Artificial intelligence and machine learning in composite materials: A comprehensive literature review and bibliometric analysis

Abstract

Artificial intelligence (AI) and machine learning (ML) are transforming composites research and engineering by augmenting experimental testing and physics-based simulation with data-driven models for analysis, design, prediction, and optimization. This paper presents a comprehensive, quantitative review of the field by performing a bibliometric analysis of 1,474 AI/ML-enabled publications from 2010–2025 across seven leading composites journals, followed by an in-depth review of over 200 representative studies. Following a systematic methodology, we develop a taxonomy that maps composite research topics to AI/ML methods to investigate cross-cutting trends, impactful and emerging topics and methods, as well as country-level publication patterns and method adoption. The analysis shows that Mechanics & Failure dominates publication volume (>700 papers, ~50% of AI/ML-enabled composite publications), while Sustainability exhibits the highest citation impact relative to its small volume. In contrast, thermal protection and thermo-mechanical topics emerge as high-novelty areas with lower impact. Methodologically, literature remains dominated by artificial neural networks (ANNs, ~26%), followed by clustering methods (~14%), with increasing use of probabilistic approaches such as Gaussian process regression (GPR) and vision-based convolutional neural networks (CNNs). Beyond trend mapping, critical gaps are identified, including a lack of leakage-safe validation, a significant “virtual-to-real gap” despite high accuracy metrics, and limited generalization under distribution shift. These issues are largely rooted in the inherently stochastic nature of composite response, driven by material and process variability. In response, emerging directions increasingly prioritize physics-informed learning, probabilistic modeling and uncertainty quantification, transfer learning, and Bayesian updating enabled by sparse sensing.