Research On Key Technologies Of Automatic Ultrasonic TOFD Imaging Inspection For Weld Of Thin Material

Weldment of thin material is wildly applied in various industries and has great influence on the development of national economy, the construction of modern national defense and the progress of advanced technology, meanwhile, the non-destructive testing has been strongly demanded by weldment of thin material. Ultrasonic TOFD (Time of Flight Diffraction) imaging inspection technology has been widely used in the field of weld quality monitoring due to its unique advantages, and it is developing in the direction of automatic testing. But the small thickness of the thin material make ultrasonic TOFD testing signal prone to aliasing, and as the automatic application foundation of the tecnology, automatic imaging technology and automatic identification technology still have some technical barriers need to be solved. Therefore, the research on key technologies of automatic ultrasonic TOFD imaging inspection for weld of thin material are carried out in this dissertation, which is supported by National Natural Science Foundation-funded project. On the basis of analyzing the principle of ultrasonic TOFD imaging and its influence factors of testing reliability, a novel dconvolution technique based on adaptive sub-band Wiener filtering is developed, which can improve the longitudinal resolution of ultrasonic TOFD imaging. And a parametric model of ultrasonic TOFD testing image is established, which can get exact measurement of geometrical information for weld of thin material using the parameters optimization identification technology. Meanwhile, according to the common characteristics of the ultrasonic TOFD testing image such as low contrast resolution and fuzzy boundaries, the research on relative image processing technologies are carried out, an image segmentation algorithm based on morphology and watershed has been developed, which can extract the region of interested(ROI) in the testing image. Combing the spatial distribution and the neighborhood distribution features of the inspection image, the methods of gray level co-occurrence matrix and local binary pattern are used to extract multi-scale global and local texture features in the ROI, respectively. The features after Gaussian normalization fusion are applied to identify the type of weld defects. The above key technologies lay the foundation of automatic quality evaluation on the welds of thin material. The detailed contents and innovative points of this dissertation as below:In chapter one, the influence of thin material weldment on the development of national economy, the construction of modern national defense, the progress of advanced technology and the importance of research on automatic ultrasonic TOFD imaging inspection technologies are introduced. The research status and development trends of this technique are systematically summaried, its limitations are analyzed and the research direction is point out. The content and chapter arrangement of the dissertation are also given.In chapter two, the principle of ultrasonic TOFD inspection and the method of sizing defects are illustrated. According to the transmission laws of ultrasonic and the interaction mechanism of ultrasonic and defects, the finite element model of ultrasonic TOFD inspection is established based on the dynamic equation of isotropic solid medium, and the model is solved using central difference method that is suitable for computing wave propagation. Meanwhile, the typical defect detection signal is simulated, and main factors are analyzed to show their effect on the reliability and precision of ultrasonic TOFD inspection results.In chapter three, a novel deconvolution technique based on sub-band Wiener filtering is proposed, which can solve the problems existing in the procedure of ultrasonic TOFD imaging inspection such as the signals are easy to be superimposed and the deconvolution reference signal is difficult to choose. By this method, an original ultrasonic TOFD detection signal is decomposed into several sub-band signals using wavelet transform, and the sub-band signals of the lateral wave, the upper diffraction wave, the bottom diffraction wave and with strong causal relationship are selected adaptively according to the coherence function, then the deconvolution reference signal, and a new signal is reconstructed by the selected sub-band signals. After that, wiener filtering is applied to the reconstructed new signal, and the ultrasonic detection signal can be separated effectively. Meanwhile, the signals after separation are used to achieve high longitudinal resolution ultrasonic TOFD imaging, so the difficulty in applying ultrasonic TOFD inspection technology to weld of thin material can be solved.In chapter four, in order to overcome the problem that the geometry information can not be measure accurately due to the insufficient lateral resolution of ultrasonic TOFD inspection image caused by the large acoustic beam spread angle. Therefore, a high-precision geometric quantity characterization technique for weld defect based on ultrasonic TOFD image parameterized model is proposed. The parameterized model of ultrasonic TOFD inspection image is established by analyzing its characteristics, and the algebraic distance is used as optimization objective function, the nonlinear parameter estimation can be converted into linear matrix to solve the model, thus the location and the size of the weld defect can be obtained accurately, meanwhile, the efficiency of above defect parameters estimation can be improve greatly.In chapter five, according to the characteristics of ultrasonic TOFD testing image characteristic such as near surface defect is easy to be shaded by lateral wave, low contrast resolution, fuzzy boundaries, the research on relative image processing technologies are carried out. At the base of image de-noising and lateral wave straightening processing, energy distribution statistics algorithm is used to eliminate the influence for the aim-region, and the combined image segmentation algorithm based on morphology and watershed has been developed. Under the premise that inherits the ability of watershed image segmentation, the influence of local minimum values for segmentation results can be eliminated using mathematical morphology method, and thus the over-segmentation problem existing in traditional watershed algorithm can be overcome. Meanwhile, the contradiction between the efficiency segmentation and segmentation precision can be coordinated effectively. It has a perfect performance for regions of interested extraction of ultrasonic TOFD inspection image.In chapter six, in order to establish the technological base needed for automatic interpretation the results obtained by the ultrasonic TOFD imaging inspection, the research on defect type recognition technology based on the multi-scale ultrasonic TOFD imaging features has been carried out. After the wavelet multi-resolution analysis for ultrasonic TOFD inspection image, combing the spatial distribution and the neighborhood distribution features of the testing image, the methods of gray level co-occurrence matrix and local binary pattern is used for extracting feature in the ROI, respectively. The feature after Gaussian normalization fusion is applied to identify types of weld defect using pattern recognition technique, and the recognition accuracy is evaluated by experiments.In chapter seven, the research results and the innovative points of this dissertation were summarized, and the future research works were also forecast.