Why do startup rejects count against OEE quality?

Because OEE measures first-pass good output against everything produced, and a unit wasted during ramp-up is as gone as one wasted mid-run. Forgiving startup scrap as 'just the first few pieces' is how quality loss escapes measurement, those pieces, multiplied across every changeover, add up to a real loss.

What is the difference between startup rejects and production rejects?

Startup rejects are defects made between the start of a run and stable running, first pieces after a changeover, warm-up scrap, purge material. Production rejects are defects made during steady-state running from drift, wear, or variation. Both are quality losses, but startup rejects point at setup and production rejects point at process control.

How does rework hide quality loss in OEE?

If a plant counts reworked units as good because they eventually shipped, the Quality factor recovers them and the loss disappears from OEE, even though the plant paid for the defect in labor and capacity. Reworked units also consume capacity twice, so honest counting is the only way the loss stays visible.

Quality Losses in OEE

Q: What are quality losses in OEE?

Quality losses are the units a line produced that were not good the first time, the gap between total count and first-pass good count. They come in two kinds: startup rejects made while a process ramps to stable running, and production rejects made during steady-state running. Both lower the OEE Quality factor.

Q: How is the OEE Quality factor calculated?

Quality = first-pass good count divided by total count. Count everything the process made, count only the units that met standard on the first attempt with no rework, and divide. Because Quality is a multiplier in OEE = Availability × Performance × Quality, even small-looking quality losses drag the whole score once compounded.

Q: What is a good OEE Quality factor?

The commonly cited world-class reference is 99% for discrete manufacturing, but that is a reference point, not an audited standard. Judge Quality against your own history and counting honesty: a strict 96% that counts every reworked unit tells you more than a soft 99.5% that forgives rework. Check that it reconciles with first pass yield.

The Quality factor's two loss types, startup rejects and production rejects, why both count, and how rework quietly erases real quality loss from the number.

Quality losses in OEE are the units a line produced that were not good the first time, the gap between total count and first-pass good count. They come in two kinds: startup rejects made while a process ramps to stable running, and production rejects made during steady-state running. Both drop the Quality factor, and rework counts as a loss even when the unit is later saved.

The Quality factor looks like the simplest part of OEE, good count over total count, and that simplicity hides two traps. The first is thinking quality loss means only scrap; it also means every unit that needed a second pass. The second is thinking defects happen only during normal running; a large share are made in the first minutes after a changeover, before the process settles. This post defines quality losses precisely, explains why startup and steady-state defects both count, works the Quality factor, and shows how rework hides real loss.

What are quality losses in OEE?

Quality losses are all output that failed to meet standard on the first attempt during planned production time. In OEE's structure, they are the third and final loss family, after time losses (Availability) and speed losses (Performance). The Quality factor captures them as a ratio: good count over total count where good means right the first time.

Nakajima's six big losses split quality into two of the six categories, and that split is the key to understanding them:

Startup rejects defects produced between the start of a run and stable running. First pieces after a changeover, warm-up scrap, purge material, the units made while settings are still being dialed in. The loss of getting the process going.
Production rejects defects produced during stable, steady-state running. The ones that show up in first pass yield once the line is settled: out-of-spec units from drift, wear, material variation, or human error.

Both are quality losses because both produce units that are not first-pass good, and OEE does not care whether the process was ramping or settled, a bad unit is a bad unit. But separating them matters for the fix, because their causes and countermeasures differ, as our companion piece on quality loss analysis works through. The six big losses guide maps all six categories, including these two, to their OEE factors.

The Quality factor subtracts every non-first-pass unit, whether made during startup or steady-state running. Together they set Quality at 96%. Hypothetical numbers.

Why do both startup and steady-state defects count?

They both count because OEE measures first-pass good output against everything produced, and a unit wasted during ramp-up is as gone as one wasted mid-run. The instinct to forgive startup scrap, "that's just the first few pieces, it doesn't count", is exactly how quality loss escapes measurement. Those first few pieces, multiplied by every changeover on a high-mix line, add up to a serious loss that a plant only sees when it stops excusing them.

The reason to still tag them separately is that they point at different problems. Startup rejects are a setup and standardization problem: undocumented settings, dial-it-in-by-feel changeovers, no first-piece verification. Their countermeasure is a standardized ramp, documented settings per product, defined startup sequence, first-piece checks, and better changeover method. Production rejects are a process-control problem: drift, wear, and variation during running, countered with statistical process control and mistake-proofing. Lump them together and you cannot tell whether to fix your changeovers or your process capability. That is why the analysis method insists on splitting them.

Both are quality losses, but they occur at different times and have different causes, startup rejects cluster after changeover, production rejects scatter through stable running. Hypothetical.

How is the OEE Quality factor calculated?

Quality = good count ÷ total count, where good means first-pass good. The steps keep the definition honest:

Count total units produced. Everything the process made during the period, good and bad, including startup pieces and units bound for rework.
Count first-pass good units. Only units that met standard on the first attempt, with no rework, repair, or re-run. This is the numerator.
Classify every non-good unit. Startup reject or production reject, with a defect code. Scrap and rework both land here; recovery does not move a unit back to good.
Divide. Quality = first-pass good ÷ total. On the hypothetical line above, 19,206 ÷ 20,000 = 96.0%.
Fold into OEE. OEE = Availability × Performance × Quality. Quality is a multiplier, so a quality problem drags the whole score, not just its own factor, a low Quality factor pulls down an otherwise healthy line. The OEE calculation works the full example, and the OEE calculator runs it.

Because Quality is a multiplier, its losses compound with the others. A line at 90% Availability and 90% Performance with a 96% Quality is not "missing 4 points", the 4 points of quality loss cost 0.90 × 0.90 × 0.04 = about 3.2 points of OEE. Small-looking quality losses are rarely small once multiplied through.

One counting rule deserves emphasis, because it is the most common way the Quality factor gets inflated: the total count is the denominator, and it must include the reject units, not just the good ones. Some plants quietly count only what passed inspection as "produced," which makes Quality read 100% by definition, there is nothing to divide against. Total count means every unit the process turned out during planned production time, defects included. If your Quality factor never moves off 100%, check whether rejects are even entering the denominator, because a Quality number that cannot fall is measuring nothing.

Does rework count as a quality loss in OEE?

Yes. Any unit that needed rework, repair, adjustment-after-failure, or a retest is a quality loss, even if it eventually ships as good product. OEE counts it against Quality because the rework consumed capacity, the unit was made, failed, and had to be handled again, and that consumed time is a loss whether or not the unit is recovered. This is the same stance first pass yield takes, and it is deliberate: it refuses to let recovery work launder a process failure out of the number.

This is where quality loss hides. If a plant counts reworked units as good, because "they shipped, didn't they?", the Quality factor recovers them and the loss disappears from OEE, while the plant keeps paying for the defect in labor, capacity, and delay. Reworked units also consume capacity twice, so they show up again as availability and performance losses elsewhere, and defect investigations can stop the line entirely, landing in the downtime log. A quality problem, counted honestly, ripples across all three OEE factors. Counted dishonestly, it vanishes, which is why the reported Quality factor is often several points higher than the true one, and why rework has to be logged at the station in real time.

Reference points, from primary sources. The commonly cited "world-class" Quality target is 99% (part of the 90% Availability × 95% Performance × 99% Quality that gives roughly 85% OEE), a reference point tracing to Nakajima's TPM work, useful context, not an audited standard (OEE.com, Calculating OEE). The cost behind quality loss is large: the American Society for Quality notes cost of quality commonly runs 15–20% of sales revenue, up to 40% at some organizations much of it the scrap-and-rework spend the Quality factor is meant to expose.

What is a good OEE Quality factor?

Benchmark against your own line's history and the honesty of your counting rule, not against a published figure. The 99% world-class reference is a useful north star for discrete manufacturing, but a strictly counted 96% that includes every reworked unit tells you more than a soft 99.5% that quietly forgives rework. Two tests matter more than any target:

Does Quality reconcile with first pass yield? They share the same first-pass logic and should roughly agree. If OEE Quality is much higher than station-level FPY, rework is probably being counted as good somewhere.
Is the trend moving? A Quality factor rising over eight weeks, with the top defect code shrinking, beats a flat high number of unknown provenance. Trend and decomposition are the honest use of the metric.

Underneath both tests is capture. The Quality factor is only as truthful as the reject and rework data behind it, and data reconstructed at end of shift systematically undercounts quality loss, startup pieces get waved off, quick reworks go unlogged. Plants that capture defects and rework at the station, feeding live counts into OEE the way Harmony computes true OEE from source signals rather than a spreadsheet (see the platform), find their real Quality factor within weeks. It is usually lower than the reported one, and that honest baseline, paired with the analysis that traces each loss to a cause, is where the improvement actually starts.