Reliability-Based Design Optimization of Composite Structures With Multi-Fidelity Surrogates

Automated Fibre Placement (AFP) manufacturing introduces defects that can strongly degrade the reliability of composite structures. While some defects are deterministic and can be predicted from process parameters, many are stochastic, including fibre-angle deviations, resin-rich zones, and local variability in fibre volume fraction, which cannot be fully characterized a priori. This paper proposes a multi-fidelity framework for Reliability-Based Design Optimization (RBDO) that explicitly propagates stochastic manufacturing defects into structural performance and design decisions. The approach combines lowand high-fidelity structural models within a Gaussian Process Regression (GPR) multi-fidelity surrogate. Structural responses are computed using the Carrera Unified Formulation (CUF), where Equivalent Single-Layer (ESL) theories provide fast, computationally efficient predictions, while Layerwise (LW) theories deliver the high-fidelity through-thickness accuracy needed for reliable stress and failure assessment. An adaptive learning strategy selects the most informative design regions, enriching the surrogate with a limited number of LW evaluations and learning correction functions that refine the ESL predictions. The RBDO formulation targets mass and strain energy minimization while satisfying probabilistic stress metrics and AFP manufacturing requirements.