The OEE percentage shown on most Australian manufacturing dashboards is a single number generated from many smaller decisions. What counts as planned downtime. What the line should theoretically produce per minute. Where rejects get counted. Most operations teams can recite the formula. Far fewer businesses can explain exactly how each input is calculated on the lines they manage.
That gap matters. OEE drives capital decisions, shift KPIs, and improvement priorities. If the inputs are not defensible, every downstream decision is built on numbers that fall apart when challenged.
A defensible OEE is one where the business can state clearly, for any line, exactly how each of the three components is calculated and where the underlying data comes from. That standard is higher than most sites currently meet.
The three questions a defensible OEE answers
Every OEE number is the product of three other numbers: availability, performance, and quality. The three components are calculated independently, from different data sources, with different assumptions. A business that uses OEE to make decisions should be able to answer the following questions on any line.
Availability
What counts as planned downtime, and what counts as unplanned? The line between the two is a business decision, not a technical one. Planned maintenance, scheduled changeovers, breaks, cleaning windows all need an explicit rule. Without that rule, two shifts can record the same event differently and the resulting availability number drifts.
Where does the timing data come from? Machine state transitions reported directly from the PLC carry timestamps at the moment the event happened. Operator entries recorded after the fact carry timestamps from when the entry was made, with the time-of-event filled in from memory. The two are not the same signal and do not produce the same number.
Performance
What is the theoretical production rate, and where did that number come from? OEM datasheets are a common source, but datasheet rates are typically measured under ideal conditions that the line does not see in production. A theoretical rate validated by observation, the rate the line achieves when nothing is wrong, gives a defensible reference. A theoretical rate copied from a brochure does not.
How is the actual production rate measured? A continuous count from the machine itself is the cleanest source. A count taken from a downstream sensor is a derived measurement. A count entered from a check sheet at shift end is not a measurement at all.
Quality
Where in the line are rejects counted? A line typically has multiple reject points: a checkweigher rejecting underweight, a metal detector rejecting contamination, a vision system rejecting label faults, a manual reject station for product that fails inspection. Each rejector produces its own count. An OEE quality calculation that ignores some of them will overstate quality.
Are the reject counts coming from the machine or from a manual entry? Machine counters give a continuous record. Manual entries are summaries.
Why the data needs to be close to the machine
Operator-entered downtime is not measurement, it is reconstruction. The reason recorded at shift end is the one that comes to mind, not necessarily the one that happened. Memory averages, simplifies, and rationalises. A downtime category entered three hours after an event is closer to a story than to data.
Machine state, by contrast, transitions in the controller in real time. A motor stops because an interlock tripped, the PLC sees the transition immediately. A reject sensor activates, the count increments at the moment the reject leaves the line. Both signals carry timestamps that cannot be edited after the fact. For analysis that is meant to drive business decisions, that level of integrity matters.
The closer the OEE data is to the machine, the harder it is to dispute. The further from the machine, the easier it is to challenge.
Why multi-OEM lines are harder than they look
A typical packaging line has six to twelve machines from different OEMs. A typical processing line has a different mix of equipment from different vendors. Each machine has its own controller, its own way of reporting state, its own counters, and its own approach to exposing data over a network.
Some machines expose a clean PackML state model that a line controller can read directly. Others expose a basic running contact and nothing else. Some report production counts via a network tag. Others expose only a totaliser that needs to be polled and differenced. Some reject sensors publish events on a fieldbus. Others close a contact on a digital output that needs to be wired and counted.
Standardising those signals into one OEE calculation is not a configuration exercise. It requires understanding how each machine actually works internally, not just connecting to its network interface. A machine that does not naturally expose what the OEE calculation needs has to be read against the grain, which means understanding the OEM's logic well enough to extract a clean signal from a less clean source.
That is the engineering layer that determines whether an OEE programme produces numbers a business can act on or numbers a dashboard can display.
What good looks like
Questions a defensible OEE should answer for any line on the site
- What is the line's planned production time for the shift, and what rules define when an event moves from planned to unplanned downtime?
- Where does the line's theoretical production rate come from, and when was it last validated against the machine running well?
- What sensors and counters feed the production count for performance, and which are read directly from the machine versus inferred?
- How many reject points exist on the line, and is every one of them counted into the quality calculation?
- For each input above, is the data taken from the machine itself or from an entry that depends on operator recall?
What this means
OEE that drives business decisions has to survive the challenge of "show me exactly how each part of this number was calculated." Most current implementations cannot. The gap is rarely in the formula. It is in the layer where machine signals become numbers, where some signals are read directly and others are filled in from memory, and where multi-OEM lines force shortcuts that are never written down.
For a business that uses OEE to make decisions, the work that matters is the work between the equipment and the dashboard. That is where defensibility is won or lost.