The OEE that comes out of the Excel sheet on the 8th of the following month is no use for managing anything — it serves to justify what has already happened. This guide shows you how to set up real-time production control with OEE by the minute: the prerequisites, the realistic phasing, the four mistakes that stall projects and what to check before choosing a solution.
The problem is not measuring OEE. It is measuring it too late.
Almost every Portuguese factory we visit already calculates OEE. Someone, somewhere, has a sheet with availability, performance and quality. The problem is the lag: the number arrives days after the stoppage that caused it. By the time the production manager sees it in the monthly meeting, the machine that stopped for 90 minutes for lack of yarn has already stopped five more times for the same reason. Monthly OEE is an autopsy. It tells you what the line died of — it never arrives in time to save it.
Real-time control reverses the logic. The gain is not in a prettier report — it is in acting while the shift is still under way. The value of visible Cycle Time lives in the shift leader who fixes the microstop at 10:15, instead of discovering it on the 8th. In a textile sector where the margin is fought over in cents per metre, that difference of hours pays for the system.
What you need before you start
Before installing terminals or talking about an API, secure the foundation. Without it, the project starts off crooked. And the foundation is not technical — it is one of definition. Most of the discussions in an OEE project are not about sensors; they are about what counts as a planned stoppage. Does the batch changeover count? Does the mandatory mid-shift cleaning count? The answer shifts the number by ten points, and if you do not fix it in writing before starting, each shift leader invents their own.
- Complete list of the workstations and machines to be monitored, with unique identification.
- Written definition of what counts as planned vs. unplanned stoppage — decided in a meeting, not left to the operator.
- A shop-floor network that can handle terminals and collectors — real coverage, not just one point in the basement.
- A short, unambiguous catalogue of stoppage reasons (maximum 12-15 codes, not 60).
- Production orders already managed in the ERP, not on loose paper.
- A project owner on the production side — not just from IT.
- Management agreement on who reads the alerts and who decides on them.
Step 1 — Map the workstations and choose the data source
Each machine emits data differently. Some have a PLC ready to connect; others require an external sensor; others only allow manual entry by the operator at the terminal. Decide the source workstation by workstation. The temptation is to automate everything at once — resist it. Start with the bottleneck workstations, where the stoppage costs the most.
KORA Productivity integrates with PLCs and collectors on the shop floor, but also accepts mobile entry when the equipment is old. On a 1990s plastic injection machine we connect the cycle signal; at a garment cutting table, the operator confirms the quantity at the terminal. Both feed the same OEE calculation. The rule is simple: do not force automation onto equipment that does not support it just to have the cleaner chart. A misread PLC signal lies with more confidence than an honest operator.
Step 2 — Configure OEE and train those who record it
OEE is three factors: availability × performance × quality. The technical part is configured in an afternoon. The hard part is recording discipline. If the operator always presses "electrical fault" because it is the first button on the screen, your Lead Time data is worth zero.
Invest time in the reason catalogue and in training whoever presses the terminal. We see in INFOS projects that OEE quality depends less on the software and more on the operator understanding that recording defends them — it shows that the machine stopped for lack of material upstream, not through their incompetence. Order the buttons on the screen by the real frequency of the reasons at that workstation, not alphabetically. A silly detail that decides whether the data is any good.
Decision matrix: automate or manual entry
| Criterion | Automate (PLC/sensor) | Manual entry (terminal) |
|---|---|---|
| Cost of the stoppage | High — bottleneck | Low to medium |
| Age of the equipment | Machine with accessible PLC | Old machine with no signal |
| Volume of microstops | Many and short | Few and long |
| Data reliability | Objective, no interpretation | Depends on discipline |
| Initial investment | Greater | Smaller |
Step 3 — Pilot on one shift, not the whole factory
Choose one line, one shift, a team that will cooperate. Run the pilot for four to six weeks. Compare the OEE the system gives with the shift leader's perception — if they diverge a lot, the problem is in the reason configuration, not the factory. Adjust before rolling out.
OEE is only worth anything when the shift leader uses it to decide today, not when the manager discusses it on the 8th of the following month.
Batch-by-batch traceability comes in here. When the brand asks for proof of compliance with the EU Strategy for Sustainable Textiles, the operational history by batch stops being a luxury — it is the difference between responding in minutes or spending a week reconstructing paperwork in a dye house in the Vale do Ave. The native integration with the ERP MULTI keeps that trail without double entry.
Step 4 — Data culture and management by exception
The system generates alerts when a workstation's OEE drops below the threshold. This only works if someone acts on them. Define who receives what and within what deadline they must respond. An alert no one reads is noise; an alert that triggers a corrective action within 30 minutes is the return on the investment.
For deeper analysis — trends by week, comparison between lines, correlation between shifts and defects — cross the operational data with Qlik Sense. The terminal serves the present; BI serves the pattern. They are different layers and do not replace one another. Confusing the two is the classic mistake: wanting the operator to read trend charts on the terminal while the machine waits. The terminal answers "what do I do now?". BI answers "why does this keep happening on Thursdays?".
Common mistakes and how to avoid them
Four stall projects with a regularity that is already tiring. The giant reason catalogue — sixty codes guarantee the operator always picks the first, and the data is born rotten. Maximum 15, reviewed after the pilot. OEE without a baseline — without measuring the starting point, no one can prove the improvement later; collect two weeks of data before changing any process, even if it hurts to see the real number.
Ownerless alerts kill management by exception: the system warns, no one reacts, and after two weeks everyone ignores the flashing screen. Each type of alert has a named owner and a written response deadline. And the whole-factory rollout on day one — it multiplies the problems across all workstations at once and turns a configuration adjustment into a widespread crisis. Pilot on one shift, adjust, then roll out. Finally, do not spend budget automating equipment that rarely stops: start with the bottlenecks, the rest does fine on manual entry.
The context the manuals do not mention
A detail you only grasp after years in Portuguese factories: the project's greatest enemy is not the technology — it is the operator who interprets the terminal as surveillance. In the mould-making industry, where Portugal is the 3rd largest producer in the world with around 472 companies and exported 80% of its 2023 output, according to CEFAMOL, the operator is highly specialised and distrusts being timed. If the discourse is "control people", the project dies on the first shift. If it is "prove that the machine stopped for lack of material that does not depend on you", the operator becomes an ally — and starts correcting the data when the system gets it wrong, because now it is their defence.
Poka-Yoke also applies to recording: make the right choice so easy that the error becomes hard. A big button for the most common reason, mandatory confirmation only where it matters, none of those seven-level menus for a man with hands covered in oil.
Before signing any contract, check the vertical fit of the solution with your sector and SKU complexity — an ERP that does not model three axes of colour-size-last in footwear will never give useful OEE per reference. In a sample collection from Felgueiras with 800 to 1,200 SKUs, an OEE aggregated by line hides exactly what you need to see: which last is strangling production.
The next step is to measure, not to buy
Run the pilot on one shift, measure the baseline before changing anything and prove the return on one bottleneck. Whoever starts with the whole factory spends more and learns less. The complete guide to MES and OEE in the Portuguese factory goes deeper into the calculation and the data architecture behind what we summarise here. Real-time OEE does not change the machine — it changes the time at which you decide to fix it. And that is where the money is.
Sources
- CEFAMOL — Associação Nacional da Indústria de Moldes, 2023 production and export data.
- AIMMAP / Metal Portugal — statistics of the Portuguese metallurgical and metalworking sector.
- European Commission — EU Strategy for Sustainable and Circular Textiles (2022).