Although geological conditions in mountain tunneling projects are usually determined by tunnel face mapping, it may vary depending on the experience and geological knowledge of the observer. The application of artificial intelligence (AI), especially deep learning models, may solve that problem in face mapping. However, it is difficult to understand how the final output is derived from deep learning. To understand this condition, it is necessary to develop a model using training data that can be easily identified by naked eyes. In this paper, an AI model was developed using face photos and digital geological data as training data. Digital geological data such as DEM (Digital Surface Model) able to emphasize features that are difficult to express with tunnel face photos alone. The evaluation results of this model were good, and it was clear that digital geological data was very useful as training data. On the other hand, it is difficult for this model to detect parts that cannot be identified by human eyes alone, such as potential discontinuities that might developed behind the slip surface. The future task is to find the relationship between the distribution of the feature area captured by AI and the actual face.
In mountain tunneling project, it is essential to grasp the geological condition accurately and decide the optimum tunnel support pattern. Geological conditions are mainly evaluated from preliminary geological survey and tunnel face mapping during tunnel excavation. Tunnel face mapping includes utilization of empirical method (i.e Rock Mass Rating, Q system, etc) and the rating for various parameters such as rock mass strength, weathering and discontinuities are determined. However, depending on the experience and geological knowledge of the observer, the judgement of parameters varies greatly. To tackle this problem, the development of a system that allows any skilled engineer to accurately evaluate the geological parameters are required.