Key figures and OEE

TrakState stores input quantities and computes from them. Finished key figures are not stored: the mean of five shifts’ OEE is not the OEE of those five shifts. Because everything is computed from sums, each figure holds over any period — and a corrected formula takes effect retrospectively rather than only on new entries.

One time base: actual runtime

Availability and performance stand on the same base, and it is the actual runtime, not the planned time.

runtime      = actual runtime − downtime
availability = runtime / actual runtime                       (capped at 100 %)
performance  = (quantity × effective cycle time) / (runtime × 60)   (capped at 100 %)
quality      = (good parts + reworked good) / quantity
OEE          = availability × performance × quality

The reason is a case from the shop floor: a shift ends after 71 minutes, target met, no stoppage. Measured against planned time that read 59 % (with 120 planned minutes) or 15 % (with 480). Nothing stood still, and the display said 59 %. A figure nobody believes does not get improved, it gets ignored. On the runtime base the same case comes out at 100 %.

What it costs, and where that is carried. The gap between plan and actual disappears from OEE. Anyone reading OEE as a measure of utilisation gets a flattering number. The gap is carried by achievement (quantity produced against quantity asked for), a figure of equal standing beside OEE — and by a mandatory comment on a shift that ended early without meeting its target.

Effective cycle time. The target cycle time is per cycle, not per part. A station that takes two parts out of one cycle needs half the time per part: effective cycle time = target cycle time / capacity. An unrecorded capacity counts as 1.

The figures

Figure	Unit	What it says
Availability	%	share of the actual runtime in which the line was running
Performance	%	ideal time for the parts produced against the time the line actually ran
Quality	%	what ended up good, rework included, against everything produced
OEE	%	availability × performance × quality
First pass yield	%	parts good without any rework, against everything produced
Rework rate	%	parts that needed rework, against everything produced
Jobs per hour	pcs/h	good parts per hour of runtime — on the same runtime as above
Achievement	%	quantity produced against quantity asked for. Not capped: over-delivery is information
Defect rate	%	defective parts (rework + scrap) against everything produced
PPM	ppm	the same defect rate in the unit a quality agreement is written in
Defective parts	pcs	counts parts, not defect occurrences
Inspected parts	pcs	what the inspections looked at, counted once per inspection
Defects found	pcs	counts defects, not defective parts
Inspection defect rate	%	defects found against parts inspected. Not capped
Inspection PPM	ppm	the same rate in ppm

Two pairs never to confuse

Defect rate and inspection defect rate are not the same figure. One measures what production reported over everything it made; the other what an inspection found in its sample. A board showing both has to say which is which.

Defective parts and defects found likewise. A part with three defects counts once as a defective part and three times in the defect Pareto. The two must never share an axis.

Why capture and the board show the same number

Every figure is defined once, in two renderings: as an expression for the capture mask, which computes live while somebody types, and as SQL for the query tag, which aggregates over many records. A test computes both ways against the same cases.

One property follows from that, and you can check it: filter the board down to exactly one record and every sum collapses onto its one summand — the figure must be exactly what the operator saw in the mask.

Part of that is that target cycle time and capacity travel into the record. Otherwise the board would recompute an old shift with a target cycle changed since.

Empty is not zero

Where nothing was captured, the figure stays empty. It is not set to 0 — “nobody captured this” and “it was zero” are two different statements. A denominator of 0 likewise yields no result and no error, but “not yet”.

The quantity may be overwritten: a miscount that cannot be corrected is a miscount that gets corrected in a spreadsheet instead. Where the quantity is filled it wins; where it is not, good parts + rework + scrap carry it. A stored 0 is a correction, not a blank.

Rounding happens at display time. The tile decides the decimals, not the definition.