Numerical Methods for Hypersingular and Near-Singular Boundary Integrals in Fracture Mechanics

The boundary integral equation is one of several equivalent forms of governing equations that can be used to compute approximate solutions to boundary value problems in elasticity and potential flow analysis. Since it determines the entire solution in terms of values only on the boundary, there are possible order-of-magnitude advantages in solution time and geometric complexity over better known 'domain-based' methods such as finite elements and finite differences. In practice, it has been hard to capitalize on these advantages. Many of the difficulties center around inability to perform certain numerical integrations. This thesis presents: (a) a systematic 'modal' method of converting singular integrals to easier integrals over 'far' surfaces; (b) an optimal quadrature method for the 'nearly singular' integration problem. An existence proof is given to show that all surface integrals arising from the 3D boundary integral can be converted to easier contour integrals if basis functions are constructed in a cartesian sense, rather than the common parametric formulations. Stokes vectors needed to make this result useful are demonstrated for the Laplace equation and for some cases of elasticity. Comparison to analytic benchmark cases shows that the method produces accurate stress intensity factors for 3-dimensional fracture analysis.