Recursive partitioning and Gaussian Process Regression for the detection and localization of damages in pultruded Glass Fiber Reinforced Polymer material

In this paper, a methodology for the detection and localization of damages in composite pultruded members is proposed. This is particularly relevant to thin-walled pultruded members, which are typically characterized by orthotropic behavior, anisotropic along the fibers and isotropic in the cross section. Hence, a method to detect and localize damage, and the influence these might have on the performance of thin-walled Glass Fiber Reinforced Polymer (GFRP) members, is proposed and applied to both numerical and experimental data. Specifically, the numerical and experimental modal shapes of a narrow flange pultruded profile are analyzed. The reliability of the proposed semiparametric statistical method, which is based on Gaussian Processes Regression and Bayesian-based Recursive Partitioning, is analyzed on a narrow flange profile, artificially affected by sawed notches with incremental depth. The numerical investigation is carried out via finite element models (FEMs) of the cracked beam, where the dynamic parameters and the modal shapes are computed. In total, three different crack sizes are investigated, to compare the results with the experimental ones. Finally, the proposed approach is further extended and validated on numerically simulated frame structures.