Machine Learning‐Based Strategy for Predicting the Multiscale Mechanical Properties of Helical Carbon Fiber Composites

Inspired by biological structures, multilevel helical configurations enhance material toughness and strength. This study develops high‐performance helical twisted carbon fiber composites (HTCFCs) and introduces a machine learning (ML)‐based multiscale performance prediction method to bridge mesoscale parameters with macroscopic properties. HTCFC specimens undergo experimental testing, validate through finite element method (FEM) simulations. A dataset of 480 material configurations is generated, incorporating helix angle, volume fraction, resin modulus, and Poisson's ratio. A three‐step ML optimization process—combining cross‐validation, random grid search, and Bayesian optimization—is applied to 16 ML models, yielding a highly accurate predictive framework. Compared to FEM, the ML model reduces computation time to 0.01%–0.03% of the original. In addition to accelerating large‐scale parametric exploration, the model also enables sensitivity‐based interpretation of structural mechanisms, supporting both efficient design and physical understanding. SHapley Additive exPlanations (SHAP) analysis identifies key structural parameters, confirming high model accuracy and alignment with experimental results. This study demonstrates that ML‐driven multiscale modeling is a fast and effective tool for optimizing next‐generation composite materials, advancing lightweight and high‐strength engineering applications.