What Actually Moves the Number

Three sensing technologies, and what each is bad at

Most buyers ask which sensor on a capsule counting machine is the most accurate. The better question is which one drifts slowest in your dust. A through-beam photoelectric array counts interruptions of an infrared beam: fast, cheap, and the first thing to lose calibration when polishing talc films the lens. A machine-vision head reads a full frame per capsule and can be trained on banded or printed shells, but it costs two to three times as much and needs its own window-cleaning discipline.

Two things I flag during selection. Vendors quote capsules per hour using size 0 hard gelatin in a clean batch, so ask for that figure re-quoted on your smallest capsule and your dustiest product. And a vision head only earns its premium when SKU count and changeover frequency are high.

The capsule failure modes I check first

Capsules misbehave differently from tablets in every way that matters here: they are light, they bounce, and they carry charge.

Static: the low-humidity failure nobody budgets for

HPMC shells, and gelatin to a lesser degree, pick up charge as they tumble down a stainless channel. Below roughly 40% RH the surface resistivity climbs and capsules cling to the channel wall and to each other. Two that cling together through the sensing plane are read as one; one that clings at the gate and releases late lands in the next bottle. I specify 45–55% RH in the counting room, ±5%, as a machine utility rather than a facilities afterthought. In a Gulf plant that means humidification against a 24 °C supply air setpoint; in Southeast Asia it more often means dehumidification, because ambient sits at 70–80% most of the year.

Talc film and the slow drift it causes

If shells are polished with talc or magnesium stearate, some of that becomes airborne at the counting head. On a photoelectric array the film lifts the baseline and flattens contrast, so accuracy is fine at 08:00 and marginal by the end of second shift. On one audit job we recorded a 0.4 percentage-point fall in exact-bottle rate across a single eight-hour shift on a talc-polished product. Moving polishing upstream and adding extraction removed most of it.

• Locked and telescoped capsules. An unopened shell is longer and heavier than a filled one, and a telescoped pair can cross the beam as one object. Reject them upstream with a polisher that discriminates length or weight.
• Broken shells and empties. A split shell weighs a fraction of a good unit and can bounce out of the sensing plane. A checkweigher catches most of these, but only if the band is tight enough to see 150 mg on a 30-count bottle.
• Bounce at the drop point. A capsule that rebounds through the beam is counted twice. Keep the gate-to-neck drop at 60–120 mm with a soft-catch funnel, not a 200 mm free-fall.
• Banded versus unbanded shells. Banded capsules are stiffer, the seam can sit across the beam, and banding fluid changes surface friction. Treat them as two recipes.

Feed and singulation: the part nobody specifies

Almost every counting complaint I have diagnosed was a feed problem wearing a sensor costume. Hold the hopper bed between roughly 30% and 70% with a vibrating fork or capacitive probe, because a deep bed compacts and bridges while a low bed starves and then surges. Use a two- or three-stage vibratory cascade with independent amplitude per stage, typically 40–80% of maximum, tuned so the last stage delivers a single-file stream rather than a ribbon. Fit a servo cut-off gate: a pneumatic gate at 40–60 ms throws in a trailing capsule on every cycle. Extract at the feed point at 150–300 m³/h per head and about 0.5 m/s capture velocity, balanced so it removes airborne talc without lifting capsules out of the channel.

Proving accuracy: a challenge test you can repeat

A spec-sheet figure means nothing without a protocol. This is the one I run at FAT and repeat at OQ, on the customer's own capsules and on their worst batch rather than a clean sample.

The number buyers misread. A per-capsule accuracy of 99.9% does not mean 99.9% of bottles are correct. On a 30-count bottle it is 0.999³⁰, so about 97.0% of bottles are exact. To get 99.5% of bottles exact you need roughly 99.98% per capsule. Always ask a vendor whether the quoted accuracy is per capsule or per container.

Documentation, data integrity and conformity

Qualification follows the IQ/OQ/PQ structure of EU GMP Annex 15, with challenge tests in OQ and drift data in PQ. Where the control system holds recipes and batch records, Annex 11 covers the computerised system, and for supply into the US, 21 CFR Part 11 governs electronic records and signatures: unique user IDs, an audit trail recording who changed a setpoint and when, and a batch report that cannot be edited afterwards. The drug CGMP obligations sit in 21 CFR 211. On the machine side I ask for a CE declaration under Machinery Regulation (EU) 2023/1230, which replaces Directive 2006/42/EC, supported by a risk assessment to EN ISO 12100 and electrical safety to EN 60204-1. ISO 13485 matters only if the line also makes a medical-device combination product.

What a capsule line actually delivers

These are the bands I quote when a buyer asks what a capsule counting machine will hold on a real floor. All figures are size 0 and size 1 hard gelatin at 30-count.

Field case: three-shift contract packer in Southeast Asia

A packer running hard gelatin sizes 0 and 1 plus one HPMC line, three shifts six days a week, called us in at 60-count HDPE bottles against 34 bottles per minute. What failed first was not the counter. The plant ran a chilled-water air handling unit with no humidification, so in the dry season the room fell to 32–35% RH and capsules clung in the channels before releasing in pairs. Polishing happened in the same room, so the lens window filmed within a shift, and the gate-to-neck drop was 210 mm, visibly bouncing the size 1 product.

We changed four things and left the detectors alone. Humidification now holds 50% ±5% RH. Polishing moved upstream behind a deduster with 250 m³/h of extraction at the head. Drop height went to about 95 mm with an anti-bounce deflector. Lens cleaning became a per-shift item with 70% IPA, and recipes were locked behind operator-level access. Measured over two weeks either side: miscount-driven rejects fell from 1.9% to 0.35%, the exact-bottle rate rose from 93.1% to 97.6%, OEE moved from 61% to 74%, and changeover dropped from 40 minutes to 24.

Procurement traps I keep meeting

Spec-sheet speed is the obvious trap; the expensive ones are quieter. Change parts get quoted for the two sizes in the enquiry, not the two you add in year two. FAT runs on your best capsules rather than your worst, which hides feed surging. Spare lens sets are left off the order, so one scratched window becomes a six-week wait. Validation documentation arrives in the builder's language only and stalls an EU or PIC/S audit by a month. Put ambient into the enquiry too: 220–240 V at 50 Hz is standard across Southeast Asia and the Gulf, but a 34 °C room at 75% RH is not the same duty as 21 °C in Germany.

Frequently Asked Questions

How accurate is a capsule counting machine in real production?
Per capsule, a well-set line sits at 99.9% or better. What matters commercially is the exact-bottle rate, which on a 30-count bottle is about 97% at that per-capsule figure. Ask for both numbers and for the protocol behind them.

Can one machine handle gelatin and HPMC shells?
Yes, as separate recipes. HPMC carries more charge and has different surface friction, so channel amplitude and gate timing need retuning and the humidity window matters more. Expect 5–10% lower speed on HPMC.

What room conditions do I need?
Hold 45–55% RH at the counting head and 20–24 °C. Below 40% you fight static; above 60% with hygroscopic fills you fight shell softening and clumping. Specify it as a utility.

How often should sensors be cleaned?
Once per shift for a talc-polished or dusty product, weekly for a clean, unpolished one, and log it. The drift is gradual enough that nobody notices until the in-process check fails.

What documentation should come with a capsule counting machine?
IQ/OQ/PQ protocols and reports built on EU GMP Annex 15, a computerised-system package meeting Annex 11 and 21 CFR Part 11 if you supply the US, material certificates for product-contact parts, a CE declaration under Machinery Regulation (EU) 2023/1230, and English-language manuals and wiring diagrams.

How many change parts do I need per capsule size?
One channel set, one gate set and one funnel per size, plus a second channel set for your highest-volume size, because that is the part that wears and gets dropped during cleaning.

Written by Helen Xu | Chief Industrial Application Engineer

Helen Xu is a Chief Industrial Application Engineer with 9 years of specialized experience in packaging machinery and liquid filling machine design, equipment model selection, and full production line process optimization. He focuses on delivering customized packaging & filling solutions for pharmaceutical, food, and chemical manufacturing industries, with mature practical expertise in GMP compliance, ISO 9001 quality management standards, and turnkey large-scale filling & packaging production line integration.