Across the Pond
summary
Tenaris deployed a computer-vision safety gate using YOLO fine-tuned on ~2,000 field images, served as a Python REST container on Kubernetes with ~300 ms CPU inference and only about two false positives in ~70,000 test images. To address 400–800 ms cross‑continent latency, a US remote worker was federated to the Italy cluster via kubelet/cloud‑provider flag changes, adjusted node monitor timings, an Nginx hostPort reverse proxy and local failover; this reduced median latency by 67% and cut timeout incidents to two over two months.
AI-written summary · about
chapters (15)
- 0:00Project intro
- 3:31Tenaris & Sigap context
- 5:40Industrial drilling pipe problem
- 8:30ML for tip detection
- 11:00Model & system requirements
- 13:53YOLO & REST architecture
- 16:43Kubernetes intro
- 19:31Intercontinental latency
- 21:17Remote worker federation
- 24:46Remote node readiness
- 26:12App deploy on worker
- 29:42Low-latency reverse proxy
- 31:31Testing & validation
- 34:03Operational results
- 35:27Q&A
related talks
40mPython, YOLO and Kubernetes: Super-Vision for an Industrial Plant
06mEdge Kubernetes in Action: From Connected Vessels to Industrial Au...
25mEdge-to-Core Machine Learning and Data Pipeline on k8s
22mBuild and Monitor Machine Learning Services in Kubernetes
37mRunning Kubernetes in a Manufacturing Line
34mHow to Survive to High Demanding Services
44mEssential-8 Security in Practice with Kubernetes
34mOn Data Locality in Kubernetes