Multivariate Statistical Sensitivity Analysis And Model Modification With Neural Network In Construction Control Of Tied Steel Arch Bridge

Tied steel arch bridges are usually characterized with reasonable stress,wide applicability and beautiful shape.With the acceleration of bridge construction in China,there are more and more this type of bridge.Construction control is an important part of bridge construction.It can not only ensure the safety in the construction process,but also make the internal force and displacement of the structure in the completed state meet the designed ideal state as much as possible.This can better meet the requirements of bridge applicability and durability.The core of construction control is the identification and correction of the error between the actual state and the ideal state of the structure.Parameter sensitivity analysis is the key step of error analysis.In the past,the parameter sensitivity analysis of tied steel arch bridges mostly stayed in the single parameter sensitivity analysis.Such analysis is incomplete.In fact,the parameters of the actual structure often affect each other.This influence is not only reflected in different parameters,but also in the same structural parameters in different parts.This paper combines substructure division,uniform experimental design,grouping experimental design and multiple stepwise regression.A tied steel arch bridge in Chengdu was investigated.The main beams and arches are divided into substructures.The parameters of each substructure were considered as independent parameters.Fang Kaitai’s uniform experimental design method and group experimental design method were used to design the numerical analysis.We established a series of finite element models.Then we employed multiple stepwise regression to analyze the calculated data,and got the significance of each parameter to the structural response.In this way,the multivariate statistical sensitivity analysis of tied steel arch bridge was completed.This analysis method not only considered the interaction between different parameters of the structure,but also the influence of variations of the same parameter in different parts.At the same time,it also greatly reduced the workload and improves the work efficiency.When there is an error between the actual bridge state and the ideal state,it is necessary to reduce the error by adjusting the cable and other methods.Due to the errors existing in the manufacturing and installation stages,the original finite element model cannot well reflect the actual structural conditions.At this time,the model needs to be modified.In this paper,the BP neural network model was established.Since there were many parameters affecting the model correction in the completion stage of tied steel arch bridge,multivariate statistical parameter sensitivity analysis was introduced.We selected some parameters of some structural parts that significantly affected the response of the concerned structure,and carried out uniform experimental design and calculation.We took the parameter change and its corresponding structural response change as learning samples.Then the BP neural network was trained.Finally,we verified and evaluated the prediction of BP neural network in both positive and negative ways.The results showed that the network can better predict the structural parameters and modify the model of tied steel arch bridge in the completion stage.