Structural Damage Identification And Reliability Evaluation Based On Improved Bayesian Method And Non-probabilistic Analysis

Damage identification and reliability evaluation consist of the core of structural health monitoring of civil structures.Presently the main challenge comes from uncertainties inherently existing in such structures.Uncertainties include the randomness or fuzziness of material and geometric parameters,modeling errors and testing noises etc.Due to it,based on the theory of Bayesian classification and the ARX model,this thesis develops two damage identification methods using time series data.The methods adopt the regression coefficients of the ARX models as sensitive features for structural damage,which can be regarded as a kind of inherent vibrational characteristics of structures and thus used for identifying damage patterns.Meanwhile,considering the influence of data fuzziness and randomness,a hybrid reliability evaluation method has been proposed to predict the reliability of a structure,in which interval arithmetic is also employed.The main contents and achievements of this thesis are described as follows:Firstly,this thesis chooses discrete data such as static displacements,modal frequency and modal assurance criterion as sensitive features for damage location.After that,a damage identification method has been developed using naive Bayesian classifier.In order to verify the rationality of the conditional independence assumption of the naive Bayesian theory,TAN classifier is also employed to determine reasonable weights for different damage features.By this means the conditional independence assumption is loosened but the TAN classification results are still identical to those given by the naive Bayesian classifier.Subsequently,damage existing in two numerical and experimental steel beams is successfully located in the interest of proving the feasibility of the proposed method.The analysis results demonstrate that the conditional independence assumption of the naive Bayesian classifier possesses satisfactory applicability.Secondly,with the aid of time series models,an improved naive Bayesian damage identification method has been given using continuous data.A five-floor planar steel frame is adopted for validating the method.Simulated seismic waves are used to excite the frame in order to acquire its acceleration responses for the establishment of a database for the dynamic responses.Subsequently the time series models are constructed and a classification database is also established on the basis of regression coefficients.Then a naive Bayesian classifier framework is utilized to achieve the identification of frame damage.Meanwhile,random vibration testing is applied to an experimental steel frame model.Random excitations are used as the inputs in order to measure the acceleration responses at the frame joints.The damage at the joints is identified using the naive Bayesian classifier,which further proves the feasibility of the proposed method.Thirdly,a probability design module based on finite element analysis is used to calculate the reliability of a numerical plane steel frame.Sensitivity analysis has also been implemented during the process.Then the random variables having considerable influence on the frame reliability are chosen to construct response surface equations,which are combined with the first-order second-moment method to evaluate the frame reliability.The results are also used as a reference for the following hybrid reliability evaluation procedure.Finally,based on the assumption of data fuzziness,the Bayesian interval estimation is carried out to obtain the fuzzy intervals of random variables.At the same time,the failure criterion is taken into account in order to propose a new hybrid reliability model simultaneously considering probability,fuzziness and non-probability properties of parameters.The hybrid model is adopted for estimating the intervals of the failure probabilities of a plane steel frame.The results have been compared with those obtained by the traditional probability theory.By this means,the feasibility and reliability of the proposed methods have been validated.