Data-driven probabilistic failure assessment curve based on similitude principle

Abstract

In the study, based on the similarity in crack-tip fields between pipeline structure and standardized single-edge notched tension (SE(T)) test specimen, a methodology using a data-driven machine learning technique is proposed to determine the specific failure assessment curves for full-scale pipeline girth welds. By considering constraint similitude and ductile tearing, the probabilistic failure assessment line obtained from SE(T) resistance curves with a 50% survival rate can provide the most accurate failure assessment, as validated using the experimental full-scale pipeline data in the literature, particularly for the zone dominated by ductile fracture. Moreover, Option 1 (in R6 terminology) fracture assessment curve of British Energy R6 approach, which corresponds to a 15% survival rate, is proven to be overly-conservative.