Metal detector daily verification SOP - three-material all-position startup checks; on failure, isolate and re-run everything since the last passing check

You are the equipment maintenance lead at a central
kitchen. The metal detector / X-ray machine is this
plant's CCP equipment. I need a "daily verification SOP"
answering two questions: what counts as a valid check,
and what happens when a check fails - especially, what
happens to everything that ran between the last pass and
the failure.

1. Startup verification (before production each day)
- Test pieces: ferrous (Fe), non-ferrous (Non-Fe), and
  stainless steel (SS); sizes per the equipment manual
  and our product profile; piece IDs logged
- Pass positions: each test piece passes through the
  aperture at [left/center/right × top/middle/bottom
  per the manual] and at the head, middle, and tail of
  the belt
- Definition of pass: every material detected at every
  position, and the reject / alarm action executed every
  time - detection without rejection counts as a
  failure
- Record: position × material × detected × reject
  executed × tester, written on the spot

2. In-production checks
- Recheck every [2] hours or at every batch change,
  whichever comes first; frequency is the stricter of
  customer requirement and plant policy
- Standard mid-belt position, all three materials

3. Failure response (four steps, order fixed)
a. Stop the flow: machine out of service, hang the
   fault tag; subsequent product goes to temporary
   manual hold - no routing around the detector
b. Isolate and trace back: from the last passing check
   record, isolate and re-run all product that crossed
   the detector. The starting point is the passing
   record - "it was probably fine in between" does not
   count, and "visual inspection instead of re-run"
   does not exist
c. Troubleshoot and restore, in fixed order: test-piece
   placement (the most common false failure) → probe
   window contamination → parameter drift → belt
   speed. Anything beyond self-repair goes to the
   equipment vendor. After restoration: three
   consecutive all-position passes before restart
d. Record and report: failure time, downtime, quantity
   isolated, re-run results - reported to QC the same
   day. This record is a standing customer-audit item

4. Verification record form (attached to this SOP)
- Daily page: date / shift / position / material /
  detected / reject executed / tester / reviewer
- Failure page: separate sheet, including the batch
  list of isolated product

Standards:
- Test pieces are periodically re-certified or replaced;
  a worn test piece's "pass" is not credible
- Test pieces live in a fixed, lockable location beside
  the machine - never pocketed, never mixed. A lost
  test piece means the verification system itself has
  failed

Constraints:
- No specific sensitivity settings - those come from
  vendor commissioning and product validation on our
  line; this SOP governs how to verify and what to do
  on failure
- No "operators should stay alert" - every step in an
  SOP is an action, not a mood

Why it is written this way

"Detection without rejection counts as a failure" is this SOP's teeth. The metal detector's most common silent failure is a jammed reject - the flap or blast valve sticks while detection functions normally: alarm lit, contaminated product passing, nothing stopped. A verification routine that watches only the alarm light never finds this failure mode until a customer complaint delivers the metal. Folding "reject executed" into the definition of pass makes one verification cycle test both subsystems - the most valuable sentence in the file.

Anchoring the traceback to "the last passing record" closes the most expensive negotiation in failure response. Once a failure surfaces, the argument is always how far back to go - production wants an hour, QC wants the whole day. Left as an unwritten consensus, it gets bargained down exactly when volume is highest. Written as a record anchor, there is nothing to negotiate. The explicit ban on "visual inspection instead of re-running" exists because metallic foreign matter is precisely what eyes cannot see; leave the option on the table and it will always be chosen.

Putting test-piece placement first in the troubleshooting order reflects field reality: a large share of "detector not working" service calls turn out to be the piece passing outside its specified position - hugging the belt edge, where sensitivity naturally drops, on a healthy machine. Checking the human before the machine avoids flying a technician across town for a false alarm, and just as often keeps a real fault from being waved off as one.

Publishing no sensitivity numbers keeps the SOP from playing fake expert. Sensitivity interacts strongly with product effect - the product's own conductivity and moisture - and another plant's settings mean nothing on our line. Valid parameters come from vendor commissioning plus validation on our product. This SOP governs process, not parameters; drawn plainly, that boundary lets the file survive scrutiny from both the vendor and an auditor.

Variables