Statistical Predictive Modelling: A Methodology to Prioritize Site Selection for Near-Neutral pH Stress Corrosion Cracking

Abstract

Pipelines are subjected to both residual and applied tensile stresses, and can form near-neutral pH SCC (transgranular stress corrosion cracking) if the pipeline is exposed to a conducive environment and is made from a material that is susceptible to SCC. This transgranular SCC is an ongoing integrity concern for pipeline operators. As part of an SCC Integrity Management Program (IMP), it is necessary to perform integrity assessments and prioritize segments of the pipelines to manage the SCC threat. Ultrasonic crack detection in-line inspection tools have proven capable of locating SCC, but reliability of these tools is not absolute and the reduced probability of detection of subcritical flaws limits options for proactive management. Hydrostatic retesting is a very effective program for removing near-critical axial defects, such as SCC, but does not provide useful information as to the location of SCC along the pipeline. NACE Standard RP0204-2004 (SCC Direct Assessment Methodology or SCCDA) outlines factors to consider and methodologies to employ to predict where the SCC is likely to occur, but the standard acknowledges that there are no well-established methods for predicting the presence of SCC with a high degree of certainty. The trend in probabilistic modelling has been to focus on establishing deterministic relationships between environmental factors, tensile stress and SCC formation, and growth; these models have achieved varying degrees of success. The Statistical Predictive Model (SPM) was previously developed to predict the likelihood of occurrence of near-neutral pH Stress Corrosion Cracking (SCC) for the NPS 10 Alberta Products Pipeline (APPL). SPM Phase 5 uses selected predictor variables representing tensile stress, environmental, pipe-related, corrosion control and operational relevant factors to determine the Probability of Occurrence of SCC. Regression techniques were used to create multi-variable logistic regression models. The results for each model are checked at locations where SCC is known to be present or absent to assess predictive accuracy, then used to prioritize susceptible segments for field excavation. The relative strength of individual predictor variables provides insight into the mechanism of near-neutral pH SCC crack initiation.