Materials Data toward Machine Learning: Advances and Challenges

Accurate Neural Network Fine-Tuning Approach for Transferable Ab Initio Energy Prediction across Varying Molecular and Crystalline Scales

Existing machine learning models attempt to predict the energies of large molecules by training small molecules, but eventually fail to retain high accuracy as the errors increase with system size. Through an orbital pairwise decomposition of the correlation energy, a pretrained neural network model on hundred-scale data containing small molecules is demonstrated to be sufficiently transferable for accurately predicting large systems, including molecules and crystals. Our model introduces a residual connection to explicitly learn the pairwise energy corrections, and employs various low-rank retraining techniques to modestly adjust the learned network parameters. We demonstrate that with as few as only one larger molecule retraining the base model originally trained on only small molecules of (H2O)6, the MP2 correlation energy of the large liquid water (H2O)64 in a periodic supercell can be predicted at chemical accuracy. Similar performance is observed for large protonated clusters and periodic poly glycine chains. A demonstrative application is presented to predict the energy ordering of symmetrically inequivalent sublattices for distinct hydrogen orientations in the ice XV phase. Our work represents an important step forward in the quest for cost-effective, highly accurate and transferable neural network models in quantum chemistry, bridging the electronic structure patterns between small and large systems.

Research on multi-source microstructure image recognition of foam ceramics using convolutional network combine with frequency domain

Foam ceramics are widely used in industrial applications due to their unique properties, including high porosity, lightweight, and high-temperature resistance. However, their complex microstructure presents significant challenges for image analysis. Traditional machine learning methods often fall short in capturing both global feature dependencies and detailed representations. To address this, a novel artificial intelligence recognition model, FD-Conv, is proposed, which combines the global information processing capabilities of Transformers with the local feature extraction strengths of convolutional neural networks. Additionally, a frequency domain block detail enhancement mechanism is introduced to improve recognition accuracy. Experimental results demonstrate that the FD-Conv model enhances recognition accuracy by at least 7.6% compared to state-of-the-art methods. Furthermore, the model effectively identifies foam ceramics with varying compositions and formulations and quantifies their microstructural phase characteristics. This research aims to advance the application of foam ceramic microstructure image analysis by improving recognition accuracy, particularly in multi-source microscopic image feature learning and pattern recognition.

Solutions for Lithium Battery Materials Data Issues in Machine Learning: Overview and Future Outlook

The application of machine learning (ML) techniques in the lithium battery field is relatively new and holds great potential for discovering new materials, optimizing electrochemical processes, and predicting battery life. However, the accuracy of ML predictions is strongly dependent on the underlying data, while the data of lithium battery materials faces many challenges, such as the multi-sources, heterogeneity, high-dimensionality, and small-sample size. Through the systematic review of the existing literatures, several effective strategies are proposed for data processing as follows: classification and extraction, screening and exploration, dimensionality reduction and generation, modeling and evaluation, and incorporation of domain knowledge, with the aim to enhance the data quality, model reliability, and interpretability. Furthermore, other possible strategies for addressing data quality such as database management techniques and data analysis methodologies are also emphasized. At last, an outlook of ML development for data processing methods is presented. These methodologies are not only applicable to the data of lithium battery materials, but also endow important reference significance to electrocatalysis, electrochemical corrosion, high-entropy alloys, and other fields with similar data challenges.

Unified Roadmap for ZIF-8 Nucleation and Growth: Machine Learning Analysis of Synthetic Variables and Their Impact on Particle Size and Morphology

Metal-organic frameworks (MOFs) have settled in the scientific community over the last decades as versatile materials with several applications. Among those, zeolitic imidazolate framework 8 (ZIF-8) is a well-known MOF that has been applied in various and diverse fields, from drug-delivery platforms to microelectronics. However, the complex role played by the reaction parameters in controlling the size and morphology of ZIF-8 particles is still not fully understood. Even further, many individual reports propose different nucleation and growth mechanisms for ZIF-8, thus creating a fragmented view for the behavior of the system. To provide a unified view, we have generated a comprehensive data set of synthetic conditions and their final outputs and applied machine learning techniques to analyze the data. Our approach has enabled us to identify the nucleation and growth mechanisms operating for ZIF-8 in a given sub-space of synthetic variables space (chemical space) and to reveal their impact on important features such as final particle size and morphology. By doing so, we draw connections and establish a hierarchy for the role of each synthetic variable and provide with rule of thumb for attaining control on the final particle size. Our results provide a unified roadmap for the nucleation and growth mechanisms of ZIF-8 in agreement with mainstream reported trends, which can guide the rational design of ZIF-8 particles which ultimately determine their suitability for any given targeted application. Altogether, our work represents a step forward in seeking control of the properties of MOFs through a deeper understanding of the rationale behind the synthesis procedures employed for their synthesis.

A materials terminology knowledge graph automatically constructed from text corpus

A scalable, reusable, and broad-coverage unified material knowledge representation shows its importance and will bring great benefits to data sharing among materials communities. A knowledge graph (KG) for materials terminology, which is a formal collection of term entities and relationships, is conceptually important to achieve this goal. In this work, we propose a KG for materials terminology, named Materials Genome Engineering Database Knowledge Graph (MGED-KG), which is automatically constructed from text corpus via natural language processing. MGED-KG is the most comprehensive KG for materials terminology in both Chinese and English languages, consisting of 8,660 terms and their explanations. It encompasses 11 principal categories, such as Metals, Composites, Nanomaterials, each with two or three levels of subcategories, resulting in a total of 235 distinct category labels. For further application, a knowledge web system based on MGED-KG is developed and shows its great power in improving data sharing efficiency from the aspects of query expansion, term, and data recommendation.

Machine learning interatomic potential: Bridge the gap between small-scale models and realistic device-scale simulations

Machine learning interatomic potential (MLIP) overcomes the challenges of high computational costs in density-functional theory and the relatively low accuracy in classical large-scale molecular dynamics, facilitating more efficient and precise simulations in materials research and design. In this review, the current state of the four essential stages of MLIP is discussed, including data generation methods, material structure descriptors, six unique machine learning algorithms, and available software. Furthermore, the applications of MLIP in various fields are investigated, notably in phase-change memory materials, structure searching, material properties predicting, and the pre-trained universal models. Eventually, the future perspectives, consisting of standard datasets, transferability, generalization, and trade-off between accuracy and complexity in MLIPs, are reported.

Data-Driven Predetermination of Cu Oxidation State in Copper Nanoparticles: Application to the Synthesis by Laser Ablation in Liquid

Copper-based nanocrystals are reference nanomaterials for integration into emerging green technologies, with laser ablation in liquid (LAL) being a remarkable technique for their synthesis. However, the achievement of a specific type of nanocrystal, among the whole library of nanomaterials available using LAL, has been until now an empirical endeavor based on changing synthesis parameters and characterizing the products. Here, we started from the bibliographic analysis of LAL synthesis of Cu-based nanocrystals to identify the relevant physical and chemical features for the predetermination of copper oxidation state. First, single features and their combinations were screened by linear regression analysis, also using a genetic algorithm, to find the best correlation with experimental output and identify the equation giving the best prediction of the LAL results. Then, machine learning (ML) models were exploited to unravel cross-correlations between features that are hidden in the linear regression analysis. Although the LAL-generated Cu nanocrystals may be present in a range of oxidation states, from metallic copper to cuprous oxide (Cu2O) and cupric oxide (CuO), in addition to the formation of other materials such as Cu2S and CuCN, ML was able to guide the experiments toward the maximization of the compounds in the greatest demand for integration in sustainable processes. This approach is of general applicability to other nanomaterials and can help understand the origin of the chemical pathways of nanocrystals generated by LAL, providing a rational guideline for the conscious predetermination of laser-synthesis parameters toward the desired compounds.