Prediction Of Material Properties Research Articles

An accelerated scheme for solving inhomogeneous elasticity in phase-field method and application to Ni-based multi-component alloys

On the way towards realistic predictions of various material properties for engineering structural materials, efficient computational simulations of microstructural evolution play an indispensable role. In the present work, an in-house phase-field code with inhomogeneous elasticity has been developed aiming to deal with precipitation processes for multi-component alloys with high efficiency. Two different numerical iterative algorithms, named as Model A and B, are implemented to solve inhomogeneous elastic equations. The resulting simulated morphology and energy evolutions under different conditions show consistency for these two algorithms. In order to perform simulation efficiently, a parallel technique based on CUDA (Computer Unified Device Architecture) architecture is used, which greatly improves the computational efficiency. After estimating the computational efficiency of different acceleration techniques for different grid sizes, Model A is extended to simulate concentration and morphology evolutions in the precipitation and rafting processes of γ′ phase in a Ni-based five-component alloy of Ni−Co−Al−Mo−W. Meanwhile, the CALPHAD method and related databases are linked to provide thermodynamic and diffusivity data. Finally, the computation costs of the parallel technique for different simulation scales are explored, testifying that this technique is capable of performing larger scale simulations for multi-component multi-phase structural materials.

Utilizing an intelligent machine learning strategy for predicting the tensile strength and ductility properties of nitrogen doped CoCrFeMnNi based high entropy alloys

The integration of machine learning (ML) within the realm of artificial intelligence has unveiled remarkable potential for predicting material properties. Particularly, determining critical properties such as yield strength (YS), tensile strength (UTS), and elongation in high entropy alloys (HEAs) has been a formidable challenge due to the extensive composition possibilities. The application of ML methods offers a viable solution for determining the optimal composition and mechanical properties of HEA materials. In this study, six ensemble ML algorithms, including random forest (RF), adaboost, xgboost, gradient boosting, bagging, and extra trees are employed to predict the tensile strength and ductility of nitrogen (N) doped CoCrFeMnNi based HEAs, which includes yield strength, tensile strength, and elongation properties. To explore the correlations between different features in the dataset, both the Pearson correlation and SHAP (SHapley Additive exPlanations) coefficient methods are employed. The gradient boosting ML model, in particular, distinguished itself with remarkable precision, achieving outstanding coefficient of determination (R2) values of approximately 98.25% for YS, 97.39% for UTS, and 94.45% for elongation. Moreover, it demonstrated lower mean absolute error (MAE) values for yield strength (32.62 MPa), tensile strength (37.48 MPa), and elongation (3.40%). This pattern strongly emphasizes its superiority in providing more accurate predictions compared to the other models. Notably, the model's prediction results for all three tensile properties are closely matched with the values obtained through experimental testing. This alignment between predictions and actual results serves as a crucial validation of the model's exceptional accuracy and reliability in precisely predicting these properties. Furthermore, the scrutinizing the impact of both prominent and less influential features on the ML model's performance is analyzed. Thereby, these findings underscore the remarkable effectiveness of the gradient boosting model as a predictive tool, emphasizing its ability to provide highly accurate and reliable predictions for HEAs properties.

Towards understanding structure–property relations in materials with interpretable deep learning

Deep learning (DL) models currently employed in materials research exhibit certain limitations in delivering meaningful information for interpreting predictions and comprehending the relationships between structure and material properties. To address these limitations, we propose an interpretable DL architecture that incorporates the attention mechanism to predict material properties and gain insights into their structure–property relationships. The proposed architecture is evaluated using two well-known datasets (the QM9 and the Materials Project datasets), and three in-house-developed computational materials datasets. Train–test–split validations confirm that the models derived using the proposed DL architecture exhibit strong predictive capabilities, which are comparable to those of current state-of-the-art models. Furthermore, comparative validations, based on first-principles calculations, indicate that the degree of attention of the atoms’ local structures to the representation of the material structure is critical when interpreting structure–property relationships with respect to physical properties. These properties encompass molecular orbital energies and the formation energies of crystals. The proposed architecture shows great potential in accelerating material design by predicting material properties and explicitly identifying crucial features within the corresponding structures.

Graph isomorphism network for materials property prediction along with explainability analysis

Deep learning, in particular graph neural network (GNN), has emerged as a powerful tool for exploring the relationship between crystal structures and materials properties. Based on the highly discriminative graph isomorphism network (GIN) algorithm, we present an improved GIN that incorporates edge features during the node feature aggregation and adjusts more weights to important nodes at both local and global levels. We show that our model can achieve comparable or better performance on the Matbench suite compared to some previously reported GNNs and the improvement is significantly greater on small datasets. Although GNNs tend to have good performance in accuracy, the black-box nature of GNNs hinders their application in materials domains to some extent. After training our model, we use t-stochastic neighbour embedding visualization and perturbation-based method to provide a straightforward and comprehensive explanation to increase the transparency and credibility of the model. We demonstrate the superiority of the presented framework in terms of accuracy and explainability. This is crucial for the design of novel materials.

Prediction of hydrogen storage in metal-organic frameworks: A neural network based approach

Gas capture, sensing, and storage systems are all within the capabilities of metal–organic frameworks (MOFs). It is common practise to choose the MOF with the best adsorption property from a large database before running an adsorption calculation. High-throughput computational research is sometimes hampered by the expense of computing thermodynamic values, slowing the progress of MOFs for separations and storage applications. When trying to predict material properties, machine learning has recently emerged as a possible alternative to more conventional methods like experiments and simulations. The H2 capacities of 918,734 MOFs drawn from 19 databases were recently predicted using ML by Ahmed and Siegel (2021). Several ML methods were utilized, and the extremely randomised tree (ERT) model emerged as the most accurate predictor of hydrogen delivery capacity in terms of both gravimetric and volumetric quantities. Interestingly we have used deep learning model (Feed-forward neural network) as well as ERT model for the prediction of H2 deliverable capacities of a huge number of MOFs developed from the previous studies and got till date best results for predictions. To verify our model’s efficacy, we also performed Grand Canonical Monte Carlo (GCMC) simulations. We show our method by forecasting the hydrogen storage capacity of MOFs during a temperature and pressure swing from 100bar/77K to 5bar/160K.

QRChEM: A Deep Learning Framework for Materials Property Prediction and Design Using QR Codes

QRChEM: A Deep Learning Framework for Materials Property Prediction and Design Using QR Codes

Probabilistic physics-guided transfer learning for material property prediction in extrusion deposition additive manufacturing

We introduce the concept of physics-guided transfer learning to predict the thermal conductivity of an additively manufactured short-fiber reinforced polymer (SFRP) using micro-structural characteristics extracted from tensile tests. Developing composite manufacturing digital twins for SFRP composite processes like extrusion deposition additive manufacturing (EDAM) require extensive experimental material characterization. Even the same material system printed on different EDAM systems can result in significant changes to the printed micro-structure, affecting the mechanical and transport properties. This, in turn, makes characterization efforts expensive and time-consuming. Therefore, the objective of the paper is to address this experimental bottleneck and use prior information about the material manufactured in one extrusion system to predict its properties when manufactured in another system. To enable this framework, we assume that changes in properties of the same material when manufactured in different systems arise solely due to microstructural changes. To that end, we present a Bayesian framework that can transfer thermal conductivity properties across extrusion deposition additive manufacturing systems. While we discuss the transfer of thermal conductivity properties, the development is such that the framework can be used for the transfer of other properties that depend on microstructural characteristics defined in the manufacturing process. These include the coefficient of thermal expansion, viscoelastic properties, etc. The framework begins by using thermal conductivity data of the composite printed in one extrusion system to probabilistically infer the constituent thermal properties of the fiber and the polymer. Next, by conducting limited tensile tests of the same material printed in another extrusion system, we infer its orientation tensor. Finally, the inferred constituent thermal conductivity properties and the inferred orientation tensor are coupled using a micromechanics model to predict the thermal conductivity properties of the composite printed in the second extrusion system. We experimentally verify the predictions and show that our method provides a reliable framework for transferring material properties while accounting for epistemic and aleatory uncertainties.

Accelerating search for the polar phase stability of ferroelectric oxide by machine learning

Machine learning emerges to accelerate first-principles calculations in materials discovery and property prediction, but developing high-accuracy prediction models requires training on large datasets of relaxed structures, which can be computationally expensive. Besides, the crystal structure prediction algorithms show promising aspects but often deviate from the ground state, leading to poor predictions. Using a crystal graph convolutional neural network, we propose an alternative and efficient method of finding the ground state by utilizing the total energy of random unrelaxed structures. The prediction model is trained on a small dataset of unrelaxed energy obtained from first-principles calculations and effectively predicts the total energy of random structures with an accuracy almost equivalent to first-principles calculations. The resulting energy landscape is then reconstructed in terms of polarity. This leads to establishing an energy landscape of a perovskite oxide in a cost-effective way, enabling a polar phase stability map of strained BaTiO3, a representative ferroelectric oxide. The proposed methodology efficiently allows to obtain the ground state by eliminating the computationally intensive procedure of structure optimization, thus accelerating the ground state search to map out the phase stability of strained perovskite materials in a practical manner.

Machine learning in metal-ion battery research: Advancing material prediction, characterization, and status evaluation

Metal-ion batteries (MIBs), including alkali metal-ion (Li+, Na+, and K+), multi-valent metal-ion (Zn2+, Mg2+, and Al3+), metal-air, and metal-sulfur batteries, play an indispensable role in electrochemical energy storage. However, the performance of MIBs is significantly influenced by numerous variables, resulting in multi-dimensional and long-term challenges in the field of battery research and performance enhancement. Machine learning (ML), with its capability to solve intricate tasks and perform robust data processing, is now catalyzing a revolutionary transformation in the development of MIB materials and devices. In this review, we summarize the utilization of ML algorithms that have expedited research on MIBs over the past five years. We present an extensive overview of existing algorithms, elucidating their details, advantages, and limitations in various applications, which encompass electrode screening, material property prediction, electrolyte formulation design, electrode material characterization, manufacturing parameter optimization, and real-time battery status monitoring. Finally, we propose potential solutions and future directions for the application of ML in advancing MIB development.

Numerical Modeling of Residual Stresses and Fracture Strengths of Ba0.5Sr0.5Co0.8Fe0.2O3−δ in Reactive Air Brazed Joints

Reactive Air Brazing (RAB) enables the joining of vacuum-sensitive oxide ceramics, such as Ba0.5Sr0.5Co0.8Fe0.2O3−δ (BSCF), to metals in a one-step process. However, damage may form in ceramic or joint during RAB. In this work, experimental microstructure characterization, measurement, and prediction of local material properties using finite element analysis were combined to enlighten these damage mechanisms, which are currently not well understood. Micromechanical simulations were performed using representative volume elements. Cooling simulations indicate that small-sized CuO precipitations are most likely to cause crack initiation in BSCF during cooling. The ball-on-three-balls experiment with porous BSCF samples was analyzed numerically to determine the values of temperature-dependent BSCF fracture stresses. The inversely calibrated fracture stresses in the bulk BSCF phase are underestimated, and true values should be quite high, according to an extreme value analysis of pore diameters.

Predictive Modeling of Tensile Strength in Aluminum Alloys via Machine Learning.

Aluminum alloys are widely used due to their exceptional properties, but the systematic relationship between their grain size and their tensile strength has not been thoroughly explored in the literature. This study aims to fill this gap by compiling a comprehensive dataset and utilizing machine learning models that consider both the alloy composition and the grain size. A pivotal enhancement to this study was the integration of hardness as a feature variable, providing a more robust predictor of the tensile strength. The refined models demonstrated a marked improvement in predictive performance, with XGBoost exhibiting an R2 value of 0.914. Polynomial regression was also applied to derive a mathematical relationship between the tensile strength, alloy composition, and grain size, contributing to a more profound comprehension of these interdependencies. The improved methodology and analytical techniques, validated by the models' enhanced accuracy, are not only relevant to aluminum alloys, but also hold promise for application to other material systems, potentially revolutionizing the prediction of material properties.

Enhancing of uniaxial compressive strength of travertine rock prediction through machine learning and multivariate analysis

Indirect methods for predicting material properties in rock engineering are vital for assessing elastic mechanical properties. Accurately predicting material properties holds significant importance in rock and geotechnical engineering, as it strongly influences decisions about the design and construction of infrastructure projects. Uniaxial compressive strength (UCS) is one of the most important elastic mechanical properties for understanding how rocks and geological formations respond to stress and deformation. However, the standard UCS test faces several challenges, including its destructive nature, high costs, time-consuming procedures, and the requirement for high-quality samples. Therefore, there is a growing demand for indirect methods to estimate UCS, which are invaluable tools for evaluating the elastic mechanical properties of materials. The study aimed to comprehensively analyze the relationships between UCS of travertine rock samples collected from the Dead Sea and Jordan Valley formations and seven different rock indices by utilizing parametric and non-parametric methods. The laboratory results indicate that the study area's travertine rock possesses high-quality and desirable properties. The results reveal that certain rock indices, such as Schmidt hammer, Leeb rebound hardness, and Point Load, strongly correlate with Uniaxial Compressive Strength (UCS). Conversely, other indices, specifically dry density, absorption, pulse velocity, and porosity, exhibit a considerably weaker or very weak relationship with UCS. The paper employs three machine learning techniques, namely the Tree model, k-nearest neighbors (KNN), and Artificial Neural Networks (ANN), to develop predictive models for rock strength. The models were trained on a dataset of rock properties and corresponding mechanical strength values. The study's results revealed that the M5 tree model is the most suitable method for predicting UCS. It demonstrates robust performance across a spectrum of metrics and boasts low prediction errors. Following the M5 tree model are the KNN, ANN, and regression methods in descending order of performance.

Optimization of mechanical properties of multiscale hybrid polymer nanocomposites: A combination of experimental and machine learning techniques

Machine learning (ML) models provide fast and accurate predictions of material properties at a low computational cost. Herein, the mechanical properties of multiscale poly(3-hydroxybutyrate) (P3HB)-based nanocomposites reinforced with different concentrations of multiwalled carbon nanotubes (MWCNTs), WS2 nanosheets and sepiolite (SEP) nanoclay have been predicted. The nanocomposites were prepared via solution casting. SEM images revealed that the three nanofillers were homogenously and randomly dispersed into the matrix. A synergistic reinforcement effect was attained, resulting in an unprecedented stiffness improvement of 132% upon addition of 1:2:2 wt% SEP:MWCNTs:WS2. Conversely, the increments in strength were only moderates (up to 13.4%). A beneficial effect in the matrix ductility was also found due to the presence of both nanofillers. Four ML approaches, Recurrent Neural Network (RNN), RNN with Levenberg's algorithm (RNN-LV), decision tree (DT) and Random Forest (RF), were applied. The correlation coefficient (R2), mean absolute error (MAE) and mean square error (MSE) were used as statistical indicators to compare their performance. The best-performing model for the Young's modulus was RNN-LV with 3 hidden layers and 50 neurons in each layer, while for the tensile strength was the RF model using a combination of 100 estimators and a maximum depth of 100. An RNN model with 3 hidden layers was the most suitable to predict the elongation at break and impact strength, with 90 and 50 neurons in each layer, respectively. The highest correlation (R2 of 1 and 0.9203 for the training and test set, respectively) and the smallest errors (MSE of 0.13 and MAE of 0.31) were obtained for the prediction of the elongation at break. The developed models represent a powerful tool for the optimization of the mechanical properties in multiscale hybrid polymer nanocomposites, saving time and resources in the experimental characterization process.

Gradient boosted and statistical feature selection workflow for materials property predictions

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

Predictions and Uncertainty Estimates of Reactor Pressure Vessel Steel Embrittlement Using Machine Learning

An essential aspect of extending safe operation of the world’s active nuclear reactors is understanding and predicting the embrittlement that occurs in the steels that make up the Reactor pressure vessel (RPV). In this work we integrate state of the art machine learning methods using ensembles of neural networks with unprecedented data collection and integration to develop a new model for RPV steel embrittlement. The new model has multiple improvements over previous machine learning and hand-tuned efforts, including greater accuracy (e.g., at high-fluence relevant for extending the life of present reactors), wider domain of applicability (e.g., including a wide-range of compositions), uncertainty quantification, and online accessibility for easy use by the community. These improvements provide a model with significant new capabilities, including the ability to easily and accurately explore compositions, flux, and fluence effects on RPV steel embrittlement for the first time. Furthermore, our detailed comparisons show our approach improves on the leading American Society for Testing and Materials (ASTM) E900-15 standard model for RPV embrittlement on every metric we assessed, demonstrating the efficacy of machine learning approaches for this type of highly demanding materials property prediction.

A machine learning framework for predicting physical properties in configuration space of gate alloys

In materials design, it is meaningful to fast and accurately predict material properties required for a specific application based on specific structures. In the present study, a framework for predicting formation energy, bulk modulus, shear modulus and Debye temperature in a large configuration space of gate alloys was proposed, by integrating two approaches, density functional theory (DFT) and machine learning (ML). DFT calculations were conducted to obtain the physical properties, as labels of dataset used in ML, for 3051 configurations enumerated. For features of ML modeling, a comparative analysis was carried out on four structure descriptors as well as their combinations with element descriptors. The predictive performance with the optimal features combination was compared with that from crystal graph convolutional neural networks (CGCNN). The predictive ability of the framework was further validated by additional data collected from Materials Project platform. Finally, the effectiveness of transfer learning is sufficiently demonstrated in the extrapolation case of gate alloys system.

ACGNet: An interpretable attention crystal graph neural network for accurate oxidation potential prediction

LiNixCoyMn1-x-yO2 (NCM) is one of the most critical cathode materials for high energy density lithium-ion batteries in electric vehicle applications. Nevertheless, capacity degradation and long-term cycle instability due to the aging of cathode/electrolyte interfaces remain significant challenges for NCM materials. Various surface stabilization techniques, including doping and coating, can be employed for NCM modifications. Traditionally, new coating materials are identified by chemical intuition or trial-and-error synthesis, which hinders the discovery speed of high-performance coating materials. A novel neural network model named Attention Graph Convolutional Neural Network (ACGNet) has been developed to predict crystals’ electrochemical stability windows from atom and bonding features and exhibits remarkable predictive performance with a mean absolute error of 0.586 V. Then, the developed model is utilized to conduct high-throughput screening of 13,943 candidate compounds for their coating potential. Among the candidates, LiPO3 exhibits prominent potential as a coating material due to its high oxidation voltage and low preparation cost. The subsequent battery assembly experiment and electrochemical characterization reveal that the incorporation of LiPO3 significantly enhances the cycle stability of NCM batteries. In summary, our model exhibits exceptional accuracy in predicting material properties and serves as a compelling example for machine learning applications in battery materials.

A Thermodynamic Strength Theoretical Model to Explain and Predict Pseudo-Ductility Behavior of SiCf/SiC Composite Material

In order to explain and predict the pseudo-ductility behavior and mechanical properties of [Formula: see text] composite material, a thermodynamic strength theoretical model based on fracture mechanics and thermodynamic method has been established. Compared with other theoretical models, the model proposed in this investigation unifies different cracks and defects based on energy method. Meanwhile, the stress–strain behavior of materials can be obtained with as few parameters as possible. Compared with previous experimental data, the correctness of this theoretical model has been verified. Some key material properties, such as elastic modulus, proportional limit stress (PLS) and yield stress (YS), have been investigated by calculating based on the theoretical model. The effect of the initial matrix porosity and the material properties of the matrix and fiber have been investigated. Results show that PLS and YS are almost not influenced by initial matrix porosity and can be seen as intrinsic parameters. Elastic modulus of the SiC reinforced fiber and the matrix can also affect material properties. Compared with the SiC matrix, PLS of the [Formula: see text] composite material is more sensitive to changes in elastic modulus of the SiC reinforced fiber. However, the improvement of elastic modulus and shear modulus of the [Formula: see text] composite material is much more difficult. For this purpose, the elastic modulus of both the SiC reinforced fiber and the matrix need to be enhanced. Results and conclusions in this investigation can provide guidance for predicting material properties of [Formula: see text] composite material and preparation in industry.

Estimation of concrete materials uniaxial compressive strength using soft computing techniques

This study addresses a critical gap in concrete strength prediction by conducting a comparative analysis of three deep learning algorithms: convolutional neural networks (CNNs), gated recurrent units (GRUs), and long short-term memory (LSTM) networks. Unlike previous studies that employed various machine learning algorithms on diverse concrete types, our study focuses on mixed-design concrete and fine-tuned deep learning algorithms. The objective is to identify the optimal deep learning (DL) algorithm for predicting concrete uniaxial compressive strength, a crucial parameter in construction and structural engineering. The dataset comprises experimental records for mixed-design concrete, and models were developed and optimized for predictive accuracy. The results show that the CNN model consistently outperformed GRU and LSTM. Hyperparameter tuning and regularization techniques further improved model performance. This research offers practical solutions for material property prediction in the construction industry, potentially reducing resource burdens and enhancing efficiency and construction quality.

Determinants and performance prediction on photocatalytic properties of hydroxyapatite by machine learning

Hydroxyapatite (HAP) have been widely used as photocatalysts in environmental remediation. However, the effects of crystallinity, crystal size, and purity of HAP on their photocatalytic performances have not been fully understood yet. In this paper, we synthesized HAP samples with different properties by controlling the calcination conditions and evaluated their photocatalytic performance in the degradation reaction of methyl orange. Combing the catalytic experiments with machine learning technique, we found that the crystallinity of HAP was the dominant parameter that determined the photocatalytic performance. And the developed machine learning method could predict the catalytic results with an error of only 2.2 %. Our work provides a route to predict material properties based its characters by experiment combined with machine learning.