On suitability of feature extraction techniques for local damage detection

Damage in the form of cracks near rivet holes in a steel channel section can be characterized by inspecting ultrasonic signals containing valuable information about these anomalies. Time-frequency representation (TFR) of time-history signal is an effective way to extract damage features out of an ultrasonic signal scattered from cracks. Several techniques are available to obtain Time-frequency representation and out of which feature extraction can be performed. However, every technique has its own advantages and disadvantage which makes it cumbersome to ascertain which specific technique is suitable to which specific problem. In present study, six TFR techniques e.g. Short Time Fourier Transform, Continuous Wavelet Transform, Wigner-Ville Spectrum, Hilbert-Huang Transform, Williams-Choi Transform and Stransform have been used to extract feature out of time-history signal obtained from finite element based wave scattering simulation of a plate with and without cracks near the rivet holes. Extracted damage features have been used to quantify the damage as a unique value by defining damage index formulation. Further, a comparison study has been carried out to assess these six techniques for their ability to give effective, reliable and consistent information about the cracks. Matlab codes have been developed to perform feature extraction and damage index calculation.