Home Search Archives Subscribe Contact Editorial Board  

Abstract              Subscribe to view full article

  Title Optimising corrosion growth predictions from ILI data using Bayesian inference
  Author(s) Michael Smith, Jonathan Martin, Matthias Peussner, and Katherine Taylor  
  Abstract CORROSION IS A KEY THREAT to pipeline integrity, with published incident statistics consistently reporting corrosion as one of the main causes of pipeline failure. Operators can monitor the corrosion threat by using in-line inspection (ILI) to detect and size corrosion features. When two or more historical sets of ILI data are available, these can be used to estimate corrosion growth rates (CGRs) along the full length of a pipeline. Two common techniques for CGR estimation are ‘box matching’ and ‘signal matching’. With box matching, corrosion feature boxes are matched between two ILIs and a population of CGRs is generated based on changes in reported depth. Signal matching involves the direct alignment, normalisation and comparison of magnetic flux leakage (MFL) signals, using a consistent sizing model and analysis interpretation.

Box matching is a convenient and well established technique, but uncertainties in the resulting CGR measurements are considerably higher than those from signal matching. When box matching is the only option for a pipeline – for example, when comparing data from different ILI vendors or technologies – the high uncertainties can be unintentionally propagated into integrity management decisions. This can result in costly, unnecessary remediation. There is therefore great value in understanding how a box matching CGR distribution should be modified in order to more closely resemble the true corrosionpactivity in a pipeline.

By explicitly modelling the uncertainties involved in box matching, a theoretical model can be produced to describe the relationship between ‘true’ CGRs (i.e. the rates actually occurring on the pipeline) and ‘measured’ CGRs (i.e. rates implied by a box matching analysis). Using Bayesian inference with the ‘PyMC3’ Python package, this model can be used to transform a set of box matching results into the most likely ‘true’ CGR distribution. The model performance is tested by comparison against signal matching results.

Through a case study for a real pipeline, it is shown that the above technique can greatly improve corrosion growth predictions, reducing repair requirements, while maintaining safety and compliance.

  Subscribe to view full article  
Home | Contact ISSN 1475 4584 Terms and Licence Agreement | Copyright | Privacy Statement