Research On Intelligent Identification Method Of Surrounding Rock Grade Based On Drilling Information

As the geological survey data cannot fully and accurately reflect the geological situation in front of tunnel boring,tunnel construction decisions are often blind,which leads to frequent geological disasters(such as sudden water and mud,landslide,rock explosion,etc.)in tunnel construction.Once the disaster occurs,the light machine will be damaged,and the heavy one will lead to massive collapse of the surrounding rock,and casualties.Some projects are even forced to be rerouted or stopped,so it is crucial to anticipate the changes of the surrounding rock in advance for safe tunnel construction.This paper focuses on the intelligent identification method of surrounding rock grade based on borehole information,and carries out research in four aspects:fracture identification of borehole images,acquisition and analysis of drilling-following parameters,comprehensive identification of surrounding rock grade by combining borehole images and drilling-following parameters,and field application,respectively.It achieves the main research results as follows:(1)It proposes a fracture detection and extraction method for borehole images based on integrated deep neural networks.The method uses migration learning to train the VGG16 classification network and ResNet50-UNet semantic segmentation network based on ImageNet dataset respectively,in which the VGG-16 classification network is used to locate and sieve the rock wall and fracture targets in the borehole image,and the Resnet50-UNet semantic segmentation network is used to extract clear and complete fractures from the borehole image binary image,based on which,an automatic analysis algorithm of RQD value of borehole envelope is established.(2)It developed a rock drilling test system based on a submersible drilling rig,established a propulsion speed,oil pressure and air pressure collection device and converted and stored the data through an automatic data collection system,which realized the automatic recording and storage of four drilling parameters,namely impact pressure,rotary pressure,propulsion pressure and propulsion speed,during the drilling process of the submersible drilling rig.Based on the developed test system,the drilling test follow-up data were collected and a sample database of the follow-up parameters and the surrounding rock grade was established.(3)A graded sample database based on the drilling parameters and the envelope rock grade,which was analyzed to obtain the correlation between the drilling parameters and the envelope rock grade.In this study,the database training sample set and prediction sample set are divided,and the support vector machine(SVM),random forest with improved base learner,Adaboost,and Bayesian optimization Lightgbm models are established respectively based on python language,and the SRG1 model for intelligent grading of perimeter rock is established by voting method.(4)This method establishes a sample database that fuses the RQD values of borehole envelope,drilling-following parameters and envelope grade,and trains the intelligent grading model of envelope based on borehole image information and drilling-following parameters based on the SRG1 model.It is applied in the field at Qiantangjiang Road Station of Qingdao Metro Line 6 and tested on the field collected data by using the established intelligent grading model of surrounding rock,in which the accuracy of Grade 2,Grade 3,Grade 4 and Grade 5 surrounding rock is 78%,78%,94%and 96%respectively,with an average accuracy of 86%.