| The overall destabilization damage of geotechnical engineering is inevitably related to the accumulation of damage and non-uniform deformation of rocks.How to realize the identification of rock deformation damage position and determine the deformation damage stage is the starting point and focus of preventing geotechnical engineering disasters.In this paper,the intelligent identification method of rock deformation damage location and stage is studied,the main contents are as follows:(1)The basic theories of digital scattering correlation methods and deep learning are systematically described.The key issues of correlation functions,search algorithms and deformation calculation methods of digital scatter correlation methods(DSCM)are firstly discussed,and the key configurations of zero-mean normalized correlation functions and three-step search algorithms are determined;then the basic theories of deep learning such as artificial neurons,multilayer perceptron and convolutional neural networks are sorted out,the characteristics and the application scenarios of activation functions,parameter update methods,convolutional neural network structures and branches are analyzed.(2)An XFEM-CNN method for intelligent recognition of rock fracture location is proposed.The rock cracking model is established by extended finite element(XFEM),deformation field data is obtained,and the data set is labeled according to the model fracture location to construct the data set;the XFEM-CNN intelligent recognition model is built with CNN as the main framework;the performance is evaluated using indicators such as subset accuracy.Numerical simulations are used to validate the method,and the results show that the XFEM-CNN model with Res Net50 convolutional neural network as the main framework has the best performance,and the subset accuracy,precision,recall,and F-score in the validation set are 94.73%,98.72%,98.71%,and 98.71%,which proves the accuracy of the XFEM-CNN method.(3)A DSCM-CNN method for intelligent recognition of rock deformation localization position is proposed.The method obtains the rock surface deformation field by DSCM,labels it according to the deformation localization zone position,and completes the dataset construction;the DSCM-CNN intelligent recognition model is built with CNN as the main framework.The method is validated using rock uniaxial compression test,and the results show that the subset accuracy,precision,recall,and Fscore of the method in the validation set are 94.19%,97.21%,96.41%,and 96.74%,which proves the accuracy and feasibility of the DSCM-CNN method.(4)A DSCM-Stacking method for intelligent identification of rock deformation damage stages is proposed.The method will obtain the rock surface deformation field by DSCM,determine the rock deformation damage stage according to the curve characteristics of the deformation non-uniform statistical index Sw,assign the corresponding labels to the deformation field cloud map,and complete the construction of the data set;build the DSCM-Stacking intelligent recognition model using Stacking integrated learning;and evaluate it using accuracy and precision indexes.The method is validated using the surface deformation field data of red sandstone specimens during uniaxial compression tests,and the results show that Stacking integrated learning can effectively improve the model recognition performance,and the accuracy of the final DSCM-Stacking method is 96.45%,the precision are above 95%,and the recognition recall rate of the near-damage stage is 100%.The XFEM-CNN method,DSCM-CNN method,and DSCM-Stacking method proposed in this paper can accurately identify the fracture location,deformation localization location,and deformation damage stage according to the rock surface deformation field. |
