Neural network modeling of jointed rock mass behavior in topocentric terrestrial field: A case study of Oktyabrsky deposit

Abstract

The article addresses a promising geomechanical trend connected with Big Data processing. Using helium release observation results, the jointed rock mass behavior in the terrestrial field is studied. The terrestrial field is assumed as a variable value governed by the positions of the major planets in the Solar system, as well as by the positions of the Moon and Sun. The observation data are used for neural network learning. The total bulk of the learning data was more than 95 thousands of observations. After learning, the neural network decision-making algorithm was analyzed, and the studies were compared with the rock mass jointing analysis data from 3D modeling of Oktyabrsky and Talnakh deposits. Interpretation of frames of faults in the study area produced more than 206 thousands of measurements of precise dip angles and strike orientations with their distribution in depth and along the horizontal. Alongside with the fault frame interpretation, helium release data were compared with roughness of walls in a vertical opening sunk in the close vicinity of the measurement site. As a result of long-term operation, the shape of the opening repeats the block structure of enclosing rock mass and, thus, can inform on its initial jointing. The wall roughness data were obtained using laser scanning and contained more than 106 thousands of measurements. The analysis and processing of all data reveals the dependence between the planetary positions in the Solar system, helium release, orientation of the main joint system and surface roughness in underground openings. In this manner, it is possible to assess deformation processes in the crust and to find their influence on rock mass behavior.