Tunnel convergence in highly stressed rock will always be an
important issue. It determines the amount of required overexcavation, influences machine and equipment selection, as well as the support concept. When underestimating the convergence, extremely costly reshaping is required. When overestimating the amount of displacements, in general the "unused" overprofile has to be filled with concrete, which is comparatively cheap. On the other hand, choosing the overprofile on the to safe side, does increase the amount of required support and displacements. Because of the importance of the prediction of tunnel closure, it is
surprising, that not much research has been done in this direction so far. Using analytical or numerical methods only does not appear to lead to sufficiently accurate results because of the huge number of parameters involved. A procedure is currently under development, which is based on
analytical functions developed by Guenot, Panet and Sulem and the modified functions by Barlow. To determine function parameters an expert system in combination with Artificial Neural Networks (ANN) can be used. The expert system consists of numerous site date, stored in the data base system DEST (Geotechnical Group Graz) numerical simulations, and monitoring data. Numerical simulations focus on the influence of support and excavation sequence on displacements, while monitoring data are used to predict rock mass structure ahead of the face.