Abstract
<title>Abstract</title> <p>Additive manufacturing (AM) and data-driven design are reshaping how multifunctional and adaptive composites are conceived, allowing load-bearing architectures to be co-designed with sensing, actuation, energy-storage and shape-changing functions. Yet the coupling between constituent selection, processing route and the resulting multifunctional performance remains difficult to navigate by trial and error. Here we combine a concise synthesis of AM-enabled multifunc-tional and adaptive composites, architected metamaterials and artificial-intelligence (AI)-assisted design with an original, reproducible case study that quantifies how well machine-learning sur-rogates predict the mechanical response of fibre-reinforced polymer composites from composition and processing descriptors. Using an open experimental dataset of 420 carbon-, glass-, basalt-and aramid-fibre composites, we trained linear, random-forest and gradient-boosting models to predict warp-direction tensile strength from thirteen constituent, textile and processing descriptors, using material-grouped cross-validation to prevent information leakage between identical materials. Ensemble models explained about 70% of the variance across unseen material families (R 2 = 0.70, RMSE ≈ 276 MPa) and reached R 2 ≈ 0.90 on a held-out set, with fabric architecture and yarn deformability emerging as dominant descriptors. We discuss how such surrogates integrate with inverse design and 4D printing to accelerate the development of multifunctional and adaptive composites, and outline the data and methodological gaps that currently limit the approach.</p>