Pattern recognition of super-alloy friction welding joint defect by wavelet packet and wavelet neural network

Detecting the super-alloy friction welding specimens of GH4169 by using the UltraPAC system, aiming at the characteristic of defects, the method of analyzing and extracting the defect eigenvalue by using wavelet packet analysis and pattern recognition by making use of the wavelet neural network is discussed. This method can realize to extract the interrelated information which can reflect defect characteristic from the ultrasonic information being detected and analysis it by the information. Constructing the network model for realizing the qualitative recognition of defects which is improved though experiment finally. The results show that the wavelet packet analysis adequately make use of the information in time-domain and in frequency-domain of the defected echo signal, multi-level partition the frequency bands and analyze the high-frequency part further which don 't been subdivided by multi-resolution analysis, and choose the interrelated frequency bands to make it suited with signal spectrum. Thus, the time-frequency resolution is rising, the good local amplificatory property of the wavelet neural network and the study characteristic of multi-resolution analysis can achieve the higher accuracy rate of the qualitative classification of welding defect.