Production Quality: Formula, Benchmarks & OEE Link

What is production quality?

Production quality is the consistency with which a line produces parts that meet specification the first time, without rework, scrap or warranty returns. It is one of the three factors in OEE, the Q in availability × performance × quality, and in my experience it is also the one where the gap between reported and real numbers is widest. Availability lies get caught within a week of automatic measurement. Quality lies can survive a quarter because the data is produced by the same people being measured by it.

After 25 years building and rolling out MES and SPC systems across automotive, pharma packaging and FMCG plants, the pattern is always the same. The plant reports 98 percent first-pass yield. The scrap bin tells a different story. The warranty data six months later tells a third one. Aligning those three numbers is where production quality work actually happens, and it starts with measuring the first one correctly.

The formula, straight

Quality Rate (%) = Good Parts ÷ Total Parts Produced × 100

The important word is good. In a strict OEE reading, a good part is one that passes inspection on the first attempt, without rework. A part that needed a touch-up, a rebuff, or a second pass through the welder is not a good part for quality-rate purposes, even if it eventually leaves the plant as sellable. This distinction is where most plants quietly overstate their number.

For OEE contribution:

OEE = Availability × Performance × Quality

A quality rate of 98 percent sounds harmless until it gets multiplied into OEE. A line at 85 percent availability and 92 percent performance running at 98 percent quality produces 76.6 percent OEE. Drop the quality rate to 94 percent and OEE falls to 73.5 percent. Three quality points are worth more than they look.

Realistic quality benchmarks by process

Usable ranges from real installations. Use them as sanity checks, not as targets.

Quality rate is deceptive because the numbers sit close together at the top end. The difference between 98 and 99.5 percent feels cosmetic. In an automotive plant running 500,000 parts per week, it is 7,500 extra defective parts per week versus 2,500. The warranty cost of those 5,000 extra parts is usually larger than the entire MES programme that would have caught them in real time.

Why reported quality is almost always too high

Four biases inflate quality numbers in practically every plant I have audited.

1. Rework gets counted as good. A part that took a second trip through the station, or five minutes of manual touch-up, ends up in the same count as a part that ran right first time. First-pass yield is the honest number. Total yield after rework is the number that usually gets reported.

2. Defect classification is loose. Operators catching a borderline part late in the shift tend to pass it through rather than stop production. End-of-shift classification on a paper sheet produces a smoothed story that does not match the actual flow of events.

3. Sample inspection misses drift. Sampling every 100 parts is efficient but catches systematic drift only after the fact. A tool wears, a process parameter drifts, and 200 parts ship with a marginal dimension before the next sample catches it. In the quality numbers, those 200 parts may never appear as defects because nobody inspected them individually.

4. Warranty returns and internal scrap live in different databases. Production reports quality rate from the shopfloor. Customer quality reports complaints separately. Finance sees warranty cost without connecting it back. Three departments, three numbers, no single source of truth. Six months later, the OEE report still shows 98 percent while the warranty line item grows.

Practical rule: if your reported quality rate is above 99 percent and it came from a manual tally, treat it as a working hypothesis, not a fact. The real number is almost always 0.5 to 2 percentage points lower once automatic part-level capture turns on.

What actually moves quality upward

Four interventions, in the order of return on effort.

1. Part-level tracking with serialisation or batch IDs. Every good and bad part tagged at source, linked to the exact machine, operator, tool and timestamp that produced it. Without this, root-cause analysis is guesswork. With it, the Pareto of defect causes writes itself every week.
2. SPC on the critical characteristics, not on everything. Statistical Process Control is powerful when focused. Three to five critical-to-quality characteristics per product, monitored with control charts that have meaningful upper and lower limits derived from process capability studies, catch drift before it produces scrap. Twenty SPC charts that nobody looks at produce nothing but paperwork.
3. Inline sensors and PLC alarms correlated with quality events. The Neoperl assembly rollout is the cleanest example from our own portfolio. Correlating PLC alarm codes with rework events produced 15 percent less scrap and 10 percent fewer stops within the first year, not through new equipment but through connecting two data streams that already existed.
4. Structured operator feedback into the quality database. The people closest to the defect see it first. Giving them a 5-second tap-tile on a touchscreen with the top ten defect categories, linked to the running order and machine, beats any end-of-shift paper sheet.

Quality in the Industry 4.0 context

Modern production quality is no longer a standalone discipline, it is an output of connected machine data, MES rules and analytics. Real examples of what this means in practice:

• Condition-based quality triggers. A humidity sensor above threshold pauses production automatically or routes parts to a secondary inspection loop instead of relying on an operator remembering.
• Model-based anomaly detection. Process parameters (temperature, pressure, torque, cycle time) correlated with historical defect data flag suspect parts in real time, before the quality inspection station sees them.
• Closed-loop feedback to the PLC. When a drift pattern is detected, parameters are adjusted automatically within the validated range. Human supervision stays in the loop but the latency drops from shifts to seconds.

None of this is science fiction. All three are running today in SYMESTIC deployments. What makes it work is not the algorithm, it is the fact that the data is captured correctly at the source. If first-pass yield is miscounted on day one, no amount of AI fixes it downstream.

FAQ

Is quality rate the same as first-pass yield?
In the strict OEE definition, yes. First-pass yield is good parts at first attempt divided by total parts produced. Rework cannot upgrade a part back into the numerator. Some plants define quality rate more loosely to include reworked parts as good, which inflates the number but removes its usefulness as a process indicator.

Does scrap count differently from rework?
For OEE quality purposes, no. A part that was scrapped and a part that needed rework both fail first-pass yield. For cost analysis the distinction matters, because rework has a labour cost but not a material cost, while scrap has both.

What is a realistic Six Sigma quality level in practice?
Six Sigma means 3.4 defects per million opportunities, which is 99.99966 percent. Real-world mid-size manufacturing plants operate at roughly 3 to 4 sigma, which is 93 to 99.4 percent. The gap between marketing claims and operational reality in this area is larger than in most metrics.

How often should quality be measured?
Continuously at part level where automation allows, every cycle where PLC alarms and sensor data exist, every sample interval otherwise. Shift-end aggregation is useful for reporting but useless for intervention.

What software measures production quality automatically?
An MES platform with quality and SPC modules, integrated with inline inspection equipment via OPC UA or digital I/O. SYMESTIC Production Metrics captures quality events at part level as a day-one capability and connects them to the production order, machine and operator automatically.

Why did our quality number drop after automating measurement?
Because for the first time it is correct. The previous number was built on operator estimates and unreported rework. The new number is lower and true. This is not a regression, it is the baseline from which real improvement finally becomes possible.

Related: OEE · Machine Availability · Operating Time · Bottleneck · Process Monitoring · MES · SYMESTIC Production Metrics

About the author

Christian Fieg

Head of Sales at SYMESTIC. 25+ years in manufacturing including Johnson Controls, Visteon, iTAC and Dürr. Six Sigma Black Belt with three years on the DMAIC side of the Johnson Controls headliner lines. Led global MES and traceability rollouts across 900+ machines in China, Mexico, USA, France, Tunisia and Russia. Author of OEE: Eine Zahl, viele Lügen (2025). · LinkedIn