Study Of Signal Characteristic And Near Surface Defect Recognition Of Friction Stir Welding Based On Ultrasonic TOFD Technology

At present, our country industry is in rapid development period, and aluminumalloy friction stir welding technology is also more and more used in aerospace,machinery, electrical appliances, automobile manufacturing, shipbuilding, high-speedtrains and other kinds of key components of the product. Due to the temperature fielddistribution of friction stir welding is difficult to control, tiny fluctuations of weldingparameters can easily lead to various types of defects (such as flash, groove,incomplete penetration, lack of fusion, holes, etc.). Therefore, the nondestructivetesting technology for friction stir welding quality is particularly important. Themethod of ultrasonic Time of Flight Diffraction (TOFD) is suitable for defectdetection in thickness direction, and the defect of aluminum alloy friction stir weldcan be detected quickly and efficiently by ultrasonic TOFD imaging detection, whichthe depth and height of defect can be measured accurately. Moreover, the weld qualityand product structure can be gkuranteed by the technology. An effiective ultrasonicTOFD method is more significant for saving manufacturing cost, improving productperformance, avoiding accidents, and so on.Ultrasonic TOFD detection technology of near surface defects detection isdiscussed in the paper. Because of the near surface detection blind area, it is verydifficult to decte near surface flaws by use of ultrasonic TOFD technology. At thesame time, the diffraction wave of the end of tiny defects is weak and mixed withnoise in friction stir weld. The low signal-to-noise ratio influence the ultrasonic TOFDdetection directly, cause of defect mistakenly identified or leak. Therefore, improvingthe testing ability for near surface defects and enhancing image quality of TOFDdetection are the objectives which must be received.Aiming at the problem of near surface blind area in ultrasonic TOFD technique,A automatic identification technology of near surface defects is proposed based onthrough wave distribution of ultrasonic TOFD and neural network. In this study, thediffraction signal of near surface defects is stimulated by optimizing the detectionparameters of ultrasonic TOFD, The diffraction signal superposition with direct wavesignal, and directly affect the amplitude distribution of the through wave pulse peak.Several key points in the part of through wave of testing signal are extracted and relationship between the amplitude distribution of key points and depth ofnear-surface defect is revealed, the characteristic numbers of amplitude distributionwhich can be used to testing near-surface defect is obtained. In addition, in order toavoid the complexity of many characteristic numbers for evaluating defect, this studyis also designed to identify the characteristic numbers for BP neural network.Moreover the characteristic numbers can be used as feature vectors for BP neuralnetwork, and the parameters of input layer, hidden layer and output layer areoptimized, the BP neural network which can be used to automatically identify for nearsurface defects is obtained. The experimental results showed that this technique canbe accurate and effective classification and automatic identification for near surfacedefects of aluminum alloy friction stir welding. This technique can also accuratelydetect the defects with buried depth of1.0mm, and the recognition ability tonear-surface defect is effectively improved.In this paper, the ultrasonic TOFD technique is adopted for testing friction stirweld. To increase the signal-to-noise ratio and enhance the image quality of D-scan,the signal of ultrasonic TOFD detection was decomposed by wavelet analysis, and thecomposition of detecting signal is analyzed in time and frequency domain. The noiseis separated from the defect signal and each dimension of the high frequency part ofthe wavelet coefficient is reconstructed by threshold quantization criterion. The usefulsignal for detection is extracted after reconstruction. The results show that thesignal-to-noise ratio of the signal was improved according to the wacelet analysis, andthe D-Scan imaging based on the processed signals becomes more accuracy.