Fractal And Wavelet Neural Network In The Ultrasonic Signal Analysis Of The Friction Welding Applications

The fraction welding ultrasonic testing signal processing and defect detection problems are studied in this paper. In the thesis, the major research and application on the method of denoising and defect detecting the ultrasonic testing signal and ultrasonic c scanning Images of friction welding joints which is made from GH4169 based on wavelet theory,WNN,theory of fractal and mathematical morphology. The main researching work and results are summarized as follows.1. Three new threshold denoising functions which avoid to be discontinuous as hard threshold function and hasn't the constant deviation of soft threshold function was proposed. Simulations show that three modified thresholding functions are proved to be effective to denoise the ultrasonic testing signal noise. The simulation results also show that the new threshold function presents better RMSE and SNR performance.2. According to the property of wavelet packets transforms, a denoising approach based on the best wavelet packet base is proposed. The new threshold function used in the ultrasonic testing signal denoising process.The result is satisfied.3. Considering the ultrasonic inspection signal in friction welded joints and making the advantage of non-linear system identification of WA and NN, a Wavelet Neural Network (WNN) used for the signal classification is constructed .The classification of welded defects are carried out. The result showed that comparing BP neural network a wavelet neural network acquired to a good identifying result.The good result is achieved and the profitable attempt is done in combination wavelet with neural networks.4. Application of morphological filtering operations on the method of denoising the ultrasonic C scanning images of friction welding is proposed. Experiment results demonstrate that the method can remove noise effectively and preserve the details of images completely.It manifests better performances than the convention alfiltering methods.5. A computing method of fractal dimension of ultrasonic C scanning image is given. As a characteristic variables of defect image, fractal dimension reveals the features of the ultrasonic C scanning image.