No abstract

The interaction of tunnel boring machines with the rock mass is highly influenced by human, technical and geological factors. Interpretation of geological observations and TBM data is currently done on a subjective basis. Technologies based on Artificial Intelligence research, can be used to automatically classify TBM data into rock mass behaviour types. Albeit first results look promising, any technology poses the threat of malicious use that deliberately harms / benefits one or another party. This paper shows how an Artificial Neural Network (ANN) can be trained to achieve the best possible rock mass behaviour classification, or how such a system can be misused to yield a more optimistic, respectively pessimistic classification to fortify the interests of one party. However, ANN also pose the chance to serve as an independent objective opinion and to improve the self‐consistency of geological classifications.

The dataset contains 1339 cone penetration tests (CPT, CPTu, SCPT, SCPTu) executed within Austria and Germany by the company Premstaller Geotechnik ZT GmbH. As a first processing step, core drillings, located within a maximum distance of approximately 50 m to the insitu tests, were assigned to these cone penetration tests, which allow an interpretation of the insitu measurements based on its grain size distribution. In a second step, the software Geologismiki was used to calculate various normalized measures, which can e.g. be used as input parameters for soil behaviour type charts. The present data can be utilized by researches for example to develop new approaches related to soil classification based on cone penetration test. Furthermore, it provides a framework for combining insitu measurements (q c , f s , R f , u 2 , V s ), normalized measures (i.e. Q t , B q , U 2 ) and soil classifications.

Digitalization will change the way of gathering geological data, methods of rock classification, application of design analyses in the field of tunnelling as well as tunnel construction and maintenance processes. In recent years, a rapid increase in the successful application of digital techniques (Building Information Modelling and Machine Learning (ML)) for a variety of challenging tasks has been observed. Driven by the increasing overall amount of data combined with the easy availability of more computing power, a sharp increase in the successful deployment of techniques of ML has been seen for different tasks. ML has been introduced in many sciences and technologies and it has finally arrived in the fields of geotechnical engineering, tunnelling and engineering geology, although still not as far developed as in other disciplines. This paper focuses on the potential of ML methods for geotechnical purposes in general and tunnelling in particular. Applications such as automatic rock mass behaviour classification using data from tunnel boring machines (TBM), updating of the geological prognosis ahead of the tunnel face, data driven interpretation of 3D displacement data or fully automatic tunnel inspection will be discussed.

No abstract