Session: 07-05: Cracks Uncertainty Modelling

Submission Number: 185828

Approaches to Model Spatial Correlation of Depth Measurements in a Crack Profile 

Modern in-line inspection (ILI) tools are capable of providing crack profiles in liquid pipelines. However, sizing uncertainty is expected in the profiles due to measurement and sizing algorithm errors associated with ILI. There is no standardized specification to categorize the profile sizing errors, other than to represent them by maximum depth and length sizing errors. In the literature, two different approaches are considered to incorporate sizing errors into crack profiles.

The first one is to consider each depth measurement reported in a crack profile to have the sizing error independent of the sizing errors of the other depth measurements. In essence, the sizing errors for all reported depths in a crack profile are assumed to be statistically independent. When crack profiles are simulated using this approach to consider the sizing uncertainty, samples of the crack profiles may become unrealistic and non-viable cracks due to the physical limitation of crack propagation. For instance, the interaction of the stress fields of cracks in proximity at their crack tips is documented in the literature. Ignoring the potential correlation due to this interaction and including such profiles to calculate the probability distribution of the burst pressure capacity of cracked pipes may result in an underestimation of probability of failure at a given operating pressure.

The second approach is to consider that sizing error in all depth measurements of a crack profile can be represented by a single random variable. This implies that the sizing error will be the same among all the depth measurements of the crack profile and the crack profile shape is preserved during the simulation of the crack profiles. While this approach could produce conservative results, this can become overly conservative due to the uniform change in all depth measurements of a crack profile, especially for a long crack. Therefore, there is a need to model the correlation between depth points of a crack profile for realistic characterizations of crack profiles to reduce the conservatism of the predicted burst pressure. This also improves the representation of the physical processes that result in observed crack profiles.

In the present study, a case study is conducted to demonstrate the impact of crack profile depth correlation on the predicted burst pressure capacity distribution using four real-world crack profiles in the public domain. Multiple approaches, such as Gaussian process and Markov chains, to model the spatial scale of correlation and to define the correlation structure in the crack profile depth measurements are discussed. Effects of different assumptions of spatial correlations on the burst pressure capacity predictions are also addressed with the case study.

Presenting Author: Smitha Koduru, Element Resilience and Risk Consulting Services Inc.

Presenting Author Biography: Smitha Koduru is a Principal Consultant at Element Resilience and Risk Consulting Services, and an Adjunct Professor at the University of Alberta, Canada. Dr. Koduru has more than twenty years of experience in application of risk and reliability concepts to evaluate structural systems and distributed infrastructure, such as buried oil and gas pipelines, facilities, underground gas storage, and authored numerous research articles and research reports related to pipeline integrity related to cracks and geohazards. She had a PhD from the University of British Columbia, Vancouver, Canada in applying reliability methods for performance-based engineering. She carries a Professional Engineer registration in the provinces of Alberta, and British Columbia in Canada.

Authors:

Smitha Koduru Element Resilience and Risk Consulting Services Inc.
Yong Li University of Alberta