Needle in the Haystack: Applied ML for Inspections of Power Infrastructure
summary
A full‑stack system is built to find rare faults across millions of aerial images captured at five frames per second from helicopter rigs over 58,000 km of transmission lines, holding about 400 terabytes of data. Machine learning runs in batched, rate‑limited workers (local GPUs and some cloud), models emit bounding boxes ingested by Django/PostGIS with GeoDjango, MPTT taxonomies, Django Simple History, an image‑grid human‑in‑the‑loop verifier, PDF generation via XHTML2PDF and portion‑based coverage math that produces actionable reports.
AI-written summary · about
chapters (17)
- 0:00Welcome
- 1:03Django app for power infra
- 2:49System overview and data capture
- 4:20Data funnel and ML pipeline
- 5:43Storage architecture and cloud
- 6:04Application architecture and inferencing
- 7:51Django ORM and data models
- 8:58Review process and image verification
- 9:40Image reports and PDF generation
- 12:11GeoDjango for geographic queries
- 14:02Projections and tower positioning
- 15:27Inspection tracking with Portion
- 16:52Customer-friendly data tables
- 17:58Self-hosted release and ISO compliance
- 19:00Security scanning and package management
- 19:23Database cleanup scripts
- 20:33Debugging Django queries
related talks
30mMaster large-scale ML batch inferencing
30mHow to Empower Utility Vegetation Management
44mML in Production
21mML Infrastructure for Autonomous Vehicles
26mFrom Data to Decision: Monitoring and Improving Water Networks using AI/ML
28mA pythonista’s pipeline for large scale geospatial analytics
44mIndustrial ML – Overview of the Technologies Available for Machine Learning
10mLightning Talk: Demystifying Challenges/Learnings in Converting an Old-school Textile Inspection Machine into a Smart System Using AI/ML