Распознавание волновых образов трещин массива горных пород на основе нейронных сетей по данным георадиолокациям = Recognition of wave patterns of rock massif cracks based on neural networks from gpr data
Статья в журнале
Русский
551.34(571.56)
трещины; массив горных пород; георадиолокация; распознавание образов; сверточные нейронные сети; автоэнкоде; cracks; rock massif ; ground-penetrating radar; pattern recognition; convolutional neural networks; autoencoder
Cracks in the rock mass significantly affect the physical and mechanical properties of rocks, and they, in turn,must be taken into account in the planning of mining operations and construction of mining structures. There arevarious techniques for detecting cracks in the rock mass using GPR data. However, the application of these techniquesis limited by the productivity of geophysical operators, as the GPR data are mainly interpreted by them manually. Tostudy the fracturing of frozen rocks from GPR data, it is possible to use artificial neural networks (ANN), which willmake it possible to analyze GPR radarograms in order to detect discontinuities and shifts of in-phase axes of GPRsignals. A significant problem in the application of ANN is the preparation of data for training (training sample). It ispossible to create a training data set using a model of GPR section of frozen rock massif with a fracture. However,the practice of using synthetic radarograms based on the model of GPR section of frozen rock massif with a crack hasshown the need for its improvement in terms of increasing the number of rock layers, the possibility of setting inclinedboundaries, taking into account the presence of syn- and antiforms. The article describes the stages of neural networkmodel development, including the creation of a training data set, selection of architecture, training and testing of theneural network model. The validation of the ANN model showed high performance of the ANN model. Nevertheless,some drawbacks are observed in the performance of the model. The developed system will significantly reduce the timecost of GPR data interpretation. Further research will be related to improving the prediction accuracy associated withthe expansion of the training data set and development of an additional ANN model.
Соколов, К. О. Распознавание волновых образов трещин массива горных пород на основе нейронных сетей по данным георадиолокациям / Соколов К. О., Шамаев С. Д. ; Институт горного дела Севера им. Н. В. Черского // Успехи современного естествознания. - 2023, N 7. - С. 109-114. - DOI: 10.17513/use.38079
DOI: 10.17513/use.38079
Войдите в систему, чтобы открыть документ