Inversion Analysis Of Damage Inducement Of Overhead Vertical Wharf In Three Gorges Reservoir Area

The overhead vertical piled wharf is the typical wharf structure in the Three Gorges Reservoir area,it has the advantages of adapting to large water level variation,good berthing stability and high loading and unloading efficiency.In the process of wharf operation,it is easy to be affected by soil sliding,uneven settlement of foundation,irregular berthing,overload and other adverse factors,resulting in wharf structure damage and even overall overturning.The existing wharf structural health monitoring methods mainly focus on the identification and location for the damaged wharf structure,which can not provide early warning and traceability for the damage inducements.Under the support of Chongqing major engineering research project "R & D and application of key technology for long-term intelligent monitoring of structural safety of inland river frame vertical wharf",the following research work has been carried out for damage inducement inversion of overhead vertical high piled wharf in Three Gorges Reservoir area(1)This paper systematically investigates the construction of vertical high piled wharf in the Three Gorges Reservoir area,analyzes and puts forward the typical structural type of overhead vertical wharf,and studies the action type,action strength,action position and other indicators of the main adverse damage inducements.On this basis,a full-scale finite element analysis model of a single section of an overhead vertical wharf is established,and the mechanical characteristics of the wharf structure under overload,out of control ship collision and bank slope instability are analyzed.The plane and axial strain distribution characteristics of pile groups are analyzed,and the stress sensitive areas of pile foundation under different damage incentives are determined.(2)According to the technical requirements of damage inducement inversion for vertical piled wharf in the Three Gorges Reservoir area,the basic idea of damage inducement inversion including pile group stress pretreatment,data dimension reduction,machine learning model construction and inversion model evaluation is proposed.On the basis of summarizing and evaluating the applicability of existing data preprocessing,data dimensionality reduction and machine learning methods,the neural network structure model based on SGD algorithm is deduced,the model evaluation methods and evaluation indexes of classification and regression problems are studied and determined,and the damage inducement inversion model based on neural network is constructed.(3)In order to meet the massive data requirements of neural network training,based on the construction of full-scale finite element analysis model,a parameterized finite element model based on beam element is established by analyzing element type,constitutive model and material parameters,boundary conditions and damage induced loads.Compared with the fullscale finite element analysis model,the maximum stress error is less than 10% It is proved that the parametric beam element model can replace the solid element model for large sample calculation.A total of 21000 typical damage cases,including overload,out of control ship collision and bank slope instability,are randomly selected.The parametric finite element model is used for parametric calculation,and the stress sample data set of pile group is constructed by means of data preprocessing,normalization and dimension reduction.(4)Based on the established back analysis model of damage inducement,classification learner and regression learner are used to back analyze the three typical damage inducements of overload,out of control ship collision and bank slope instability.The results show that the accuracy of identifying the position of overload is 0.98,and the strength identification index R2 is 0.99;The results show that the recognition accuracy of ship collision position is 0.94,and the recognition index R2 of ship collision strength is 0.99;The identification index R2 of slope action intensity reaches 0.99;The accuracy of mixed sample type identification is 1.0,the accuracy of position inversion identification is 0.94,and the strength identification index R2 is0.98.Through the analysis of the inversion results,it can be seen that the neural network inversion model has good performance and can effectively inverse the damage inducement effect.