| Ultrasonic testing has high value in use and plays the key role in field of NDT. Many researches on ultrasonic testing do exist, but most of them are focused on qualitative recognition and evaluation, research on quantitative recognition is still in primitive stage. This paper, which is set in quantitative recognition for bonding quality of composite plate, designs an opened ultrasonic testing system which detects de-bonding by means of ultrasonic echo and analyzes methods of signal feature extraction, which play the key role in quantitative recognition for de-bonding of composite plate.Besides time domain and frequency domain analysis, this paper proposes a new time-frequency analysis method named empirical mode decomposition (EMD) to extract signal features. EMD reveals features of echo in time-frequency domain effectively, improves resolution of signal analysis and expands more space for speculating signal parameters. It is difficult for signal processing methods to directly find signal features which can recognize bonding quality quantitatively (?)ithout deep research on transmission principle of ultrasound in composite plate. The paper extracts a lot of signal parameters in time and frequency domain as well as time-frequency domain, evaluates the feature effectiveness of high-dimensional signal parameters with the help of feature effectiveness evaluation thesis. Thus, invalid signal parameters are excluded and a new low-dimensional set of effective features is available.This paper applies exhaustive searching strategy, evaluation criterion of geometry distance measurement and filter algorithm to feature selection, a three-dimensional set of effective features is searched and verified to be the best one by Support Vector Machine (SVM). The paper proposes an advanced method for feature extraction, provides a new way for researches on quantitative recognition and feature extraction, and strengthens the foundation of 10-level quantitative recognition for de-bonding in composite plate. |
PDF Full Text Request