Data-Driven Stress Intensity Factor Solutions for Axial Outside Surface Cracks in Thin-Walled Cylinders

In the flaw evaluation, fracture mechanics methods require calculation of stress intensity factor, K, in fitness-for-service (FFS) codes, such as API 579 and ASME BPVC Section XI. For a surface crack in a cylinder, the K calculation becomes to compute the influence coefficients G0 and G1 in the FFS codes. API 579 provides finite element analysis-based tabular data of G0 and G1 for selected cylinder size (t/R_i), crack aspect ratio (a/c), crack depth ratio (a/t), and crack tip locations. Recently, curve-fit solutions of G0 and G1 were obtained for surface cracks at the deepest and surface points. For an arbitrary cylinder size, however, three-parameter interpolations are still needed for estimating the G0 and G1.

To simplify the K factor calculation, the present authors (PVP2022-86164) developed a data-driven K factor solution for axial outside surface cracks in thick-walled cylinders (D/t ≤ 20) based on the API 579 tabular data of G0 and G1 for axial outside semi-elliptical surface cracks at the deepest and surface points using the machine learning technology. This paper further develops a data-driven K factor solution for axial outside surface crack in thin-walled cylinders (D/t ≥ 20) using the artificial neural network (ANN). The sigmoid activation function is adopted in the optimal learning algorithm for the ANN model to learn from the API 579 tabular data of G0 and G1 and to predict the K solutions at the deepest and surface points for an arbitrary cylinder or crack size. The proposed ANN model contains three input variables, one hidden layer with five neurons, and one output variable. The data-driven solutions of G0 and G1 at the deepest and surface points of the axial outside semi-elliptical surface cracks are validated by the existing closed-form solutions for thin-walled cylinders.