Defects Per Unit

Defects per unit is the metric that catches what pass/fail misses. A part can have three small defects and still ship. A part can fail one critical check and never ship. Scrap rate and first-pass yield treat both as binary; DPU counts the defects themselves. On complex products with many features, this difference matters. The conversation about quality is much different when you can see that the average unit has 0.4 defects than when you only know that 96 percent of units passed.

"Pass-fail counts the parts. DPU counts what went wrong with them."

How defects per unit works

The arithmetic is total defects divided by total units. The interesting questions live in the count of defects and the choice of inspection point.

The choices that shape the number

• Inspection scope. All inspections during the process, or just final inspection? Process-wide DPU captures the rework loop; final-only DPU captures what escapes.
• Defect taxonomy. A clear list of defect modes is required. "Defect" without modes is unactionable.
• Same unit, multiple defects. DPU counts them all. This is the metric's signature feature and the reason it differs from yield.
• Sample versus full count. Many shops inspect a sample. DPU based on samples is a statistical estimate; full inspection gives a clean count but costs more.

DPU also serves as the bridge between operator-visible defect counts and the more abstract Six Sigma metrics. The defect log that produces DPU is also the data feed for DPMO and for pareto analysis of failure modes. A shop that has clean DPU tracking has the foundation for any quality improvement project that follows.

Where defects per unit fits on the shop floor

Imagine a 20-person sheet metal fab shop building enclosures for an industrial controls customer. The enclosures have about fifteen quality-critical features each: weld joints, painted surfaces, dimensional tolerances, and threaded holes. The shop reports a 96 percent final yield and considers quality acceptable.

The customer asks for DPU data as part of an audit. The shop sets up a defect log at final inspection with fifteen defect modes derived from the quality plan. The first month of data shows DPU at 0.4. The 96 percent yield number was correct: most units shipped. But on average, every shipped unit had close to half a defect found and corrected before it left the dock. The hidden rework cell handling those corrections is six labor-hours a day.

The Pareto on the defect log is even more useful than the headline DPU. Two modes (paint flow and one threaded-hole burr issue) account for 70 percent of defects. A spray-booth tuning project and a deburring fixture revision drop those two modes by 80 percent over the next quarter. DPU falls from 0.4 to 0.12. The rework cell shrinks to less than two hours a day. None of this was visible in the 96 percent yield headline. DPU surfaced the work.

Common mistakes with defects per unit

• Reporting the headline without modes. A DPU number alone is just a thermometer reading. The defect-mode breakdown is what tells you where to operate.
• Comparing across product complexity. A 0.5 DPU on a 20-feature product is different from 0.5 DPU on a 5-feature product. Without complexity normalization the comparison is unreliable.
• Counting only formal inspection defects. Operator-fixed defects that never appear on a log are invisible to the metric. Capture them or accept that DPU understates.
• Using sample-based DPU as if it were a full count. Sampling produces an estimate. Treat it as one and report the confidence interval where it matters.
• Treating DPU as a target. It is a result of process control. Pushing the number without fixing the failure modes leads to under-reporting, not better quality.

Defects per unit and related Lean tools

Defects per unit is the count-based companion to scrap rate, which measures the share of unrecoverable output. DPU normalizes for unit volume but not for product complexity; defects per million opportunities adds the complexity normalization for cross-product comparison, and parts per million is the customer-facing defective-parts view. Together with first-pass yield, these metrics give a shop a complete view of where defects are happening and how to prioritize the work to remove them.