Estimating corrosion growth rates is a non-linear, multivariate problem with many potentially compounding factors. CUI is especially problematic, since visual inspection is not possible or at the very least an expensive undertaking. Studies suggest that in the refining and chemicals industry CUI causes between 40% and 60% of all pipeline maintenance cos
Established standards such as API RP 581, used as a rationale for risk-based inspection (RBI), are based on limited data and statistical models that do not account for non-linearity. This can lead to more uncertainty, risk and additional cost. The highly empirical nature of these corrosion estimation techniques come with assumptions that may not take into account the specific conditions, or parameter combinations, a pipeline is likely to experience. As a result, estimations may be based on extrapolation, introducing error.
Data & AI solutions
Machine learning (ML) can uncover hidden patterns within structured or unstructured, historical or real-time datasets, such as operating temperature, environmental conditions, insulation type, pipe design and insulation condition, providing valuable insight for the RBI process.
Deep machine learning models as well as inference systems are developed to predict rates of corrosion based on input parameters such as operating temperature and environment. The model was trained with 70% of the data (inputs vs known corrosion rates) and 30% of the data was used to test the model for accuracy.
Several models were developed to predict CUI growth rates, ranging from ANFIS inference systems over Bayesian networks to several types of ANN.
Business case & beneifts
Over 20 percent of major oil and gas incidents reported within the European Union (EU) since 1984 have been associated with corrosion under insulation (CUI) as a contributing factor. Being able to estimate CUI could provide industry with the analytical capabilities to intervene in pipeline corrosion before a catastrophic failure occurs thereby reducing HSE liability and lost production or at least provide better predictive analysis for risk based assessment.
Increased flexibility & accuracy around scheduling of inspection intervals, enabling more targeted inspections and ability to predict corrosion hotspots while better quantifying uncertainties.