A general rock failure criterion can be represented by τ = a1(a0 + σN)β, (δτ/δσN) = tanφ where τ and σN are shear and normal stress and tanφ is the instantaneous slope of the tangent to the failure envelope at point (σN,τ). Triaxial/shear test data can be simulated at any predefined conditions. Three circles measuring the strength of the tested rock at failure characterise the envelope: the uniaxial compression, uniaxial tension and biaxial compression/tension circles. The criterion is defined through three models. The non-linear reduced regression model, is used to fit a set of N data points: σN(i), tanφi and τi to obtain the three parameters of the criterion. In uniform distributions of triaxial test data, the slope angle φ of the tangent to the envelope at an unknown point (σN,τ) is approximated by (δR/δσm) ≈ (R(i+1) - R(i))/(σm(i+1) - σm(i)) = sinφi. In uniform distributions of shear test data, the slope angle φ of the tangent to the envelope at a known point (σN,τ) is approximated by (δτ/δσN) ≈ (τ(i+1) - τ(i))/(σN(i+1) - σN(i)) = tan φi. Calculated values of tanφi, σN(i), τi for N-1 data points, they are fitted into the non-linear reduced regression model to obtain the criterion parameters. The second regression model, the variable power polynomial, takes the form R = a0 + a1σmγ + a2σm2γ, (δR/δσm) = sinφ = q. The non-uniform distributions of triaxial test data σm,Rm are firstly fitted to the polynomial for modelling irregular data. The power γ is systematically searched at order n ≤ 2. Optimal γ is reached once the condition 0 ≤ q ≤ 1 starts to be satisfied. This gives rise to a set of σN(i), τi and tanφi, which is fitted into the non-linear reduced regression model to obtain the criterion parameters. Optimal fitting can be checked graphically from the degree of conformance between the two sets of scatter plots of σ1-σ2 calculated from both the polynomial model and the criterion. The third regression model is used to linearise the non-uniform distributions of shear test data, (σN,τ) by log/log transformation and translation. The model takes the form Y = ρ0 + ρ1X + ρ2X2 where X = log(σN + τ) and Y = log τ. The associated variable t is searched systematically until ρ2 tends to zero. At this optimal condition, the three parameters of the criterion are evaluated in turn a0 = t, a1 = exp(ρ0) and β = ρ1. Experimental data are fitted to the criterion without a priori information related to the tested rock.
Read full abstract