Drone and crawler ingestion
Imagery and video from aerial and in-line platforms with flight and telemetry logs preserved.
AI intelligence system
PipeIntegrity
From a single inspection to longitudinal asset integrity.
A vision system that detects and localises pipeline deterioration, then tracks each defect across inspection cycles so maintenance is prioritised by progression rather than by appearance.
- Sector
- Oil & Gas · Asset Integrity
- Year
- 2024
- Duration
- 9 months
- Team
- 7 people
- Platforms
- Web · Field tablet
- Status
- In production
Cycle-to-cycle
Defect identity
Same flaw recognised across campaigns
GPS-located
Every finding
Positioned on the asset, not just in a frame
Rate of change
Ranking basis
Not present severity alone
Terabyte-scale
Campaign ingestion
Per inspection run
01 — The problem
Drone and crawler inspections generated terabytes of imagery per campaign. Analysts reviewed it, wrote a report, and filed it. The next campaign started from scratch.
This meant the single most useful fact about a defect was unavailable: whether it was getting worse. A stable twenty-year-old corrosion patch and a crack that had doubled since spring looked identical in a report that only described the present.
02 — What we did
We made identity across time the core problem. Defects are localised to real-world coordinates on the asset, so the same physical flaw is recognised as the same entity across campaigns even when captured from a different angle, altitude and light.
Once a defect has a history, prioritisation changes completely. The system ranks by rate of change and projected time-to-threshold rather than by current severity, which is what an integrity engineer actually needs to schedule work.
A defect without a history is a photograph. With one, it is a forecast.
Interface
What it looks like in use
Two of the screens that carry the most weight in daily use, rebuilt here from the production design system.
CR-1187 · KP 42.318GROWING
2020 · 12mmMONITOR
2022 · 19mmMONITOR
2024 · 34mmSCHEDULE
Threshold in 14 months
A defect with a history is a forecastThe same physical flaw is matched across campaigns, so ranking is driven by growth rate rather than present appearance.
Line 4 · campaign 2024-Q3GPS LOCATED
KP 42.3 · crack · growingPRIORITY 1
KP 61.7 · corrosion · stablePRIORITY 6
KP 18.2 · coating lossPRIORITY 9
318TRACKED DEFECTS
5CAMPAIGNS
Findings sit in physical spaceEvery detection is positioned on the asset rather than in a frame, which is what makes cross-campaign identity possible.
Capabilities
What the system does
Grouped by the job each set of capabilities exists to do, rather than by which team built it.
Capture and ingestion
01 · From the fieldGeospatial registration
Frames registered against the asset model so every pixel has a position on the pipeline.
Capture quality gating
Blur, exposure and coverage gaps flagged during the campaign while the crew is still on site.
Field tablet review
Crews confirm coverage and mark points of interest before leaving the location.
Detection and measurement
02 · What is thereCorrosion and crack detection
Segmentation models trained per surface and coating type, since a coated line and bare steel behave differently.
Dimensional estimation
Defect extent estimated in physical units using capture geometry rather than reported in pixels.
Classification
Findings typed against the client's integrity taxonomy so they map onto existing procedures.
Confidence reporting
Low-confidence detections routed to analyst review instead of entering the record unflagged.
Longitudinal intelligence
03 · The differentiatorHistorical comparison
Each defect matched to prior campaigns by position and shape signature, with the match evidence shown.
Progression tracking
Growth rate computed per defect across the full inspection history.
Time-to-threshold projection
Projected date at which a defect reaches an intervention threshold, with uncertainty stated.
Maintenance prioritisation
A work list ordered by projected risk and access cost rather than by current appearance.
Architecture
How it is put together
Layer by layer, with the reason each one exists — because the reason is usually the interesting part.
01Ingestion pipeline
Bulk campaign upload with automatic association of imagery to telemetry and asset segment.
- S3
- Batch processing
- EXIF/telemetry fusion
02Registration
Photogrammetric alignment of frames to the asset model producing per-detection world coordinates.
- Photogrammetry
- PostGIS
03Vision models
Segmentation and classification per surface class with calibrated confidence outputs.
- PyTorch
- Segmentation
- Calibration
04Identity matching
Cross-campaign defect matching on position, shape descriptor and surrounding context.
- Spatial matching
- Shape descriptors
05Integrity console
Map-based workspace with defect history, progression charts and the prioritised work list.
- React
- MapLibre
- Deck.gl
Technology
Vision
- PyTorch
- Segmentation models
- Photogrammetry
Geospatial
- PostGIS
- MapLibre
- Deck.gl
- GDAL
Backend
- Python
- FastAPI
- PostgreSQL
- Celery
Platform
- S3
- Kubernetes
- GPU nodes
- Airflow
Outcome
What changed
Measured against how the operation ran before, not against a benchmark chosen after the fact.
-
Inspections stopped being disposable. Each campaign now adds to a defect history instead of replacing the last report.
-
Prioritisation is based on movement. Engineers schedule against growth rate and projected thresholds.
-
Coverage gaps are caught on site. Quality gating happens while the crew can still re-fly a section.
-
Findings sit in physical space. Every defect has coordinates on the asset, not a frame number.
How it ran
-
Months 1-2
Data and taxonomy
Assessed historical campaigns and mapped the client integrity taxonomy onto model outputs.
-
Months 3-4
Registration
Built geospatial registration first, because cross-campaign identity depends on it entirely.
-
Months 5-6
Detection models
Per-surface segmentation with confidence calibration and analyst review routing.
-
Months 7-8
Progression engine
Cross-campaign matching, growth computation and threshold projection.
-
Month 9
Console and rollout
Map workspace, prioritised work list and integration with the maintenance system.
Related work
From the same practice
Manufacturing & Production Quality
LineTrace
From defect detection to defect causality.
A multimodal system that correlates line video, machine telemetry, batch records and operator actions to explain why a defect happened, not merely that it did..
2025 · 10 months
Telecommunications & Network Operations
NetCause
Not alarm detection. Failure hypothesis reduction.
A correlation engine that turns an alarm storm into a short ranked list of probable root causes by reasoning over network topology and fault propagation rather than alarm counts..
2025 · 8 months
Pharmaceutical & Regulated Manufacturing
BatchResolve
From deviation to evidence, with the audit trail intact.
A deviation investigation assistant that assembles batch records, equipment telemetry, inspection imagery, SOPs and historical deviations into an evidence-linked case file a quality unit can defend to a regulator..
2025 · 11 months
Something similar?
Running into the same problem PipeIntegrity solved?
Tell us what you run. We will reply within two business days with what we would build for your situation — and, just as usefully, what we would leave out.