A metal detector is a critical control point when it's the last step that can catch metal before the food is sealed and shipped. Its critical limit isn't "no metal", which nobody can monitor. It's this: the detector is on, every pack goes through it, and it rejects each of its test pieces every time it's checked. The test pieces are the smallest ferrous, non-ferrous and stainless steel spheres the machine reliably finds in your product, worked out when it's set up. FDA treats a hard or sharp object 7 to 25 mm long in ready-to-eat food as adulteration, and smaller ones where infants or the elderly are among the consumers, so test pieces sit far below that line. Below are the limit, the checks, what to do when one fails, and a worksheet.
The short version
- The limit: detector on, and all three test pieces rejected at every check.
- Test pieces: ferrous, non-ferrous and stainless steel, sized for your product and aperture.
- Check at start-up, at a set interval, at each product change, after repairs, and at the end.
- A failed check puts everything since the last good check on hold.
- Open and look at every reject.
- TaroStack keeps each check with its batch, and the lot on hold. More below.
Is it a CCP?
| Where metal comes from | Example | What should catch it first |
|---|---|---|
| Ingredients | Wire, staples, shot, harvester parts | Supplier specification, sieve, magnet |
| Equipment | Blade chips, mixer bolts, sieve wire, can-opener shavings | Blade and sieve checks, maintenance |
| People and repairs | Tools, drill swarf, a nut left behind | Clean-up and sign-off after every repair |
| Packaging | Carton staples, foil | No staples in the room |
The detector is a CCP if your hazard analysis finds metal reasonably likely and the detector is the control you rely on to catch it. Under FDA's preventive controls rule it's a process control like any other, with a limit, monitoring, corrective action, verification and records. It doesn't replace the first-line controls in the table. A detector finds a blade tip. It doesn't find the reason the blade broke. The written program for all of them is in foreign material control program.
The limit, and the test pieces
A made-up line packing bagged granola. Its sizes are its own.
| Critical limit | Detector on; every pack passes through; each test piece rejected into the locked bin at every check |
| Test pieces | 2.0 mm ferrous, 2.5 mm non-ferrous, 3.0 mm stainless steel |
| How | Each piece sent through inside a pack of product: at the leading end, the middle and the trailing end |
| When | Start of the run, every hour, each product change, after any stop or repair, the end of the run |
Why three metals. Detectors find iron most easily and stainless steel least, and most food equipment is stainless.
Why sizes differ by product. Wet, salty or warm food (fresh bread, cheese, meat) gives a signal of its own, called product effect, and the detector has to be set less sensitive to ignore it. A larger aperture is less sensitive too. So nobody can give you sizes from a table. Whoever sets up the machine establishes what it reliably detects in your product, and those become your test pieces. Customers and audit standards often set maximum sizes as well: check theirs.
Why inside a pack. A test piece waved through an empty aperture proves little. It should travel as the product does, and be rejected as a pack is.
FDA's own line is much larger than any test piece. CPG Sec. 555.425 treats hard or sharp objects 7 mm to 25 mm long in ready-to-eat food as adulteration, and smaller ones when the food is for infants, the elderly or others at special risk. A detector that finds 2 to 3 mm spheres is working well inside it.
When a check fails
- Stop the line.
- Put everything packed since the last good check on hold. This is why the interval matters: hourly checks mean an hour of product at most.
- Find the cause: the wrong product setting, sensitivity changed, or a reject device that didn't fire.
- Fix it, and pass all three test pieces again.
- Send the held product back through the working detector.
- Write down what failed, what was held, and what happened to it.
Rejects. Every rejected pack goes into a locked bin, and someone opens each one. Real metal is a finding: what is it, where did it come from, and is the rest of the batch safe? A false reject is information too. Packs put back on the line unexamined are how a working detector fails.
Verification. Have the detector serviced and calibrated on a schedule, keep the certificate, and have a second person review the check records. Under FDA's rule, monitoring and corrective action records are reviewed within 7 working days.
A morning of checks
| Time | Check | Ferrous 2.0 mm | Non-ferrous 2.5 mm | Stainless 3.0 mm | OK? |
|---|---|---|---|---|---|
| 7:00 | Start of run | Rejected | Rejected | Rejected | Yes |
| 8:00 | Hourly | Rejected | Rejected | Rejected | Yes |
| 9:00 | Hourly | Rejected | Rejected | Not rejected | NO: act |
| 9:25 | After repair | Rejected | Rejected | Rejected | Yes |
| 10:00 | Product change | Rejected | Rejected | Rejected | Yes |
At 9:00 the stainless piece went through. The reject flap's air line had worked loose: the detector saw the piece and nothing pushed the pack off. The 412 packs made since 8:00 were held, the line was fixed, and all 412 went back through at 9:40. None was rejected.
This is general information, not legal advice. The log on its own is in metal detector log, glass and hard plastic in glass and brittle plastic policy, and how a limit is written in critical limits for CCPs.
Get the worksheet
Download the metal detector CCP worksheet. Three sheets and a read-me: the hazard analysis for metal from receiving to packing, the plan for the detector as a control point, and a monitoring log. It's an ordinary Excel file with no macros, and it works in Google Sheets, LibreOffice and Numbers.
Where this stops working
The 9:00 check failed. Which packs were made between 8:00 and 9:00? If the answer is "the ones on that pallet, probably", and half of that pallet left on the 8:30 truck, the hold has become a recall. A paper log records that a check failed. It doesn't say where an hour of product went.
How TaroStack does it
TaroStack keeps the checks with the batch they belong to.
Each run is a batch with a lot code. Your checks are fields on its record: test pieces rejected, time, initials.
Put the lot on hold when a check fails, and it's kept out of what you can sell until someone releases it.
If some has shipped, one report lists which customers got that lot and how much. Starting a recall holds what's left.
Lot labels print with the code, so the pallet says which batch it is.
| The failed check | Paper log | In TaroStack |
|---|---|---|
| Which packs since the last good check | That pallet, probably | The batch's lot |
| Keeping them off the truck | A sign on the pallet | On hold |
| What already shipped | Bills of lading | One report |
| Proof the check was done | A clipboard | On the batch record |
TaroStack doesn't run the detector or set its sensitivity. Recipes, batches, lots and recalls are on every plan, from $9.99 a month (5 recipes) or $39 a month (50 recipes, with allergen tracking). The first 14 days are free, with no credit card. If you'd like a hand setting up, ask, and we'll do it with you.
