Prediction of Falling Rocks from a Tunnel Face by Multimodal Deep Learning

Abstract

In mountain tunneling work, falls of rocks caused by oversight of tunnel face conditions have been a problem. Therefore, the authors have developed technologies to predict falling rocks using a part of rock properties. However, the accuracy is assumed to be not enough because other factors are not considered. In this paper, in order to verify the possibility of improving prediction accuracy, the authors predicted falling rocks by combining images and existing rock properties. As a result, the model combining images and the numerical values of the rock properties showed higher accuracy than the other models.