Probabilistic characterization of effective orthotropic laminate properties from directional guided-wave velocities using Gaussian-process surrogates and Bayesian calibration

Abstract

This study develops a two-stage probabilistic framework for separating specimen-level orthotropic behaviour from direction-dependent guided-wave perturbations in a glass-fibre laminate. Directional A₀ group velocities at 70kHz are measured on two concentric paths using a dry-point contact system and predicted using a Wave Finite Element laminate model. In Stage 1, both rings are used jointly to infer one common orthotropic reference model. In Stage 2, the remaining directional response is represented by a smooth common angular scaling field and ring-specific local deviations, without prescribing affected directions. Gaussian-process surrogates enable adaptive Markov-chain Monte Carlo inference, while the likelihood distinguishes experimental variability, surrogate uncertainty, and ring-specific model discrepancy. A local parameter-sensitivity analysis restricts the effective scaling to stiffness components observable through the selected A₀ feature. Synthetic recovery tests and baseline-assisted/baseline-free comparisons are used to assess identifiability and the possible absorption of broad anomalies into the inferred reference. The experimental results identify directional perturbations that are consistent with the known impact-influenced propagation paths, while the inferred scaling fields and moduli are interpreted as effective guided-wave-equivalent quantities rather than direct local material measurements.