Research Of Feature Eaxtraction Technology And Damage Identification Method For Guided-wave-based Structural Health Monitoring

For simple and complicate structures and variational workingenvironment, guided-wave-based structural health monitoring is researchedsystemically. Effective feature extraction techniques and damageidentification methods are proposed, being more adaptive to the actualengineering practice.Some research conclusions and developing trend about guided-wave-based structural health monitoring, interaction with damage of individualmode of guided waves, signal processing techniques for extracting featuresof guided wave signals and damage identification methods are summarized.The advantages and disadvantages of them are discussed and the mainresearch contents, approaches and framework of this thesis are introduced.The interaction with notch and hole in a thin aluminum plate of thefundamental symmetric (S₀) mode and the fundamental anti-symmetric (A₀) mode which are activated at low frequency is researched using dynamicfinite element method. The displacement of node is decomposed along andvertical to the direction of damage-scattered wave propagation usingcylindrical coordinate system. The radial in-plane displacement (Uh), theout-plane displacement (Uz) and the in-plane displacement vertical to theradial (Uv) are used to describe the damage-scattered S₀mode, A₀mode andfundamental shear (SH₀) mode, respectively. Their energy distributions atthe sensor nodes with different angles are calibrated in wave field around thedamage. Results indicate that the damage-scattered energy in some damagecases does not vary monotonically with the monotonic variation in thedamage dimension, so the damage size can not be deduced only dependedon the damage-scattered energy. It is concluded that when the S₀modeinteracts with the through-thickness notch or hole, not only the S₀mode isscattered by the damage but also the SH₀mode which is mode-converted bythe S₀mode appears. When the A₀mode interacts with the through-thicknessnotch or hole, only the A₀mode is scattered by the damage. For the littlefrequency-thickness product (less than1MHz·mm), the S₀mode is moreadaptive for damage identification using active sensor network. Research ofthe interaction with damage of Lamb waves is helpful for the furtherresearch of feature extraction technology and damage identification method. A cylindrical model is proposed and a corresponding damageidentification method is promoted to realize the guided-wave-based healthmonitoring of revolving cylindrical structure. Reflection coefficient andresidual coefficient of the captured longitudinal (L) wave in the proposedcylindrical model are defined and calibrated corresponding to differentdamage cases based on dynamic finite element analysis. Simulation resultsdemonstrate that the residual coefficient can be utilized as feature parameterto interpret different damage cases. The promoted damage identificationmethod is validated experimentally. A circular thin aluminum plate with twocentral circular piezoelectric transducers (PZTs) is used to model a radialslice of the proposed cylindrical model. The first intrinsic mode functions(IMFs), extracted from the captured wave signals as signal features, arecompared between the benchmark (without damage) and damage case (withdamage) to calibrate the residual coefficient of the S₀mode. Experimentalresults consist with the simulation results, indicating the efficiency of thepromoted damage identification method and building foundation for thefuture research about detecting revolving cylindrical structure.Aiming on extracting effective singal feature for damage identificationin the working environment with broad band noise, a split spectrumprocessing (SSP) algorithm is proposed. The time of flight (ToF) of the damage-scattered S₀mode is estimated by comparing the instantaneousamplitude variation degree (IAVD) of the captured wave signal frombenchmark with that from a damage case, so as to realize theguided-wave-based damage identification. Wave signals captured from anintact aluminum plate (benchmark) are firstly acquired in ideal environment(without noise). Then a notch is artificially introduced into the aluminumplate and wave signals captured from the notched aluminum plate (damagecase) are acquired in different environments (without and with noise).Experimental results demonstrate that even though the captured wave signalfrom the damage case has a low signal-to-noise ratio (SNR), the IAVD ofthe captured wave signal extracted using the SSP algorithm is almost notinfluenced by the noise. As result, the proposed SSP algorithm is capable ofestimating the ToF of the damage-scattered S₀mode with high precision,regardless of the broad band noise disturbance, leading to the successfulidentification of the notch in the aluminum plate using triangulationalgorithm.With the objective of avoiding damage identification error, which iscaused by time unsynchronization of activating and sampling signals and theactivated signal differences between benchmark and damage cases, amethod of calibrating damage index (DI) based on the correlation of signal energy spectrum is promoted for the complex structure. Two schemes aredeveloped to calibrate DI and the calibrated DI is utilized as featureparameter to represent damage information. Scheme I: for a sensing path,analyze the correlation of energy spectrums of captured wave signalsbetween benchmark and damage cases, and the correlation coefficient istaken as the calibrated DI. Scheme II: for a sensing path, calibrate thecorrelation coefficient of energy spectrums of the activated wave signal andcaptured wave signal, and the variation of correlation coefficient for damagecase relative to that for benchmark is taken as the calibrated DI. Inexperiment, integrate the calibrated DI using scheme I and scheme II with adamage diagnostic imaging algorithm, respectively, to estimate theprobability of damage presence, locating a cone hole in a composite platewith stiffener. With a weight distribution function, the calibrated DI valuesfor individual sensing paths are mapped to individual grids in the inspectedarea covered by a senor network, constructing a probability image for thepresence of damage. Results demonstrate that the promoted method iscapable to avoid the influence of the time synchronization and scheme II iscompetent for identifying damage in the complex strucre more precisely,avoiding the error induced by the activated signal differences betweenbenchmark and damage cases. To realize damage identification without referring benchmark andidentify multi-damage with numbered sensing pathes in a sensor network,time-reversal Lamb waves combined with a damage diagnostic imagingalgorithm is developed. The distortion between the original activated toneburst and the reconstructed waveform is analyzed to calibrate atime-reversal-based DI which is utilized as feature parameter to representdamage information. The time-reversibility of Lamb waves and theefficiency for identifying damage of the time-reversal-based DI isinvestigated and evaluated in aluminum plates without and withthrough-thickness notches (with different lengths) based on dynamic finiteelement method. An experimental evaluation is conducted for the developeddamage identification method at the different environmental temperatures.The successful locations of dual notches in an aluminum plate demonstratethat the developed method is not only benchmark-free, but also competentfor identifying multi-damage with numbered sensing pathes.In this thesis, reasearches of feature extraction techniques and damageidentification methods with and without referring benchmark for simple andcomplicate structures and variational working environment build afoundation for abroad application of guided-wave-based structural healthmonitoring in actual engineering practice.