| Joint roughness coefficient(JRC)is an important index to evaluate the shear strength of rock mass,and has a significant influence on the mechanical and seepage characteristics of the rock mass.Therefore,the measurement and calculation of JRC have always been an important research direction in geotechnical engineering.At present,the most commonly used JRC measurement method is the comparison method proposed by Barton,that is,estimating the JRC range by comparing its profile obtained using a contour curve analyzer with ten standard joint profiles.Although this method is simple to operate and portable,it has strong subjectivity,a low degree of automation,high requirements on the experience of engineers,and can not obtain the quantitative estimation of JRC.However,more accurate measurement methods,such as roughness statistical parameter method and fractal dimension method,usually need to use 3D scanners for joint profile scanning and coordinate extraction,which have high requirements on the scanning equipment,complicated process,high workload,and cannot obtain results immediately in the field.To make up for the limitations of the existing JRC measurement methods and achieve a reasonable balance between convenience and accuracy,this paper converts the JRC measurement into signal analysis and identification,reflects the roughness characteristics of joints through easily measurable strain signals and sound signals,and proposes a JRC measurement method based on friction signal processing and deep learning.The main research contents are as follows:(1)Establish the friction signal database required for the classification model.The coordinates of real two-dimensional joints,the strain and the sound signal generated by the sliding of the plectrum on the joint profile are collected through a large number of experiments.The redundant information of the signal is removed by means of reduction,resampling,moving average filtering,and detrending.A joint friction signal database labeled with JRC is established to provide data support for deep learning.(2)Research on the JRC measurement method based on deep learning.On the basis that the strain signal and sound signal can reflect the roughness characteristics of the joint curve,the advantages of deep learning in data analysis and pattern recognition are utilized to build a variety of deep learning models for signal feature extraction and recognition based on the structure of the deep learning model built for the randomly generated joint coordinate.The classification effects of the models are then compared and analyzed.(3)Research on the JRC measurement method based on audio fingerprint recognition.Aiming at the problem of low recognition accuracy of the deep learning models,a JRC measurement method based on audio fingerprint recognition is proposed based on the in-depth research and comparison of a variety of audio fingerprints and retrieval algorithms.This method can return the curve shape and roughness coefficient of the similar joints through the retrieval of the audio fingerprint.On the basis of the traditional standard joint curve comparison method,the audio fingerprint comparison completed by the machine is added,which reduces the subjectivity and improves the accuracy of the estimation results.The results show that the friction signal can reflect the roughness characteristic of a joint to a certain extent,and the estimation of JRC by friction signal is a fast and feasible JRC measurement method.Without manual intervention,the JRC prediction model proposed in this paper can achieve a classification accuracy of up to 55%,and the introduction of manual comparison can further increase the accuracy of this method.Compared with the existing JRC measurement methods,the method proposed in this paper has the advantages of simple operation,low equipment requirements,and a high degree of automation.It can provide reliable JRC estimation when the joint coordinates are unknown and provide a new idea for rapid JRC measurement in engineering applications. |
PDF Full Text Request