OEE in manufacturing: calculating, measuring and improving

A machine can be running without producing effectively. Short stops, reduced speed and rejected products can easily hide behind a total count. OEE, short for Overall Equipment Effectiveness, makes those losses visible. Not as a scorecard for the factory, but as a starting point for targeted improvements.

At Vlegel Technology in Genk, the software engineering team for machines, data and AI, we therefore start with what the data actually means. For factories and machine builders, an OEE dashboard is only useful when it is clear what a machine measures, which time counts and how losses are recorded.

What does OEE measure?

OEE combines availability, performance and quality. Availability describes how much of the planned production time the machine is actually running. Performance compares the quantity produced during that run time with what would be possible at the ideal cycle time. Quality indicates how much of the output meets requirements first time.

Together, these factors distinguish three types of loss: not running, running too slowly and making defective products. The overall percentage points you in the right direction, but the underlying factors tell you where to look. Low availability calls for a different approach from high reject rates.

Calculating OEE correctly

Use ratios between zero and one for the calculation. Availability = run time ÷ planned production time. Run time is planned production time minus downtime within that period.

Performance = (total units produced × ideal cycle time per unit) ÷ run time. Use consistent time units and a validated ideal cycle time for each product. Divide by run time here, not planned production time: otherwise you count downtime twice.

Quality = good units ÷ total units produced. Then: OEE (%) = availability × performance × quality × 100.

With 90% availability, 85% performance and 95% quality, OEE rounds to 72.7%. This does not mean the machine ran perfectly for 72.7% of the time. It means that good output, converted into time at the ideal cycle time, equates to 72.7% of planned production time.

Agree on measurement rules before building a dashboard

Define planned production time and how you account for changeovers, breaks, breakdowns and microstops. Also agree on when a product counts as good and how rework is recorded. Without consistent definitions, identical situations can produce different scores.

Check counters, timestamps and machine states too. A missing signal does not automatically mean downtime. Record short stops consistently under either availability or performance, without counting the same loss twice. Only compare machines or shifts when the product mix and measurement rules allow it.

Connecting machines and structuring data

Depending on the control system, you can read machine data via OPC UA and make it available to other systems via MQTT. The two protocols have different roles: OPC UA provides structured access to machine data; MQTT supports publishing and receiving messages.

A Unified Namespace (UNS) can bring that data together in a shared structure for factory, line and machine. Publish not just counter values, but also context such as product, order, machine status and timestamp. This lets dashboards and analytics use the same data without building a separate connection for each application. A UNS does not replace agreed data quality rules.

From loss to action

Start with one machine or line and investigate the main loss. Discuss downtime reasons with operators, check speed losses against the cycle time and link rejects to product and process data. Make a targeted change, then check what changes under comparable conditions.

OEE does not tell the whole story. MTTR helps track repair times; MTBF provides insight into the average operating time between failures. Use these metrics alongside OEE. A higher OEE on one machine is not a goal in itself if it creates inventory or waiting time elsewhere.

AI can summarise recorded losses or highlight patterns for further investigation. Our boundary is clear: AI only reads, it does not write to the machine control system or change any settings. A person reviews the analysis and decides what action to take.

Want to discuss whether your OEE measurement is useful on the shop floor? Book a 15-minute call with Vlegel Technology. We will look at which machine data is available, which definitions are missing and what makes sense as a first step.