Reliable damage localization in cantilever beams using an image similarity assessment method applied to wavelet-enhanced modal analysis

Abstract Modal analysis is a powerful tool for the detection of damage in structures such as cantilever beams. The Continuous Wavelet Transform (CWT) is often used to locate damage, but may lead to unsatisfactory results when methods designed and tested successfully on simulated beams are applied to experimental tests. In the present work, a novel approach is presented that allows to reliably detect the correct damage locations along a beam in an experimental setup, using only finite-element simulated data as a reference. The novel method is based on two main innovations: (1) an image similarity assessment method making an efficient use of the full wavelet scalogram, and (2) an ensemble approach accounting for uncertainties in materials properties. The image processing method is shown to dramatically outperform three reference methods constructed for comparison, two of them based on signal processing approaches proposed in literature, combined with a neural network processing step. This superiority is demonstrated through a systematic assessment both of simulated data, where the noise level was varied for sensitivity analyses, and of experimental data. When combined with the ensemble approach the new methodology is shown to have a 100% damage recognition rate on the experimental data studied.