Build a Missing Part Detection System with Computer Vision

Build a missing part detection system that checks every assembly before it ships, from the first labeled image to a live camera on the line. Aadhav Sivakumar, Edge AI Engineer, built this end-of-line check for a carburetor with nine part types to verify: nuts, valves, set screws, and labels. It runs on an NVIDIA Jetson beside two industrial cameras, so it can theoretically run inside a factory without reliable internet. You can pull a nut off and the HMI screen flags it as missing. This type of system is deployed by more and more manufacturing who need to catch missing components before they move further down the production line. Along the way, Aadhav shares the choices that make the model reliable. He trains it on missing parts as well as present ones, varies camera angle and lighting in the dataset while keeping out the upside-down shots that would confuse it, and uses an earlier version of his own model to label new images in bulk. He also explains how neural architecture search matches model size to the frame rate you need, and how the finished Workflow runs on the edge behind an operator screen anyone on the floor can read. = Chapters = 0:00 Intro: End-of-Line Part Checks 1:00 Live Demo: Missing Part Alerts 4:03 Labeling Present and Missing Parts 6:32 Building a Robust Training Dataset 9:03 Auto Labeling with Your Own Model 13:22 Segmentation Labeling with SAM 3 15:57 Training RF-DETR Segmentation 17:10 Augmentation and Train/Test Splits 21:10 Neural Architecture Search and mAP 24:49 Building the Inspection Workflow 26:36 Deploying to NVIDIA Jetson = Resources = RF-DETR Seg launch: https://blog.roboflow.com/rf-detr-segmentation/ RF-DETR on GitHub: https://github.com/roboflow/rf-detr Neural Architecture Search launch: https://blog.roboflow.com/train-with-neural-architecture-search/ Device Manager docs: https://docs.roboflow.com/deploy/device-manager Missing component detection with Roboflow: https://roboflow.com/ai/missing-component-detection Join a future live session: https://roboflow.com/webinar