Innovation Of The Global Sensitivity Analysis Method Based On Elementary Effect And Its Engineering Application Study

There are extensive sources of uncertainty in the practical engineering field to be considered,which causes corresponding uncertain effect on the model response of the structure.How to quantitatively evaluate the influence of uncertainty factors on model response is particularly significant for model evaluation and model updating.Global Sensitivity Analysis(GSA)as an uncertainty evaluation method can effectively evaluate it.The current GSA application for complex structural models is generally based on the surrogate model combined with the variance method.However,the surrogate model sometimes can’t fully describe the original model,and the variance method has the disadvantage of large computational demand.Besides,the coefficient of variation for the parameters of the building structure is usually relatively small,so the spatial domain of the multi-dimensional parameter distributional space is also very limited.The excessive sampling amount and analysis of the variance method can easily cause waste of computational resources.So,researchers must study how to improve the efficiency of the GSA methodsBased on the abovementioned background,this thesis firstly introduced the elementary-effect-based GSA method to perform GSA for the building structures,which can help with GSA under moderate sampling demand.Firstly,this thesis analyzed the disadvantages of the Morris method’s traditional trajectory and radial sampling strategy in the process of Euclidean distance optimization and its random sampling process.Then,this thesis proposed a core-based sampling strategy with multi-layer Latin hypercube sampling to improve it,which improves the sampling and optimization efficiency of the elementary-effect method.For the new sampling method,this thesis used the seven-parameter g-function under two settings to verify its three optimization steps one by one.Then this thesis applied the new method on a cantilever column structure for GSA as well as the trajectory method and the radial method for verification purposes.Through the above investigation,we have verified the efficiency of the new method,which shows the sensitivity recognition accuracy of the newly proposed method is 4.7 times that of the trajectory method and 11.5 times of the radial method.Next,this thesis illustrated the superiority of the GSA method over the local sensitivity analysis(LSA)in the non-linear and non-monotonic situation with the mathematic function example.GSA method has stronger consistency and stability in solving non-linear function models and non-monotonic g-function applications,which can solve the application shortage of the LSA method.Then this thesis applied the new method to a steel frame structure,whose seven structural parameters were selected for GSA exploration.The GSA value curve of each parameter in the full push-over process of the steel frame and the corresponding LSA results were solved.Finally,the LSA result of the steel frame structure was obtained and compared with the GSA result.The conclusion shows that the relationship between the uncertain parameters of the steel frame structure and the output base shear is nearly linear,so there is little difference between GSA results and LSA results for some model analysis problems with strong linear relationship and model parameters with small variance in practical application.