Data‐Driven High‐Fidelity Reconstruction of 3D Carbon Fiber Nonwoven Preforms for Polymer Composites: Integrating Micro‐ CT Characterization With Stochastic Modeling

Abstract

A data‐driven framework integrating micro‐CT characterization with stochastic modeling is developed to generate statistically equivalent 3D virtual fiber architectures for carbon fiber nonwoven preforms. Key microstructural descriptors, including fiber length, orientation, curvature, and bundle width, are extracted from micro‐CT data to serve as inputs for the stochastic fiber growth algorithm. Multi‐scale calibration and validation confirm the statistical equivalence between the virtual preform and the physical specimen. The calibrated model parameters include fiber length distribution (lognormal, μ  = 1.9, σ  = 0.633), orientation distributions (polar angle: normal at 90° with σ θ  = 1.1°; azimuthal: bimodal at 15°/155° with σ φ  = 45°), and curvature autocorrelation coefficients ( R Δθ  = 0.7, R Δφ  = 0.5). The spatial hashing strategy reduces computational cost, enabling efficient generation of large‐scale preforms. This modeling framework allows the geometric reconstruction of fibrous materials with different fiber volume fractions and provides a sound geometric basis for evaluating permeability and mechanical properties of the composites.