Stress-strain curves for brittle rocks show three characteristic stress levels: crack initiation, long-term strength, and peak strength. Damage-controlled testing at low confining stresses has shown that the long-term and peak strengths are sensitive to the amount of induced damage, i.e., the greater the amount of damage, the lower the long-term and peak strengths. These tests also showed that the brittle-failure process is characterized by a loss of cohesion as friction is mobilized. Excavation of a circular test tunnel in massive brittle rock resulted in failure around the tunnel. The back-calculated strength for the failed rock around the tunnel is approximately one-half of that measured in laboratory tests. Crack-induced damage of Lac du Bonnet granite, both in the laboratory and in situ, begins when the load exceeds approximately one-third of the unconfined compressive strength. However, the stress level associated with failure is a function of loading path. In the laboratory, where the loading path monotonically increases, the ultimate strength of an unconfined sample is 225 MPa. Numerical studies suggest that in situ the loading path around the tunnel is more complex, involving stress increase and decrease and stress rotation. For this loading path, failure initiates at a stress between 100 and 120 MPa. Conventional frictional failure criteria did not adequately predict the extent of brittle failure measured around the circular tunnel. The results from the damage-controlled laboratory tests and the microseismic monitoring carried out during tunnel construction indicate a constant-deviatoric-stress criterion is a reliable indicator for predicting the onset of damage. This criterion was also found to give a reasonable prediction for the maximum depth of failure around the test tunnel. The fundamental assumption in the constant-deviatoric-stress criterion is that at low confining stresses, such as those which occur around underground openings, the brittle-failure process is dominated by cohesion loss.
