Technical Condition Evaluation Of Assembled Simply Supported Beam Bridge Based On BP Neural Network

Assembled Simply Supported Beam Bridge has the advantages of convenient installation and fast construction.It is the most widely used beam bridge at present,and is also the first choice for small and medium-span beam bridges.There are a large number of assembled simply supported beam bridges in China,and with the increasing number of new bridges,more and more bridges need to be inspected regularly to evaluate the damage degree of bridges.However,due to the differences in the professional knowledge background and practical engineering experience of different bridge experts,the results of technical condition evaluation of the same bridge will be quite different and discrete,and will also mislead the follow-up management and maintenance of the bridge.Therefore,a disease evaluation model based on BP neural network learning is established in this study.The expert’s experience of disease evaluation is learned through BP neural network,and the disease scale is evaluated by training convergent neural network model.This will greatly reduce the impact of subjective factors on bridge disease evaluation by assessors,and make the evaluation results of bridge technical condition closer to the real technical condition of the bridge.Firstly,the inspection reports of 200 RC and PC assembled simply supported girder bridges are collected,and the disease data are preliminarily sorted out by a unified disease record method.The disease data of the unified format are sorted out,and the disease data lacking key information are eliminated.The disease data of the unified format are analyzed by experts,and the disease evaluation scale is carried out according to the evaluation rules of the Standard for Technical Condition Evaluation of Highway Bridges,which was implemented in 2011.Since then,the preliminary collection of disease data and expert analysis and processing have been completed.Then,according to the determined learning sample vector,the disease information needed by the theory is extracted from the preliminary expert evaluation data.A BP neural network evaluation model was established to compare the effects of the number of hidden layers,neuron nodes of different hidden layers,number of learning samples,different learning efficiency,different momentum factors,different expectation errors,different transfer functions on the evaluation effect of BP neural network.Finally,the convergence network with the minimum prediction error under the influence of different training variables is saved,and the accuracy of the evaluation model is verified by the disease data of Jinsha River Bridge and Qijiang River Bridge.Comparing with the disease scale given by experts,the results show that,except for 75% of the crack evaluation accuracy of the bridge deck pavement,the overall disease evaluation accuracy is over 80%,which validates the reliability of the evaluation models effectively.