Edge AI and IIoT
Why SCADA and PLC data are not enough to optimise your process
Almost every production site has a PLC that controls the machine and a SCADA system that shows what is happening. That data is reliable and there is a lot of it. Anyone who tries to optimise with it still gets stuck. Here is why, and what you put next to it.
What PLC and SCADA are built for
A PLC runs a fixed program: measure, compare with a setpoint, adjust. SCADA shows those values, raises alarms and lets an operator step in. In the ISA-95 model they sit at the lowest levels, close to the machine. The ERP with orders and planning sits far above. That separation is deliberate, because control has to keep running whatever happens in the office.
Where it falls short
- Context is missing. The PLC knows a pump runs at 80 percent. It does not know which order is on it, which batch of raw material goes in or what power costs at that hour.
- Fast signals disappear. A historian usually does not store every reading. Filters keep only values that differ enough from the previous one. That is fine for a trend, but a short spike that announces a bearing problem is gone.
- Measurement points are missing. A PLC measures what it needs to control. Vibration, current per motor or an image of the product are usually not in it.
- Tags have no meaning. A value is called DB12.DBW40 or TT_104. What it is, in which unit and on which machine, lives in someone's head or in a spreadsheet.
- The logic only reacts at a limit. A PLC program does what it says when a value gets too high. It does not see a value creeping up for three weeks.
Why you do not just send the data to the cloud
An obvious step is to send everything to a cloud platform and analyse it there. For a decision that has to go back to the machine, that route is too long, and the line has to keep working when the connection drops. Every connection from the shop floor to the outside is also an entry point you have to secure. The ISA/IEC 62443 standards cover that security of industrial automation. In the Netherlands, the Cyberbeveiligingswet (the Dutch NIS2 law) has applied since 15 August 2026, with requirements for more than 8,000 organisations in 18 sectors.
What edge AI adds
Edge AI is computing power next to the machine, in the cabinet or at the line. An industrial computer reads the PLC through OPC UA or Modbus, takes in the extra sensors the PLC does not have and runs a model there. Only the result goes up: an alert, a prediction or an hourly summary. What such a model does:
- Spot deviations in vibration, current or temperature before an alarm goes off.
- Estimate a value you do not measure directly, from the values you do measure.
- Judge a product with a camera and send the verdict back to the line.
- Suggest a setting that uses less energy or material at the same quality.
Control stays in charge
The model does not take over control. It passes on advice, a value or a sorting signal, and the PLC works within its own limits. Emergency stop and safeguarding stay in the safety chain certified for that. Switch the model off and production keeps running.
Where to start
One machine with a problem that costs money.
- Pick a failure that keeps coming back or a quality problem you discover too late.
- Check which data already exists in the PLC and the historian, and which measurement is missing.
- Put an edge computer next to it that reads along, without changing anything in the PLC program.
- Let the model only watch for a few weeks first and compare its alerts with what happened.
Frequently asked questions
Does our PLC program need to change?
Not for reading along. The edge computer reads the values through OPC UA or Modbus. Only when you want to send a result back do you agree with your PLC programmer where it comes in.
How is this different from a digital twin?
A digital twin is a model of your installation that runs live alongside it. Edge AI is the computing power at the machine that collects data, checks it and gives a first verdict. A twin often runs on data that comes in through the edge.
Do we need a cloud for this?
No. The model runs locally. You decide what goes outside.
Sources
- ISA-95: koppeling tussen besturing en bedrijfssystemen · Niveaus van de fysieke machine tot het ERP
- AVEVA: exception en compression in de PI-historian · Filters die bepalen welke metingen worden opgeslagen
- ISA/IEC 62443: beveiliging van industriële automatisering · Normenreeks voor OT-beveiliging
- Rijksoverheid: Cyberbeveiligingswet vanaf 15 augustus 2026 van kracht · Ruim 8000 organisaties in 18 sectoren, 7 juli 2026
Our approach
From advice to management, with one partner
The same rhythm for every project.
Advice
We start with a conversation, not with code. First clarity on what you need and what you do not.
Build
We build and integrate to measure, with technology that fits you. You own your data and your code.
Manage
We stay involved: monitoring, adjusting and growing with your business.
Which machine costs you the most right now?
Tell us what is there and what goes wrong, and we will tell you which data already exists and what needs to be added.
Call 085 083 5775A 30-minute intake