Error Study On Construction Control Of Long Span Steel Box Girder Cable-Stayed Bridge

With the increasing span of cable-stayed bridge, the influences of geometric nonlinearity are becoming more and more evident during construction, so it is necessary to find the geometric error control method to adapt the non-linear characteristics of large-span cable-stayed bridge. In this thesis, take Sutong Bridge as the background, whose span is 1088m, the sources of error and control measures are studied.There are two main tasks in the construction control of Large-span cable-stayed bridges which are the prediction and adjustment of structure. For the errors correction of the structural parameters, the main parameters causing the structure deviation are firstly determined, and these key parameters are identified by using the corresponding theory and methods, at last the correct estimates are get. In addition, the deviation of alignment and stress may occur in the future construction, and they are predicted by using these methods. According to the errors that are identified in current stage and predicted in the next stage, these errors are controlled by taking the corresponding measures.In this thesis, the error characteristics and the sources of large-span cable-stayed bridge during construction are researched. Using the geometric nonlinear analysis software NLABS to simulate the full-bridge construction process, the stress-free alignment of the steel box girder and the stress-free length of the stayed cable are identified under the base state. According to the Geometry Control theory, the main parameters are identified by using sensitivity analysis methods to identify the structural parameters of the impact of the structure during the construction process and bridge. these parameters contains the main beam weight, elastic modulus and so on.Comparing with several theoretical approaches which are used to analyze bridge construction control, BP neural network is suitable for large-span cable-stayed bridge in the application of construction control because of its multi-parameter nonlinear mapping features. The basic principles of BP neural network are stated. Aiming at its disadvantages such as falling into local minima and slow convergence, the measures are proposed with the algorithm of the combination additional momentum method and adaptive learning rate method to improve BP neural network. By setting up the multi-parameter identification model and the alignment prediction model with MATLAB neural network toolbox, the simulation results of network are analyzed and compared. It is feasible that the error analysis and control is carried out by using the improved BP neural network algorithm in the large-span steel box girder cable-stayed bridge during construction.