probabilistic assessments for seam cracks
TRANSCRIPT
ê Improve integrity management of seam weld anomalies• Reduce risk from seam weld anomalies in
low toughness vintage pipe – other models are unconservative
• Reduce cost by avoiding unnecessary digs• Quantify benefits of maintenance actions,
such as pressure changes, hydrostatic tests and in-line inspection
ê Compile vintage pipe fracture toughness test data from across the industry
ê Enable Consortium members to present a united front to regulators
3Crack Model Consortium
ê An improved version of the PRCI MAT-8 model for seam weld anomalies:• realistic crack front profiles – better accuracy• predict the sharpening of blunt seam weld flaws
(ex. lack-of-fusion features) – reduced conservatism• crack geometries that deviate from hoop stress plane
(ex. hook cracks)ê The most comprehensive database of vintage pipe fracture toughness in the industry
ê ILI tool performance characterization: probability of detection (POD), probability of identification (POI), sizing
ê A software tool that:• implements a probabilistic version of the new PRCI MAT-8 model• evaluates mitigation: hydrostatic pressure tests, pressure reductions, future in-line
inspections • uses cloud computing to speed up calculation time
Deliverables of the Consortium 4
ê The PRCI MAT-8 fracture model represents the best available technology to predict burst pressure in pipe joints with longitudinal cracks.
ê The model will be improved:• Realistic crack front profiles• Fatigue sharpening at the tip of seam weld flaws
ê This modeling framework will continue to be refined, calibrated, and validated in the Consortium.
5Model Development and Analysis
Fatigue initiationRealistic crack front
13Sizing Error
Measured Size
ILI S
ize 𝑦 = 𝑚𝑥 + 𝛽 + 𝑒
𝛽
1.5 mm
1.5 m
m
𝑚 =𝑎𝑏
𝑎𝑏
𝑦
𝑥
𝑒
slopeintercepterror
Unity plot
15Sizing Error
ê Model produces regression lines that could describe the data we have
ê Thousands of regression lines were calculated, only 50 are shown.
ê Lines close to the center are more likely – Intervals are calculated from these
ê Each model parameter is a distribution