Ultrasonic Inspection And Signal Recognition Of The Defects In Friction Stir Welding Joints

Friction stir welding(FSW) is an emerging solid-phase welding technology,and the Nondestructive testing(NDT) technology of the defects of the FSW joints is also at its initial stage.Ultrasonic testing has become an important means to inspect defects in the FSW joints due to its profound quantitative and location capability.Since there is limit on qualitative capability for ultrasonic testing,this issue has become one of the research challenge of ultrasonic testing.Taking the ultrasonic echo radio frequency(RF) signals of the aluminum clad penetration defect,channel defect and lack of penetration(LOP) in the FSW joints as research object in this paper,the time-domain and frequency-domain features of the defects echo signals in ultrasonic inspection were analysed.And then wavelet transformation was used to perform feature extraction and the defects classification performance was evaluated.At last,a artificial neural network(ANN) whose network inputs were the extracted feature vectors was created to recognize the defects type.The research result showed that the time-domain and frequency-domain waveform of the defects echo signals present obvious characteristics,which can be used for qualitative analysis of the defects.For the echo static waveform of the channel defect signal in ultrasonic testing,there are several consecutive wave peaks fluctuation within the wave width range.But for the other two kinds of the defects,there is only a obvious main wave peak within the wave width range. For the dynamic echo waveform of the defect by travering scan that is perpendicular to the weld line,aluminum clad penetration defect and LOP defect have similar characteristic, that is,as transducer moves away from the weld line,the amplitude of the defect echo wave goes up to a peak and remains unchanged at the peak for a while,then drops down into the valley.But the situation of the channel defect is different,the amplitude drops down first into the valley and then goes up gradually.These charactersitcs can be used to recognize channel defect. The average values of the principal frequency of the defect echo signal rank from high to low as follows: aluminum clad penetration defect defect,LOP,channel defect. Additionally, the difference between the figures of the power spectral density(PSD) of channel defect and the others is that there are several obvious wave peaks around the principal frequency.It's benefical to achieve the feature extraction of the defects signals with wavelet transform theory.Three feature extraction methods based on wavelet packet(WP) signal component node energy,WP node coefficients,wavelet decomposition of the PSD of the defects echo signal were used to extract the features of the three types of defects.To evaluate the classification performance of the feature extraction methods above by classification criteria based on Euclidean's distance,and the result showed that the feature extraction method based on wavelet decomposition of the PSD of the defects echo signal has the best classification performance.It's effective to achieve the defects of FSW joints recognition by ANN with extracted feature input vectors.A back propagation(BP) neural network whose inputs vectors were the features above was created and trained in this paper.The result of networks recognition experiments showed that the network whose input vector is the feature base on wavelet decomposition of the PSD of the defects echo signal has the best classification capability.The BP network's rate of the defects recognition is 85.71%,and rate of the LOP and channel defects recognition are both 100%,but the rate of the aluminum clad penetration defects recognition is just only 33.33%.