Three Different Questions Called Defect Rate
When two people in the same factory quote a defect rate, they are often answering different questions. One is counting rejected pieces. Another is counting individual faults on those pieces. A third is counting faults against the number of chances a zipper has to fail. The three numbers can differ by a factor of five on the same lot, and none of them is wrong. The error is quoting one while comparing it to the other.
Before any target is set, the factory must fix three definitions: what counts as a unit, what counts as a defect, and what counts as an opportunity. Without those three, a defect rate is a word rather than a measurement.
Defining the Unit
The unit is the thing that is accepted or rejected. For a zipper it might be a single finished zipper, a set of two matched zippers, a garment-length cut piece, or a carton. The choice changes the number substantially, so it must be stated on the report. Most factories use the single finished zipper for process control and the carton for shipment decisions, and both are legitimate as long as they are labelled.
Defective Unit Rate
The simplest and most widely understood metric is the percentage of inspected units that fail at least one requirement. The formula is defective units divided by units inspected, multiplied by one hundred. For low rates, multiply by one million instead and report parts per million. Data source: inspection records with a pass or fail result per unit, and the inspection standard version in force on that date.
This metric answers "how many pieces will I have to hold back". It does not answer "how many problems are in each piece", which is why a lot with many single-fault rejects and a lot with a few multiple-fault rejects look identical here.
Defects per Unit
Defects per unit, usually written DPU, divides the total number of recorded defects by the number of units inspected. It counts faults rather than pieces, so a zipper with a bent slider and a loose tape end contributes two. The formula is total defects divided by total units. Data source: defect code records at repair stations, final inspection and customer returns, coded to a common list.
DPU is the right metric for repair capacity planning, because repair time scales with the number of faults, not with the number of rejected pieces. It is the wrong metric for shipment acceptance, because a customer rejects pieces.
DPMO and the Opportunity Count
Defects per million opportunities normalizes for complexity, which lets a two-slider bag zipper be compared with a simple closed-end zipper. The formula is total defects divided by the product of units and opportunities per unit, multiplied by one million. The opportunity count is a design decision and must be fixed for a product family. A workable seven-point definition for a standard zipper is:
- Slider function: moves freely, locks and unlocks correctly.
- Puller attachment: secure, correctly oriented, no sharp edge.
- Chain engagement: teeth interlock without splitting under load.
- Tape integrity: width, weave and edge condition within specification.
- Length and end treatment: within tolerance, top and bottom stops present and secure.
- Appearance: shade, plating colour and surface finish match the approved standard.
- Marking and packing: brand, size, lot code and count correct.
DPMO is powerful and easy to abuse. Changing the opportunity count changes the score without any change in quality, so publish the count and freeze it. Data source: the same coded defect records used for DPU, plus the product specification that defines the opportunity list.
First Pass Yield and Rolled Throughput Yield
First pass yield is the share of units that pass a step without rework or repair. The formula is units entering minus units scrapped and reworked, divided by units entering, multiplied by one hundred. Rolled throughput yield multiplies the first pass yield of each step in sequence, so five steps at ninety-five percent each give about seventy-seven percent overall.
Rolled throughput yield is the metric that reveals hidden rework. A factory can report ninety-nine percent final yield while rolled throughput yield sits near eighty percent, because the difference is being absorbed quietly at repair benches. Data source: step-level input, output, scrap and rework quantities from production reporting.
Denominator Discipline
Most inflated or deflated defect rates come from the denominator, not the numerator. Three rules prevent it. First, the denominator must be units inspected, not units produced; if only part of a lot is checked, the rate applies to the checked part. Second, the numerator and denominator must cover the same lot and the same period. Third, re-inspected units must not re-enter the denominator, or the rate will fall simply because failures were examined twice.
A second discipline concerns sample size. A rate calculated from twenty pieces carries very wide uncertainty. If zero defects are found in a sample of n pieces, the upper bound of the ninety-five percent confidence interval is approximately three divided by n. Twenty pieces with no defects is therefore consistent with a true rate near fifteen percent, which is not evidence of a capable process.
Reporting That Survives Comparison
A defect rate report is only useful if it can be compared across months, products and customers. Five elements make that possible:
- The unit definition and the opportunity count, stated in the header.
- The inspection stage: incoming, in-process, final, outgoing or customer.
- The sample size and the sampling method, including whether sampling was random.
- The defect code list version, so that a change in coding is visible.
- The product family breakdown, because a mix shift can move the total with no process change.
YCC keeps one defect code list shared by molding, stamping, plating and assembly, so that a fault found at final inspection can be traced back to the step that created it.
What the Metric Cannot Tell You
No defect rate identifies a cause. It measures an outcome after inspection has already intervened, so it is always a filtered view of the process. A stable rate can hide a rising internal failure rate offset by better sorting. Pair the defect rate with first pass yield and with the cost of failure, and the picture becomes actionable rather than decorative.