An Exploratory Study on Data-Driven Vibration Based Damage Detection and Characterization for a Population of Composite Aerostructures

This study investigates the effectiveness of four robust to uncertainty data-driven methods of machine learning type for random vibration response-only damage detection and characterization in a population of composite aerostructures under various uncertainty factors. The employed methods are based on the Multiple Model (MM) and the Hyper-Sphere (HS) frameworks using Multiple Input Single Output AutoRegressive with eXogenous excitation Transmittance Function (MISO TF-ARX) models that may account for excitation uncertainty. The methods capabilities are explored via numerous Monte Carlo simulations using digital, Abaqus based, models allowing for maximum flexibility in experimentation. Based on these, a population of 90 Carbon-Epoxy square hollow beams with lightweight composite aerostructures properties is employed taking into account manufacturing, temperature and excitation uncertainty. In addition, early-stage debonding and delamination damages are inserted to a portion of the population at two levels per damage and at two distinct locations for the methods evaluation. All considered damages affect slightly and similarly the structural dynamics overlapping with the effects caused by uncertainty in order to increase the diagnosis (detection & characterization) difficulty and explore the methods' performance limits. The results indicate almost perfect detection in all considered damage scenarios except from the low level debonding, while damage characterization (type, location and level), which is performed via a hierarchical classification scheme, is very promising.