Flaw Detection And Sparse Signal Reconstruction For Ultrasonic Nondestructive Testing

Ultrasonic nondestructive testing(NDT)techniques have been widely used for damage detection in structures such as buildings,bridges,pipelines,etc.With the help of ultrasonic NDT methods,we can obtain valuable information about whether damages exist in structures,pinpoint their localizations if there is any,and even quantitatively measure their size,making in-time structural evaluation more easily.Compared to global structural health monitoring(SHM)and some localized SHM techniques,ultrasonic NDT can detect small defects or flaws in structres more effectively and have been a sigficant part of morden SHM systems.The includion of ultrasonic NDT into SHM systems is tremendously valuable,making the full-scale(from global to local)monitoring of structures possible,and will improve the accuracy of structural safety evaluation.Although ultrasonic NDT techniques have been vastly developed,their applications to real structures are somehow limited,partly due to incompetence of the required signal processing techniques.One of the many common challenges is noises in NDT signals,particularly the structure noise often encountered during the testing of many coarse-grained materials.Another issue is with GW inspection,in which signals are dispersive and multi-mode.Furthermore,mode conversion and mode-overlapping impose even more challenges.In view of these problems,several signal processing techniques have been proposed and investigated in this thesis,with focus on the pulse-echo NDT technique.The detailed research effects are highlighted as follows:(1)From a mathematical point of view,the models for flaw signal as well as noise were analyzed in details.Due to different mechanism of scattering,the differences in spectral characteristics between flaw signal and noise were highlighted.In the meantime,the NDT signal model also shows its connection to the concept of sparse signal representation(SSR).One SSR technique called Robust Sparse Bayesian learning(RSBL),which will be adopted in the following chapters,is introduced in detail.(2)Regarding to the troublesome structure noise seen in many bul k-wave NDT applications,a signal processing method using template matching concept was proposed,based on the different spectral characteristics of flaw echo and structure noise,under the assumption that Rayleigh scattering rule is valid.The known spectral expression of flaw echo was used as the template,thus flaw detection can be realized by calculating the matching coefficient between the template and local spectrum of the signal.This method has been quantitatively investigated by analyzing synthetic ultrasonic signals and validated by experimental study.(3)Based on the concept of SSR,a method for signal de-noising and flaw signal reconstruction RSBL was proposed in Bayesian framework for bulk-wave NDT,including the optimized dictionary design,the sparse decomposition of signal and post-processing schemes for flaw signal reconstruction.Results from numerical study shown that the performance of the proposed method was good even for signals with low signal-to-noise(SNR)ratio.Additionally,a comparative study between the proposed method and another two SSR methods for noisy signal processing was conducted.Results confirmed that the proposed method has superior performance.(4)The proposed Bayesian method was extended to the processing of narrowband GW signal,which are often the case for GW inspection using PVDF sensors.Since narrowband GW signal shows insignificant dispersion,it is viable to use Gabor pulse model to approximate GW echoes.Based on the different propagation patterns of different GW modes,a technique using graphical representation for GW mode identification as well as flaw localization was also proposed.Results from numerical investigation and experimental study showed that GW signal can be accurately reconstructed even with strong noise.Flaw localization results were also of high precision.(5)For dispersive multimode GW signal,a signal processing method using Chirp model was developed for damage detection and localization.Starting from the consistence between signal dispersion and Chirp model,the design of over-complete chirp dictionary was introduced and the procedures for GW mode identification as well as flaw localization were presented.The accuracy of the signal reconstruction was quantitatively studied by numerical simulations.Based on the merits of this method,the feasibility of using 3 sensors for GW damage localization was demonstrated by an experimental study.