Delamination area quantification in composite structures using Gaussian process regression and auto-regressive models

After detecting initial delamination damage in a hotspot region of a composite structure monitored through a data-driven approach, the user needs to decide if there is an imminent structural failure or if the system can be kept in operation under monitoring to track the damage progression and its impact on the structural safety condition. Therefore, this study proposes delamination area quantification by stochastically interpolating global damage indices based on Gaussian process regression and taking into account uncertainty. Auto-regressive models are applied to extract damage-sensitive features from Lamb wave signals, and the Mahalanobis squared distance is used to compute damage indices. Two sets of laboratory tests are used to demonstrate the effectiveness of this methodology—one in carbon–epoxy laminate with simulated damage under temperature changes to show the general steps of the procedure, and a second test involving a set of carbon fiber–reinforced polymer coupons with actual delamination caused by repeated fatigue loads. Various levels of progression damage, measured by the covered area of delamination, are monitored using piezoelectric lead zirconate titanate patches bonded to the structural surfaces of these setups. The Gaussian process regression proved to be capable of accommodating the uncertainties to relate the damage indices versus the damaged area. The results exhibit a smooth and adequate prediction of the damaged area for both simulated damage and actual delamination.