Artificial intelligence (AI) is profoundly transforming the paradigms and methodologies of advanced materials research and development. This review systematically examines cutting−edge advances in AI applications across materials composition/structure design, property prediction, synthesis optimization, and industrial implementation. By integrating data−driven approaches, physics−informed modeling, and autonomous experimental systems, AI has enabled high−accuracy cross−scale performance prediction, inverse design of materials with extreme properties, intelligent optimization of synthesis processes, and non−destructive defect detection, significantly accelerating development cycles while overcoming performance bottlenecks. The work highlights breakthroughs in representative case studies including high−throughput screening of stable crystals, targeted development of radiative cooling materials, and optimization of electrolytes for high−voltage batteries, while elucidating how techniques such as few−shot learning, transfer learning, and physics−constrained algorithms address challenges in data scarcity and multiscale modeling. Looking forward, the synergistic convergence of AI with quantum computing and generative design will propel materials innovation toward an accelerated transition to advanced paradigms characterized by data−driven workflows, autonomous decision−making, and intelligent iteration.

Artificial intelligence (AI) technology is profoundly transforming the research paradigms in the field of materials science, driving the analysis methods for material microstructures to shift from traditional human−experience−dominated approaches to data−driven intelligent recognition. AI−based microstructure recognition and quantification, characterized by high accuracy and efficiency, have significantly advanced the development of high−throughput microstructure analysis techniques. This review focuses on the emerging field of AI−assisted microstructure analysis of metallic materials. Following the development from qualitative analysis toward refined quantitative analysis of microstructures, it systematically summarizes the research progress in traditional machine learning algorithms, deep learning−based classification, object detection, and semantic segmentation algorithms for the classification, recognition, and quantification of metallic material microstructures. Particular emphasis is placed on the current state of widely adopted semantic segmentation algorithms. Meanwhile, addressing the challenges faced by semantic segmentation in this domain, such as high microstructural complexity and limited annotated samples, the innovative strategies proposed by researchers in data augmentation and model architecture improvements, along with their enhanced performance, are discussed. Finally, the existing limitations and future directions of AI−based microstructure analysis methods are summarized and outlooked.

Dual−phase steel with improved formability (DH steel) is developed as an evolution of conventional dual−phase steel (DP steel) to meet the increased ductility requirements associated with the fabrication of complex−shaped automotive components. Currently, DH steel with a tensile strength of 980 MPa has reached mass production, while the development of DH steel with a tensile strength of 1180 MPa has attracted significant research interest. In this study, a performance−driven machine learning methodology was employed to design the chemical composition and processing parameters of 1180 MPa−grade DH steel. Additionally, interpretable machine learning techniques were used to elucidate the fundamental relationships between the microstructural characteristics and mechanical properties. Initially, leveraging data extracted from the literature, a composition and process−performance predictive model was developed using a neural network algorithm. Subsequently, a multi−objective genetic algorithm was implemented to efficiently design the chemical composition of the novel DH steel. Following this, based on orthogonal experimental data concerning the processing parameters of the newly designed DH steel, a random forest algorithm was applied to construct predictive models for tensile strength and fracture elongation, with processing parameters serving as input variables. An optimized set of preparation process parameters was determined using a multi−objective genetic optimization algorithm. The resulting parameters are as follows: a coiling temperature of 510°C, an annealing temperature of 860°C, an annealing duration of 160 s, a slow cooling temperature of 715°C, an over−aging temperature of 340°C, and an over−aging duration of 110 s. The resulting DH steel demonstrated an exceptional balance between strength and ductility, achieving a tensile strength of 1214 MPa and an elongation after fracture (A₈₀) of 15.5%. Finally, SHAP analysis was conducted to reveal the influence patterns of microstructural features on mechanical performance, thereby providing theoretical insights to guide the design and microstructure−performance optimization of advanced high−strength steels.

With the continuous advancement of technologies in data acquisition, deep learning, and model generation, data−driven methods have provided a powerful tool for predicting the properties of fiber−reinforced composites, leveraging their unique advantages in uncovering high−dimensional nonlinear relationships, constructing surrogate models, and processing multimodal data. This review systematically reviews recent progress in this field, categorizing digital characterization methods into four types: collection of intrinsic material parameters, image−driven feature extraction, physics−informed feature engineering, and cross−scale data−driven techniques. It summarizes the modeling strategies and prediction accuracy of data−driven models in predicting the mechanical, thermal, acoustic, and electrical properties of composites. The engineering significance of interpretability analysis and uncertainty quantification techniques is elaborated, highlighting their roles in enhancing model transparency and quantifying prediction risks. This review aims to provide a comprehensive perspective—from theoretical foundations to engineering applications—for the deeper application of data−driven methods in predicting the properties of composites.