Internal corrosion is one of the 9 main pipeline threats identified in ASME B31.8S that can lead to pipeline failures. However, managing internal corrosion can prove difficult because it’s often unknown whether internal corrosion is present and if present, whether it’s active or passive. The OneBridge Solutions team has collaborated with pipeline operator partners to apply data science and machine learning to identify areas of active internal corrosion and quantify the severity of these identified areas. To this end, two data science models were developed.
Problem
Assessing internal corrosion with inline inspection (ILI) is challenging, as detection and sizing of internal corrosion can be greatly hindered by the cleanliness of the pipe. Individual pit-to-pit based growth rates can provide false positives when assessing for active corrosion on a pipeline due to measurement uncertainty. Additionally, it's harder to obtain field measurements of internal corrosion because of its location. Therefore, operators often implement mitigation strategies based on the assumption of active growth. Alternatively, pipeline operators may employ no mitigation strategies under the false assumption of passive growth, which could lead to a pipeline failure.
Solution
Two models were developed to identify the prevalence and severity of active internal corrosion utilizing data science and machine learning. Both models analyze the change between two or more inline inspections and therefore, require ingestion and alignment via CIM.
Inline Inspection Ingestion & Alignment
First, the inline inspection reports are uploaded into CIM and our ingestion algorithm, powered by machine learning, assigns every feature a standardized classification.
Then, CIM's data science-driven alignment algorithm allows for automated simultaneous alignment of multiple ILIs, regardless of a difference in the inline inspection service provider. By analyzing the aligned weld/joint patterns across ILIs, all modifications of the physical pipe can be tracked by the change in the patterns. Odometer values of the aligned ILIs are then interpolated to a common normalized odometer space based on a single suitable ILI that is used as a reference.
Internal Corrosion Model #1: Analyzing Pipe Replacements as Large Coupons
- Identifying new sections of pipe: Once the inline inspection results are aligned, the model identifies all new sections of pipe and counts the new anomalies present.
- Calculate linear density: The linear density is then calculated by dividing the above values by the total length of the pipeline.
Internal Corrosion Model #2 Applying Statistics to Calculate a “Growth Score”
- Calculate Corrosion Growth Rate (CGR): Once the inline inspection results are aligned, match the metal loss anomalies and calculate a pit-to-pit corrosion growth rate.
- Bin Anomalies: The model finds “bins” with 50+ anomalies, not to exceed 400 ft. Analysis is conducted on matched and unmatched anomalies.
- Calculate a growth score: A growth score or metric is calculated for each of the selected populations from the product of the normalized population mean, standard deviation, and skewness.
- Identify Active Areas = any area with a positive growth score.
- Determine Severity = the number of active areas above the threshold of one standard deviation.
- Calculate Density = total number of corrosion regions with a growth score > zero divided by the total length of corroded pipe in miles.
Results are reported for both matched and unmatched anomalies and include all locations of the identified active areas per pipeline along with active corrosion severity and density. (Unmatched anomalies are considered to have grown from zero depth.
Results
The following provides results from one pipeline operator partner where the IC models were applied to 27 pipeline systems. Figure 1 provides a graphical representation of new pipe contained within a pipeline, acting as a large coupon, with active corrosion per IC Model #1. Girth welds are shown as dotted vertical lines.
Figure 1 (above): Example of a new pipeline section with active corrosion identified by IC Model #1
Figure 2 (above): Identifying corrosion bins by analyzing the change between two ILIs using IC Model #2.
Ranking Active Corrosion on a Pipeline
- The internal corrosion model outputs all areas of active corrosion with normalized odometer ranges.
- In Figure 2 above, the red vertical lines indicate the active area.
- The bin with the largest growth metric value is shown with vertical orange lines. Several overlapping bins with positive growth metric contribute to this area.
- The color scale of the anomalies corresponds to depth normalized to the depth scale of the selected ILI. The high value of the growth metric indicates that the growth in this area is a statistical outlier compared to the other regions across the pipeline.
Figure 3 (above): Areas of active internal corrosion with their corresponding growth metrics (scores) per IC Model #2
Figure 4 (above): Identifying active areas of corrosion on a pipeline. Areas to the right of the red line are above the threshold.
Ranking Pipelines by Internal Corrosion Threat
- The results from all analyzed pipelines systems are integrated in a summary table and rankings are produced based on active corrosion density and severity. An example of active corrosion ranking based on the two measures described above is shown in Table 1, along with counts of internal repairs. A slight positive correlation is observed between repair counts and the two growth metrics
- This analysis was based on only matched anomalies.
- There are 3 more ranking tables that are produced by the model: pipelines ranked by the active corrosion density based on the matched anomalies, and pipelines ranked by the active corrosion severity and density based on unmatched (new) anomalies. Four different and complementary rankings give operators more freedom in making decisions about the most concerning pipelines.
Table 1: Pipelines ranked by active corrosion severity
Conclusion
As opposed to simulated data, the algorithm for Model #2 has been adapted to be used on real ILI data at scale and can be applied to pipelines with one or multiple ILIs. The model has been modified for better selection of active growth cases based on a revision with ~500 pipelines with several thousand ILIs. This model can reliably identify active corrosion areas with minimal influence from ILI system error.
Experience has shown that these two internal corrosion models determine areas of active corrosion with high accuracy, which can help operators evaluate their internal corrosion management programs and mitigation strategies. And because these internal corrosion models rely on analyzing inline inspection data, they can be applied to external corrosion as well.
Stay tuned for more updates on how OneBridge Solutions is applying data science and machine learning to solve the toughest pipeline integrity problems.
Like to learn more? Click here to download the paper presented at last year’s PPIM conference.
