Automatic Classification of Geology Using TBM Data

For the optimization of TBM driving parameters, the complementation of the geological tunnel documentation, and the minimization of risk in zones of highly fractured rock, a computer-based system that is able to automatically make decisions on the type of geological conditions is desirable. We propose a classification system that correlates sensor data of the TBM with geological assessments in order to provide such an estimation. The system relies on a machine learning method from statistical pattern recognition. By using Bayes classification and techniques for the approximation of probability density functions, an estimation of the geological class-membership of a new TBM data set can be derived from the statistical model. The system learns the structure of the data adaptively, and without a priori usage of domain- or expert knowledge, which makes it employable for a wide range of TBM tunnel projects. Recent experiments showed that a classification system based on this method is appropriate for practical application. Our approach is currently being realized as an additional software component of the tunnel documentation system 2doc by Pöyry Infra GmbH Salzburg.
Helmut A. Mayer

Last modified: Nov 14 2008