Technical Condition Evaluation Of Simply Supported Bridges Based On Support Vector Machine Method

As China becomes a bridge construction superpower,it is imperative to improve the bridge operation and management level.The evaluation of bridge technical condition is a common means of bridge detection and considered as an important basis for decision-making.However,it is often interfered with subjectivity.Therefore,the following research is made on this issue.(1)Collect relevant literature,analyze the method and process of assessing the technical condition of simply-supported beam bridge,and summarize the four main sources of subjectivity: missing documents caused by leaks in file management,differences in detection and recording methods adopted by different bridge inspectors,evaluation differences caused by different evaluators' personal experiences and their understanding of a large number of natural language descriptions in standards and bridge inspection reports,and errors caused by a large number of manual operations.(2)Analyze the demand for technical condition evaluation results during bridge operation and management.In other words,when facing the same detection report,people always expect to obtain a more accurate result.And it is generally believed that the evaluation results of experts are more authoritative and more accurate.(3)According to the demand analysis,an evaluation method of bridge technical condition based on support vector machine(SVM)is proposed.Aiming at the sources of evaluation subjectivity,this paper further analyzes the feasibility of this method and explains the theoretical basis of SVM;(4)Taking the factors of component weight,disease hazard and frequency of occurrence into consideration,five disease indicators,including the crack,the concrete falling and losing angle of main beam,the plate bearing positional traverse,void or shearing over the limit etc,has been selected and discussed as typical diseases.According to Standards for Technical Condition Evaluation of Highway Bridges,the actual detection report and existing research theory,this paper constructs feature vectors for SVM calculations,the disease and collect the scoring samples from the experts in bridge management authorities;(5)Write a computer program in Python,and input the expert scoring samples into SVM.Evaluate the model in terms of its precision,accuracy,recall rate and F1-score,and obtain the optimal values of C,gamma and other parameters through grid search with layered k-fold cross-validation.Finally,the model will be assigned and the expert samples will be used for training in order to build a SVM classifier for rating disease indicators;(6)Taking the left part of HWZ Bridge as an example,the evaluation results of the machine are compared with that of the general practitioners and experts.The results show that the machine evaluation has a high level of accuracy,and is closer to the score of experts.Therefore,it is concluded that using the method of SVM to evaluate the technical condition of simply-supported beam bridge can effectively reduce the subjectivity.