Investigation On Creep And Shrinkage Of Bridge Prediction Model Based On Uncertainty Quantification

Currently,for long-span prestressed girder bridges,the long-term deflection of main girder has been widely concerned by the academic community.The deformation and cracks of prestressed concrete bridges increase with time,resulting in the deterioration of their working conditions.The study show that shrinkage and creep effect is one of the main reasons affecting the time-dependent of bridge structure.The shrinkage and creep prediction model currently used has the risk of underestimating or overestimating the value of structural response.Therefore,this paper bases on the shrinkage and creep database established by Bazant of Northwestern University.The residual model is used to evaluate the normative model,and the residuals are mathematically analyzed and the mathematical model of the residual is established.In addition,the concept of uncertainty coefficient isproposed using the ratio method.In addition,use it as a modified model for long-term deformation of concrete beams.The main research results obtained in this paper are:(1)The selection of the surrogate model is introduced on the structural response of the simply supported beam and the high dimensional nonlinear problem of the rubber concrete water resistance test data.Firstly,the response of the simple supported beam is mainly studied.The results show that the Gaussian process regression(GPR)still has a good fitting effect when the sample is small.Secondly,taking the rubber water-resistance data as example,combined with the traditional response surface method,the comparison of the results obtained by the two methods can be acquired that in the case of small samples and highly nonlinear,GPR has higher accuracy than the traditional response surface,so GPR has an advantage in dealing with such problems.(2)Based on the shrinkage and creep test database established by P.Bazant,the residual method is employed to evaluate the CEB-FIP90 model prediction model.The residuals change with the parameters of the shrinkage and creep,indicating that: For the creep function,53 groups with positive residual values,accounting for 44.5% of the total number of data,66 groups with negative residuals,accounting for 55.5%,resulting in a slightly higher estimation for CEB-FIP90 model.For shrinkage strain,153 groups with positive residual values,accounting for 66.2% of the total number of data,78 groups with negative residuals,accounting for 33.8%,which can lead to underestimation of CEBFIP90 model.The residual model is developed by using the residual data.The verification using the experimental data proves that the residual model can effectively reduce the deviation between the experimental value and the model value.Finally,to quantify the uncertainty of long-term deformation of concrete structures,based on the ratio between experimental data and predicted values,an uncertainty coefficient model is proposed.(3)The study of long-term deformation of concrete beam by Faber in B.Espion database is carried,and NO.1 beam is chosen as the research object.Firstly,the forward uncertainty quantification theory is used,and the prior is used to generate 20 sets of data points.Based on these,the finite element model is established.The results show that the experimental values are less accurate than the simulated values.The long-term deflection values calculated by different parameters are quite different.Subsequently,the Gaussian process meta-model is established by using the difference between the above simulation data and the experimental data.The posterior distribution of the uncertainty coefficient is obtained by Bayesian inference.Combined the posterior distribution of the uncertainty coefficient with the ANSYS analysis,the results show that the values obtained by the posterior distribution are closer to the experimental values.Inadditionally,the model error is reduced compared with the parameter model of the prior distribution.Uncertainty.In summary,the backward uncertainty quantification combined experimental data with simulation can obtain more accurate model parameters,reduce the uncertainty of the model and improve the prediction accuracy of the prediction model.