AI & Automation

How to Measure OEE with Machine Sensors: A Practical Starting Guide

November 11, 2026 · 5 min read · Etzal Group Team

A production manager is asked how well a line is running and answers with an impression. Overall equipment effectiveness, or OEE, replaces the impression with a figure built from three measurable things: whether the machine was running, how fast, and how much of the output was good. Collected automatically, it shows where capacity is being lost.

The three factors

Availability. Of the time the machine was planned to run, how much did it actually run?

Availability = run time / planned production time

Losses here are breakdowns, changeovers, waiting for material and other stops.

Performance. While it ran, how close was it to its ideal speed?

Performance = (total count x ideal cycle time) / run time

Losses here are slow running and the many short stops that do not get recorded as downtime.

Quality. Of everything produced, how much was good the first time?

Quality = good count / total count

Losses here are scrap and rework.

OEE = Availability x Performance x Quality

An example: a shift has 420 planned minutes. The machine ran for 360. Availability is 86%. In that time it made 1,600 parts with an ideal cycle of 0.2 minutes, so performance is 320 divided by 360, or 89%. Of those parts, 1,540 were good, so quality is 96%. OEE is 0.86 times 0.89 times 0.96, about 73%.

The single figure is a summary. The three factors, and the reasons behind each, are what you act on.

Signals you need from the machine

Two signals carry most of the calculation:

  1. Run state. Is the machine producing, idle or stopped?
  2. Count. How many cycles or parts?

Ways to obtain them:

  • From the controller. Modern machines expose state and counters through industrial protocols such as OPC UA or Modbus.
  • From electrical signals. A current sensor on the main motor distinguishes running from stopped. A stack light can be read with simple inputs.
  • From added sensors. A photoelectric or proximity sensor at the output counts parts. This works on machines of any age.

A small gateway collects these signals, stamps them with time and sends them on.

What still needs a person

  • Planned production time. The schedule, shifts and planned breaks.
  • Ideal cycle time. Per product. Use the demonstrated best sustainable rate, not a brochure figure.
  • Downtime reasons. The system knows the machine stopped. An operator selects why from a short list on a tablet or panel.
  • Reject counts, unless inspection is automated.

Keep the reason list short, around ten to fifteen entries, and include "other" with a comment. A list of sixty reasons produces guesses.

Definitions to settle first

Agree on these before collecting anything, because they change the number:

  • Are breaks and planned maintenance excluded from planned time?
  • How long must a stop last before it counts as downtime and not a performance loss? Five minutes is a common line.
  • What is the ideal cycle time for each product?
  • Does reworked product count as good?

Write the definitions down and apply them identically across lines, or the figures cannot be compared.

Reading the results

Look at losses, not only the score.

  • A downtime Pareto. Reasons ranked by total minutes. A few causes usually account for most of the lost time.
  • Short stops. Their number and total duration are frequently a surprise.
  • Speed loss by product. Some products consistently run below the ideal.
  • Changeover times compared between shifts and teams.
  • Quality loss by cause and by time in the run, such as start-up scrap.

Then pick the largest loss and work on it. Measure again. This loop is the purpose of the whole exercise.

Displaying it where it helps

  • A screen at the line showing the current shift against target, in large figures.
  • A daily review of yesterday's losses with the team, taking ten minutes.
  • A weekly view for management of trends and the top causes.

Operators who see the data and help explain it treat it as their tool. Data used only to assign blame is quickly distorted.

Mistakes to avoid

  • Chasing the number. OEE can be raised by redefining planned time. That improves nothing.
  • Comparing unlike processes and drawing conclusions about the people running them.
  • Ignoring short stops because each seems trivial.
  • Too many reason codes.
  • Unrealistic ideal cycle times, which make performance look permanently poor or, set too low, produce figures above 100%.
  • Starting with every machine. Begin with the constraint, the machine that limits output for the line or plant.

A first step

Choose one bottleneck machine. Capture run state and count automatically, add a tablet for stop reasons and rejects, agree the definitions and run for four weeks. The first Pareto of downtime reasons usually pays for the pilot by itself.

We connect machines of different ages and brands, build the data collection and set up the displays in our IoT and Control projects. If you want to know where a line is losing capacity, tell us what machines it has and what data exists today.

Summary

OEE multiplies availability, performance and quality. Capture run state and counts automatically, let operators supply reasons and rejects, fix the definitions before you start, and use the loss breakdown to choose what to improve. Begin with the bottleneck.

Frequently asked questions

What is a good OEE score?

It depends heavily on the industry and the process, so comparing with your own past is more useful than chasing a universal figure. Many plants that start measuring honestly find their OEE lower than they assumed. The trend and the size of each loss category matter more than the single number.

Can OEE be measured on old machines without a modern controller?

Yes. A current sensor on the main motor shows running and stopped states, and a simple sensor on the output can count parts or cycles. That is enough for availability and performance. Quality data is usually entered by operators or taken from inspection records.

Why is automatic data collection better than manual logs?

Manual logs miss short stops and round the long ones. Small stoppages of a minute or two are rarely written down, yet together they often form the largest loss. Automatic capture records every stop with its exact duration and frees operators to record only the reason.

Author

Etzal Group Team

Etzal Group Team, helping businesses grow with technology. Contact us.

Related service

WhatsApp AI Chatbot

We build, train, and run a WhatsApp AI agent that answers customers 24/7, qualifies leads, and follows up.