Digital AIUpdated 24 October 20268 min read

Edge Computing in Manufacturing: Use Cases and Skills

Edge computing in manufacturing runs software and AI models on hardware next to the machine instead of in a distant data centre. This guide covers when edge beats cloud, the typical architecture, hardware options, security, and the skills UK engineers need.

Edge ComputingAIManufacturingReal-Time ProcessingIndustrial IoT
Technician using a handheld diagnostic tool beside a server rack in a data centre

Key takeaways

  • Edge computing runs processing and AI beside the machine for low latency, lower bandwidth use, resilience and data control; the cloud still handles training and fleet-wide reporting.
  • A typical stack is sources, connectivity (OPC UA or MQTT), an edge runtime running containers, and an upstream broker or cloud.
  • The edge advises and records; the PLC keeps deterministic, safe control of the machine.
  • Platforms include Siemens Industrial Edge, NVIDIA Jetson modules, industrial PCs and microcontrollers running LiteRT.
  • Security and lifecycle management matter: segment networks, avoid internet exposure and plan updates, as NCSC guidance and IEC 62443 emphasise.

Edge computing in manufacturing means running software, data processing and AI models on computers located at or near the machine, such as an industrial PC in the cabinet, instead of sending every reading to a remote cloud. It gives faster responses, lower network traffic and continued operation if the internet connection drops.

Why run AI at the edge instead of the cloud?

Four practical reasons push work to the edge:

  • Latency. A vision system checking parts at line speed cannot wait for a round trip to a data centre. Decisions in tens of milliseconds need local processing.
  • Bandwidth. A single camera or a vibration sensor sampling at kilohertz rates produces far more data than you want to upload. The edge device filters and summarises, sending only results and exceptions.
  • Resilience. If the site's internet link fails, local analytics and buffering continue, and data is forwarded when the connection returns.
  • Data control. Many manufacturers prefer that raw process data, images and recipes stay on site, sending only aggregated figures to the cloud.

The cloud still has jobs: training large models, comparing sites, long-term storage and fleet dashboards. Most real architectures are hybrid: train in the cloud or on a workstation, run (infer) at the edge, and report upwards.

Wall-mounted industrial PC built on an ATX motherboard
Wall-mounted industrial PC built on an ATX motherboardPhoto: David Lippincott for Chassis Plans, CC BY 3.0, via Wikimedia Commons
Industrial Edge simply explained.

What does an edge architecture look like?

A typical factory edge stack has four layers:

  1. Sources: PLCs, drives, sensors and cameras on the plant network.
  2. Connectivity: drivers and protocols that read the data without disturbing the control system, usually OPC UA or MQTT. Our guide to OPC UA for industrial IoT explains the options.
  3. Edge runtime: a small server, industrial PC or gateway that runs applications, often as containers, such as data collection, OEE calculation, anomaly detection or a vision model.
  4. Upstream: an MQTT broker, historian, MES or cloud platform that receives results.

The edge sits on the information technology side of the control network boundary, not inside the control loop. The PLC remains responsible for deterministic, safe machine control; the edge device advises and records.

Which edge hardware and platforms are used?

Options range from rugged commercial products to hobby boards used for learning.

  • Siemens Industrial Edge bundles hardware, software and a central management system, supports Siemens, third-party and custom apps, and lists use cases such as OEE monitoring, remote maintenance and intelligent maintenance. Siemens says it also supports deploying AI models for tasks such as defect detection, and cites IEC 62443-4-2 certification for the device security.
  • NVIDIA Jetson modules are compact GPU modules for edge AI; NVIDIA positions the Orin family for manufacturing and robotics, with the entry-level Orin Nano listed at up to 67 TOPS (a vendor figure).
  • Industrial PCs and gateways from many vendors, running Linux and Docker, are the workhorse for data collection and light analytics.
  • Microcontrollers can run very small models for vibration or sound classification using LiteRT (the successor to TensorFlow Lite), which is useful for battery-powered sensors.
  • Raspberry Pi and similar boards are fine for learning and prototyping, but not for production in harsh cabinets without proper industrial packaging.

Choose by the workload: simple data forwarding needs a gateway, image analysis needs a GPU or accelerator, and a fleet of sites needs central management of updates and certificates.

How do you decide between edge, on-premises server and cloud?

Ask four questions about each workload. How fast must the answer arrive? If a reject gate needs a decision inside one conveyor pitch, run it at the edge. How much raw data is produced? Gigabytes of images per shift favour local processing. What happens if the link fails? If production must continue, keep the logic local. And who needs the result? A maintenance planner across five sites needs a central view, so summaries should still flow upwards.

A sensible first rule is to keep anything time-critical or bandwidth-heavy at the edge, keep long-term history and model training central, and keep anything safety-related in the PLC and safety system, not in either.

What are the manufacturing use cases?

  • Visual inspection at line speed, with the model running on a GPU module beside the camera; see our computer vision inspection guide.
  • Predictive maintenance where a gateway computes features from high-frequency vibration data and sends only alerts and summaries.
  • OEE and downtime tracking calculated locally and published to dashboards.
  • Energy monitoring with local anomaly detection on meter and compressor data.
  • Protocol translation between older equipment and modern systems.

What does a first edge project look like?

Pick one line and one question, for example "which of these three filling machines loses the most time to micro-stops?" Install a small gateway on the plant network, read the machine state bits over OPC UA or MQTT without touching the PLC program, and calculate availability locally. Publish the result to a dashboard and a broker. Once the data is trustworthy, add a second app, such as anomaly detection on motor current. This staged approach gives value early, tests your network and security design on something low-risk, and builds the team's confidence before a vision or control-adjacent project. Our article on IO-Link and smart sensors shows how to get richer data from the sensors themselves.

What are the risks?

More devices mean more to manage. Plan for software updates, certificate renewal, monitoring, backups and decommissioning. Keep the edge network segmented from the control network and limit what can connect. The UK's National Cyber Security Centre published an item in August 2026 highlighting disruptive cyber activity and the risk from internet-exposed systems and edge devices, a reminder that an edge gateway reachable from the internet is a liability. Review the NCSC operational technology guidance and, for industrial products, the ISA/IEC 62443 series before you connect anything.

Also remember that a model deployed at the edge can drift. Version your models, log predictions, and agree who retrains them.

What skills do edge and industrial AI roles need?

A strong profile combines automation knowledge with a few IT skills: networking, Linux, containers, MQTT and OPC UA, plus basic Python and model deployment. These skills sit between OT and IT roles, which is why they are hard to hire for.

Start free

Start with the free AI for Industrial Automation course to see where edge fits beside PLC and SCADA, work through Docker and Containers for Engineers, then try the Machine Learning Engineer job simulation. If you want a structured self-paced route with graded projects, the EDWartens AI Module is available; check its page for the current price.

Start free

Learn AI and machine learning free, then practise the job

Free job simulation

Machine Learning Engineer

Train, evaluate and document a model, and catch target leakage.

The first five days are free, no card needed.

Live UK jobs

AI and ML jobs in the UK

Updated daily, with licensed visa sponsors marked.

Browse all engineering jobs

Frequently asked questions

What is edge computing in manufacturing?

It is running software and AI models on computers at or near the machine, such as an industrial PC in the cabinet, rather than in a remote data centre. It gives faster responses, reduces network load and keeps working if the internet link fails.

What is the difference between edge and cloud computing in a factory?

Edge processes data locally with low latency and keeps raw data on site. Cloud offers large-scale storage, model training and multi-site comparison. Most factories use both: train in the cloud or on a workstation, run models at the edge, and send summaries upwards.

What hardware is used for edge AI in factories?

Common choices are industrial PCs and gateways, GPU modules such as NVIDIA Jetson for vision, managed platforms such as Siemens Industrial Edge, and microcontrollers for tiny models. The right choice depends on the workload and on how many sites you must manage.

Is edge computing safe for control systems?

It can be if it is segmented from the control network, kept off the public internet, patched and monitored. The edge should advise or record, while the PLC keeps safe control. Follow NCSC operational technology guidance and the ISA/IEC 62443 series.

Which skills do I need for edge computing jobs?

Networking, Linux, containers, MQTT and OPC UA, Python and model deployment, combined with PLC and SCADA knowledge. Free courses on EDWartens UK cover industrial data, Node-RED, Docker and MLOps.

Sources

  1. Siemens Industrial Edge
  2. NVIDIA Jetson modules (vendor figures)
  3. LiteRT (formerly TensorFlow Lite), Google AI Edge
  4. Operational technology collection, National Cyber Security Centre (checked 24 October 2026)
  5. ISA/IEC 62443 series of standards, ISA
  6. MQTT: the standard for IoT messaging

Share this article

About the author

EDWartens Team

EDWartens Instructional Team

The EDWartens instructional team brings together UK-based automation engineers, certified Siemens and Rockwell specialists, and industry practitioners with decades of combined experience in PLC programming, SCADA engineering, and industrial AI.

View all articles by EDWartens

Ready to become job-ready?

Start with a free course today, or train live in Siemens PLC, HMI and SCADA on the Professional Module.