Manufacturing & Maintenance 7 min read Published 24 August 2026

Shop floor data collection: the plumbing under every number your plant reports

By Frank Guo · Technology & Product Leadership, addanode

TL;DR — Every OEE figure, downtime report and production dashboard is only as good as its shop floor data collection — and in many South African plants that layer is still paper and end-of-shift memory, which is why the numbers upstairs and reality downstairs disagree. Real collection draws on four sources: PLC tags (read, never written), retrofit sensors on machines with nothing to read, operator terminals for the human-judgment facts (stop reasons, quality calls), and existing meters (energy, air, water) that plants forget they own. The staged path: automate counts and run-state first, keep operators for reasons only, buffer everything through load shedding — and never let software that merely re-types spreadsheets call itself data collection.

The four signal sources, and when each applies

Source	What it yields	When it's the right tool
PLC / controller tags	Machine state, counts, speeds, fault codes — over Modbus, OPC-UA, or vendor ports	Machines from roughly the 2000s onward with accessible controllers; read-only, no logic changes
Retrofit sensors	Output counts (photo-eyes), run/stop (current clamps), rejects (gate sensors), vibration/temperature	Older machines, or where the PLC is a black box — the machine retrofit route
Operator terminals	Stop reasons, changeover types, quality holds, comments	Everything a machine can't know; the rule is machines record what and when, people record why
Existing meters	Energy, compressed air, water, steam per line or department	Almost every plant has them, almost none reads them continuously — free signals for cost-per-unit

Why the clipboard version always decays

Manual collection fails on a schedule you can predict: week one is diligent, month one is patchy, quarter one is end-of-shift reconstruction. Not because operators are careless — because recording competes with producing, and producing wins. The decay is worst exactly when data matters most (breakdowns, rush orders, short staffing), so manual records are systematically blind to the plant's worst hours. Our 190-complaint downtime research and dashboard-failure analysis both trace to this single root: numbers people typed, versus numbers machines produced.

What good collection looks like in a South African plant

Automatic for facts, human for judgment. Counts, run-state, rates and energy come from signals; operators tag reasons on a terminal in seconds. Nobody transcribes anything, ever.
Read-only on the controls. Collection must never touch control logic — reading tags and clamping sensors keeps production risk at zero and OT security intact (see connecting without exposing PLCs).
Buffered through load shedding. Records that vanish during outages describe a different, calmer plant than the one you run. Edge buffering with timestamps is non-negotiable here.
One layer, many consumers. The same collected stream should feed OEE, downtime Pareto, andon, energy-per-unit and whatever MES you may buy later — collect once, use everywhere, over standard protocols.

A staged path that doesn't bet the plant

Stage 0 — baseline manually with our free downtime templates for a fortnight. Cheap, imperfect, and enough to pick the pilot line.
Stage 1 — automate the bottleneck line: counts + run-state + reason terminal. Weeks of effort; the trust dividend arrives with the first shift report nobody can argue with.
Stage 2 — widen and deepen: more lines, energy meters, quality gates — priorities set by what stage 1's data exposed.
Stage 3 — integrate upward only when something upstream (ERP, MES, group reporting) demonstrably needs the feed.

Frequently asked questions

What is shop floor data collection?

The systematic capture of production facts at their source — machine states, counts, downtime, quality, energy — using machine signals and operator input, instead of paper forms and spreadsheets. It's the foundation layer under OEE, downtime analysis, andon and MES; when it's weak, every number above it is negotiable.

What software is used for shop floor data collection?

The meaningful distinction isn't brand but architecture: software that reads machines (via OPC-UA, Modbus, MQTT and sensor inputs) versus software where people re-type what they remember. Insist on the first kind — and on your data being exportable over standard protocols, so the collection layer outlives any single application.

Can data be collected from old machines without PLCs?

Yes — that's routine retrofit work: photo-eyes and proximity switches for counts, current clamps for run/stop, gate sensors for rejects. No controller needed, no control changes made, and the data quality matches what modern machines report. Age of machine is almost never the real blocker.

How do we get operators to buy in?

Collect automatically so operators aren't blamed-by-clipboard, keep their input to a few taps for reasons, and — decisively — show them the data working for them: a maintenance call that arrives faster, a changeover allowance corrected, a shift report that defends them. Buy-in follows usefulness, not memos.

What does shop floor data collection cost?

It rides the production-monitoring cost structure: indicatively R40,000–R120,000 per line to instrument and connect, plus a modest platform fee — details in our price guide. Existing PLC tags and meters reduce the sensing bill; the reason terminals are the cheapest, highest-value add-on.

Primary sources

Stop typing your production numbers. Start reading them.

Tell us about your lines — old, new or mixed. We'll propose a collection layer that starts on one line, reads what exists, and never touches your control logic.

Book a Discovery Call Machine retrofit →