This repository contains a demo setup for NebulaStream showing a use-case for Smart Manufacturing.
Three video sources are modeled that show engine wiring, pipe clipping, and underbody screws from the AutoVI dataset.
NebulaStream reads these video sources, uses an ML inference operator to detect anomalies in the images (i.e. manufacturing errors) and sends the detected anomalies to a Kafka Sink.
For better insight into the demo, we further provide a simple WebApp that can show the three video streams.
Everything is set up and connected using docker compose and works out of the box if you have docker installed.
Requirement: Have docker installed
- Clone this repository
- Open terminal in the repository root folder.
- Run once to set up your local system:
bash ./demo/scripts/build.sh(slow) - Run to start demo
bash ./demo/script/start.sh(quick) - Navigate to http://localhost:8085 and select desired conntections.
- Stop everything: Cancel the running terminal with
CTRL+C
To change the speed at which the images are being send, go to ./demo/config/datagen-*.yaml and adapt the value period.
period defined how often an image is being sent in millisecond.
Restart the system after you changed the period.
To change the kafka sink, go to ./demo/queries/query-engine-wiring.yaml and adapt the values brokers and topic.
You must restart the query to have the change reflected.
- AutoVI dataset: Carvalho et al., The Automotive Visual Inspection Dataset (AutoVI): A Genuine Industrial Production Dataset for Unsupervised Anomaly Detection, 2024 (https://doi.org/10.1016/j.compind.2024.104151)