128

Big data, machine learning and artificial intelligence: these buzzwords invariably create a lot of hype but do they actually live up to it? TC Energy’s threat management team set out to answer that question by curating a comprehensive dataset of pipeline integrity digs in-ditch findings to create a machine learning model for predicting SCC on its transmission pipeline assets. In collaboration with Pipeline-Risk (PLR), results from 1800+ digs since 2012 containing approximately 27,000 data points were consolidated. A classification-based machine learning model was trained on a subset of the consolidated dig data, and its performance was then evaluated using the remaining, unseen portion of the dataset. The results exhibited an impressive predictive efficacy of up to 90% for predicting likelihood of SCC on TCE transmission pipelines. This model is now utilized in assisting in SCCDA dig site selection workflows and prioritizing ILI assessments resulting in significant efficiency gains. The most important factors contributing to the success of this model were a robust, and highly organized enterprise data infrastructure, SME-driven criteria and an effective team structure allowing a comprehensive threat management approach. This paper will present key data strategies to ensure your organization is optimally positioned with people and processes to maximize value created by the computational prowess of machine learning.

KEYWORDS: SCC Threat Management, Machine Learning

Read/Download PaperGradeAvailable after the event