Probabilistic Interpretation of Welding Residual Stress in Fracture Assessment

Abstract

Residual stress is an unavoidable by-product of welding processes that can negatively
influence performance and longevity of structural components. Due to the multifaceted
interaction of phenomena taking place during a multi-pass weld process, variances (in
their many different forms) are likely to occur and contribute to scatter in experimental results.
The presence of variance in weld residual stress measurements is often treated conservatively
in many engineering disciplines including design, manufacturing, in-service inspection, and
fitness-for-service assessments. Although safety requirements remain a priority for industrial
practice, probabilistic and data-driven approaches are becoming increasing available for optimising performance of components and their related functions.
The primary aim of this thesis is to quantify variability in weld residual stress measurements
of girth-welded components commonly used in a range of industrial applications including onand off-shore piping, pressure vessel components and in nuclear application. The investigation
utilises observed statistical distribution models of measurement data to predict through-thickness
weld residual stress distributions for use in safety assessment calculations. The presented
probabilistic methods use Monte Carlo simulation to model variability of parameters to determine
susceptibility of failure using a failure assessment diagram approach. The implications regarding
general use in structural integrity assessments involving fracture are demonstrated using case
study pipes presented in Chapters 3 and 5.
Variance of the combined transverse residual stress database (perpendicular to girth weld
direction) are shown statistically to be normally distributed. This interpretation was considered
in a fracture assessment context using different probabilistic models to simulate observed
variability of measurement data. Influence of pipe parameters on weld residual stress data of
austenitic welds are also investigated using statistical methods indicating support of current
treatment advice, presented in Chapter 4. The austenitic database was found to possess overall
improved normality correlation compared with ferritic and combined measurement datasets.
Consequently, this research is aimed to promote engagement of probabilistic applications in
integrity assessments to improve acceptance criteria to make informed decisions based on realistic
estimates of variability.

Date of Award	5 Dec 2023
Original language	English
Awarding Institution	University of Bristol
Supervisor	Harry Coules (Supervisor) & Isabel Hadley (Supervisor)