Ultrasonic Signal Classification And Identification Study Based On Empirical Mode Decomposition And Neural Network

Ultrasonic nondestructive examination is a kind of most widely used methods of nondestructive examination, which status is very important in field of nondestructive examination. In field of ultrasonic nondestructive examination, Ultrasonic spectrum analysis can gain more available information than conventional detection methods, which supply probability to improve objectivity and reliability of result. At the same time, conventional ultrasonic method is difficult to identify the types of defects, even variety of scanning method were used, identification of defect types still need highly skilled artificial technique. So it require some further research work to make use of the these information from spectrum analysis in defects pattern recognition and defects qualitative evaluation, automatic identification and intelligence examination aspect, and the work has great significance in development of ultrasonic nondestructive examination.Firstly, the basic principles of EMD decomposition method and neural network were introduced in this paper, ultrasonic echo signal was analyzed in time domain and frequency domain by EMD, and then the parameters of time and frequency domain characteristics were analyzed and summarized. On the basis, some analysis was made in this paper, including selection principle of neural network input characteristics parameters and how to select part characteristics parameters of time domain and frequency domain as input of neural network, and come to an ideal characteristic parameters combination.Secondly, based on the basic theory of neural network technology and combined with neural network input characteristics vector, according to specific characteristics of ultrasonic echo signal, identification defect diagnosis system for ultrasonic echo signal based on BP was built up, and the specific structure of BP neural network was designed.Finally, using the measuring data of former research subject, the training samples and validation samples are built based on the time and frequency domain characteristics of echo signal. Finally, the BP neural network is trained on MATLAB, and then an experiment is made to verify the validity of neural network for diagnosing the defect type of ultrasonic echo signal. The result shows that BP neural network used for diagnosing the defect type of ultrasonic echo signal is effective and practicable.