Multi-model Updating Of Finite Element Models For Bridge Structures

Structural model updating is to obtain a refnined model with sufficientaccuracy by updating a coarse model using the measured structural responses.However, under the influence of numerous uncontrollable factors, the structuralresponses and identified modal parameters present themselves with uncertainty, andthe unique updated models obtained by currently available methods are not real inpractice. In this thesis, the factors that influence the modal analysis are studied. Bytesing a bridge model with a vast amount of test settings, the certainty of measuredmodal data is quantified. An efficient modal updating procedure is proposed, andmulti-model updating of bridge structures is achieved by incorporating the modalparameter variation. The main content of this thesis is as follows:Typical time-domain and frequency-domain modal parameter identificationmethods, i.e., Peak Pick method, Frequency Domain Decomposition method,Stochastic Subspace Identification method, and Eigen-system RealizationAlgorithm method are investigated to estimate the effect of different parametersettings in these methods on the modal parameter identification. The accuracy androbustness to noises of these methods are compared. Numerical studies show thatPP and FDD method miss some frequencies at the10-15%noise level, but for SSIand ERA method, that happens only for30-40%or higher noise level.A total of216group and2160modal tests are conducted on the BinzhouYellow River highway bridge model by varying the excitation pattern, sensorlocation, sensor type, number of sensor and modal parameter identificationalgorithm, and the variation of the lower5modal frequencies is obtained.Experiments show that excitation pattern, sensor location and modal parameteridentification algorithm have significant influence on the modal parameteridentification, and the other two have minor influence. The modal frequenciesvariation is significant, and they don’t follow the six distributions considered inthis paper.The model updating method based on response surface method is investigated.Response surface functions are constructed using orthogonal design, centralcomposite design, and uniform distribution design combined with polynomial function, radical basis function and BP neural network. It shows that the centralcomposite design with polynomial function delivers the best accuracy. Multi-modelupdating is conducted to the finite element models of the Binzhou Yellow Riverbridge and Ningbo Yongjiangte bridge, and the updated models with statisticalcharacteristics is obtained. The multi-model updating method proposed in thispaper is fast and effective in implementation, which is a good choice to the casewhen multi models are needed to preserve their variation.