Pipeline anomaly detection at scale
An end-to-end computer vision system detecting corrosion, structural defects and leaks across large volumes of pipeline imagery — built on Azure Machine…
The decision
Where do we send an inspection crew next?
The problem
Thousands of miles of pipeline, much of it through terrain nobody visits often, inspected by manual checks and periodic drone flyovers reviewed by hand. The failure mode that matters is not a missing photograph — it is a defect that was photographed and then missed at the end of a long batch.
Outcome
A production pipeline performing automated anomaly detection at scale, with human review and continuous retraining behind it — reducing reliance on fully manual inspection while leaving domain experts in control of the decisions that carried a cost.