A prediction of rock fall from the tunnel face by convolutional neural network

Abstract

Tunnel construction requires accurate prediction of occurrence of rock fall from the tunnel face, by evaluating rock mass properties, such as weathering grade, crack distribution, and others. Those evaluations are commonly based on subjective visual inspections, the results of which are likely to vary from person to person. Therefore, to achieve consistent determination of that, we developed a quantitative analytical method applied with image analysis based on engineering geology. In this method, occurrence of rock fall from tunnel face with high and/or low developmental level of weathering and crack can be predicted with a probability of approximately 80%, on the other hand, that with moderate level of weathering and crack can be done only with that of 40-60%. For improving probability, we attempted prediction of rock fall by convolutional neural network, and the result showed approximately same value as that by the above method.