Machine learning (ML) approaches allow engineers and data scientists to develop models that learn from collected data and generalize inferences for new situations where data is not already available. While several algorithms and approaches within the ML domain are widely known and currently being applied in pipeline integrity management, others are less known and not being applied as widely in pipeline integrity management as in some other industries. Moreover, a lack of industry-wide guidance on best practices for applying ML to pipeline integrity decisions may be an impediment to understanding...
Machine learning (ML) approaches allow engineers and data scientists to develop models that learn from collected data and generalize inferences for new situations where data is not already available. While several algorithms and approaches within the ML domain are widely known and currently being applied in pipeline integrity management, others are less known and not being applied as widely in pipeline integrity management as in some other industries. Moreover, a lack of industry-wide guidance on best practices for applying ML to pipeline integrity decisions may be an impediment to understanding and adopting ML approaches.
To help address this need, this report describes a set of best-practice guidelines for applying ML to pipeline integrity management that are accompanied by a stress corrosion cracking susceptibility case study.
The core guidelines include a step-by-step process describing best practices for applying ML to pipeline integrity management. Additional informative materials on common ML methods were developed to supplement these guidelines and included in the appendices. A comprehensive review of existing published ML applications in pipeline integrity management was also compiled to provide insights regarding promising opportunities for ML applications, as well as a discussion of common challenges and potential solutions in the pipeline industry that were noted in the literature. Finally, a set of demonstration ML models predicting stress corrosion cracking susceptibility were developed according to the step-by-step procedures laid out in the guidelines to support targeted in-line inspections, direct assessment, and risk modeling. The models were trained and validated using anonymized pipeline attributes and inspection data for approximately 30,000 km of transmission pipelines from six operators in Canada and the United States.
This report is intended to assist pipeline operators and analysts in the development of new ML models, as well as in the evaluation of their ML current practices. Additionally, the guidelines are intended to assist regulators and other stakeholders in the evaluation of ML models produced by industry to ensure appropriate adherence to best practices.