Artificial Intelligence in Materials Science and Engineering: Current Landscape, Key Challenges, and Future Trajectorie

TL;DR

This review addresses a central challenge in materials science: how to accelerate discovery and design using AI while managing data quality and interpretability. It surveys traditional, deep, and generative ML methods, emphasizing data representations, uncertainty quantification, and integration with physics. Key contributions include gap analyses across data, modeling, multiscale design, sustainability, and standardization, as well as future directions like universal material representations, knowledge graphs, and autonomous experimentation. The work highlights the practical impact of AI-enabled materials science, including faster screening, informed design, robustness, and sustainable manufacturing, while outlining necessary infrastructure and collaboration to realize these gains.

Abstract

Artificial Intelligence is rapidly transforming materials science and engineering, offering powerful tools to navigate complexity, accelerate discovery, and optimize material design in ways previously unattainable. Driven by the accelerating pace of algorithmic advancements and increasing data availability, AI is becoming an essential competency for materials researchers. This review provides a comprehensive and structured overview of the current landscape, synthesizing recent advancements and methodologies for materials scientists seeking to effectively leverage these data-driven techniques. We survey the spectrum of machine learning approaches, from traditional algorithms to advanced deep learning architectures, including CNNs, GNNs, and Transformers, alongside emerging generative AI and probabilistic models such as Gaussian Processes for uncertainty quantification. The review also examines the pivotal role of data in this field, emphasizing how effective representation and featurization strategies, spanning compositional, structural, image-based, and language-inspired approaches, combined with appropriate preprocessing, fundamentally underpin the performance of machine learning models in materials research. Persistent challenges related to data quality, quantity, and standardization, which critically impact model development and application in materials science and engineering, are also addressed.

Figures (15)

• Figure 1: Classification of Machine Learning Techniques and their Applications in Materials Science and Engineering
• Figure 2: A diagram illustrating various types of machine learning algorithms, categorized into different groups such as Regression, Regularization, Deep Learning, and others. Each group is further subdivided into specific methods or techniques, demonstrating the hierarchical structure of machine learning approaches. Each group node is color-coded and spatially arranged to reflect its conceptual relation to the central theme of algorithm types.
• Figure 3: Comparative architectures of deep neural networks and their applications in materials science. FCNNs consist of fully connected layers suitable for tabular data. CNNs utilize convolutional and pooling layers for spatial data, such as microstructure images. GNNs operate on graph structures, making them ideal for modeling atomic structures. RNNs, including LSTMs, are designed for sequential data, enabling analysis of time-series measurements.
• Figure 4: This diagram illustrates the Transformer model's architecture adapted for materials informatics applications. It showcases the flow from input representations—such as atomic sequences or material compositions—through embedding layers and positional encodings, into the encoder comprising multi-head self-attention and feed-forward networks. The decoder, equipped with masked self-attention and encoder-decoder attention mechanisms, processes this information to generate outputs like predicted material properties or novel compositions. This architecture enables the modeling of complex relationships within material data, facilitating tasks like property prediction and generative material design
• Figure 5: Schematic overview of generative AI models applied to materials discovery. This diagram depicts the operational flow of three key generative models: (1) Generative Adversarial Networks (GANs), where a generator creates synthetic data and a discriminator evaluates its authenticity through adversarial training. (2) Variational Autoencoders (VAEs), which learn compact latent representations of data via an encoder-decoder structure to generate new material samples. (3) Diffusion Models, which progressively add noise to data and then reverse this process to generate realistic new samples. These models facilitate the generation of novel materials data, contributing to the discovery of new materials with enhanced properties.