GCPNet: An interpretable Generic Crystal Pattern graph neural Network for predicting material properties.

Chemical heuristics have been fundamental to the advancement of chemistry and materials science. These heuristics are typically established by scientists using knowledge and creativity to extract patterns from limited datasets. Machine learning offers opportunities to perfect this approach using computers and larger datasets. Here, we discuss the relationships between traditional heuristics and machine learning approaches. We show how traditional rules can be challenged by large-scale statistical assessment and how traditional concepts commonly used as features are feeding the machine learning techniques. We stress the waste involved in relearning chemical rules and the challenges in terms of data size requirements for purely data-driven approaches. Our view is that heuristic and machine learning approaches are at their best when they work together.

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

The use of machine learning methods for accelerating the design of crystalline materials usually requires manually constructed feature vectors or complex transformation of atom coordinates to input the crystal structure, which either constrains the model to certain crystal types or makes it difficult to provide chemical insights. Here, we develop a crystal graph convolutional neural networks framework to directly learn material properties from the connection of atoms in the crystal, providing a universal and interpretable representation of crystalline materials. Our method provides a highly accurate prediction of density functional theory calculated properties for eight different properties of crystals with various structure types and compositions after being trained with 10^{4} data points. Further, our framework is interpretable because one can extract the contributions from local chemical environments to global properties. Using an example of perovskites, we show how this information can be utilized to discover empirical rules for materials design.

In condensed matter physics and materials science, predicting material properties necessitates understanding intricate many-body interactions. Conventional methods such as density functional theory and molecular dynamics often resort to simplifying approximations and are computationally expensive. Meanwhile, recent machine learning methods use handcrafted descriptors for material representation, which sometimes neglect vital crystallographic information and are often limited to single property prediction or a sub-class of crystal structures. In this study, we pioneer an unsupervised strategy, drawing inspiration from natural language processing to harness the underutilized potential of Crystallographic Information Files (CIFs). We conceptualize atoms and atomic positions within a crystallographic information file similarly to words in textual content. Using a Word2Vec-inspired technique, we produce atomic embeddings that capture intricate atomic relationships. Our model, CIFSemantics, trained on the extensive material project dataset, adeptly predicts 15 distinct material properties from the CIFs. Its performance rivals that of specialized models, marking a significant step forward in material property predictions.

Multi-output ensemble deep learning: A framework for simultaneous prediction of multiple electrode material properties

A method has been developed for building fully atomistic models of chemically crosslinked PVA. It combines a dynamic crosslinking approach followed by a high‐temperature annealing procedure. The crosslinking procedure allows to control both crosslinking density and the amount of free crosslinking sites. It also takes into account some network defects, such as dangling tails and wasted loops, which have been experimentally observed on chemically crosslinked polymer networks. This procedure is applied to PVA which is chemically crosslinked with two different linear polyols. Several material properties such as glass transition temperature, Young's, shear and bulk modulus were predicted from the equilibrated structures and found to agree well with available experimental data.magnified image

Instead of performing expensive experiments, it is common in industry to make predictions of important material properties based on some existing experimental results. Databases consisting of experimental observations are widely used in the field of Material Science Engineering. However, these databases are expected to be noisy since they rely on human measurements, and also because they are an amalgamation of various independent sources (research papers). Therefore, some conflicting information can be found between various sources. In this paper, we introduce a novel truth discovery approach to reduce the amount of noise and filter the incorrect conflicting information hidden in the scientific databases. Our method ranks the multiple data sources by considering the relationships between them, i.e., the amount of conflicting information and the amount of agreement, and as well eliminates the conflicting information. The scalable Gaussian process interpolation technique (SGP) is then applied to the clean dataset to make predictions of materials property. Comprehensive performance study has been done on a real life scientific database. With our new approach, we are able to highly improve the accuracy of SGP predictions and provide a more reliable database.

Pb-Sn solders are known to coarsen and deform heterogeneously under shear thermomechanical cycling. Solder joint failure often occurs within these coarsened shear bands. The connection between microstructure and materials properties is well documented, which leads to the conclusion that an improved understanding of the interaction between stress and microstructural evolution will better enable the prediction of materials properties over time. In this paper, we simulate microstructural evolution of Pb-Sn solders using a novel two-phase version the Potts model, a kinetic Monte Carlo model that is coupled to a finite element mechanics model. Using this model, possible mechanisms for heterogeneous coarsening are explored.

As environmental pollution worsens, related environmental regulations are becoming stricter. Recently, regulations on internal combustion engine vehicles have been tightening, leading to increased attention towards hydrogen and electric vehicles as alternatives. Especially within a decade, electric vehicles have emerged as prominent eco-friendly vehicles, with lithium-ion batteries garnering significant attention as energy storage devices. Among them, all solid-state batteries, boasting advantages such as high energy density and output, are continuously researched as the ultimate solution, free from issues like explosions. The battery manufacturing process typically comprises the pole plate process, assembly process, and formation process. To enhance battery performance and manufacturing efficiency, process optimization is essential, and simulation can reduce resource and time wastage.Finite Element Method (FEM) is a prominent simulation technique that can simulate various processes to derive optimal designs or processes. Predicting problems that may occur during electrode manufacturing processes and deriving relevant alternatives can save time and costs.Accurate simulation requires high-quality mechanical properties of materials. However, for sulfide-based solid electrolytes, exposure to air leads to material property changes, and their thickness, often below 100μm, makes deriving material properties through measurement nearly impossible.To estimate electrode material properties, Xu derived force-deformation curves (F-D Curve) through measurements for the entire lithium-ion battery (LIB) and calculated the material properties of the anode-separator-cathode assembly as representative properties. Wierzbicki estimated the mechanical properties of cylindrical-shaped lithium-ion cells using homogenization techniques. However, in such cases, it is impossible to obtain data for each component, resulting in weaknesses in optimizing the process for all solid-state batteries.To obtain data for each component, Cheng measured the elastic modulus using nanoindentation after sintering the NMC powder to derive the material properties of NMC, a key cathode material. However, for sulfide-based batteries, such as those using solid electrolytes, approaching through the above methods is impossible due to their reaction with air.In this study, we utilized Crystal Graph Convolutional Neural Network (CGCNN), a machine learning technique, to derive the material properties of Li6PS5Cl, a representative material among sulfide-based solid electrolytes. Materials are composed of a crystal structure where atoms are arranged in a certain pattern. Due to variables such as the type of atoms, composition ratio, and arrangement, predicting outcomes is challenging without measurements. However, CGCNN inputs structural information of bulk materials to predict basic properties accurately, offering a promising alternative. After modeling the material structure, a certain strain was applied to calculate the bulk modulus and shear modulus. Using this data, CGCNN was trained to predict micro-scale bulk modulus and shear modulus, from which Young’s modulus and Poisson’s ratio, representing key mechanical properties, were derived.Based on this study, utilizing CGCNN, the material properties of Li5PS5Cl were predicted, resulting in Young’s modulus of 28.07 MPa, Shear Modulus of 11.02 MPa, and Poisson’s ratio of 0.27. This study successfully derived the challenging-to-measure material properties of solid-state batteries. This serves as fundamental data for process optimization in the mass production process of solid-state batteries, contributing to enhanced productivity and cost reduction. This research is expected to significantly contribute to the ultimate goal of optimizing the manufacturing process of solid-state batteries and improving productivity, ultimately leading to cost reduction.This work was supported by Technology Innovation Program (project name : Equipment technologies for 50cm2 all solid state battery cell, project Number : 20012349) funded by the Ministry of Trade, Industry & Energy (MOTIE, Korea).This research has been conducted with the support of the Ministry of Trade, Industry and Energy(MOTIE), Republic of Korea as “Development of a battery pack case for electric vehicles that deltas thermal runaway”(P0024522) in the small and medium business win-win innovation leap project.This research was funded by the ‘New Faculty Research Support Grant’ at Changwon National University in 2024.

A reliable predictor is very useful for real-world industrial applications to forecast thefuture behavior of dynamic systems. A smart predictor, based on a novel recurrent neuralfuzzy (RNF) scheme, is developed in this paper for multi-step-ahead prediction of materialproperties. A systematic investigation based on two benchmark data sets is conducted interms of performance and efficiency. Analysis results reveal that, of the data-drivenforecasting schemes, predictors based on step input patterns outperform those based onsequential input patterns; the RNF predictor outperforms those based on recurrent neuralnetworks and ANFIS schemes in multi-step-ahead prediction of nonlinear time series. Anadaptive Levenberg–Marquardt training technique is adopted to improve therobustness and convergence of the RNF predictor. Furthermore, the proposed smartpredictor is implemented for material property testing. Investigation results show thatthe developed RNF predictor is a reliable forecasting tool for material propertytesting; it can capture and track the system’s dynamic characteristics quickly andaccurately. It is also a robust predictor to accommodate different system conditions.

The success of machine learning (ML) in materials property prediction depends heavily on how the materials are represented for learning. Two dominant families of material descriptors exist, one that encodes crystal structure in the representation and the other that only uses stoichiometric information with the hope of discovering new materials. Graph neural networks (GNNs) in particular have excelled in predicting material properties within chemical accuracy. However, current GNNs are limited to only one of the above two avenues owing to the little overlap between respective material representations. Here, a new concept of formula graph which unifies stoichiometry‐only and structure‐based material descriptors is introduced. A self‐attention integrated GNN that assimilates a formula graph is further developed and it is found that the proposed architecture produces material embeddings transferable between the two domains. The proposed model can outperform some previously reported structure‐agnostic models and their structure‐based counterparts while exhibiting better sample efficiency and faster convergence. Finally, the model is applied in a challenging exemplar to predict the complex dielectric function of materials and nominate new substances that potentially exhibit epsilon‐near‐zero phenomena.

Recent machine learning guided material research - A review

Abstract: Material research is an ever-growing field with the invention of new materials almost every day. It is not feasible for everybody working in the field to test each and every material on the Universal Testing Machine (UTM). This poses a limitation that has been sought to be tackled for quite some time now. One approach to tackle this is to use machine learning to predict the properties of the materials, thus eliminating the need of extensive time requirement and cost of performing experiments on the UTM. In this paper, the authors have trained and tested a machine learning based model that can predict material properties on the basis of composition and given temperature.

Internal stress in polycrystalline materials is an intrinsic attribute of the microstructure that affects a broad range of material properties. It is usually acquired through experiment in conjunction with continuum mechanics modelling, but its determination at nanometre and submicron scales is extremely difficult. Here, we report a bottom-up approach using atomistic calculation. We obtain the internal stress in polycrystalline copper with nanosized grains by first computing the stress associated with each atom and then sorting the stress into those associated with different self-equilibrating length scales, i.e. sample scale and grain cell, which gives type I, II and III residual stresses, respectively. The result shows highly non-uniform internal stress distribution; the internal stress depends sensitively on grain size and the grain shape anisotropy. Statistical distributions of the internal stresses, along with the means and variance, are calculated as a function of the mean grain size and temperature. The implementation of this work in assisting the interpretation of experimental results and predicting material properties is discussed.

The prediction of elastic material properties via an adjoint formulation in this talk an efficient computational formulation to determine the acoustic properties of an elastic material, given its response to a time-harmonic excitation, is presented. Such problems arise in biomedical imaging using techniques such as Magnetic Resonance Elastography. The problem is posed as an inverse problem and its solution is computed using a mathematical programming algorithm (from mathematical programming). The functional to be minimized is the norm of the difference of the ‘‘measured’’ and a ‘‘trial’’ velocity field. At every iteration the trial velocity field is determined by solving the Helmholtz equation. The derivatives of the functional with respect to material properties, which form the input to the optimization algorithm, are computed by solving an adjoint problem. Using this formulation, the cost of computing derivatives is independent of the number of parameters used to represent the material properties, and thus the overall problem is rendered computationally tractable. A description of this methodology and an illustrative example to indicate its effectiveness will be presented.