Why Sampling Misses the Bad Round: 100% Weight Inspection

Every ammunition plant inspects. The real question is how many rounds get inspected. For decades the standard answer has been a sample: pull a set number of rounds from a lot, measure or fire them, and accept or reject the whole lot on the result. Sampling is well understood, it is written into military specifications, and for many characteristics it is the only practical option. A cartridge cannot be fired and then shipped.

Weight is different. Weighing does not destroy the round, and it can now be done at line speed. That changes the math. This post shows why, using a published recall as the worked example.

What sampling is designed to do

Acceptance sampling answers a statistical question: given what we found in the sample, how confident are we about the lot? It is built to control the rate of defects across many lots. It is not built to find the individual defective round.

Military quality programs are open about how sampling is used. A 2008 U.S. Army article on quality control at Lake City Army Ammunition Plant quoted the plant's quality assurance lead: "We test random samples and have a 98 percent lot acceptance rate." The same article warned that "when the standard is not met, the result could be disastrous and result in injury or even death."

The sampling method itself is still being studied. A 2023 paper presented at the Donald R. Keith Memorial Conference at West Point, Analysis of Statistical Sampling Procedures at an Army Ammunition Plant, looked at destructive acceptance testing of .50 caliber ammunition. It concluded that "the current process for testing ammunition does not provide the most accurate risk assessment due to a biased sample."

Neither source says sampling is wrong. Both make the point this post is about: a sample gives a probability, not a guarantee about every round.

A worked example: one bad round in 430

The AFTE recall and safety warning list includes a 1992 Winchester notice for a lot of 22 Subsonic. It reported that "approximately 1,024 double powder charge weight cartridges may have been produced out of 440,000 rounds manufactured" in that lot.

That is about one defective cartridge in every 430. For a manufacturing defect that is rare. For the shooter who chambers one of those 1,024 rounds, it is not rare at all.

Now suppose that lot is sampled at random and inspected perfectly. The chance that a sample contains no double charges at all works out like this:

Calculated as (1 − 1,024/440,000) raised to the sample size. It assumes random draws and perfect detection.

Even a 500-round sample would miss the problem almost one time in three. And if a sample does turn up one bad round, the lot is quarantined after the fact. Every round in it then needs to be sorted, reworked or scrapped.

This is the basic limit of sampling. When defects are rare and severe, the sample size needed for confidence grows to the point where weighing the whole lot is the simpler option.

What 100% weight inspection changes

A loaded round's weight is the sum of its parts: case, primer, powder and projectile. Change any of them and the total changes. That is why weight is such a useful single measurement. Our AccuWeigh 360 is built to catch "a missing primer, a light or heavy powder charge, a wrong or defective projectile, damaged brass."

The figures below are from the AccuWeigh 360 product page:

• Resolution: it weighs "every projectile, case or loaded round to within a tenth of a grain."
• Throughput: the AccuWeigh 360 is modular. It starts at 3,600 rounds per hour with one weigh cell and scales past 12,000 rounds per hour as weigh cells are added. One weigh cell handles 28,800 rounds in an eight-hour shift.
• Limits: set "as tight as 0.1 grain or as wide as ten grains for large rounds."
• Records: it stores "up to 25,000 preset configurations and a record of every part weighed."
• Labor: "84% lower QC labor," at "six times the throughput of hand weighing."

With a single weigh cell, at 28,800 rounds a shift, an AccuWeigh 360 would weigh every round of a 440,000-round lot in a little over 15 shifts. Configured with enough weigh cells to run past 12,000 rounds per hour, the same AccuWeigh 360 would weigh that lot in under five eight-hour shifts. Multi-cell configurations are quoted per line.

Inline beats end-of-line

Where the inspection sits matters as much as how many rounds it covers. A sample pulled after a lot is complete tells you about a problem only after the whole lot has been made.

An AccuWeigh connected to a loader works differently. In the product page's words, "it catches a bad run as it happens instead of after a whole lot is loaded." If a powder measure starts drifting, the first out-of-limit rounds are rejected and the trend shows up in the data. The operator can stop the line after a handful of rounds rather than thousands.

Rejecting what the machine cannot trust

Inspecting every round is only worth it if the decision on each round is trustworthy. Our patent-pending firewall decision logic is built to fail safe: "when a reading cannot be trusted, the part is rejected rather than assumed good."

The weigh platform is designed to twelve sigma. On our technology page we define that as "roughly one component or round weighing a tenth of a grain outside its limit and still passing in twelve billion." That is a design target, and we state it as one.

Every round on the record

Recall notices are written around lot numbers. The more precisely a manufacturer can show which rounds were measured and what each one weighed, the more defensible its quality claims and the narrower any recall can be drawn.

AccuWeigh keeps "a record of every round for the life of the unit." Through the optional app, "weight data by lot number or by individual round" is available, and an optional remote data feed can send results to a central system. With that record, a plant can show every round was weighed, not just state that the lot passed a sample.

Where sampling still belongs

We are not arguing that sampling should be dropped. Ballistic and pressure testing, primer sensitivity, and other destructive tests can only be done on a sample. Weight cannot detect everything, either. A round with the right weight of the wrong powder, for example, has to be controlled upstream through powder handling and lot control.

The practical approach is to use each method where it is strongest:

1. Measure every round for the characteristics a non-destructive check can catch: weight, dimensions and visual defects.
2. Sample for the characteristics that can only be tested destructively.
3. Record both, so a lot's history covers every round, not just the sample.

For visual and dimensional defects, our AccuVision 360 works as "a virtual case gauge for every round, at 6,000 to 10,000 rounds per hour." For more background on weight as a QC measure, see our earlier post, Weight as a quality control measure.

Put a number on your line

Tell us your calibers, components and volume per shift, and we will show you what weighing every round would look like on your floor, including how many weigh modules it takes. Tell us what you load. We will spec the line.