Verify Part Presence: Improving Quality Control with Computer Vision

​A missing bolt or an unseated seal is easy to overlook at the end of an assembly line, but can be expensive later. When defects escape the factory, it results in additional costs from reworks, warranty claims, safety recalls, and more.

​In this live session, Aadhav Sivakumar, Edge AI Engineer at Roboflow, will show how computer vision can be used to automate inspections and catch missing components in a manufacturing line. He’ll demonstrate a live end-of-line part presence check he built from the ground up. The system watches a carburetor kit and an intake manifold, verifies components that should be there, and alerts a human viewer when a part is missing.

​We will cover:

• ​Training a model to see missing components: See the strategies used when annotating and training a custom computer vision model to detect the absence of components.

• ​Making it run in real time: Learn about the strategies and technology used to ensure the system runs within the computing constraints of an NVIDIA Jetson, like TensorRT compilation, camera input resolution, and Neural Architecture Search (NAS).

• ​Designing the Human Machine Interface (HMI): How to present live detections so an operator can tell at a glance what the model sees and why a part was flagged.