Using Machine Learning As Part of External Corrosion Direct Assessment to Overcome Data Limitations

Abstract Operators need to keep their pipelines fit for purpose, maximise life and control costs. To achieve these aims, effective Integrity Management is essential, with a clear understanding of the relevant threats to the pipeline asset. External corrosion is one of the main threats faced by operators, costing millions annually in identification, mitigation, and repair. ILI is widely used to identify and size external corrosion. Where in-line inspection (ILI) is not possible, knowledge-based models reliant on data and assumptions for multiple variables are used. Combining the variables that are believed to contribute to corrosion with above ground surveys, are used to identify corrosion “hotspots” for in-field investigation. The process is known as External Corrosion Direct Assessment (ECDA). However, ECDA can present significant uncertainties, often requiring multiple iterations of costly excavations to obtain reasonable confidence in pipeline condition — a major limiting factor to the ECDA approach. To improve data resolution, Virtual-ILI (V-ILI) has been used to incorporate data collected over many years by ROSEN ILI. The Virtual-ILI, when combined with other relevant data such as rainfall, soil type and coating, contains information of corrosion trends across thousands of pipeline segments. Through machine learning algorithms, trained on this historical data, the incorporation of V-ILI has the potential to substantially reduce uncertainty.