INTELLIGENT IDENTIFICATION OF ROCK DEFORMATION LOCALIZATION BY DSCM-CNN METHOD

Identifying rock deformation localization is of great significance for the early warning and prediction of rock damage and geotechnical disaster. In this paper, a DSCM-CNN model is proposed for the intelligent recognition of rock deformation localization by combining digital speckle correlation methods (DSCM) and convolutional neural networks (CNN). The maximum shear strain field nephograms of rock specimens during uniaxial compression tests are obtained by DSCM, and then the nephograms are labelled according to the position of the deformation localization zones to complete construction of the dataset. The proposed DSCM-CNN intelligent identification model is trained by the training dataset. The method is validated by uniaxial compression tests on red sandstone specimens. The proposed DSCM-CNN intelligent identification model can automatically identify the rock deformation localization zones. The subset accuracy, precision and recall are 94.19%, 97.21%, and 96.41% respectively, which proves the feasibility of the DSCM-CNN model. The proposed intelligent identification DSCM-CNN model provides a new idea for intelligent monitoring of rock deformation localization.