What is OEE? Measuring and improving production efficiency
Learn how to calculate OEE, collect reliable machine data and tackle losses from downtime, reduced speed and rejects.
OEE stands for Overall Equipment Effectiveness. It shows how much of your planned production time is spent making good products at the ideal production speed. An OEE of 100% means: no downtime, no speed losses and no rejects. In practice, the breakdown is what matters most: where are you losing production capacity, and which cause can you tackle?
At Vlegel Technology in Genk, we prefer to look at the losses behind the number, not just the number itself. As a software engineering team working on machines, data and AI for factories and machine builders, we start with reliable signals and clear definitions.
Calculating OEE: three factors
OEE = availability × performance × quality. Multiply the factors as decimal values. With 90% availability, 95% performance and 98% quality, OEE is approximately 83.7%. That overall score brings different losses together, but does not yet tell you what needs to change.
Availability = run time / planned production time. This factor shows how much of the scheduled time the machine is actually running. Breakdowns and changeovers during that time reduce availability. Agree upfront which periods count as planned production time. Otherwise, the same production run can get a different score simply because the planning definition has changed.
Performance = ideal cycle time × total units produced / run time. This compares actual output with what is technically achievable during run time. Slower cycles and minor stops cause performance losses. Use a validated ideal cycle time for each product or recipe, not an arbitrary production target.
Quality = first-pass good units / total units produced. Rejects and products needing rework do not count as first-pass good units. Make sure production and quality records use the same order, period and product definition.
Reliable data starts with shared definitions
For a useful measurement, you need at least machine states, timestamps, production counters, quality data and the ideal cycle time. Manual recording can be a starting point. Automatic recording makes short stops and recurring patterns easier to spot, provided the signals are interpreted correctly.
A PLC signal saying ‘machine active’, for example, does not necessarily mean products are being made. Check the records with operators and the maintenance team. Also agree when a short stop counts as a performance loss and when it counts as an availability loss. The same loss must not be counted twice.
Connecting machines via OPC UA and MQTT
You can read machine data via OPC UA and make it available to other applications via MQTT. The right connection depends on the control system and existing infrastructure. Keep data collection separate from machine control: an OEE application does not need to write settings to the machine.
A Unified Namespace (UNS) can organise that data in a shared structure, for example by factory, line and machine. Give values context too: timestamp, unit, product, order and data quality. That lets dashboards and analyses use the same definitions. A UNS does not replace those agreements; without context, a counter is just a number.
From measurement to improvement
Start with one machine or production line and check that the records match reality. Then look at which losses have the biggest impact. For downtime, investigate breakdowns, changeovers and waiting. For performance losses, look at cycle times and minor stops. For quality losses, look for recurring defects and start-up problems.
Choose a specific cause, make a change and measure again under comparable conditions. Do not compare lines blindly: product mix, cycle times and planning rules all affect the result. A higher local OEE does not automatically mean more good finished products either, if the bottleneck is elsewhere on the line.
AI can summarise recorded stops or flag patterns for further investigation. We keep the boundary clear: AI only reads, a human decides. A correlation is not a proven cause, nor an instruction to adjust the machine automatically.
Want to discuss how to measure OEE reliably on your machines? Book a 15-minute call with Vlegel Technology. We will look at which signals are available, which definitions are still missing and what makes sense as a first step.
