conductivity, solubility, density. The results showed that the HAPPY-based network could achieve higher prediction R-squared score two-fold faster training times. further tested robustness versatility with an augmented dataset. Additionally, present topo-HAPPY (Topological HAPPY), extension incorporates topological details connectivity, leading improved solubility temperature score.

Language: Английский

---

Explainable machine learning for 2D material layer group prediction with automated descriptor selection

Materials Today Chemistry, Journal Year: 2025, Volume and Issue: 44, P. 102567 - 102567

Published: March 1, 2025

Language: Английский

---

Computational Data-Driven Materials Discovery

Trends in Chemistry, Journal Year: 2021, Volume and Issue: 3(2), P. 79 - 82

Published: Jan. 12, 2021

Language: Английский

---

Gradient boosted and statistical feature selection workflow for materials property predictions

The Journal of Chemical Physics, Journal Year: 2023, Volume and Issue: 159(19)

Published: Nov. 16, 2023

With the emergence of big data initiatives and wealth available chemical data, data-driven approaches are becoming a vital component materials discovery pipelines or workflows. The screening using machine-learning models, in particular, is increasingly gaining momentum to accelerate new materials. However, black-box treatment methods suffers from lack model interpretability, as feature relevance interactions can be overlooked disregarded. In addition, naive training often lead irrelevant features being used which necessitates need for various regularization techniques achieve generalization; this incurs high computational cost. We present feature-selection workflow that overcomes problem by leveraging gradient boosting framework statistical analyses identify subset features, recursive manner, maximizes their target variable classes. subsequently obtain minimal redundancy through multicollinearity reduction performing correlation hierarchical cluster analyses. further refined wrapper method, follows greedy search approach evaluating all possible combinations against evaluation criterion. A case study on elastic material-property prediction classification metallicity illustrate use our proposed workflow; although it highly general, demonstrated wider subsequent material properties. Our Bayesian-optimized models generated results, without techniques, comparable state-of-the-art reported scientific literature.

Language: Английский

---

Predicting the formation enthalpy and phase stability of (Ti,Al,TM)N (TM = III-VIB group transition metals) by high-throughput ab initio calculations and machine learning

Acta Materialia, Journal Year: 2024, Volume and Issue: 276, P. 120139 - 120139

Published: June 25, 2024

Language: Английский

---

Elemental Reactivity Maps for Materials Discovery

Chemistry of Materials, Journal Year: 2025, Volume and Issue: unknown

Published: Feb. 21, 2025

Language: Английский

---

Unsupervised Machine Learning Prediction of a Novel 1:3 Intermetallic Phase with the Synthesis of TbIr₃ (PuNi₃-type) as Experimental Validation

Journal of the American Chemical Society, Journal Year: 2025, Volume and Issue: unknown

Published: April 18, 2025

Crystal structure classification of binary intermetallic structures with 1:3 stoichiometry was done machine learning algorithms. The data set included 97 features and a total 2366 reported compounds adopting six different types. An unsupervised method based on principal component analysis (PCA) followed by clustering using the K-means applied to cluster belonging With recommendation engine, we predicted expansion clusters then identified cluster/structure-type overlap. PuNi3-type among clearly segregated types according model, novel representative, TbIr3, selected for experimental validation, this structure. final supervised predictions were partial least squares discriminant (PLS-DA), support vector (SVM), XGBoost, confidently predicting that TbIr3 belongs accuracies 96.6, 99.8, 99.9% respectively. Successful crystal segregation attributed descriptors comprising both compositional structural features. Given phase could be controversial due extensive study Tb-Ir diagram reports in two types, conducted independent validations confirm existence Subsequent theoretical validation explained Ir-Ir contacts are primary stability factor compared other

Language: Английский

---

Machine Learning based prediction of noncentrosymmetric crystal materials

Computational Materials Science, Journal Year: 2020, Volume and Issue: 183, P. 109792 - 109792

Published: May 30, 2020

Language: Английский

---