What OEE is, how it is calculated, and why one number is not enough
The line ran all day. At the end of the shift the output is below what was expected, and nobody can say exactly where it went. This is usually the point where a single metric gets proposed: OEE.
OEE — Overall Equipment Effectiveness — shows how much of a machine's or a line's planned time was actually turned into good output. It is the product of three separate ratios: availability, performance and quality.
Availability measures how much of the planned time was worked. Planned production time is compared with the time actually run. The difference is downtime: breakdowns, die changes, waiting for material, setup. For the ratio to mean anything, “planned time” has to be defined. Are planned maintenance, breaks and out-of-shift hours inside the plan or outside it? Availability computed before that question is answered cannot even be compared between two shifts.
Performance measures the speed at which the running time was used. The quantity produced in the running time is compared with what the ideal cycle time would have produced in that same time. The loss here is usually of the invisible kind: running slow, micro-stops of a few seconds, manual intervention. The ratio rests on the ideal cycle time. If that time is not an agreed number per product, the performance ratio is arguable too.
Quality measures how much of the output was right the first time. It is the share of total output that meets the specification. What happens to reworked output is decisive here. If rework counts as good, the extra time and material it consumed disappear from the metric entirely.
The three ratios multiply, and the single number hides the cause. OEE is their product. Because it is a product, a drop in one pulls the whole figure down, and none of the three substitutes for another. The real issue is this: the same OEE value can come out of very different combinations. A line that stops often but runs fast and clean between stops, and a line that almost never stops but runs slow, can show the same number. What needs to be done in those two cases is entirely different.
So the single number is a monitoring metric, not an action metric. OEE should always sit on the screen with its three components. If it is not visible which component the drop came from, the metric only says “not good” — it does not say where to look.
There are things the metric does not cover. OEE looks at the planned time of one machine or one line; idle time caused by having no orders, imbalance between stations, and stock waiting in front of the line do not enter the ratio. A high OEE therefore does not mean the workshop as a whole is running well — the metric only describes the inside of the time it defines. How often the product was changed over in that time does not show up on its own either, and on a line with frequent changeovers availability and performance blur into each other.
The number becomes useful only when stop reasons are coded.
Availability on its own says “we were down this many minutes.” To become useful, those minutes have to be distributed across reasons. The reason list is kept short; a long list does not get chosen at the line and collapses into one heading within days. The headings must not overlap: if two headings both look right for the same event, the minutes split in two and both totals shrink.
The planned-versus-unplanned distinction lives inside that list. A die change is planned, and its duration can be shortened; a breakdown is unplanned, and its cause lies elsewhere. Merge them under one heading and it becomes invisible which side to work on.
The reason is chosen at the moment of the stop, by the person living through it. A reason field filled in from memory at the end of the shift points at the most memorable cause, not the most frequent one.
Certain things have to be recorded for the figure to mean anything.
Behind an OEE value stand: planned production time per shift; the start time, end time and reason for every stop; total quantity produced; an agreed ideal cycle time per product; the good and defective quantities; and a decision on which side rework is written.
All of it has to come from one set of definitions. If shifts record under different definitions, the resulting number compares definitions rather than shifts. For the same reason, placing the OEE of two lines running different products at different cycle times side by side is misleading. The meaningful comparison is a line against its own history.
Turning the metric into a scoreboard also breaks the measurement. When the number is used to judge people, stop records thin out and the figure rises while the loss stays where it was. Recording discipline holds only when it is clear that the number describes the line.
Starting from nothing, the three components do not have to be built at once. Once stop durations and reasons are recorded, availability appears; add output counts and cycle times and performance follows; add the good-versus-defective split and quality joins them. At every step the metric explains a little more, and which record is still missing becomes visible.
At Miletus, OEE starts as a definition job rather than a screen job: stop reasons are written together with the shift supervisor, cycle times are fixed per product, and the record moves onto a tablet at the head of the line. The engineering side frames the problem; software makes it visible. On the production line the measurement is set up on one line first, then carried to the others under the same definitions. A single number produces a decision only when it is read together with its three components and their reasons.