Probabilistic Corrosion Growth Rate (PCGR) models are typically studied on one or two Inline Inspections (ILIs) of a small sample size and use evaluation metrics that are operator specific. This comprehensive empirical assessment aims to expand the evaluation of PCGR models by examining anomalies that had distinct Box-to-Box (B2B) matches in four successive ILIs, resulting in 57,678 individual external corrosion anomalies from twelve pipelines of differing products, vintages, coatings, material specifications, and operating environments. This paper evaluates several PCGR models which are a combination of well-known industry models and simulation-based regression techniques. These models were evaluated using a historical train-test split on each studied anomaly set, where three prior ILIs were used to simulate the depth of each anomaly projected to the year of the fourth ILI. The simulated depths were compared to the measured depths of the final ILI. Statistical metrics and graphical techniques were used to measure a model's predictive ability to effectively simulate the mean depth and uncertainty. The models presented in this paper are starting points for oil and gas transmission pipeline operators with ILI data to model corrosion growth with time. This work provides empirical guidance on how to use multiple corrosion ILI B2B matches to predict corrosion depth growth.
Traditional Corrosion Growth Rate (CGR) models used in the integrity assessment of corroded pipelines are deterministic. A common Magnetic Flux Leakage (MFL) inline inspection (ILI) tool performance specification on general corrosion anomaly depth is +/− 10% Wall Thickeness (WT) at 80% confidence which corresponds to a standard deviation of 7.81% WT. 1 Probabilistic Corrosion Growth Rate (PCGR) models incorporate these large measurement uncertainties and provide more realistic reliability assessments as shown in Figure 1.
Studies have shown that incorporating the growth of anomaly length does not significantly affect the accuracy of burst pressure predictions.2 Due to these studies, the focus of this paper is on empirically finding the best model to forecast the depth of a corrosion anomaly into the future while incorporating measurement error. Several different probabilistic models exist such as Growth by Rule which assumes a linear growth and a heuristic corrosion start date, Monte Carlo simulation-based regression models such as Linear Regression and Power Regression, and novel developments from PRCI such as EC 1-2 and EC 1-10. 3 4 5Linear Regression and Power Regression models were both selected after a literature review demonstrating that both models can appropriately be used to deterministically model corrosion in different situations. A study done by Romanoff on 8 different ferrous pipe alloys and steels underground suggested the use of a Power Regression to model corrosion penetration depth.6
