The tablet's mechanical reality
Before the machine, the tablet has already decided part of your defect rate. Friability above 0.8 % sheds fines that bridge pockets and trip the camera; a thickness tolerance wider than ±3 % lets a tablet sit proud and crush at the seal; capsules need a gentler orbital feeder than compressed tablets. Build the upstream controls first:
- Metal check and dedusting before the blister hopper, not after.
- Tablet thickness and hardness logged per batch; feed parameters locked to the result.
- Capsule versus tablet feeding configured as two recipes, not one compromise.
Feeding architecture and why dedusting belongs before the blister
Dust generated at the feed is the quiet killer of seal integrity and inspection reliability. On a 200 bpm line we measured hopper dust load at 6–11 mg/m³ without extraction and 1.2 mg/m³ with a 350 m³/h vacuum hood upstream of the feeder. The cleaner feed cut camera false-rejects from 14 to 3 per 10,000 and reduced seal-zone contamination complaints.
| Condition | Dust load (mg/m³) | False rejects / 10k |
|---|---|---|
| No extraction | 6–11 | 14 |
| 350 m³/h hood upstream | 1.2 | 3 |
Inspection and rejection logic
Missing-tablet detection must cover every pocket at full speed. The decision that costs OEE is rejection granularity: rejecting the whole blister for one missing tablet is simple but wastes good product; single-tablet ejection is elegant but adds a station and its own failure mode. We size this against false-reject cost.
- Camera coverage: 100 % of pockets, dual-side where tablets are small.
- Sensitivity / specificity: target 99.5 % / 99.8 % on a clean feed.
- False-reject cost: at a 1.5 % false-reject rate and 400 bpm, you scrap ~360 blisters per shift of good product.
Format changeover: SMED applied to blister tooling
Changeover is where the calendar dies. A traditional blister change — forming plate, sealing plate, feeder track, lidstock core, print roller, and format camera recipes — ran 95 minutes. Applying SMED took it to 34:
| Step | Before (min) | After (min) |
|---|---|---|
| Tool preparation (off-line) | 22 | 8 |
| Plate swap + poka-yoke | 41 | 14 |
| Recipe load + first-article check | 32 | 12 |
Poka-yoke format parts (keyed, colour-coded) stopped the recurring "wrong plate" restart. Parallel work — one operator swapping while a second pre-stages tooling — was the single biggest saving.
In-process control plan
| Parameter | Method | Frequency | Limit |
|---|---|---|---|
| Seal integrity | Vacuum decay sample | 30 min | 0 leakers / 20 |
| Missing tablet | Camera log review | Continuous | < 2 / 10k |
| Forming depth | Pin gauge | 60 min | ±0.15 mm |
| Print / code | Vision verify | Continuous | 100 % read |
| Reject confirmation | Physical count | Shift | Confirmed removed |
Anonymised case: EU GMP solid-dose site, 3-shift
A 12-format facility ran 268 bpm effective against a 400 bpm nameplate. After moving dedusting upstream, re-cutting changeover to 34 min, and locking IPC to the table above, effective output reached 357 bpm with defect ppm down from 1,900 to 410 over six months. The packer did not change; the process around it did.
Validation: IQ/OQ/PQ and data integrity
EU GMP Annex 11 and 21 CFR Part 11 put the recipe management, audit trail and user levels inside the validated scope. PQ must run on the worst-case product and speed, not the easiest. Require: documented IQ of the line, OQ of each critical parameter with proven acceptable ranges, PQ across three consecutive batches, and an audit trail that cannot be silent-edited. Missing-tablet logic and reject confirmation are critical parameters — they belong in the OQ, not the service manual.
Regional notes
- EU: Annex 1 cleanroom grading and Annex 11 data integrity are the binding ceiling; budget validation time accordingly.
- Southeast Asia: 30–34 °C and 80 % RH push WVTR budgets; size the web for the climate, not the catalogue.
- Middle East: 45 °C ambient and Arabic + Latin print need a wider code quiet zone and proven first-read rate.
Selection mistakes
- Buying to nameplate speed and ignoring changeover minutes.
- Treating missing-tablet detection as a checkbox rather than a critical parameter.
- Forgetting dedusting upstream of the hopper.
- Skipping poka-yoke format parts, then repeating wrong-plate restarts.
FAQ
Should I reject the whole blister or eject one tablet?
Whole-blister reject is simpler and safer for low-value products; single-tablet ejection pays back only where scrap cost and speed justify the extra station. Size it against false-reject cost.
How fast can a real changeover be?
34–45 min with off-line tool prep, keyed format parts and two operators. Above 90 min, you are losing more to changeover than to downtime.
What goes in the OQ for this machine?
Seal window, forming depth, missing-tablet sensitivity/specificity, code read rate, and reject confirmation — all with proven acceptable ranges.
Is dedusting really needed before the blister?
Yes. Feed dust drives false rejects and seal contamination. A 300–400 m³/h hood upstream typically cuts false rejects by 70 %.
How does EU Annex 11 affect the packer?
Recipe management, audit trail and user access become validated functions. The PLC platform and its change control must be documented.
Can the same packer run capsules and tablets?
Mechanically yes, but as two recipes with different feeder settings. Lock parameters per product to protect tablet crush and capsule bridging.
What PQ evidence does a regulator expect?
Three consecutive validated batches at worst-case product and speed, with IPC records and confirmed reject handling retained in the batch record.
Why does humidity change my defect rate?
High RH raises WVTR risk and can soften some tablet coatings, increasing fines. Size the web and the environmental control together.