Anchorage Quality Classification Based On Small Samples

The application of bolt support in complex environments is very easy to cause damage,affect the safety of the engineering structure,and cause huge harm,so it is very important to test its safety and reliability.In actual engineering,because the anchoring of broken bolts is often less than that of intact bolts,the anchoring of broken bolts is a small sample.This paper mainly studies the classification of anchoring quality of small samples.The main work of this paper is as follows:(1)Aiming at the problem of non-destructive testing of bolt anchoring quality,the method of stress wave detection is mainly used to detect the quality of bolt anchoring,and the acceleration signal of bolt anchoring is collected through the sensor,and the collected signal is stored as bolt anchoring data.In addition,there are redundant features for the original bolt data,which will cause the classification performance of the classifier to be poor and the classification time-consuming problem.Using the kernel principal component analysis method of radial basis kernel function,the feature extraction work of the data set is carried out.Feature extraction of the acceleration signal of anchor rod anchoring is carried out,and then the extracted features are used to represent the characteristics of the acceleration signal of anchor rod anchoring.(2)Aiming at the problem of anchoring quality data classification,random forest algorithm is used to classify the anchoring data after feature extraction.By comparing the classification performance of different classifiers on bolt data,it is found that the random forest algorithm has the best classification performance.Therefore,the random forest algorithm is selected to classify the quality of bolt anchoring grouting.In view of the fact that the actual bolt data is unbalanced,there is a problem of misclassification of negative samples to positive samples,so anchor bolt data sets with different balance rates are constructed,and random forest algorithm is used for classification research.(3)Aiming at the problem that the random forest classifier has poor classification performance on unbalanced anchor bolt anchoring data.Based on the random forest algorithm,the support vector machine is used to analyze and evaluate the small sample data of bolt anchoring,and the feature weight coefficient is obtained,and the features of the small sample are weighted.Then combined with the advantages of the costsensitive random forest algorithm,the cost-sensitive random forest algorithm weighted by the support vector machine is used to complete the classification of the bolt data,and the method is used to complete the classification of the bolt data.(4)Perform experimental verification on the improved random forest algorithm.After experiment and feature extraction,the anchor bolt anchoring data obtained is used to construct sample sets with different balance rates,which are classified by random forest,support vector machine and random forest improved by cost-sensitive technology.The result is that the traditional random forest algorithm can only identify part of the defective bolt data,and the improved random forest algorithm can identify most of the defective bolt data.The experimental results show that the classification of complete and defective bolt anchoring based on the random forest algorithm can solve the problem of small sample bolt data.This paper mainly classifies and detects the anchoring quality of small samples of bolts,introduces the nuclear principal component analysis algorithm through the stress wave detection method to extract the features of the bolt data,selects the most effective features for the random forest algorithm,and weights them from the sample features.And the algorithm itself improves the random forest and verifies it on the bolt data obtained from the experiment.The result shows that the improved random forest algorithm can identify more small samples of bolt anchoring data.