Research On Ultrasonic Defect Detection Method Of Metal Components Based On Machine Learning

Metal components are widely used in aerospace,railway,military and other fields.Nondestructive testing of metal components is necessary that defects will affect the performance and service life of metal components.Ultrasonic nondestructive testing technology is extensively used in defect detection of metal components.However,there are some interference factors such as surface echo oscillation and ultrasonic scattering,which make it difficult to detect the actual position and size of the defect.Based on the above questions,this paper carries out the research on nondestructive testing of the defects in metal components.The ultrasonic pulse echo technique can be used for the defect detection.However,the "dead zone" created by the front-wall echo may overlap the defect echo,which interferes the detection of near-surface defect.Based on the difference of the wave characteristics of the defect echo signal,this paper proposed a method to identify the primary echo and the secondary echo of the defect through Support Vector Machine(SVM),and the accurate calculation of near surface defect position could be achieved.Firstly,the simulation experiments verified the ability of SVM to identify the primary and secondary defects echo,and then the verification experiment was conducted on the near surface defects of the 440 C stainless steel bearing inner ring.The experimental results showed that the proposed method can effectively identify the primary and secondary echoes of defects,and detect the actual position of defects.The average accuracy of classification reached 95%,and the prediction error of near surface defect position was less than 0.2mm.Ultrasonic nondestructive testing technology can evaluate the defect size of metal components,but it is difficult to detect the actual size of the defect from the ultrasonic C-scan image due to the influence of the transverse resolution of the probe.In this paper,a new method to determine the actual size of defects was proposed by combining Gradient Boosting Decision Tree(GBDT)and SVM.Firstly,the maximum amplitude of defect echo signal of 304 stainless steel component was obtained by the extreme value theory,and then the descending gradients k was set to characterize the defect size.Secondly,GBDT was used to extract features of defect signals and non-defect signals under different gradients,and the extracted features were used to classify the signals by SVM.Finally,the quantitative detection of the actual size of the defect was achieved by analyzing the classification accuracy of different descending gradients.The experimental results revealed that the proposed method could accurately estimate the actual size of the defect,and the detection error was within 10% with the size of the sound beam larger than the size of the defect.