A novel method for feature extraction of multi-layered structure ultrasonic echo signals

A novel method named wavelet-fractal dimension of ultrasonic echo signals in the feature extraction was developed. The method is made up of two main steps. Firstly, the ultrasonic echo signals are decomposed into varying scales by dyadic wavelet transform in order to protrude the signal components on damage. Wavelet transform is capable of signifying the detail of signal in both time region and frequency region. The wavelet packet technique focuses on each point of signal with alterable window to effectually analyze the nonlinear dynamical characteristic of damage structures. Secondly, combining fractal dimension with wavelet transform, a wavelet fractal neural network is established. The results demonstrate that in different states the fractal dimensions of multilayered echo signals at various frequencies are quite distinguishable, thus the fractal dimensions of signals is able to serve as the character of structural damage detection, to identify the different structural states with neural network.