Quantitative Evaluation Of Surface Defects Based On Laser Ultrasound And Its Application

Type identification of surface defects and near-surface defects is a fundamental problem in quantitative laser ultrasonic non-destructive testing.In this area of work,the use of ultrasound imaging to reconstruct the geometry of multiple defects allows multiple types of defects to be detected,however,this method requires complex off-line data processing for long-term global ultrasound scanning and data synthesis stage,so the real-time method does not meet the current high-speed,online testing needs of the industry.However,using the feature extraction algorithm,the time-frequency feature is extracted directly from the signal.Because it does not need to wait for global ultrasound scanning and data synthesis,the non-imaging detection method can be used for timely detection,which is suitable for the requirement of non-stop detection in production practice:a)The theory of laser excited ultrasonic surface wave is studied.By using three-dimensional finite element laser point source,thermoelastic effect model,by rail as an example,numerical simulation of the interaction of different crack depth and surface wave,analyzed the spectral characteristics of different depth of the defect echo,and summarized the relationship between defect depth and echo amplitude and phase;b)The theory of wavelet packet soft threshold denoising is studied.Based on laser ultrasonic simulation signal,wavelet basis,decomposition level and threshold function are selected based on SNR,correlation coefficient and root mean square error respectively.A denoising scheme for wideband and non-stationary defect echo is selected;c)A feature fusion algorithm based on wavelet packet transform for defect ultrasonic signals in time and frequency domain is proposed,and the characteristics of wavelet packet energy and wavelet packet local entropy are visualized.The wavelet packet fusion feature has good separability,and its separability measure value reaches 9.84,which is higher than other characteristics;d)The identification method of linear discriminant analysis and nonlinear separability measure based on support vector machine,performance evaluation carried out by rail surface defects as an example of wavelet packet fusion feature,the support vector machine 50 percent off cross validation accuracy 98.7354%,demonstrates the excellent ability of wavelet packet fusion feature signal time-frequency information mining.