To verify count accuracy on an automated counting machine, run reference batches with a known count through the machine under normal production conditions, compare the machine’s count with the known count, and calculate the error rate over enough parts to be statistically meaningful. Then repeat a short version of that check routinely, so drift is caught before a short shipment reaches a customer.
This article is part of our automated counting guide, which explains how image-based counting machines work and where they fit. Here we focus on proving and keeping accuracy once a machine is on your floor.
Why count accuracy needs verifying
A counting machine is only as useful as the trust people place in its number. Short counts lead to customer complaints and credits; over counts give product away. Both are invisible until someone recounts, so a simple, repeatable verification routine is what turns a machine count into a number shipping, quality and customers will accept.
What causes miscounts
| Cause | What happens | How to prevent it |
|---|---|---|
| Touching or stacked parts | Two parts read as one, or one large blob is split | Feed rate and part spreading set for the part; image analysis tuned for that part type |
| Debris and fragments | Chips, flash or broken pieces counted as parts | Size and shape limits on what counts; keep the counting surface clean |
| Lighting or optics changes | Dirty lens or backlight, ambient light, a failing lamp | Regular cleaning; enclosed lighting; a daily reference check |
| New part variants | A new color, finish or size behaves differently | Add and verify each new part type as its own recipe |
| Running too fast | Parts arrive faster than they can be separated | Verify accuracy at the speed you actually run, not a slow demo speed |
| Operator handling | Parts added or removed during a count | Simple screens, clear start and stop, and a count log |
How to run a count accuracy verification
- Prepare reference batches. Count each batch manually twice, by different people if possible, until both counts agree. Include your smallest, largest and hardest-to-separate parts.
- Run at production conditions: the normal feed rate, lighting, operators and batch sizes.
- Count each batch several times through the machine, re-mixing between runs.
- Record every result against the known count, by part type.
- Calculate the error rate as total miscounted parts divided by total parts counted, and review any miscount to find its cause.
How many parts is enough?
A machine that counts 5,000 parts without a single error has not been proven perfect. A useful rule of thumb (the “rule of three”) says that zero errors in n parts gives about 95% confidence that the true error rate is below 3 ÷ n.
| Parts counted with zero errors | Error rate below (about 95% confidence) |
|---|---|
| 500 | 0.6% |
| 1,000 | 0.3% |
| 5,000 | 0.06% |
| 30,000 | 0.01% (100 per million) |
If you do see errors, the observed rate is simply errors divided by parts: one miscount in 5,000 parts is 0.02%, or 200 parts per million. Decide the accuracy you need from the cost of a short or over count to your customer, then size the verification to prove it.
Keeping a counter accurate day to day
- Start-of-shift check: count one small reference batch per part type in use. It takes a minute and catches most problems.
- Clean on a schedule: the counting surface, lens and lighting, more often in dusty environments.
- Log counts and checks: a counting machine that records each batch gives traceability when a customer questions a shipment.
- Re-verify after changes: a new part type, a moved machine, a lighting change or a software update.
- Watch the trend: occasional reference check misses that start to repeat are an early warning.
How this compares with counting by weight
Counting scales estimate count from total weight divided by an average piece weight, so their accuracy depends on how much individual parts vary in weight, and on the sample used to set the average. Image-based counting counts each part directly, so piece-weight variation does not affect it. The verification method above works for both, and is a fair way to compare them on your own parts. See automated counting vs counting scales for the full comparison.
For owners and operations leaders
Accurate counting pays back through fewer credits, fewer recounts and less labor; Sciotex counting machines can pay back in a few weeks. A simple verification routine protects that payback and gives your customers a number they can trust.
Want to see how accurately your parts count?
Send us your parts for free sample testing. Sciotex counting machines are designed, programmed, built and supported in the US, with no subscription or support contract required.
Frequently asked questions
Run reference batches with a known count, verified by two manual counts, through the machine at normal production speed several times, then divide the total miscounts by the total parts counted.
Enough to prove the accuracy you need. With zero errors, about 95% confidence that the error rate is below 3 divided by the number of parts counted, so 5,000 error-free parts supports an error rate below 0.06%.
Common causes are touching or stacked parts, debris, dirty optics or changed lighting, new part variants that were not set up, running faster than the parts can be separated, and handling during a count.
A quick reference batch at the start of each shift for each part type in use, plus a full verification after any change to parts, lighting, location or software.
Image-based counting counts each part directly, so it is not affected by piece-to-piece weight variation, which limits counting scales. Verifying both on your own parts is the fairest comparison.