Recognition Of Welding Crack Based On Acoustic Emission Signal Characteristics

Welding crack is the crevice formed by the breakage of metal atom bonding force in the local area of welding joint.As the most harmful defect in welding parts,the structure quality and service safety are affected by the crack.In order to effectively restrain welding crack and improve the quality of welding structure,it is necessary to identify the crack accurately.The stress wave produced by the change of crack state in welding process is a transient,sudden and unrepeatable acoustic emission(AE)signal.It can characterize the physical phenomena of welding crack generation and propagation in real time.Therefore,an identification method of welding crack based on AE signal characteristics is propoesed using AE testing technique.Characteristic information of AE signal that can quantitatively characterize the welding crack state is acquired to realize identification of welding crack state.It has important engineering application value and theoretical research significance to provide prior information and technical means for restraining welding crack and guaranteeing welding structure quality.In this paper,AE testing technique is used to achieve welding crack detection.An identification method of welding crack based on AE signal characteristics is studied.Based on the acquisition of crack AE signal in welding process,the decomposition method,the feature construction method of AE signal and the crack state identification method are studied.It provids basic theory and key technology for quantitative extraction of AE signal characteristic information and accurate identification of crack state.The main research contents are arranged as follows:(1)Based on the theory of synchrosqueezed wavelet transform(SST)algorithm,the method of SST decomposition based on frequency characteristics is studied.The signal energy distribution in the time-scale plane obtained by wavelet transform is reconstructed into the time-frequency plane.According to the distribution characteristics of instantaneous frequency,the coefficients in the frequency band are extracted and reconstructed.The AE signal can be decomposed into the mode of the corresponding frequency band.The method is applied to the AE signal of hybrid excitation source in welding process under complex noise background.The crack AE signal has been extracted effectively.(2)The characteristic construction method of welding crack AE signal is studied.The time-frequency transformation is performed to the characteristic components obtained by SST.The principal component analysis is also used to reconstruct the three-dimensional time-frequency data into covariance matrix.According to the cumulative contribution rate of variance,the principal component is selected to be reconstructed as the enhanced signal.The approximate entropy algorithm is introduced to the reconstructed signal.The approximate entropy eigenvector of the AE signal is constructed,which can quantitatively describe and characterize the crack.The feature construction method is applied to the AE signal of welding crack,of which the feature vector can be used as a numerical index for quantitative identification.(3)Considering the problems of small sample,state evolution,multi information input and nonlinearity in welding crack identification.An identification model of welding crack based on HMM and SVM is proposed to achieve crack identification.Firstly,the HMM model is screened to get the two possible states which are closest to the crack evolution process,and then the final recognition result is obtained by SVM classifier.The model not only makes full use of the state transition and dependency ability of the HMM model,but also makes full use of the strong binary classification ability of SVM under the condition of small sample.(4)AE signal testing experiment during welding tensile process and crack AE signal testing experiment during welding cooling process are designed.The characteristic frequency range of AE signal produced by welding part deformation and fracture state are obtained by tensile test.According to the characteristic frequency band of signal,the AE signal generated during the evolution of welding cold crack state are decomposed,constructed and identified.The validity of the identification method of welding crack is verified.