—— Wira
A type test is not a quality control test, and the sample size reflects that. Three samples are not there to give a statistical estimate. They are there so that the worst of the three can be judged, and so that a single assembly error does not decide the verdict for a whole design.
Understanding why the sample count is what it is changes how the samples are selected, which is the part that most laboratories get wrong.
| Test | Typical sample count | Selection rule |
| Dielectric strength | 3 | All must pass, no averaging |
| Temperature rise | 1 to 3 | The worst position is chosen, not the average |
| Mechanical endurance | 3 | Each sample run to its own end point |
| Impact | 5 positions on 1 sample | Positions, not samples, are the replicate |
| Dimensional | 3 to 5 | Spread matters more than the mean |
| Environmental sequence | 3 | One control, two exposed, in some standards |
The pattern is that replication is used where variability is expected and not where the test is deterministic. Dielectric strength is deterministic in principle, so three samples are a check on assembly, not a statistical sample. Impact is variable by position, so the standard replicates positions on one sample instead.
For a temperature rise test, the worst case is the pole carrying the most current, in the mounting position with the least ventilation, at the top of the rated ambient. That combination is a design decision, and the laboratory has to be able to justify it in the report. Taking the average of three poles is not a permitted shortcut.
The same logic applies to a mechanical test: the sample with the tightest fit, or the one from the least favourable mould cavity, is the one that matters.
Report each sample separately, with its serial and its result. A report that gives a single average for three samples has thrown away the information the third sample existed to provide. Where a standard allows a retest after a single failure, say so and report both results, because the retest is part of the record.
A test sample set prepared to the standard, with the worst case identified in writing before the test starts, is what separates a type test from a demonstration, and the site index lists the equipment used for each test family.
Three samples cannot establish a failure rate. If zero of three fail, the 95 percent upper confidence bound on the failure proportion is about 63 percent, which is not a quality claim. What the sample does prove is that the design and the process are capable of producing a conforming unit at least once, and that the assembly sequence does not contain an outright error.
That is why the count is small and the selection rule is strict. Three samples chosen from one cavity, one shift and one operator prove less than three samples spread across the process, and the report should say which was done. Where a standard allows a retest after one failure, the retest sample comes from the same population, not from a fresh one.
Three is also a practical number in a laboratory. A dielectric test takes 1 minute per sample, a temperature rise test takes 3 hours to reach stability, and an endurance run takes a week, so a set of 3 can be run in parallel on one bench without a second shift. Raising the count to 10 would multiply the laboratory time by 3.3 and would not change the confidence interval enough to matter at this sample size.
For scale, a typical set for a plug and socket product runs the same 3 samples through a 1 minute dielectric test at 2 kV, a 3 hour temperature rise test at 1.1 times rated current, and a 10,000 operation endurance run. Three very different durations are decided by one sample count, which is why the count is written in the product standard and not chosen by the laboratory.
Standards referenced: IEC 60884-1; ISO 2859-1.
IEC60061-3 E27 Lamp cap go gauge and no go gauges
IEC60061 B22 Go No Go Gauge of 7006-11-8 & 7006-10-8
Ipx3 Ipx4 Hand-Held Spray Nozzle
IEC 60529 Ingress Protection Test Probe Kit of Access Probes
Manufacture High Precision IEC60950 Telecom Test Probe
