Identifying exceptional electrocatalysts from the vast materials space remains a formidable challenge. Machine learning (ML) has emerged as a powerful tool to address this challenge, offering high efficiency while maintaining good accuracy in predictions. From this perspective, we provide a brief overview of recent advancements in ML for electrocatalyst discoveries. We emphasize the applications of physics-informed ML (PIML) models and explainable artificial intelligence (XAI) to electrocatalyst development, through which valuable physical and chemical insights can be distilled. Additionally, we delve into the challenges faced by PIML approaches, explore future directions, and discuss potential breakthroughs that could revolutionize the field of electrocatalyst development.

Tuning electrolyte bulk properties, fundamentally the Zn-ion solvation structures, is key to addressing degradation issues in aqueous Zn-ion batteries (AZIBs). The common practice is to add water-soluble organics as a cosolvent. However, a comprehensive fundamental understanding of the cosolvent effect on electrolyte bulk properties is still lacking. In this work, using ethylene glycol (EG) as the cosolvent and 2M ZnSO₄ as the base aqueous Zn-ion electrolyte, we report from a computational perspective how the cosolvent affects aqueous electrolyte bulk properties such as conductivity and pH. To ensure reliability of the computational results, we have used experimental ion conductivity data to validate our computing methods. Further, we show new hybrid solvation models that encompass H₂O, cosolvent and anion, e.g., EG-Zn(H₂O)₅²⁺ and EG-Zn(H₂O)₄²⁺-SO₄^2-. Based on these cosolvent-involved Zn-ion solvation models, the experimental pH trending has been successfully explained. Our work offers new insights into the cosolvent effect on aqueous Zn-ion electrolyte bulk properties and solvation structures.

Understanding the impact of the primary structure of peptides on a range of physicochemical properties is crucial for the development of various applications. Peptides can be conceptualized as sequences of amino acids in their biological representation and as molecular architectures composed of atoms and chemical bonds in their chemical representation. This study examines the influence of different biological and chemical representations of peptides on the local interpretability and accuracy of their respective prediction models and has developed “feature attribution” methodologies based on these representations. The effectiveness of these methodologies is validated through physicochemical analyses, specifically within the context of peptide aggregation propensity (AP) prediction, with training datasets derived from high-throughput molecular dynamics (MD) simulations. Our findings reveal significant discrepancies in the attribution extracted from sequence-based and chemical structure-based representations, which has led to the proposal of a co-modeling framework that integrates insights from both perspectives. Empirical comparisons have demonstrated that the contrastive learning-based co-modeling framework excels in terms of effectiveness and efficiency. This research not only extends the applicability of the attribution method but also lays the groundwork for elucidating the intrinsic mechanisms governing peptide activities and functions with the aid of domain-specific knowledge. Moreover, the co-modeling strategy is poised to enhance the precision of downstream applications and facilitate future endeavors in drug discovery and protein engineering.

Accurately estimating the state of charge (SOC) of lithium-ion batteries is essential for optimizing battery management systems in various applications such as electric vehicles and renewable energy storage. This study explores advancements in data-driven approaches for SOC estimation, focusing on both conventional machine learning and deep learning techniques. While traditional machine learning methods offer reliable performance, they often encounter challenges with high-dimensional data and adaption to complex operational conditions. In contrast, deep learning models provide enhanced capabilities in nonlinear modeling and automated feature extraction, leading to improved accuracy and robustness. Through comprehensive evaluations across diverse scenarios, this research identifies key technical challenges and outlines future directions, including distributed training, incorporation of physical data, development of dynamic neural networks, and the establishment of standardized benchmarking protocols. These insights aim to guide the creation of more precise, efficient, and adaptive SOC estimation models, thereby advancing the reliability and effectiveness of battery management systems.

In materials science, point defects play a crucial role in materials properties. This is particularly well known for the wide band gap insulators where the defect formation/compensation determines the equilibrium Fermi level and generally the doping response of a given material. Similarly, the main defect trends are also widely understood for regular metals (e.g., Cu and Zn discussed herein). With the development of electronic structure theory, a unique class of quantum materials - gapped metals (e.g., Ca₆Al₇O₁₆, SrNbO₃, In₁₅SnO₂₄, and CaN₂) that exhibit characteristics of both metals and insulators - has been identified. While these materials have internal band gaps similar to insulators, their Fermi level is within one of the main band edges, giving a large intrinsic free carrier concentration. Such unique electronic structures give rise to unusual defect physics, e.g., when the acceptor defect formation in n-type gapped metal results in the decay of the conducting electron to the acceptor states. In concentration limits, such electron-hole recombination can compensate for the energy needed to break chemical bonds and form acceptor vacancy, often leading to off-stoichiometric compounds. Such unusual physics, however, makes these quantum materials distinct from traditional compounds. Motivated by this, herein, we establish a minimal level of theory needed to account for the complex interplay between electronic structure and analyzing defects in gapped metals that can be utilized for their design in different practical applications.

Machine learning has emerged as a critical tool for processing the complex and large-scale datasets generated in the steel industry. However, a single machine learning model struggles to capture all relevant information owing to the variety of steel grades, thereby limiting its extensibility and broader industrial application. Furthermore, most machine-learning models are “black boxes” with low interpretability. Therefore, this paper proposes a novel strategy for industrial big data analysis. First, a data classification model was developed using unsupervised clustering techniques to automatically divide the dataset into four distinct classes. Simultaneously, key physical metallurgy (PM) variables were calculated and incorporated as input features to improve property prediction. Next, an interpretable knowledge graph was constructed for each class, connecting the relevant features with the PM variables. Using these graphs, a graph convolutional network (GCN) model was developed for each class to predict the steel properties. The results demonstrate that this approach delivers better predictions than models without automatic data classification. Furthermore, compared to traditional deep learning models, GCN models based on interpretable knowledge graphs provide superior prediction accuracy and significantly improved interpretability and extensibility.

Cathode materials are crucial in potassium (K) batteries, directly impacting their performance and lifespan. In this study, we used a combination of geometrical-topological (GT) analysis, bond valence site energy (BVSE), Kinetic Monte Carlo (KMC), and first-principles calculations to screen potential cathode materials for K-ion batteries among inorganic phosphides. Through GT analysis, we screened 143 K- and P-containing compounds and identified 30 with two- or three-dimensional K-ion migration pathways. BVSE further narrowed down 13 compounds with K-ion migration energies below 1 eV. KMC simulations of ionic conductivity led to the selection of K₃Cu₃P₂ for detailed first-principles calculations. It was demonstrated that K₃Cu₃P₂ possesses a reversible capacity of 72.47 mAh·g^-1, minimal volume change (1.47%), and a charge compensation mechanism involving Cu and P. Its low migration energy barrier contributes to a high ionic diffusion coefficient and conductivity of 1.87 × 10^-3 S·cm^-1 at 25 °C, making K₃Cu₃P₂ a promising candidate for stable and efficient K-ion diffusion in cathode applications.

Magnesium (Mg) alloys have attracted considerable attention as next-generation lightweight thermal conducting materials. However, their thermal conductivity decreases significantly with increasing alloying content. Current methods for predicting thermal conductivity of Mg alloys primarily rely on computationally intensive first-principles calculations or semi-empirical models with limited accuracy. This study presents a novel machine learning approach coupled with multiscale computation for predicting thermal conductivity in multi-component Mg alloys. A comprehensive database of 1,139 thermal conductivity measurements from as-cast Mg alloys was systematically compiled. A multiscale feature set incorporating elemental characteristics, thermodynamic properties, and electronic structure parameters was constructed. Key features, including atomic radius differences, enthalpy, cohesive energy, and the ratio of electronic thermal conductivity to relaxation time, were identified through sequential forward floating selection (SFFS). The XGBoost algorithm demonstrated superior performance, achieving a mean absolute percentage error (MAPE) of 2.16% for low-component ternary and simpler Mg alloy systems. Through L1 and L2 regularization optimization, the model’s extrapolation capability for quaternary and higher-order novel systems was significantly enhanced, reducing the prediction error to 13.60%. This research provides new insights and theoretical guidance for accelerating the development of high thermal conductivity Mg alloys.

In this study, we innovatively proposed a deep learning model architecture to address the industry challenges in the detection of porosity in magnesium alloys. Magnesium alloys, known for their lightweight and high-strength characteristics, are extensively utilized in aerospace, automotive, and biomedical fields. However, the absorption of hydrogen during the production process leads to the formation of pores, which not only reduce the material’s strength and durability but may also cause premature failure of the material. The formation of pores typically occurs during the solidification stage of magnesium alloys, where hydrogen dissolved in the molten metal is released upon cooling, forming tiny gas pores. The presence of these gas pores significantly affects the mechanical properties of the material, potentially leading to crack initiation and propagation under high stress. Therefore, accurate detection and quantification of pores are crucial for enhancing the quality control of magnesium alloys. Our developed model integrates window-shaped perception blocks with convolutional neural networks, enhanced by aggregated sensing layers (ASLs) on long-range connections. Extensive training on real samples demonstrated that our model outperforms mainstream algorithms such as U-Net and TransUNet across various evaluation metrics, particularly in fine target detection tasks under complex scenarios. Specifically, our model achieved a Dice coefficient of 74.77% and an Intersection over Union index of 71.00%, significantly surpassing other models. Moreover, the method also demonstrated superior accuracy in pore edge prediction, effectively mitigating issues of oversegmentation and undersegmentation, especially for small and irregular pores. An ablation study further confirmed the effectiveness of each component, with the ASL module showing particular strength in feature extraction and reducing upsampling loss. In summary, this research highlights the significant potential of deep learning technology in material defect detection and provides an efficient, automated solution for practical production, contributing to advancements in materials science and industrial quality control.

The improvement in energy storage performance of ferroelectric (FE) materials requires both high electric breakdown strength and significant polarization change. The phase-field method can couple the multi-physics-field factors. It can realize the simulation of electric breakdown and polarization evolution. It is widely used to reveal the modification mechanism and guide experimental design. Starting with the models of electric breakdown and polarization evolution, this work reviews the latest theoretical progress on FE materials with high energy storage performance. Firstly, the enhancement mechanisms of electric breakdown strength are analyzed. Subsequently, the improvement strategies at domain scales are analyzed. Finally, this review summarizes and looks ahead to the development of theoretical models, such as machine learning models.

The development of efficient and stable hydrogen production technologies is crucial for global clean energy transition. Solid oxide electrolysis cells (SOECs) have emerged as a promising technology for green hydrogen production due to their high efficiency, low-cost catalysts, and excellent adaptability to renewable energy sources. However, significant challenges remain in materials design, interface engineering, and system integration. This perspective reviews recent advances in artificial intelligence (AI)-guided SOEC development, focusing on machine learning approaches for design of key materials. Furthermore, we highlight how AI technologies can address the key challenges in both single-cell performances and system-level integration with renewable energy sources. Looking forward, we outline strategic directions for advancing AI-driven SOEC development toward commercial implementation, which may offer valuable insights and experiences within the field of energy conversion and storage.

Over the past decades, machine learning has kept playing an important role in materials design and discovery. In practical applications, materials usually need to fulfill the requirements of multiple target properties. Therefore, multi-objective optimization of materials based on machine learning has become one of the most promising directions. This review aims to provide a detailed discussion on machine learning-assisted multi-objective optimization in materials design and discovery combined with the recent research progress. First, we briefly introduce the workflow of materials machine learning. Then, the Pareto fronts in multi-objective optimization and the corresponding algorithms are summarized. Next, multi-objective optimization strategies are demonstrated, including Pareto front-based strategy, scalarization function, and constraint method. Subsequently, the research progress of multi-objective optimization in materials machine learning is summarized and different Pareto front-based strategies are discussed. Finally, we propose future directions for machine learning-based multi-objective optimization of materials.

Density-functional theory (DFT)-based atomistic simulation methods have been essential in studying the structure-property relationships in heterogeneous catalysis. However, for complex catalytic processes, such as iron-based Fischer-Tropsch synthesis (FTS), the temporal or spatial scales involved are generally too large to perform DFT calculations. Recently, the development of machine learning potentials (MLPs) has demonstrated the capability for atomistic simulation on a large scale and long duration, and the rise of large atomic models (LAMs) is gaining much attention with unified descriptors incorporating a wide range of chemical knowledge and fine-tuning methodology for efficiently deploying the model to downstream tasks. In this work, we construct a MLP named fine-tuned Fischer-Tropsch deep potential (FT$$ ^2 $$DP) model, which is fine-tuned from upstream DPA-2 LAM on a downstream dataset focused on the iron-based FTS process. We further applied this model to investigate iron-based FTS in both surface reactions and reconstructions of edge sites combined with the double-to-single transition state optimization method and the local genetic algorithm. Our work demonstrated the capability and efficiency of our model for iron-based FTS simulations, while revealing the reaction mechanism on common active sites containing [Fe$$ _4 $$C] squares, and the abundant formation of [Fe$$ _4 $$C] squares on several reconstructed surfaces. These insights highlight the potential of utilizing LAM for atomistic simulation for iron-based FTS processes and other complex catalytic reactions.

Optimizing sampling efficiency is crucial for solving complex material design challenges, especially with a limited experimental budget. This study focuses on improving sampling efficiency by reducing the search space for carbide-free bainitic steels through the uncertainty-aware modeling of constraints. These constraints include avoiding the formation of undesirable competing phases such as carbides, ferrite, and martensite, as well as accounting for practical limitations on phase transformation durations. Experimental data, obtained through dilatometry and metallography, inform most constraints, except for the presence of carbides. To model these constraints, we use machine learning (ML) models trained on a combination of newly acquired experimental data and experimental data from the literature. Predicting properties in unexplored regions of the design space can lead to inaccuracies. Thus, reliable uncertainty quantification is essential to avoid excluding parts of the design space due to overconfident erroneous predictions. To address this, we employ conformal prediction (CP), a distribution-free framework that provides calibrated post-hoc uncertainty estimates for the different ML models, ensuring reliable extrapolations without prematurely excluding viable design regions. This approach achieves a reduction ranging from 80% to more than 99% depending on the strictness of the employed criteria reduction in the search space, greatly enhancing sampling efficiency without compromising reliability.