Starting…
Key numbers
Wall of the opening (r = a)
Lines: Kirsch at r = a. Markers: BEM at element midpoints (which lie on the chord, a·cos(π/N) from the centre, plotted at their polar angle).
Radial line
Contour map
Analytical solution (Kirsch)
Exactly as implemented in js/kirsch.js. Internally everything is tension-positive; r is the distance from the hole centre, θ is measured counter-clockwise from +x, q = (a/r)2.
Total stresses (far field + internal pressure p, p > 0 pushing on the wall):
Displacements induced by the excavation / pressure (G = E / [2(1+ν)], κ = 3 − 4νeff):
Rotated in-situ state. The entered σxx, σyy, τxy are the unrotated state; with rotation β (counter-clockwise, degrees) both Kirsch and BEM receive σ′ = R(β) σ R(β)T, i.e. (the same in either sign convention, since switching convention negates the whole tensor):
The Kirsch formulas above are evaluated with this rotated far field. For a circular opening the solution is rotationally covariant, so the readings at points that rotate with the stress (Roof, Wall with “follow” on) are independent of β in the analytical solution; any change of the BEM values with β therefore measures the rotational dependence of the discretisation (the mesh is not rotated).
Plane stress is obtained by the substitution ν → νeff = ν/(1+ν) in κ (G keeps the true ν); plane strain uses νeff = ν. The stresses do not depend on the elastic constants.
At the wall (r = a, q = 1): σr = −p, τrθ = 0 and, tension-positive,
So for τxy = 0: σθ(0°) = 3σyy − σxx + p and σθ(90°) = 3σxx − σyy + p.
In the rock-mechanics convention (compression positive), all stresses are the negatives of the tension-positive ones, with p still a positive push. With principal far-field stresses σ1 ≥ σ3 (compression positive) the wall tangential stress ranges between
the maximum occurring where the wall normal is perpendicular to the σ1 direction. With σv = 20, σh = 10 MPa, p = 0: 50 MPa at θ = 0° (springline) and 10 MPa at θ = 90° (crown). A tensile σθ,min (3σ3 < σ1, i.e. k < 1/3) shows as a negative number in this convention. Internal pressure alone gives σθ = +p (tension) in the tension-positive convention, i.e. −p when compression is positive.
Cartesian components and principal values are obtained by the standard transformation and by σ1,2 = (σxx+σyy)/2 ± √[((σxx−σyy)/2)2 + τxy2]. In this page the principal values are computed from the displayed tensor, so σ1 ≥ σ3 in whichever convention is selected. The displacements are those induced by the excavation (field before excavation excluded), in mm after the unit conversion.
Method
The boundary of the opening is approximated by N straight elements with constant (piecewise-uniform) boundary quantities. The mesh is rotated by −π/N so that an element midpoint lies at θ = 0. Two formulations are available:
Direct BEM (Somigliana identity): unknowns are the boundary displacements and tractions. The element influence integrals combine the constant line-load and constant displacement-discontinuity element solutions of Crouch & Starfield (1983); all integrals are analytic, no numerical quadrature is involved. The wall hoop stress is recovered from the tangential strain of the boundary displacements (central difference of neighbouring midpoint displacements) and Hooke’s law with the known tractions.
Fictitious stress method (Crouch & Starfield, 1983): the unknowns are fictitious constant line loads on the boundary of the infinite plane, determined so that the boundary tractions are satisfied. It is not a displacement discontinuity method. Stresses at the wall are the stresses of the fictitious-load solution evaluated on the rock side of each element midpoint.
The problem is solved as the induced problem by superposition: the boundary traction applied on the wall is t = −p·n − σ⁰·n, where σ⁰ is the far-field stress and n the wall normal; stresses in the rock are then the far field plus the induced stresses (total stress), displacements are those induced by the excavation.
Accuracy. Both methods converge at first order: errors halve when N doubles. In the direct method the piecewise-constant boundary displacement jumps at the element end points; these jumps act as stress singularities at the nodes, so field values close to the wall are badly wrong (this is not a quadrature problem). Measured error relative to the stress scale: about 135 % at 0.1 element lengths L from the wall, 44 % at 0.25 L, 12 % at 0.5 L, 1.4–1.8 % at 1 L and below 0.7 % from 2 L outward. For the direct method the radial chart therefore omits BEM points closer than 1.5 L to the wall, and the contour map masks BEM cells within 1.5 L (hatched band).
The fictitious stress field is smooth right up to the wall (2–7 % error within 1 L), so nothing is masked, but it is about 2.5 times less accurate than the direct method further out (N = 120: 0.65 % against 0.24 % at r = 1.5a). Its wall hoop stress on a circle is almost exact (about 10−4 %): this is a circle-specific superconvergence and should not be expected on other shapes. For the direct method at N = 120 the wall hoop stress error is about 0.95 % and the wall displacement error 0.59 % (fictitious stress: 0.0005 % and 1.05 %).