Dissertation
Dissertation > Industrial Technology > Automation technology,computer technology > Computing technology,computer technology > Computer applications > Information processing (information processing) > Pattern Recognition and devices > Image recognition device

Research on Characterization of Weld Defect Based on Ultrasonic TOFD

Author ChiDaZuo
Tutor GangTie
School Harbin Institute of Technology
Course Materials Processing Engineering
Keywords non-destructive characterization ultrasonic TOFD weld defect signal processing image processing
CLC TP391.41
Type PhD thesis
Year 2007
Downloads 631
Quotes 12
Download Dissertation

It is necessary to non-destructively characterize weld defect in evaluating the reliability, security and residual life of large components. Although it had already been proved the ideal detection tool in thick wall weld testing, ultrasonic time of flight diffraction (TOFD) has technological limitations. Due to the influence of imaging mode, testing system and acoustic characteristic of the material under testing, characteristic information of the defect is often obscure in the raw TOFD image. So, accurate quantitative data cannot be obtained immediately from the original testing results. In this paper, forward problem of ultrasonic TOFD imaging is studied through the behavior analysis of ultrasound propagation and action on the defect. Based on this, inverse problem of non-destructive characterization is studied by using proper signal and image processing technique, and then the accuracy of weld defect characterization is improved.For forward problem, through the description of ultrasound propagation behavior, sound pressure field in the specimen is calculated based on Multi-Gaussian Beams theory. Meanwhile, in order to characterize the acoustoelectric elements of the ultrasonic inspection system, system efficiency factor is determined by performing a reference experiment. Based on these, inspective models for several typical defects are established by employing Kirchhoff approximation theory. By using the models, ultrasonic TOFD A-scan lines are simulated and D-, B-scan foreground images are synthesized. The experimental results show that the simulated dada are in better accordance with the measured ones. Forward synthesis can do help to optimize the testing parameter and solve inverse problem of non-destructive characterization.For inverse problems, the suppression of background clutter is studied firstly. According to the feature of D-scan image, a clutter suppression method based on statistic energy distribution of the image is presented. The experimental result shows that the method operates rapidly for its simple algorithm, but the time base flutter influences on it greatly. In order to improve the adaptability of clutter suppression to cater for the tested image under field engineering situation, recursive least-square (RLS) adaptive filter is adopted, by which clutter wave with fluctuation can be removed robustly. The non-destructive testing result shows that the signal of near surface defect mixed with lateral wave can be separated effectively by RLS adaptive filter and the inspective range is enlarged.Based on preprocessing of clutter removal, noise suppressing for D-scan image is studied. According to the strong correlation of defect signals in adjacent A-scan lines, a modified wavelet based algorithm is presented. The presented algorithm employs the merits of both multi-resolution of wavelet and cross-correlation of adjacent signals. By using the modified algorithm, the traditional wavelet based denoising methods including wavelet modular maximum (WMM) and nonlinear wavelet shrinkage (NWS) are improved. The experimental results show that processed by improved WMM method, the isolate and irregular noise points can be suppressed effectively and singular point representing the defect can be detected easily and accurately in the decomposed maximum modulus sequences. The improved NWS denoising method is very robust and unsusceptible to the mother wavelets, threshold modes and estimation methods, and has greater adaptability for complex noise. The improved methods are superior to the traditional ones in noise suppression and defect signal recovery. By using the presented methods, information of buried depth of the defect in the weld can be obtained definitely.In order to obtain lateral location and more accurate buried depth information of the defect in the weld, enhancement of B-scan image is studied. According to the dynamic relation between the probes and defect tip in the course of testing, B-scan imaging model and synthetic aperture focusing (SAF) algorithm model for image reconstruction are developed. For the reason of frequency band and noise, resolution shows contradict to signal-to-noise ratio in the reconstructed image. To solve the problem, an improved Wiener filter technique acting as preprocessing is presented. The results show that time resolution of the image is enhanced through minimizing the duration of ultrasonic impulse by Wiener filter and amplitude capture. The lateral resolution of the image is enhanced greatly by SAF. So, measurement can be taken on reconstructed image rapidly and accurately.Defect extraction from D-scan image is studied based on preprocessing of clutter and noise suppression. Traditional threshold based segmentation methods are somewhat ineffective for D-scan image, in which gray difference is often ambiguous and the gray level of target is often nonuniform. The experiments show that entropy based segmentation method is suitable for the image, and local threshold is effective in order to avoid under-segmentation of small targets which are composed by a small amount of pixels. In order to obtain sub-images, an automatic block partition method based on statistic energy distribution of the image is presented. The postprocessing of mathematical morphological closing and opening operation can fill the gaps and delete speckles in the binary image. In the ultimately processed image, defect can be detected automatically.

Related Dissertations
More Dissertations