Empirical Modeling of Mechanical Damages, Weathering and Boundary Conditions in GFRP Plates Simulation for Robust Machine Learning Training

The data is in Linux SquashFS format, can be opened with 7-zip on all platforms and mounted on Linux. The file contains all the draft versions of the scientific paper as well as the data, scripts and ANSYS projects used to create the paper. Abstract Modern data‑driven structural health monitoring requires large high‑fidelity datasets, making simulation essential when experimental data are limited in scale. This work proposes an efficient methodology for constructing a linear finite element model of glass fiber reinforced polymer (GFRP) plates to accelerate data generation . A baseline ANSYS model was built under linearization assumptions and calibrated through localized material parameter updating (modal updating). Various empirical strategies were explored to replicate physical imperfections by integrating manufacturing-induced edge defects from guillotine cutting, such as resin irregularities and fiber degradation, alongside sub-surface damage mechanisms like delamination and matrix cracking via localized parametric modifications. The efficacy of these approaches was systematically compared, evaluating the quantitative improvements each approach yielded regarding the model’s fidelity and predictive accuracy. The resulting model incorporates effective damping and reproduces experimental modal characteristics with high accuracy. By avoiding computationally expensive nonlinear simulations, the proposed framework enables rapid generation of large‑scale simulation datasets for robust machine‑learning‑based SHM.