Edge AI: artificial intelligence running at the machine
AI models run on an industrial computer next to the line: visual inspection with cameras, anomaly detection from vibration and current, results sent to the PLC and monitoring software. Uptrol delivers this project by project.
Why process at the machine
- Manual visual inspection is tiring, varies between shifts and leaves no evidence.
- Streaming video to the cloud costs bandwidth and adds latency, too late to reject a bad part.
- Plants do not want images of their lines to leave the local network.
- Surface defects vary too much to describe with fixed image-processing rules.
How Uptrol does it
- 01
Feasibility
Review real defect samples, line speed and the smallest defect to catch; agree acceptance criteria.
- 02
Imaging and data capture
Choose camera, lens and lighting; capture images or sensor signals on the line.
- 03
Train and optimise
Combine classic image processing with deep learning; optimise the model to run in time on the edge hardware.
- 04
Deploy and monitor
Connect to the PLC to reject bad parts, run in parallel with manual checks, track accuracy after handover.
Typical use cases
Visual inspection
Missing parts, wrong assembly, scratches, print defects, foreign objects.
Code reading and counting
Barcodes, QR, DataMatrix, expiry dates; counting parts on conveyors.
Anomaly detection
Learn a machine’s normal vibration and current and flag anything unusual.
Local processing
Images and raw data stay in the plant; only results and defect images leave.
Line integration
OK/NG results to the PLC via digital I/O, Profinet, EtherNet/IP or Modbus TCP.
Quality traceability
Defect images stored, statistics per shift, feeding the quality part of OEE.
Every Edge AI project starts with a small test on real samples. Uptrol only proposes a full deployment when the test meets the criteria both sides agreed.
A good fit when
- Defects are visible in images and real defect samples are available for testing.
- Manual inspection is a bottleneck or a source of customer complaints.
- Critical machines need early warning but have no vibration monitoring yet.
What do you need to monitor or automate?
Tell us about your plant, station or server room. We will reply with a concrete approach and next step.