From SCADA to the Edge: Where Should Your Control Data Be Processed?

Every industrial site that adds edge computing runs into the same architectural question: if the edge node can process data, what is the SCADA system still for?

The answer is that they solve different problems, and the useful way to think about a modern plant is not edge versus SCADA but a set of tiers, each with its own latency budget, its own data volume and its own reason to exist.

The Traditional Picture

Classical industrial control is a hierarchy. Sensors and actuators sit at the bottom. PLCs and controllers read them and execute deterministic control loops in milliseconds. Above that, SCADA polls the controllers, presents the plant to operators, logs historical data and raises alarms. Above that sit MES and ERP systems, dealing in production orders and business processes.

It works, and it has worked for decades. Its limitation is the polling model at the SCADA layer. SCADA asks each controller for the values it has been configured to ask for, at an interval that the network and the server can sustain. Everything else the machine knows is discarded.

That was a reasonable trade when storage and bandwidth were expensive. It is a poor trade when the questions being asked — why does this machine degrade, which parameter predicts this failure, what actually happened in the four seconds before the fault — need data at a resolution the polling interval never captured.

What the Edge Tier Adds

An edge node is a computer next to the process. It is not a replacement for the controller and it does not run the safety-critical loop. What it does is process data at the source, at full resolution, and decide what deserves to travel further.

Four things make that worthwhile.

Latency. A decision that must be made in milliseconds cannot involve a round trip to a remote server. Machine-vision inspection rejecting a part on a moving line is the clearest example: by the time a cloud service replies, the part is gone.

Bandwidth and cost. A vibration sensor sampling at kilohertz produces more data in a day than a site's uplink can carry in a week. Processing locally and transmitting features — RMS levels, spectral peaks, anomaly scores — instead of raw waveforms reduces the volume by orders of magnitude while keeping the information that matters.

Availability. A remote pumping station, a desalination plant, an offshore facility — connectivity is not guaranteed. Local processing continues, buffers results, and synchronises when the link returns. A system that stops working when the link drops was never suitable for that site.

Data protection. Some data should not leave the plant. Processing locally and exporting only aggregates is a straightforward way to satisfy that requirement.

Dividing the Work

A workable division looks like this.

Controller tier — microseconds to milliseconds. Deterministic control loops, interlocks, safety functions. This stays exactly where it is. Nothing about edge computing changes the argument for a real-time controller executing the control loop.

Edge tier — milliseconds to seconds. Protocol conversion between fieldbus and Ethernet; high-rate data acquisition and feature extraction; machine-vision inspection; local anomaly detection; store-and-forward buffering; local HMI; data normalisation before anything is sent upstream. This is where an industrial PC earns its place.

Plant tier — seconds to minutes. Supervisory control, operator interfaces, alarm management, the historian, plant-wide coordination. SCADA's job, and still SCADA's job.

Enterprise tier — minutes to months. Cross-site analytics, model training, production planning, reporting. Cloud or data centre, working on curated data rather than raw signals.

The pattern is consistent: the further from the process, the longer the acceptable latency and the more aggregated the data.

What the Edge Node Has to Be

Because it sits on the plant floor rather than in a server room, the hardware requirements are unlike a server's.

Fanless and sealed. Dust, vibration and no maintenance access. Die-cast aluminium chassis acting as the heatsink.

Fieldbus-capable. The edge node's most common job is talking to equipment that predates Ethernet. RS-232/422/485 ports for Modbus RTU, and multiple Ethernet ports so that the OT network and the IT network stay separate. Our IBOX-1226 provides three 2.5 G Intel ports and six COM ports; the IBOX-3226 scales that to 12th-generation Core processors with up to 64 GB of RAM.

Enough compute for the analysis, not more. Protocol conversion and data logging need very little — an IBOX-1326 on an N100 is ample. Machine vision or on-device inference needs considerably more, and that is a sizing exercise, not a default.

Resilient to power loss. Auto power-on after AC recovery, a watchdog timer, and storage that tolerates unexpected shutdown. Plant power is not clean.

Remotely manageable. An edge node inside a cabinet on a factory floor should not require a physical visit for a software update.

Segmentation Is Not Optional

Connecting operational technology to information technology networks is exactly how OT networks get compromised, and the edge tier is where the boundary is enforced.

In practice that means the edge node has separate physical interfaces for the OT and IT sides, the OT side is not routable from the corporate network, and a firewall appliance sits at the boundary with explicit rules about what may cross it. Compact multi-port appliances such as the N1141 or N3161 make this practical at the cell or line level rather than only at the site perimeter.

Treat every connection as deliberate. The value of plant data does not justify an unmanaged path into the control network.

Start with One Question

The mistake is deploying edge infrastructure as a strategy. The successful pattern is narrower: pick one problem that the current architecture cannot answer — a recurring fault nobody can explain, a quality check that is still manual, a remote site whose data arrives days late — and solve that.

The infrastructure you build to solve it will tell you what the next one needs.

If you are working out what belongs at the machine and what belongs upstream, we can help you match the hardware to the answer — fanless industrial PCs with serial and multi-Ethernet I/O, compact network appliances for segmentation, and platforms sized for local analytics where the workload justifies it.

Related Articles:

Understanding Edge Computing: The Future of Data Processing

Industrial PCs: The Backbone of Industry 4.0 Innovation

RS-232, RS-422 and RS-485: Why Serial Still Runs Modern Industry

Contact Us

IWILL MENA

Email: sales@iwillmena.com

Call: +852 914 61951