Research On Ultrasonic Identification Method Of Micro Defects Based On Signal Processing

Aeronautical important metal components such as turbine disks will produce some tiny defects in the manufacturing process.When ultrasonic testing is performed for such defects of tens to hundreds of microns,the echo signal of the defect is weak or has a similar amplitude to the noise signal,resulting in a small defect.The defect signal is polluted by the noise signal or even buried in the noise signal,and it is not easy to identify.Aiming at the problem of ultrasonic identification of tiny defects in plate-like components,a series of researches have been carried out using digital signal processing methods.The specific work content is as follows:Firstly,the composition of the ultrasonic pulse reflection signal of small defects is analyzed,the ultrasonic echo signal is mathematically modeled,and the pulse reflection signal of the small defect test block is collected by full wave train using ultrasonic feature scanning technology.Aiming at the composition characteristics of ultrasonic echo signals of small defects,a method of calculating signal correlation coefficient sequence is proposed to identify small defect signals.The simulation signal is obtained by the method of finite element analysis.After verifying the validity of the method with the simulation signal,the correlation coefficient sequence of the actual detection signal is calculated to realize the identification and location of the small defect signal.Then,according to the requirement of intelligent identification of small defects,a defect identification method based on neural network is studied.The spectral signal of a small number of data points on the small defect test block is input into the BP neural network as a training set for training.After evaluating the performance of the network by various indicators,the BP neural network is used to identify the detection signal of the test block,and the recognition accuracy of the network is verified by observing the imaging results of the identified test block.Finally,in order to eliminate imaging noise and enhance the imaging display of defects,a cluster analysis method is studied.The time-frequency diagram of the coda in the echo signal is selected as the characteristic parameter,and input into the K-means clustering algorithm for cluster analysis,which eliminates the misjudgment caused by noise,enhances the imaging display of small defects,and improves the imaging quality.In addition,a dedicated cluster analysis imaging human-computer interface for small defect test blocks is also designed and written.The results show that the signal processing method of calculating the correlation coefficient sequence can effectively identify the small defects in the echo signal,and the relative positioning error is basically below 10%,and the maximum positioning error is only 0.3mm;the small defect identification based on BP neural network The method can identify small defects in flat-bottom holes with a minimum size of φ0.1mm;the cluster analysis algorithm can effectively eliminate the interference of imaging noise and enhance the imaging display of small defects,which improves the image quality.