Application of Neural Networks in Rock Mass Stress Assessment by Photoelasticity
S. A. Neverov1, A. A. Neverov1, A. I. Konurin1, M. A. Adylkanova2, D. V. Orlov1
1Chinakal Institute of Mining, Siberian Branch, Russian Academy of Sciences, Novosibirsk, Russia
2D. Serikbaev East Kazakhstan Technical University, Ust-Kamenogorsk, Kazakhstan
Keywords: Photoelasticity, optical pattern, isochromatic curves, contour lines, modeling, sensor, experiment, borehole, stress-strain behavior, rock mass, neural networks, geomechanical data
Abstract
The optical polarization method with ring-shaped photoelastic sensors, digital photography of isochromatic patterns and their clarification using neural networks is developed for the stress measurement in rock mass. The case-studies of the photoelasticity application in solving various problems of elasticity and rock pressure analysis are reviewed. As a result of a lab-scale experiment, a data set of 15 000 isochromatic images is collected. The machine learning algorithm was a convolutional neural network, the Inception module. The authors recommend using downhole sensors for the continuous stress monitoring in underground mines and integrating the obtained data in a digital model with the help of IoT.
|
