Modal Parameter Identification Of Large-scale Civil Engineering Structures Based On Ambient Excitation

During the past few decades, along with the growth of China’s economic development,infrastructure construction has developed rapidly and a lot of long-span bridges andlarge-scale space structures have been built. These large-scale civil engineering structures areprone to be damaged during their service lives, caused by factors such as complex conditionloads, fatigue effect, corrosion effect and material aging. So structural health monitoring(SHM) is very necessary. The ambient excitation method to identify large-scale structuremodal parameter is widely used for structural health monitoring. Different from traditionalone, modal parameter can be identified from output-only data in ambient excitation method.But due to noise interference, inadequately excitation or uncertain environmental conditions(for example, temperature, humidity, wind loads and traffic loads), it would meet someproblems in ambient excitation method, such as low Signal to Noise Ratio (SNR), modalparameter changes under different environmental conditions. Because of these problems,algorithms for SHM may work in numerical or laboratory model, but difficult to be applied topractical engineering. This dissertation focus on the core problems of output-only modalparameter identification and the Yamen Bridge, Dukeng Bridge and GNTVT Benchmark aretaken as engineering background. The main work and conclusions include as follows:(1)An adaptive noise reduction method based on Genetic Algorithm (GA) and SingularValue Decomposition (SVD). The row number of reconstruction matrix (p) and the order ofeffective rank(k) both are difficult to determine for noise reduction based on singular valuedecomposition. An improved adaptive noise reduction method is proposed, using theoptimization function of GA, the two key parameters can be determined adaptively. Twonumerical simulation signals with different frequency components are employed. The resultsshow that can beN4orN3(N is the length of data), is twice as the number ofdominating frequency. For measured signal, the frequency components are more complicatedin order not to miss the true frequency components, when dealing with measured signals,should be more than twice as the number of dominating frequency, but can still beor.(2)Investigation of the key parameters in NExT/ERA algorithm. The number ofextraction point for fast Fourier transform(NFFT), the length of data used in modaldentification and the dimension of Hanke matrix are the three key parameters in NExT/ERA algorithm. The three key parameters are investigated through the output-only data of twobridges and the results show that:①NFFT shouldn’t be too large, or it will reduce the timesof average,1024or2048is suitable;②the length of data shouldn’t be too less, it should atleast longer than60times as the first natural vibration period (T1) of the structure;③aempirical equation is proposed for determination of the demension of Hankel matrix and theequation is verified by the output-only data of two bridges.(3)Several mode accuracy indicators are introduced. Though numerical simulations, theConsistent-Mode Indicator for Observability matrix (CMI_O) is investigated and verified, andthe suitable range of value for distinguishing the physically true modes from the spuriousmodes is proposed.(4)For Operational modal analysis, the key issue is how to identify the weakly excitedmodes and distinguish the physically true modes from the spurious modes. An improvedmultiple reference DOFs stabilization diagram algorithm based on NExT/ERA(M-NExT/ERA) is presented. By setting different reference DOFs in each group of data,NExT/ERA is used to identify modal parameters. Damping ratio, CMI_O and ModalAmplitude Coherence (MAC) was used as threshold to identify the most accuracy modalparameters. A numerical simulation of8DOFs shear model is employed, the results show thatM-NExT/ERA is reliable to identify modal parameter accurately under the conditions ofnon-white noise excitaion and strong noise interference.(5)Development and engineering application of MIDP toolbox for modal parameteridentification. Base on MATLAB, a modal parameters identification toolbox named "MIDP"is developed. Data loading, data pre-process, modal identification and data post-processfunctions are involed in MIDP. MIDP and ARTeMIS are bothe used to identify the modalparameter of Z24bridge(Continious bridge) in Swiss.The results show that the modalparameters identified from the two methods are in good agreement and MIDP is reliable. Atlast, MIDP is applied to modal identification of Yamen Bridge(Cable-stayed bridge) andDukeng Bridge(Continious-Arch ccomposite bridge).(6)Variability analysis in dynamic properties of high-rise structure under differentenviromental conditions. Basded on the measured data of GNTVT Benchmark, the influencelaw of dynamic properties caused by different environmental conditions is analysed. Theresults show that modal frequencies decrease as the temperature rise, but it is hard to observethe trend of damping ratio. Temperature to frequency influence mechanism of GNTVT isinvestigated through the updated ANSYS model. The results show that both the material property and internal force state of GNTVT will change along with different temperature.Modal frequency decrease as temperature rise, and the change of material property is the mainreason of the change of modal frequency.(7)NLPCA-SVR model between envriomental factors and modal frequency.NLPCA-SVR mothod is proposed, NLPCA is first applied to eliminate the correlation amongenvriomental factors and extract principal components from the envriomental factors fordimensionality reduction. The predominant feature vectors in conjunction with the measuredmodal frequencies are then fed into a support vector algorithm to formulate regression models.Grid serach method (GSM), Genetic Algorithm (GA) and Flying fruit optimization algorithm(FOA) are used to determine the hyper-parameters of the SVR models.The results show thatNLPCA-SVR model is better than OriginalData-SVR model, it can accurately fit and predictthe changes of frequencies along with temperature changes.