Study On Structural Damage Identification Based On Phase Space Reconstruction

Vibration-based damage detection technique has been the most widelyinvestigated technique. This method is advantageous because it imposesminimum effect on the operation of the structures and has the potential todevelop into a long-term real time monitoring technique. According towhether or not the feature extraction depends on theoretical model, thevibration-based technique can be divided into two categories, namely themodel-based method and the model-free method. In contrast to model-basedmethods, model-free methods don’t rely on a specific theoretical model, whichis either not available or not easy to establish, and hence they are potentiallyapplicable to complex structures. Damage-sensitive features for model-freemethods can be extracted from frequency domain, time domain a ndtime-frequency domain. This dissertation is to use time-domain data to exractdamage feature by incorporating nonlinear time series analysis, and to detectdamages in nonlinear structures by phase-space-based features. The maincontent is as follows.(1) The procedure of phase-space reconstruction using univariate timeseries is studied. The C-C approach, mutual information and false nearestneighbors algorithms are investigated to determine time delay and embeddingdimension for phase-space reconstruction. The noise effect on the determinationof these two parameters is studied, and de-noising is necessary before phasespace reconstruction.(2) The trajectory and geometrical feature of chaotic attractor areintroduced, and the geometrical feature of chaotic attractor is used to test thechaotic characteristic of measured time series. The effect of variable selection,length of data and reconstruction parameters on the computation of thegeometrical features of phase-space attractor is studied.(3) Attractor prediction error is employed as the damage-sensitive featureto identify damage in a three degree-of-freedom spring-mass system, for which both linear and nonlinear behaviors are considered. The results show thatattractor prediction error can indicate the presence of damage for both cases.(4) FEM model of a three-story frame structure is built and a nonlinearelement is introduced in the bottom story. Displacements at each story are usedto reconstruct the phase-space, and the attractor prediction error is employedagain to indicate possible damages. The results show that attractor predictionerror can not only indicate the presence of damage, but also qualitativelydescribe the extent of damage.(5) The existence of chaotic property of votex-induced vibration ofdeep-water risers is proved by testing the experimental data from MARINETEKCompany. Identification of riser damage is performed using the aforementionedmethod to the numerical simulation of riser response by Large-Eddy-Simulation.Damages are simulated by stiffness reduction. The results testify theapplicability of the method in marine riser damage detection.