Research On Ultrasonic Guided Waves Based Structural Damage Identification Method Considering Uncertainty

Ultrasonic guided wave has been widely used in the non-destructive inspection of steel bridges,pipelines,aircraft wings,wind turbine blades and other structures due to its excellent advantages such as low energy attenuation,large detection area,and sensitivity to small damage.Ultrasonic guided wave can detect small damage and its development,which is an important part of a full-scale structural health monitoring system.However,a large amount of uncertainty exists in the guided wave-based nondestructive testing and structure damage identification,resulting in inaccurate or even wrong results.Therefore,the research on the guided wave-based nondestructive testing considering the uncertainty is of great significance to improve the accuracy of structural damage identification.The purpose of this paper is to apply uncertainty methods to the guided wavebased non-destructive testing,further complete this technique,and explore new uncertainty strategies that can be used for guided wave signal feature extraction and damage identification.To this end,this thesis focuses on the uncertainty in ultrasonic guided wave signal processing and damage identification,and studies the guided wave dispersion characteristics,feature extraction of complicated wave packets,multi-damage localization and damage imaging,respectively.The details are as follows:(1)Based on the guided wave propagation model,a comprehensive wavenumber analysis of the received complex frequency response is conducted,which reveals the relationship of the dispersion characteristics and the propagation distance with the complex frequency response,and then a complex Sparse Bayesian Learning(SBL)method for extracting the dispersion characteristics of guided wave is proposed.The proposed method utilizes the sparse characteristic of the dispersion curves relative to the whole wavenumberfrequency domain to perform sparse Bayesian learning and then estimates all dispersion curves from the received signals.The effectiveness of this method is verified by numerical simulations and experimental studies.(2)Due to the disability of conventional time-frequency analysis in feature extraction of complicated wave packet,such as mode-converted packet,dispersive packet,and overlapping packet,a sparse Bayesian learning method to extract the propagation distance of complicated signal is proposed.Based on the sloved dispersion characteristics and the propagation distance,an over-complete dictionary matrix considering multimode,mode conversion and dispersion is established,and then the time-domain signal is sparsely decomposed to obtain the propagation distance of every complicated wave packet.The results of aluminum plate detection show that this method not only extracts the propagation distance of each mode in the mode-converted wave packet but also separates overlapping wave packets.(3)Due to the excessive uncertain overlapping areas obtained by the triangulation method,it is difficult to obtain the accurate damage locations,and then a multivariate SBL-based multi-damage localization method for plate-like structures with known propagation velocity is proposed.According to the sparse characteristic of the occurred damages relative to the whole structure,a vector constituted by the propagation times from each possible damage location to all sensors is established,and then all vectors are combined to form an over-complete dictionary matrix.Multivariate SBL is performed on the propagation times extracted from all received signals to obtain multiple damage locations.By sharing the propagation velocity and error parameter,this method fuses all propagation times to achieve the purpose of reducing the uncertainty of multidamage localization.(4)For multi-damage localization with unknown propagation velocity,a Gibbs sampling method is proposed to identify propagation velocity and multiple damage location and quantify their uncertainty.According to the parameter linearization model of the triangulation method,the posterior probability distribution function(PDF)of each unknown parameter is derived by Bayes theorem,and then the joint posterior PDF of all parameters is approximately obtained by the Gibbs sampling.This method can not only identify multiple damage locations and propagation velocity,but also quantify their uncertainty,thereby improving the accuracy of damage assessment.(5)Due to the low resolution and contrast of conventional dalay-and-sum imaging when identifying the cracks in the top plate of orthotropic steel bridge deck,a variational Bayesian principal component analysis based dalay-and-sum imaging is proposed.Through Bayesian principal component analysis,the scattered wave packets are extracted from the delayed signals and at the same time,the interference of noise is suppressed to improve the imaging resolution.The good performance of the proposed method is demonstrated by numerical signals and experimental detection signals.