Research On Defect Feature Extraction And Intelligent Recongnition Of Copper And Steel Re-melt Deposit Welded Joints Based On Ultrasonic Testing

As the defect identification results of ultrasonic testing are lacking of accuracy and reliability, the defect identification of copper and steel re-melt deposit welded joints were regarded as the research subject and the feature extraction of the defect waveform, the separability comparison for defect kinds, the methods of intelligent identification and the comprehensive evaluation of recognition results were analyzed in this article.Mechanical processing was used to make different reference blocks and the natural defects were made by changing the welding process parameters. Manual ultrasonic testing, radiation testing, automatic ultrasonic testing and destructive testing certification (such as cutting, turning etc.) were used to identify and collect the waveform signals of different defect samples.The defect waveform singals were intercepted and normalized to prevent varies of disturbance factors. The Daubechies8 wavelet packet decomposition at three levels was applied to time-frequency feature extraction as waveform Bx, peak Bf, energy distribution Ef, energy percentage E of the defect waveform signals.Euclidean distance was used to analysis the separability comparison for time-frequency feature which was better than the time and frequency independently feature of defects with different kinds.Artificial neural network (ANN) was applied to identify the common defects as gas pore, slag, shrinkage porosity and incomplete fusion of copper and steel re-melt deposit welded joints. The selection of defect samples and network structure was optimum designed. The BP neural network (NN) ensembles were used to improve the generalization ability of the single NN, and the results show that neural network ensemble composed of three single BP with improved BP algorithm Scaled Conjugate Gradient (SCG) is more feasible than the single BP NN with algorithm SCG and Levenberg-Marquardt (L-M) for the identification of different defects and its recognition accuracy of each kind and the synthesis reach to 95% and 96.25% respectively which is satisfied the realistic applications.Intelligent ultrasonic testing system of copper and steel re-melt deposit welding was established based on the Matlab Graphical User Interface (GUI).The system is composed of three parts:data preprocessing for feature extraction, data management for data query and intelligent identification for off-line identification of the defect waveforms, and it can realize the qualitative and quantitative analysis of the defects with ultrasonic C-scan testing.