Bottle Inspection Machine: Selection and Validation

Choose a bottle inspection machine from your defect list, not from the technology on offer: transmitted-light vision for cracks, chips and surface faults, X-ray for dense inclusions in dark or printed glass, polarised light for annealing strain. Then validate it with a seeded set of known defects, each container passed at least ten times, and record detection per defect instead of one overall catch rate. Treat the result as a measuring instrument with a stated capability limit. No inspector screens out every defect, and the threshold it works to is a decision you have to make and document yourself.

Write the requirement before you ask for offers

Machine suppliers answer the enquiry they receive. Ask only to "inspect bottles" and you get a rate and a footprint that cannot be compared with anything. Nine points belong on paper first.

• Defect types and sizes. Cracks, chips, stones, unmelted glass, bubbles, foreign bodies, scuffs, moulding faults and finish bore problems are separate detection problems. Size matters as much as type: a unit that catches a two millimetre stone every time may not see a half millimetre one, so state which is required.
• Location on the container. Finish, body wall, base and the inside of the base are each viewed by different optics at different angles. Say whether the base and heel, the usual blind spots, are in scope.
• Container appearance. Colour, opacity, wall thickness, internal pressure, decoration and coating decide what light or a sensor can get through. Plain flint is the easy case. Dark amber, frosted or heavily printed glass is much harder and may call for another detection principle.
• Container condition at the inspection point. Straight from forming, after annealing, after washing and drying, or after warehouse storage are four different situations. Moisture, dust, condensation and coatings all generate signals that read as defects.
• Rate and line position. The station must not become the bottleneck, and it must sit where a reject can leave the flow cleanly. Include space, headroom, services and noise.
• Fate of a rejected container. Scrap, hand re-inspection, return to the supplier and re-work each put a different cost on a false reject, and that cost shapes the threshold.
• Who operates and maintains it. If only a specialist can change a setting, the machine will stay on its delivery settings while the container drifts away from them. Decide the calibration routine at selection, not during commissioning.
• Customer or market requirement. It may be a written defect specification, a sampling scheme with an acceptable quality limit, or a demand for inspection records. It fixes the evidence you will need later.
• Proof of performance. Without an answer here, acceptance is a demonstration, not a validation.

What each detection principle finds and where it is blind

No single principle covers every defect class. The table sets the three main ones against the same questions, and the notes after it explain the reasoning.

Vision

Cameras and controlled lighting, usually shining through the wall from behind, pick up anything that alters the path of light. That makes vision the strongest choice for cracks, chips, moulding faults and scuffs, and it handles finish bore and height checks well. Its limit is contrast. Dark amber, frosted or heavily decorated glass leaves less light to work with, and a simple side-looking layout struggles with the base and heel. Moisture and dust must be filtered out or false rejects climb.

X-ray

X-ray measures how much radiation the container absorbs, so it responds to density instead of light. It picks out dense inclusions that vision can lose against decoration, and it does so through opaque and printed glass. A crack, though, is glass next to air with almost no density contrast, so X-ray should not be specified for cracks or chips. It also brings shielding, interlocks, local regulatory obligations and trained operators, and it is generally slower per container than a vision station giving the same coverage.

Polarised light and infrared

Polarised light shows the internal strain left by annealing, which exposes containers that look perfect but are likely to break later. Infrared absorption identifies moisture or washing residue that a camera sees only as an ambiguous mark. Both complement vision and neither reliably finds a physical crack or inclusion alone.

Combining them

Transmitted-light vision plus polarised stress inspection covers most visible and annealing-related requirements. An X-ray unit alongside adds the dense inclusions, and a moisture or residue sensor at the infeed covers washed containers. Let the defect list and the glass colour pick the combination, not an appetite for the most technology on the line.

What an inspector will not do for you

An empty bottle inspector, also sold as a glass container inspection system or vision inspection station, applies one fixed measurement to every container at a speed no person can hold, with no fatigue. That is why 100 percent inspection of the visible surface and wall has become normal practice.

It does not define an acceptable container. It compares what it sees against a limit somebody entered. It also does not fix the process: a line making many of one defect will show a high count of that defect, and the count describes the forming or annealing conditions, not the machine.

Two further limits tend to surface late. First, the machine was validated against the defects people imagined when buying it, yet in service it meets whatever the line really produces. Moulding, decoration or handling faults that were never specified may lie outside its optical or sensor envelope, and adjustment cannot bring them in. Keep the original validation record and add each new failure type to the sample set, so coverage is known instead of assumed.

Second, rejects are evidence. Inspect the reject bin on a fixed schedule, log a reason against the reject counter, and compare that log with the defect types the station is meant to catch. Otherwise it becomes a black box that stops the line now and then.

How to read a quoted inspection speed

A rated speed nearly always refers to one container size, one colour, one defect specification and an assumed reject rate, and it is given per lane or per station. A four-station machine is not four times as capable on every defect, because some faults are better seen from a particular angle and every station makes its own reject decision. Ask each bidder for the rate on your container, at your defect sizes, with the reject rate they assumed. The same unit can be quoted fast with a loose limit or slow with a tight one.

What production feels is the net rate: rated speed minus stops, re-inspection of rejects and manual intervention. A fast machine that needs attention every twenty minutes inspects fewer containers per shift than a slower one that runs unattended. Request the maintenance interval and spare-parts list together with the rate, and count daily specialist attention as hidden labour.

Size the inspector against the line as well. Too slow and it throttles everything. Far too fast and capacity sits idle, though the line is protected from a downstream stoppage. A modest margin over line rate, with accumulation on both sides to absorb short stops, is usually right.

Setting the threshold between false rejects and escapes

Any sensitivity setting produces two errors. A false reject (false positive, over-rejection) throws away a sound container. An escape (false negative, missed defect) lets a faulty one through. Pushing one down pushes the other up, and no setting brings both to zero. That is true of any measurement made under noise, whoever built the machine.

The two errors cost very different amounts. A false reject costs a container and some handling labour, and on a running line false rejects normally far outnumber escapes. An escape costs whatever the defect does downstream: a returned batch, a stopped filling line, a complaint or a safety incident. A sound rule is to fix the setting so escapes of your defined critical defects stay at a level you can state, then accept the false reject volume that follows as a known cost.

Three controls keep that decision from eroding:

1. Record a reason for every rejection against a fixed list of defect codes, so a rising count can be traced to the container, the machine or the setting.
2. Re-check a sample of rejects by hand at intervals, which tells a real increase in defects apart from threshold drift.
3. Put the threshold under written change control: who may alter it, on what evidence, and how the change is logged.

A limit quietly loosened to keep the line moving is the most common way an inspection station turns decorative. Where acceptance criteria use sampling language, the usual reference is ISO 2859-1 with an acceptable quality limit agreed between the parties and defects classed as critical, major and minor. We name ISO 2859-1 and AQL only as recognised frameworks for writing a requirement, with no claim about any plant, container or machine.

Building and running a seeded defect set

A seeded defect set, also called a challenge set or known reject set, is the only credible validation tool. Containers with defects of known type, size and position go through the machine in a set sequence, and its response is logged container by container.

Design the set; do not gather it at random. It needs:

• every defect type in the specification;
• each type at more than one size, one near the minimum to be detected and one comfortably larger;
• the same defect at more than one position, since a shoulder crack and a body-wall crack look different to the optics;
• at least one sound container from the same mould for each defective one, so false rejects are measured alongside catches.

Run each container at least ten times, preferably more. Detection of a marginal defect is a matter of probability: found on seven passes out of twenty is not the same result as found on all twenty, and only repeated passes show the difference. Log a detection count per container and give the set a unique identity that is reused at every re-validation.

Control the set like a gauge. Store it apart from production, confirm before each use that the defects are present and unchanged, and re-examine it whenever a defect appears to have grown. A set damaged or swapped without a record is worse than none, because the numbers it yields look confident and describe nothing real. When a new defect type turns up on the line, add it and revise the whole capability statement.

Evidence to require at the witnessed trial

Never accept a machine on a demonstration run with the maker's own sample bottles. Three items should be in hand before sign-off, and all are easy to request beforehand and hard to extract afterwards.

• A trial on your containers, at the maker's site or yours, using a defect set you supplied or watched being assembled. For each defect it should show whether repeated passes caught it and at which setting. A single total catch rate hides a few defects that are missed every time.
• A written capability statement naming the detection principle; the container size, colour and decoration the figures apply to; the defect types and minimum sizes demonstrated; the false reject rate measured in the trial; and the test conditions. A figure with no container and defect size attached is not a capability statement.
• The documentation package: operating and maintenance manual, calibration procedure and interval, spare-parts list, safety documents for the principle used, and operator training material. With X-ray, settle radiation safety paperwork and local regulatory requirements before installation. If the equipment bears a marking claimed by its maker, get the maker's own documents instead of trusting a line in an offer.

Splitting the work between machine and manual sampling

A machine relocates manual effort; it does not remove it or stand in for a sampling standard. The machine takes what it can measure on every container, normally visible wall and finish defects plus the dimensional checks it was set up for. Manual sampling draws a defined sample from the accepted stream and concentrates on what the machine cannot see: base, heel, decorated area, finish bore, and the inner surface once the container is turned over.

Write the split into the inspection procedure as a table so no item is claimed twice or dropped. Review both sets of results together. The most telling case is a clean machine shift alongside base chips in the manual sample, which shows the base lies outside the machine's envelope and that the envelope, not the setting, must change.

Frequency and sample size can follow a recognised scheme such as ISO 2859-1 with an agreed acceptable quality limit and defect classification, again cited as a framework and not a product claim. In practice, draw from accepted containers and not from rejects, draw at a fixed point in the shift, and write the acceptance limit before seeing results. The same written classification underpins the check on a finished batch before loading and the inspection when a delivery reaches your plant; the machine is one instrument for enforcing it, not a substitute for either routine.

After acceptance: supplier specification, records and drift

With a station running, what you can ask of a container supplier changes. It is fair to hold a supplier to the defect specification the machine enforces, as long as that specification is written and stable. It is not fair to hold them to an unseen limit adjusted on the line without notice, and most disputes start there. Issue the written specification, the critical, major and minor classification and the agreed sampling scheme before using the machine as grounds for a claim.

Records come next. A useful record shows the number inspected, rejects by defect code, manual samples and their outcomes, and each threshold change with its reason. That turns a supplier conversation into a factual one and feeds the periodic review. Note that a machine generates records, not certificates; what a certificate contains and which tests sit behind it is explained in our guide to quality certificates for glass bottle orders.

Drift is the change most often neglected. Lamps age, lenses gather dust, sensors shift, containers change and limits get adjusted. Set a re-validation interval at which the full seeded set is run again and compared with the acceptance record. If the detection count on that reference set has dropped, the machine no longer delivers the coverage originally accepted, even though it powers up every morning. Plan maintenance, spares and re-validation on one schedule and track uptime as a production metric.

What to send for a detection proposal

Three inputs produce a usable proposal instead of a general machine description: the defect types to be caught with the smallest size that matters for each, the acceptance requirement (how defects are classified and which are critical), and the line speed the station must match. Add colour, wall thickness and decoration, which sway the choice of principle more than anything else, and the container's state at the inspection point, dry and clean or fresh from washing.

From that we can lay out which principles cover which defects, where blind spots remain, what the seeded set should contain, and how to set the threshold so tolerable escapes and affordable false rejects are both written down before an order is placed. If you are still choosing the container itself, start with our glass bottle and jar ranges.

Frequently asked questions

Which glass bottle defects does a vision machine detect?

Set up in the usual way, it catches cracks, finish and body chips, scuffs, moulding faults, stones and bubbles, and it checks finish bore and some dimensions. Expect poorer results at the base and heel, confusion between inner-surface and outer-surface defects, and a larger minimum detectable defect in dark amber, frosted or heavily printed glass.

Can an X-ray inspector detect cracks?

No. This is the most frequent misunderstanding at selection. X-ray is suited to stones and foreign bodies, including in opaque and printed containers. Any crack or chip requirement has to be met optically.

Should I pick vision or X-ray for empty bottle inspection?

Work from the defects. Cracks, chips and surface faults point to vision, with polarised light as an option for annealing stress. Dense inclusions or wall thickness distribution, especially in dark or heavily decorated glass, bring X-ray in. If both groups matter, specify a combined station and do not compromise on a single principle.

How are false reject rate and escape rate measured?

With one seeded set, run at the intended production setting. The matched sound containers yield the false reject rate over repeated passes. The defective ones yield the escape rate, stated per defect type and size, never as one average.

How big should a seeded defect set be?

Coverage sets the size, not a fixed count. Because every container is run many times, a modest, well-designed set produces a large body of evidence.

Does the machine replace manual AQL sampling?

No. People keep sampling the zones the machine cannot view while it handles everything measurable on each container, and the two results are read side by side.

What belongs in the acceptance test?

A trial on your own containers with a witnessed defect set, a capability statement tied to container size, colour and defect size, and the full documentation package covering calibration, spare parts, safety and training.