Why "we'd need a new PLC" is usually wrong

Plenty of South African plants run machines that are twenty or thirty years old and still perfectly productive. When OEE comes up, the assumption is that measuring it means a control-system upgrade — a new PLC, new wiring, a shutdown, and a budget that kills the idea on the spot. That assumption is almost always wrong.

OEE is just three things: was the machine running when it should have been (availability), did it run at rate (performance), and was the output good (quality). Every one of those is observable from the outside. An old line doesn't speak OPC-UA, but it still draws current when it runs, still indicates state on a stack light, still moves product past a point you can count, and still rejects bad parts somewhere. You don't need the machine to report — you need to observe it. That's a sensing problem, not a controls upgrade.

Key idea: you're not integrating with the old controller. You're adding a thin, non-invasive layer of sensors that watches the machine from outside and sends what it sees to the platform. The machine itself is untouched, so there's no shutdown and no risk to a line that already works.

The signals an old line already gives you

  • Run / stop state → motor current. A clamp-on (CT) current sensor on the main drive tells you, with no wiring into the machine, whether it's running. Current draw also hints at load and rate.
  • State → stack lights. Most lines already have a red/amber/green tower. Tapping those signals gives you running, stopped and fault state directly.
  • Counts → proximity or photo sensors. A proximity switch or photo-eye watching product pass a point gives you a good-enough piece count without touching the machine's logic.
  • Quality → reject gate or manual entry. If the line ejects rejects, count them at the gate; if not, capture scrap at the operator terminal. Either way you close the quality leg.
  • Downtime reasons → operator terminal. The single most valuable add-on: a small touchscreen where the operator picks why the line stopped. That's what turns "we were down 90 minutes" into "we lost 90 minutes to changeover and material starve."

The step-by-step path

  1. Define the question per line. Decide exactly what you must measure — availability, performance, quality, and which downtime reasons matter — before anyone buys hardware. The decision drives the sensing, never the other way round.
  2. Audit the existing signals. Walk the line and list what's already there: motor, stack light, count point, reject gate. You'll usually find you need fewer sensors than expected.
  3. Fit a small edge device with sensors. A clamp-on current sensor, a count sensor and a stack-light tap, wired to an edge gateway mounted in or near the panel. Non-invasive, installable in a normal maintenance window — often without stopping production.
  4. Add the operator terminal for downtime reasons. Configure a short, plant-specific reason-code list with the people who run the line, so it's fast to use and actually gets used.
  5. Buffer at the edge, then sync. The edge device stores data locally and forwards it to the platform. When load shedding hits or the network drops, nothing is lost — it catches up when power and connectivity return.
  6. Baseline, then improve. Run for a couple of weeks to establish the true current OEE — usually lower and more revealing than anyone guessed — then use the reason-coded downtime to attack the largest loss first.

What it costs vs a PLC swap

A controls retrofit means engineering, new hardware, re-wiring, validation and downtime. The edge approach is a handful of sensors, a gateway and a terminal per line, installed around production rather than instead of it. As an indicative guide we put a single old line in the R70,000–R120,000 band once-off (see our OEE cost guide) — a fraction of a PLC project, and it pays back from the downtime it makes visible. Indicative ZAR; scope your line for a real number.

The honest caveats

Externally-sensed OEE is excellent for the metric that matters — availability, performance, quality and reason-coded downtime. What it won't give you for free is deep process data the machine never exposed (precise temperatures, recipe parameters). If you need those, that's a separate, targeted sensing decision — not a reason to replace the controller. And a current clamp infers rate rather than reading a perfect part count; for most lines that's plenty, and where it isn't, you add a dedicated count sensor. The goal is a trustworthy OEE number that drives action, not laboratory precision.

This is the core of how we deliver OEE management at addanode: read the machine you have, don't replace it. Because we build both the edge hardware and the addaNet platform in-house, we can sense an old line non-invasively and have it on a live dashboard alongside your newer equipment — one OEE picture across the whole plant, old and new.

Frequently asked questions

Can I really measure OEE without touching the machine's control system?

Yes. A clamp-on current sensor reads run/stop without any wiring into the controller, a proximity or photo sensor counts product passing a point, and a stack-light tap reads state. The machine's logic is untouched, so there's no shutdown and no risk to a line that already works.

Do I need to stop production to install it?

Usually not. Non-invasive sensors and an edge gateway can typically be fitted in a normal maintenance window, and often while the line keeps running. That's a major contrast with a PLC retrofit, which means a planned shutdown and re-validation.

How accurate is current-based OEE compared with reading the PLC?

For availability and reason-coded downtime it's effectively as good — the machine is either drawing current or it isn't. For counts, a current clamp infers rate; where you need an exact part count you add a dedicated count sensor. The result is a trustworthy OEE number that drives action, which is the point.

What happens to the data during load shedding?

The edge device buffers locally and syncs when power and connectivity return, so no production record is lost during an outage. Designing for load shedding from the start is essential for South African sites — cloud-only systems with no edge buffering lose data every time the power drops.

How many old lines can I do this way?

As many as you like, but start with your worst bottleneck. The same edge-sensing recipe standardises across similar lines, so the first one carries the learning and each additional line is faster and cheaper to bring on.