Prediction of stress-driven rock mass damage in spent nuclear fuel repositories in hard crystalline rock and in deep underground mines

Nuclear plants have existed since the 1950s, and they provide 11 % of the world's electricity production. Worldwide, 30 countries are operating 448 nuclear reactors for electricity generation, and 57 new nuclear plants are under construction in 15 countries. Measured by deaths per terawatt hour, nuclear power is the safest method to provide energy, but it does produce a range of radioactive waste, which must be disposed of safely and responsibly. The deep geological repository is currently the only acceptable long-term solution for high-level nuclear waste. The two most common causes of rock mass failure are structurally controlled gravity-driven failure and stress-induced failure. Usually, surface and near-surface rock excavations are subject to structurally controlled gravity-driven problems, but in deep rock spaces, the in-situ stress of the rock mass increases and the risk of stress-driven problems grows. The five most common stress-driven damage mechanisms are i) rockburst, ii) spalling, iii) convergence, iv) shearing and v) seismic. Excessive convergence is rarely a problem in hard, massive rock mass. In this thesis, the remaining four mechanisms are addressed. The goals of the research were to discover the damage-reducing capability of thin sprayed concrete liners, to define the strength of long rock joints, and to develop a real-time risk management concept. Numerical modelling was used to design an in-situ concrete spalling experiment, the ICSE. Laboratory scale mortar rock joint replicas were used to study the scale effect, and large 2.00 m by 0.95 m (ASPERT) and 0.50 m by 0.25 m rock joints were sheared to validate the methods. A new real-time formulation of the Geotechnical Risk Management (GRM) concept was studied using both example cases and case data. New methods were developed for the photogrammetric capture of rock joint surfaces and shear testing of large rock samples. The numerical modelling predictions for the in-situ experiment show that the thin concrete liner produces up to 3 MPa of support pressure and using polyaxial Ottosen criterion the liner is not damaged during the heating stage. Both the replica shear tests and the large shear tests results show a weak negative scale effect. Based on the initial analyses using example data, Bayesian networks appear compatible with the Observational Method, and the approach is ready to be tested using real data. The three main conclusions each address the stress-driven damage prediction and mitigation. The stress-driven damage can be reduced using support pressure generated by thin concrete liners. A new method was developed to capture rock joint geometry using photogrammetry and to manufacture mortar replicas for laboratory scale shear testing. The use of Bayesian networks, together with the real-time geotechnical risk management concept, was demonstrated. The results contribute towards predicting stress-driven damage in deep underground spaces.