Recording of GTC 2020 Online Session from Seagate presenting their Edge RX Platform for Smart Manufacturing using Machine Learning at the Edge in their disk factories. They present the real challenge and solution of finding faults in the production of 1 Billion+ read/write heads for disks using supervised deep learning on images.

This is a good example of the challenges that machine learning can help to solve in the manufacturing industry for high volume and high quality products. The talk explains well the problem space, solution approach and practical insight into deployment, model management, operations, scaling and performance considerations on the "Edge" / factory site.

The software and hardware stack used includes Keras, Tensorflow, NVIDIA TensorRT, Kubernetes and NVIDIA GPUs for learning and inference.

Check out GTC at https://www.nvidia.com/en-us/gtc/

Presenters:

- Addis Woldesenbet, Seagate Technology
- Chao Feng, Seagate

Session Description:
Proving the advantages of AI on a pilot manufacturing line has challenges, but it's been demonstrated by several companies in various industries. Deploying such technologies on a global scale across three continents requires solving a host of additional challenges, requiring collaboration between data scientists, manufacturing operations, and IT solutions providers. We at Seagate will share the lessons we learned from the past year of innovation.

Chapters:

- Introduction to Seagate: 00:00
- Seagate Edge RX Platform: 03:42
- Overview of Challenge and Solution: 05:13
- Technical Introduction to Problem Space: 08:23
- Automated Inspection Rule Based Approach: 10:58
- Approach with Supervised Machine Learning: 12:55
- Model Training - Segmentation Model: 14:44
- Model Deployment Workflow: 19:47
- Usage of NVIDIA TensorRT Inference Server: 21:32
- Inference Performance Comparison TensorFlow Serving vs. TensorRT: 23:58
- Scaling to Production: 25:31
- Overview Factory Machine Learning Platform: 27:16
- Seagate Edge RX Details: 33:00
- Application Architecture - Client Side: 37:37
- Application Architecture - Server Side: 39:26
- Summary: 43:26