Study Of Ultrasonic Detection For Bonding Flaws Of Thin Plate Based On Time-Frequency Analysis And Artificial Neural Network

Inner-flaws detecting technology of thin plate of composite materials is a newly arisen and many subjects intersectional technology which has been successfully applied in production of thin plate of composite materials and many other fields. This paper is around inner-flaws detecting technology of thin plate of composite materials, and ultrasonic non-destructive testing for flaws is studied using four eigenvalues by time-frequency features of the ultrasonic echo. We study ultrasonic detection for bonding flaws of thin plate based on time-frequency analysis and artificial neural network by combining the Artificial Neural Network(ANN) with four eigenvalues.Echo signal of ultrasonic detection takes information of interface bonding quality. How to extract the information is the key that bonding quality can be quantitatively recognized. Common signal processing methods always analyze echo signal in time or frequency domain solely. So helpful information is not extracted roundly and exactly. This paper analyzes the characteristics of the echo signal, and based on this, we firstly define characters of attenuation coefficient of echo and energy of shortcoming signal models in time domain. Then, we extract characters by the maximum entropy spectrum estimation and wavelet analysis in frequency domain. In maximum entropy spectrum analysis, the spectrum move method is applied to estimate the center frequency of echo and center frequency deviation is a character. In wavelet analysis, we use wavelet transformation to process the echo signal. The echo signal is decomposed by db4 wavelet. And then detail signals of different grades are reconstructed. According to the characteristic of echo signal in frequency domain, the frequency energy of detail signal is a character. Based on those methods, the four eigenvalues are established. We design an artificial neural network pattern recognition algorithm to judge bonding quality by an improved BP algorithm. Simulation results show that the algorithm is very exact for quantitative recognition of bonding quality. Time-frequency features and Artificial Neural Network pattern recognition are applied in bonding quality ultrasonic detection of thin composite plate is very helpful for design and realization of digital ultrasonic detection instrument.