Research On The Method Of Feature Extraction For Plate Bonding Flaw Based On The Improved EMD Algorithm

Because of its high specific strength, high specific modulus, excellent damping performance and simple process, the composite plate has been widely used in the real life. If we can judge the quality of composite plate in advance, it can effectively reduce the huge loss. The professionals can only judge whether the composite plate is free from bonding or not, but can not accurately determine the flaw degree of the composite plate. In view of such limitations, this paper researchs on the method of feature extraction for plate bonding flaw, and then the eigenvalues are used to identify the composite plates with different debonding grades. It will lay the foundation for the automatic recognition of the machine.However, a lot of information that can be used to evaluate the quality of the composite plate is concealed in the ultrasonic echo signal detected by the detector in some form. So the echo signals under the ultrasonic detection require the use of time-frequency method to analyse. In order to cognize the physical characteristics better which are represented by signals, we must transform some information feature in signals into a form which is easy to understand. Aiming at the problem of mode mixing and endpoint effects caused by the traditional Empirical Mode Decomposition (EMD) method, this paper has proposed the improved method of EMD based on the noise assisted and the improved method of EMD based on the mirror continuation. The experimental simulation results show that these two methods can effectively suppress the phenomenon of mode mixing and endpoint effects. In addition, at the time of denoising the echo signals, a new method based on the cubic spline interpolation is proposed in this paper. The experimental results show that this method can effectively filter out the high frequency interference signals in the echo signals, and achieve the purpose of improving the decomposition effect of EMD.Through the analysis of the characteristics of the echo signals and the principle of ultrasonic detection, it is known that the energy of the echo signals increases with the increasing of the debonding grades. So we choose the time domain energy and the IMF1 energy moment as the eigenvalues. In addition, with the different debonding grades, the wavy degrees of the echo signals are not the same. Therefore, we choose waveform index and weighting waveform index as the eigenvalues. Using the improved EMD method to analyze the echo signals of ultrasonic detection, we can extract the eigenvalues that can identify the flaw grades of composite plate. Finally, we use the SVM method to train the extracted eigenvalues and identify the composite plates in ten different debonding grades. The recognition results show that the recognition accuracy of the plates is high, and the expected goal is achieved.