Bilinear Time-Frequency Transformations in the Analysis of Damaged Structures

The performances of six time-frequency transforms applied to the assessment of structural damage are compared and their properties discussed from a theoretical point of view. Some examples of applications to synthetic and real signals are presented. Specifically, their diagnostic sensitivity when processing accelerometer signals recorded during the impulsive loading of damaged aluminum beams is discussed. The test programme involved different position and damage levels artificially produced on the beams. The results obtained outline the importance of finding criteria for a fully automatic interpretation of the time-frequency transforms. Also the problems associated with the use of neural pattern recognition in the diagnostic interpretation of time-frequency maps of signals detected on structures are illustrated. An application shows the results obtained in the neural classification of structural defects by applying techniques based on invariant moments