A physics-informed framework for modeling as-manufactured variability and permeability in Automated Dry Fiber Placement using in-line laser profilometry

This work presents a generalized framework for constructing high-fidelity “as-manufactured” virtual preforms, integrating in-line laser profilometry data with a Bivariate Gaussian Process (BGP). The methodology is demonstrated here for Automated Dry Fiber Placement (ADFP), where process-induced variability differentiates physical preforms from their nominal designs and hinders accurate performance prediction. To bridge this gap, the probabilistic framework captures spatial correlations in tow geometry while strictly enforcing the conservation of local fiber volume through a physical coupling between tow width and thickness. These constraints ensure that stochastic representations remain physically admissible. To evaluate the framework, a large-scale Monte Carlo analysis is performed to quantify the in-plane permeability of dry fiber preforms, with representative cases validated against Computational Fluid Dynamics (CFD) benchmarks. Results demonstrate that the physical coupling mechanism induces a systematic geometric flow restriction: tow thinning during compaction reduces in-plane permeability by approximately 4.4% relative to idealized geometries. Based on these findings, a quantitative Process Design Curve is derived, establishing a predictive relationship between manufacturing variability and functional performance degradation. By embedding manufacturing signatures into digital models, this methodology provides a predictive framework with relevance to future virtual qualification workflows.