Bring Deep Learning to Your Keyence Cameras

Train deep learning models on images from your Keyence CV-X and VS series systems and run them next to the line you already have. Roboflow catches the variable defects built-in tools miss, and sends pass or fail results back to the same PLC.

A grid of 16 identical silver metal machined parts on a black tray, with one part highlighted by an AI detection showing "missing thread - 96%" on a corner hole.

Keyence Cameras AI Vision Solution

Keyence vision systems give factory floors icon-driven setup and rugged cameras. But the built-in tools have a ceiling: when parts vary, lighting drifts, or defects refuse to follow a pattern, teams need deep learning they can train on their own data. Roboflow lets you do that with the images your Keyence systems already capture.

The Keyence CV-X series (opens in a new tab) is Keyence's flagship all-in-one vision system, with camera options up to 21 MP and 64 MP for high-resolution inspection. The newer VS series (opens in a new tab) packs built-in AI and optical zoom into an IP67 smart camera, and the IV series vision sensors cover simpler OK/NG checks.

  • Featured: CV-X series controllers and cameras, VS series smart cameras
  • Attributes: All-in-one controller plus camera, icon-driven setup, factory-hardened
  • Integration: Images over FTP or from your archive, results to PLCs over EtherNet/IP, PROFINET, or OPC UA

How to Use Roboflow with Keyence Cameras

Keyence cameras run through Keyence controllers, so Roboflow works with the images they produce. There are three common ways to set it up:

  • Send images over FTP: Keyence publishes an FTP image output guide for the CV-X. Point that output at a PC on the line, and a small bridge app sends each image to your Roboflow Workflow and writes the result to the PLC. Our guide to connecting an FTP smart camera to Roboflow and a PLC walks through the full setup.
  • Train on your image archive: Save OK and NG images from the controller to a PC or SD card, label the defects your current tools miss, and train a model before you change anything on the line.
  • Add a camera where rules fail: For the one station that needs deep learning, add a GigE camera and an edge device running Roboflow next to the existing Keyence system, and keep the rest of the line as it is.

Tip: save the borderline images too, the barely good and barely bad parts. They teach the model where your pass or fail line sits better than obvious examples do.

Example App: Casting Defect Detection with Keyence Images

Here is how a foundry or machining plant can use images from an existing CV-X station to catch casting defects that vary too much for rules:

  • Set the CV-X controller to send every image to a PC on the line over FTP.
  • Collect images across shifts and lots, and label porosity, cracks, cold shuts, and inclusions in Roboflow.
  • Train an RF-DETR (opens in a new tab) object detection model on those images in Roboflow.
  • Build a Workflow that flags any casting with a defect in a critical zone and sends a pass or fail to the PLC.
  • Run it on an edge device next to the line, so the result reaches the reject gate in time.

See casting defect detection for the full use case and metal surface defect detection for machined faces. Short on defect images? Start with few-shot defect detection.

More Applications for Keyence Cameras

The images Keyence controllers capture are yours. Your team owns the model, the data, and the ability to retrain as your parts change.

Run vision models on Roboflow hardware

Roboflow Edge computers and the AI1 Smart Camera arrive ready to run your models. Each one ships with RoboflowOS, the operating system we build and maintain, so fixes and security patches reach your devices over the air without a site visit.

Roboflow Edge computers
  • Settings that survive a restart

    RoboflowOS rebuilds static addresses, gateway, DNS and camera-port tuning from a persistent store each time the device starts.

  • Updates over the air

    Security patches, camera driver fixes and NVIDIA platform upgrades come in from the Roboflow platform, separate from your models and Workflows.

  • Tested on the hardware you get

    We build the inference runtime too, so we test every OS image against it on the same boards we ship.

  • Locked down by default

    The OS image is read-only and ships without SSH, a desktop or a local login. Your team reads logs, checks streams and views telemetry in Deployment Manager, the Roboflow tool for managing devices remotely.

  • Five years of updates included

    Every appliance comes with five years of RoboflowOS updates from the purchase date, plus a one-year hardware warranty.

RoboflowOS on NVIDIA Jetson

  • Smart Camera

    AI1

    The AI1 puts a camera, a light and a computer in one housing. It pairs an 8 MP (3840 x 2160) camera and LED ring with an NVIDIA Jetson Orin NX 8GB. Teams use it to prove out an edge vision project on a fixed inspection view, counting parts a few inches away or spotting defects dozens of feet away.

    Memory
    8 GB
    Cameras and networking
    Integrated 8 MP camera and LED ring1 x GbE
    1 stream
  • Roboflow Edge

    RF-NX16

    The RF-NX16 handles stations with several cameras. It has six PoE ports on a shared 150 W budget, two more GbE ports and an NVIDIA Jetson Orin NX 16GB.

    Memory
    16 GB
    Cameras and networking
    6 x GbE PoE with a 150 W total budget2 x GbE
    1–4 streamsRF-NX16 Spec Sheet
  • Roboflow Edge

    RF-AGX64

    The RF-AGX64 runs higher-resolution cameras and bigger models. It has 64 GB of memory on an NVIDIA Jetson AGX Orin, four PoE ports and two 10 GbE ports.

    Memory
    64 GB
    Cameras and networking
    4 x GbE PoE2 x 10 GbE
    4+ streams, line scanRF-AGX64 Spec Sheet
  • Roboflow Edge

    RF-THOR128

    The RF-THOR128 is our most powerful appliance and runs on the NVIDIA Jetson T5000 with 128 GB of memory. It has one 5 GbE port and a 100 GbE QSFP28 port that breaks out to four 25 GbE links, and its cameras connect through a PoE switch.

    Memory
    128 GB
    Cameras and networking
    1 x 5 GbE and 1 x 100 GbE QSFP28Breaks out to 4 x 25 GbE
    8+ streams, edge VLM inferenceRF-THOR128 Spec Sheet

One GbE PoE camera port

Already running Keyence cameras? Connect them to an RF-NX16, RF-AGX64 or RF-THOR128 and manage the whole station from Roboflow.

Not sure which one fits your line?

Send us your cameras, Workflow and line rate, and we’ll create a recommendation for you.

AI for Keyence Cameras Workflows

Use Keyence cameras? Explore Workflows you can get started with to create computer vision apps in minutes. Easily deploy computer vision models with Roboflow Inference alongside your Keyence system. Integrate AI by connecting to existing PLC and HMI solutions via standard protocols like MQTT or OPC-UA.

How to Setup Keyence Cameras with AI

Teams often add Roboflow when a Keyence inspection hits its accuracy ceiling. Train a model on images archived from your own CV-X or VS system, then run it at the edge on an NVIDIA Jetson or industrial PC alongside your existing Keyence hardware, no rip-and-replace required.

Talk to a Vision AI engineer about your Keyence inspections.

Bring us the inspection your Keyence tools can't hold, and we'll map a working solution.

Ask us about:

  • Solution architecture that works alongside your Keyence controllers and PLCs
  • A live demo on images from your own Keyence systems
  • Deployment options: edge, on-prem, or air-gapped, with results sent to your PLC or MES
  • ROI modeling against escapes, false rejects, and manual inspection time

Over 80,000 organizations build with Roboflow.

  • Rivian
  • Pella
  • Chobani
  • USG Corporation
  • BNSF Railway
  • American Woodmark
Start where you are

Additional resources