Study of cross-correlation signals in a data-driven approach for damage classification in aircraft wings

This paper discusses, experimental results of classifying several mass adding in a wing aircraft structure, using cross-correlated piezoelectric signals, represented by principal components. Piezoelectric signals are applied and recorded at specific points of the structure under analysis. Then, statistical features are obtained by means of principal component analysis to the correlation between excitation and response signals. Unsupervised learning is implemented to the reduced feature space, in order to identify clusters of damaged cases. The main result of this paper is the advantage resulting from using cross-correlated signals, evaluated through the performance of clustering indexes. Experimental data are collected from two test structures: i.) A turbine blade of a commercial aircraft and ii.) The skin panel of the torsion box of a wing. Damages are induced adding masses at different locations of the wing section surface. The results obtained show the effectiveness of the methodology to distinguish between different damage cases.