Study On Nondestructive Testing Method Of Bolt Anchorage Quality Based On Improved Stacking Auto-Encoder

Rock bolt support is widely used in all kinds of foundation construction because of its low construction cost,rapid support effect and reliable support effect.However,rock bolt support is a kind of implicit support.If the bolt anchor system has defects,the bolt support capacity will decline,which will affect the project quality and even cause people’s life and property safety accidents.Therefore,in order to ensure the stability of rock bolt support and prevent the occurrence of safety accidents,it is of great significance to study the defect classification of bolt anchor system.The deep learning method overcomes the shortcomings of the traditional artificial classification method,such as the poor accuracy and slow classification speed,and requires the classification personnel to have higher professional technical level and experience.Relying on the deep neural network,it constantly learns more abstract features of the data,reduces the impact of human subjective factors on the classification effect,so that the classification results have credibility.This paper uses the Stacked Self-Encoder(SAE)algorithm combined with the Softmax multi-classifier network to achieve classification of different anchoring defect types.In order to further improve the accuracy of classification,Shuffled Frog Leaping Algorithm(SFLA)optimization algorithm during SAE network pre-training was improved.The improved SFLA-SAE-Softmax intelligent network is established to classify the bolt defects based on the experimental data and simulation data obtained by the stress wave bolt nondestructive testing method.The main research contents are as follows:(1)By expounding the propagation theory of guided wave in bolt and multi-layer structure,the dispersion equation of guided wave in bolt and bolt anchoring medium surrounding rock system and the dispersion curve of longitudinal guided wave in two media are obtained.This paper analyzes the dispersion characteristics,multimode characteristics and the selection of the frequency end of the excitation wave,selects the Hanning window modulation sine wave in the frequency band of 20 k Hz-30 k Hz as the excitation wave,and according to the dispersion curve,the propagation speed of the longitudinal guided wave in the frequency band is 5 000 m/s in the bolt and 2 800 m/s in the bolt anchorage medium surrounding rock system,this lays a foundation for the transient analysis and calculation of anchor structure.(2)Using ANSYS / LS-DYNA finite element software,the anchorage models of different types of anchorage defects are established,and the transient response of different anchorage structures is analyzed and calculated.The length of the anchor,the anchorage length,the location and size of the defects in the model are determined,the accuracy of the model is verified,and the simulation data of the intelligent classification of the anchor are obtained.(3)According to the defects such as insufficient density of anchorage agent and anchorage cavity in the application of anchor system engineering,four kinds of experimental defect identification and classification anchor models are constructed.Based on the stress wave nondestructive testing method,a test platform of anchor system is built to obtain the test data of intelligent classification of anchor.(4)The structure of SAE network and its training mode are described.The classification network of SFLA-SAE-Softmax is established to realize the classification of different types of anchorage defects,and its classification effect is evaluated by multiple classification evaluation indexes.The evaluation results show that the network has good classification performance in the classification of different types of anchorage systems.In order to further improve the ability of SFLA optimization and improve the accuracy of classification,the optimization method of SFLA algorithm is improved,the improved SFLA model is proposed,the improved SFLA-SAE-Softmax classification network is established,and the classification effect is verified by using the different simulation data obtained based on ANSYS / LS-DYNA finite element software and the test data obtained based on the test platform It shows that the improved SFLA-SAE-Softmax network not only improves the convergence speed and convergence accuracy of the pre training of the SAE network,but also improves the classification accuracy of the whole network.It has a good applicability in the classification of bolt anchoring quality with different defects.