Comparative Analysis Of Similarity Measures Of Bridge Damage Characteristics

Traditional bridge damage feature evaluation methods are mostly based on the recognition accuracy of the classifier.The calculation process is complicated,and the evaluation result is related to the performance of the classifier.The field of pattern recognition is based on the similarity measurement theory to achieve feature evaluation and selection.However,there are few applications of this theory in the evaluation of bridge damage characteristics.Corresponding research has been carried out based on the above problems,and the main research contents are as follows:1.In order to better screen out effective test data,the existing exploratory data analysis method was quantitatively improved in combination with statistical principles,and the screening criteria were determined based on the health monitoring data analysis of a largespan suspension bridge.2.In order to reduce the noise of the test data,the empirical mode decomposition and its improvement methods are compared and analyzed based on the simulation signal.A better method is selected to decompose the test data,and then the signal is analyzed according to the effective coefficient method.Refactoring.Finally,based on the reconstructed signal,the corresponding damage feature is constructed by combining Fourier transform and HilbertHuang transform.3.Based on the similarity measurement theory,the metric function analysis method and the random forest method are used to evaluate the damage characteristics: in the metric function analysis method,by quantifying the difference between samples of different working conditions,the damage warning characteristics and damage location characteristics are evaluated.In the random forest method,the classification model and regression model are used to evaluate the importance of the three types of features.A numerical simulation model,namely the three-vibrator dynamic system is established,and the specific application of the characteristic evaluation method is explained.4.Based on the shaking table test of a certain T-shaped rigid frame model bridge,the measured data is screened and the feature is extracted,and then the feature evaluation and analysis are performed using the metric function analysis method and the random forest method.The results show that: for the same type of features,two The evaluation results of this method are relatively close.In addition,the evaluation results of the T-shaped rigid frame model bridge and the three-vibrator dynamic system are compared and analyzed.The results show that the performance of each feature in the two types of problems has a good consistency.