1BGC Engineering, Victoria, Canada, 2Pembina, Calgary, Canada, 3Cambio Earth, Calgary, Canada, 4Cambio Earth, Vancouver, Canada, 5Cambio Earth, Victoria, Canada
Inertial Measurement Unit (IMU) bending-strain data collected during in line inspections (ILI) are widely used to monitor external force related and specifically landslide induced deformations. Previous work demonstrated that machine learning (ML) can effectively prioritize individual bending-strain features associated with landslide interactions from a single inspection. However, identifying movements between inspections remains a significant challenge, as accurate run-to-run comparison requires precise alignment of ILI datasets. Chainage misalignment, odometer drift, wheel slip, and differences in vendor processing can introduce sufficient misalignment to obscure or falsely indicate pipeline movement. This study describes an integrated workflow combining automated run-to-run alignment with ML-based strain feature classification to enable source-attributed detection of pipeline movement between inspections. The workflow first aligns ILI datasets in a multi-stage framework that estimates a smooth warping compensating for both systematic chainage drift and localized distortions, while preserving genuine pipeline deformation. This allows subtle changes in bending strain to be distinguished from alignment errors. Once aligned, bending-strain signals from successive runs are differenced to isolate candidate pipeline movement features, which are screened using a convolutional neural network (CNN) classifier to identify changes associated with landslide-related deformation. By combining signal alignment with automated ML-based screening, the workflow enables efficient comparisons between inspections acquired under different conditions, including direct comparison of inspections processed by different vendors. The approach extends previous ML-based strain feature classification into a complete run-to-run change detection framework, providing pipeline operators with an automated method to identify, prioritize, and monitor new or evolving geohazard-related deformation. This workflow will reduce manual review effort and support earlier identification and enhanced monitoring of landslides.
Available after the event



