Glass Bottle Label Inspection: Post-Label Verification

The Three Interfaces a Post-Label Inspector Has to Define

A post-label inspection station is described by suppliers in the language of cameras, resolution, frame rate and lighting. None of those words decides whether the station will work. What decides it is three interfaces, and until all three are written down, every quotation that arrives will be answering a slightly different question and none of them can be compared. The three are the physical interface between the label and the container, the data interface between what is printed and what is supposed to be printed, and the removal interface between a reject decision and the container leaving the line.

The physical interface is the one most people picture first, and it is genuinely the hardest to make repeatable. The label is not a flat object. It is a thin film or paper sheet wrapped onto a curved, and often tapered or oval, surface. It may be transparent, metallised, embossed or matt, and the surface under it may be flint, amber, frosted or decorated. Every one of those variables changes how light returns from the label to the sensor, and therefore changes whether the edge of the label can be found at all. A station specified without a named label family and a named container surface is a station that will be tuned on the day it arrives and tuned again on the next production run.

The data interface is the one buyers most often leave out, and it is the interface that actually prevents the recalls that matter. The labeler applies whatever the reel or the print head gives it. It has no way of knowing whether that reel belongs to the product currently running, whether the date code printed on the label corresponds to today’s batch, or whether the barcode encodes the pack size that was ordered. Those are not imaging questions at all; they are comparison questions, and they require the station to hold a reference: the expected artwork, the expected code string, the expected format, and the job record for the run. Without that reference the station can only judge whether a label looks well applied, which is the least valuable of the three checks.

The removal interface is the third, and it is the one that turns a detection into a control. A station that identifies a defective pack is only useful if the pack is taken off the line, and the mechanical arrangement of that removal determines what the detection tolerance is allowed to be. If the station sits where a reject can be pushed into a chute without touching neighbours, fine tolerances are affordable. If it sits in the middle of a dense mass of containers with no gap, the reject may take two or three sound bottles with it, the cost of a reject rises sharply, and the station has to be set to reject less, which quietly reduces its value. The interfaces are therefore not three separate conversations; they constrain each other, and the sequence in which they are settled is the subject of the rest of this page.

Why a Correctly Applied Label Still Needs Its Own Verification

The common objection to post-label inspection is that the labeler is already set up and running correctly, so a second check is a duplicate. The objection rests on a confusion between two different things. The labeler is a mechanical device that places an object at a position and, if a print head is fitted, marks it. Post-label inspection is a verification of identity. The labeler can perform its mechanical task perfectly and still deliver a container that must never be shipped, and no amount of adjustment at the labeler can prevent that, because the failure did not originate in the labeler’s mechanism.

The failures that make post-label inspection worth its footprint are the ones that end in a recall rather than in scrap. The first is the wrong label for the product: the previous reel was not fully consumed, a changeover was interrupted, or two similar SKUs share a line and the wrong artwork was loaded. The second is the missing label, which an operator notices on a stopped line but which can pass unnoticed at speed when a single container loses its label in a run of thousands. The third is the wrong code: a date or batch string that is legible, correctly positioned and completely incorrect for the batch being filled. The fourth is the unreadable code, where the label is present and correct in content but the printed or pre-printed barcode fails to scan at the customer’s distribution centre, which converts a production problem into a chargeback. The fifth is displacement: the label is right, but it is skewed, wrapped around the shoulder, applied over the base radius, or overlapping a competitor’s or a previous label.

The distinction that matters commercially is between a defect of application and a defect of content. A defect of application can be reduced by better mechanics and by better containers. A defect of content cannot be reduced by mechanics at all, because the mechanism has no knowledge of content. A line that has invested heavily in the labeler and nothing in verification has optimised the half of the problem that is easy to control and left the half that generates recalls entirely to human attention at line speed, which is precisely where human attention is weakest.

Label Inspection Point and Defect Type Table

The table below is the working core of this page. It takes each inspection point separately, states the defect it is there to catch, explains why that defect matters commercially rather than cosmetically, names the sensing approach that normally fits it, sets out how the reject decision is drawn, and states where on the line the check belongs. No single row is sufficient on its own, and a station built to satisfy only the rows that need the simplest optics will reliably catch the defects that cost the least.

Which Label Family Your Imaging Has to Survive

Imaging is a comparison of light returned from a surface against an expectation. Anything that changes the way light returns changes the difficulty of the check, and label materials change it more than any other single variable on the line. A station that works perfectly on a matt paper label on amber glass may be unusable on a clear film label on the same container, not because it is a poor station but because the two label families present completely different optical problems.

Paper labels, whether uncoated or coated, are the easiest family because they are opaque and they scatter light diffusely. Their edges are visible against glass, their surface is uniform, and the main difficulties are the ones introduced by the container rather than by the label: a dark glass body under a light application window, a strong highlight from a bright light source, or dust and glass fines sitting on the surface. Coated paper is only slightly harder, and the family as a whole tolerates a simple facing camera with a diffuse light source.

Film labels in polypropylene or polyester are harder in a way that is easy to underestimate. A metallised or bright silver film acts as a mirror, so a diffused light source produces a specular streak across the label that can erase the very code the station is meant to read. A clear film used as a no-label-look decoration is the hardest case of all, because there is no contrast between the label edge and the glass unless the edge itself is printed or unless the illumination is arranged to create a controlled reflection difference. These two cases generally require polarised illumination, a domed diffuser, or a deliberate dark-field arrangement, and they should be named in the enquiry rather than discovered at commissioning. A matt white film sits between the two extremes and is usually straightforward.

Shrink sleeves introduce a third set of problems. A sleeve covers the whole circumference, so there is no glass against which to measure position, and the seam line is a genuine feature of the product that must be excluded from the defect mask or the station will reject every bottle. Sleeve material is often thinner and glossier than a label, and after shrinking it conforms to the container profile, which means the surface the camera sees is no longer cylindrical. The practical consequences are that the reference datum has to be derived from the container rather than from the label edge, and that the station has to be taught the seam position for each format rather than for each container type.

A fourth group of variables sits on top of the label material rather than inside it. Condensation from a cold fill or from a cooling tunnel puts a fine water film over the label and destroys edge contrast. Glass fines from upstream handling sit on the surface as small bright specks that resemble print defects. Static attracts light dust to film labels. A strong warm highlight from an unshaded lamp can saturate a sensor across an area larger than the defect being looked for. None of these is a reason to abandon the check, but each of them is a reason to state the container condition at the inspection point in the enquiry, because it changes the lighting arrangement and sometimes the sensing principle.

Where the Station Belongs: Straight After the Labeler or Before the Case Packer

The placement question is presented as a choice between two positions, and in practice it is a choice between two different jobs. A station placed directly after the labeler is close to the source of the defect, so a trend can be fed back to the labeler while the faulty reel or the drifting head is still on the machine. A station placed immediately before the case packer sees the pack in its final state, after every downstream operation that could have damaged the label, but it is far from the cause and its feedback arrives too late to correct anything except the next shift.

The distance also decides how much of the line a reject takes with it. Close to the labeler, containers are usually separated and single file, so a reject can be removed cleanly with little risk of taking a sound neighbour. Near the packer, containers may be accumulated in mass flow or grouped in a case, and a reject at that point is expensive or impossible without stopping the line. The general preference is therefore to place the primary check close to the labeler, where the reject is cheap and the feedback is useful, and to place a lighter confirmation check near the packer only where a downstream operation genuinely threatens the label.

There is one structural case where the primary check has to move downstream regardless. If the label is applied before a shrink tunnel, or before a full-body sleeve is shrunk, the label or sleeve only reaches its final geometry after the heat step, so a check placed before the tunnel is checking an intermediate state. In that situation the station belongs after the tunnel, and the labeler is controlled by a separate, simpler check that confirms the sleeve or label has been presented and tacked. The two checks then answer two different questions rather than the same question twice.

A third consideration is the state of the container. A station placed after a pasteuriser, a cooling tunnel or a cold-fill capper may be looking at a wet, cold surface with condensation running down the label. Some imaging arrangements can be made to work through a light water film; others cannot. This is not a reason to move the station, because the check still has to happen after the process that threatens the label, but it is a reason to specify the wet condition in the enquiry rather than to specify a dry container and hope.

Vision, Light Inspection and Code Verification Are Three Different Jobs

Three methods are used in this area, and they are frequently discussed as though they were competitors. They are not. Each answers a different question, each has a blind spot the other two do not share, and the usual correct answer is some combination of them rather than a choice between them.

Machine vision, meaning an image compared against a geometric or pattern model, is the tool for position, presence, skew, wrap and surface relief. It is at its best when the question is where something is or whether something is there, and at its weakest when the question is what something says, because a model that accepts reasonable variation in print will also accept a wrong character that happens to fall inside the variation. A vision station is therefore the right instrument for the label edge and the wrong instrument for the date code.

Code reading and code verification, meaning the decoding and grading of a printed bar code or 2D code, is a separate function and a separate algorithm. Reading answers whether the code can be decoded once; verification grades how likely it is to be decoded on a different scanner, at a different angle, on a different day. For a manufacturer whose real exposure is a chargeback at a distribution centre, the grade is the useful number and a successful read is not, because a code that reads on the factory scanner is not necessarily a code that meets the customer’s requirement. Optical character verification of the date and batch string is a third variant of the same idea, and it compares a printed string against an expected string rather than grading a symbol.

Human light inspection belongs in the same discussion rather than outside it. At low volumes, at the start of a new format, and during changeover windows, a trained inspector under controlled lighting will catch a class of defects that no installed system is looking for, because a person can notice that something is wrong without having been told in advance what to look for. The economic threshold is a matter of volume and of the number of format changes rather than of technology, and the practical arrangement is usually a defined manual check for the first containers of every run whether or not an automatic station is installed. What a manual check cannot do is examine every container at production speed for a full shift, and that limit is what the automatic station exists to overcome.

Why a Post-Label Inspector Should Be Set to Over-Reject

Every inspection decision is a trade between rejecting something sound and passing something defective. Reducing one increases the other, and no setting eliminates both, because the decision is made on a noisy measurement. For a label check the two errors have very different costs, and the asymmetry is more extreme than in almost any other inspection area on a glass bottle line.

A false reject costs one container, one label, and the value of the contents if the container is already filled. On many lines the pack is still empty at this point, so the cost is a bottle and a label and the labour of dealing with it. An escape costs a pack that reaches a customer with the wrong label, no label, an unreadable code or the wrong batch code. The consequences of that escape are a customer complaint, a chargeback, a selective recall that is expensive precisely because it cannot be limited if the code was wrong, and in the case of a regulated product a conversation with an authority. The ratio between those two costs is not close, and it is the reason a post-label station should generally be set to reject generously rather than to pass generously.

There is a second reason that is specific to this inspection point. Most of the defects being looked for are not hazardous in themselves; they are administrative. A crooked label is a commercial failure, not a safety failure. But the same station is also the only place where a wrong label, a wrong batch code or an unreadable symbol can be caught before the pack is sealed into a case, and once it is in the case the pack is no longer individually visible. The further downstream a defective pack travels, the more expensive it becomes, and the packer is the last point at which it is still a single unit rather than part of a shipment. A station that under-rejects to protect its own reject rate is trading a cheap cost against an expensive one.

The discipline that keeps this decision honest is to write it down before the station is ordered. The written form should state, for each defect type, whether the requirement is zero escapes or a stated maximum escape rate, and it should state the false reject volume that will be accepted as the consequence. Without that statement, the setting will drift in the direction of whatever is complained about most, and the first complaint is almost always about the reject bin filling up.

Writing the Tolerance Down: Position, Angle and Readability Limits

An inspection tolerance is not a property of a camera. It is an agreement about how much variation in the printed or applied result is acceptable, and it has to be written before anyone tunes the station. The reason is practical rather than philosophical: a tolerance agreed in advance produces a stable line, and a tolerance discovered by turning a dial until the rejects stop produces a station that passes a different product every week.

For position and skew, the tolerance should be expressed against a stated datum, in millimetres for offset and in degrees for rotation, and the datum has to be something the station can actually find on the container rather than something printed on a drawing. For a container with a mould seam or an embossed mark, the seam is often the most reliable datum. For a fully sleeved container there may be no datum on the glass at all, in which case the datum has to come from the container outline. The tolerance should also state separately what happens when the label is present but rotated, because a rotation and an offset produce different downstream consequences and are often confused in a single reject count.

For code quality, the tolerance should be a grade or a measured contrast level rather than a pass or fail on whether the code scans. A grade threshold allows a supplier conversation about a reel that is trending towards unreadable before it becomes unreadable, and it gives the line a warning that a print head needs cleaning. A simple scan test gives no warning at all, because the code scans perfectly until the day it does not. For printed date and batch strings, the tolerance should be an explicit legibility criterion, agreed with the customer where the customer’s own specification applies, and it should be applied character by character rather than to the string as a whole.

For surface defects such as wrinkles, bubbles and lifted edges, the tolerance is usually best expressed as a length or an area rather than as an intensity, because a length or an area can be argued about with a supplier and an intensity cannot. A rule such as a lifted edge longer than a stated length in the sealed area, or a wrinkle covering more than a stated area of the face panel, is enforceable and reviewable. An intensity threshold that only exists inside the machine’s software is not.

Every tolerance in the list has a second half that is usually forgotten: what changes it. A change of label material, a change of container colour, a change of lighting lamp, a change of print head or a mechanical change at the labeler all invalidate the setting that was agreed. The written tolerance should therefore be accompanied by a short change-control note stating who may alter it, what evidence is required before an alteration, and how the alteration is recorded. A tolerance without a change rule is a tolerance that will be adjusted quietly and will no longer describe the product.

Line Speed, the Reject Interface and the Replacement Logic

The station has to keep up with the line without becoming its constraint, and the rate claim that arrives in a quotation almost always needs to be read carefully. A rate is meaningful only when it is attached to a container size, a label family, a defect list and an assumed false reject rate. The same hardware quoted at a high rate with a loose tolerance and at a lower rate with a tight one is the same hardware, and the number that matters in production is the net rate after stops, after the handling of rejects and after any manual intervention the station requires.

The mechanical fit matters as much as the rate. The station needs standoff, headroom, a stable container presentation and enough conveyor length downstream for a reject to be removed without touching a neighbour. Where containers pass the station in mass flow rather than in single file, the presentation is not repeatable enough for a positional measurement and the station has to move, or the flow has to be converted to single file before it. Where the container is tall and narrow relative to its base, a gentle guide is usually needed to stop the top of the container swaying, because a two degree sway at the shoulder of a tall bottle is a large movement at the camera.

The reject interface should be settled with the same discipline as the detection itself. Three decisions belong in the written specification rather than in the commissioning week. The first is the actuation method, which on a glass line is normally a pneumatic pusher or an air jet, and which has to be fast enough to act between two containers at the line rate. The second is the confirmation, meaning a sensor that proves the intended container actually left the conveyor, because a reject decision without a confirmation is a decision that cannot be audited. The third is the destination: a locked reject bin, a return lane for rework, or a separate accumulation for inspection, each of which implies a different handling routine and a different record.

Spares and replacement logic are the part of the specification that decides what the station looks like in year three. Lighting is the first consumable to age, and a light source that has dropped in output will reduce the contrast margin on every check the station performs, so the light is normally replaced on an interval rather than on failure. Lenses and windows collect glass dust and need a cleaning routine and a cleaning interval, not an instruction to keep them clean. The camera and the processor are the most reliable parts of the system and should not be the first suspects when detection performance falls away. The specification should state which parts are field replaceable, which of them are format specific, and whether a change of label family requires a new part or only a new recipe.

What This Page Does Not Cover: Empty Bottle Defects, Adhesion and Reject Handling

This page covers the verification of an applied label and of the content printed on it. It deliberately does not cover three subjects that are often merged into it during a supplier conversation, and separating them is what keeps a station specification short enough to be enforced.

The first is the inspection of the container itself. Cracks, chips, stones, unmelted glass, finish faults, insufficient annealing and wall thickness distribution are container defects, and they are detected by a different principle at a different point on the line, in most cases while the container is still empty. A label station cannot substitute for that check, and a container inspection machine cannot read a bar code. Where a buyer is choosing that equipment, the selection criteria, the defect sample set and the acceptance method are set out on the empty container inspection page linked at the top of this page, and the two stations should be specified separately even when they are bought from the same supplier.

The second is adhesion and durability. Whether a label remains attached through condensation, abrasion, a chilled display or a hot warehouse is a question about the adhesive, the container surface, the application pressure and the storage environment, and it is answered by a peel or immersion test rather than by a camera. A visual check for a lifted edge at the inspection station is a useful early warning and is not a substitute for that test. The third is the hardware that removes a rejected container from the line, which has to be sized to the container, the rate and the available space, and which is a mechanical subject in its own right.

Two further boundaries are worth naming for readers arriving from elsewhere. Where the container and the artwork are being chosen rather than checked, the container families and their applications are set out on glass bottle quality inspection, which covers the general inspection framework and the records that travel with a batch, of which this label check is one component rather than the whole. And where the container carries a child resistant closure and the governing question is a regulated test protocol rather than a printed label, the applicable subject is child resistant closure testing, which sits in the closure and compliance chain and shares only the requirement that every check should have a written acceptance criterion.

Questions Buyers Ask About Glass Bottle Label Inspection

Do I need a label inspection system if my labeler is already reliable?

Reliability of application and correctness of content are two different things, and the labeler only controls the first. A reliable labeler will still apply the wrong reel if a changeover is interrupted, will still run a container with no label if the supply fails for a few seconds, and will still print a date code that belongs to the previous batch. Those failures are the ones that end in a recall rather than in scrap, and none of them can be prevented by improving the labeler mechanism. The question is therefore not whether the labeler is reliable but whether the line can afford a single escape reaching a case.

What is the difference between reading a barcode and verifying it?

Reading answers whether the code decodes on the scanner that is being used, at that moment. Verification grades the symbol against a defined quality scale and predicts whether it will decode on a different scanner, at a different angle, in a different light and after a period of storage. For a factory whose exposure is a chargeback or a rejection at a distribution centre, the grade is the useful measurement, because a code that reads perfectly on the factory scanner can still fail on the customer’s. The usual arrangement is to verify on a sampled basis and to read on every container.

Where on the line should the inspection station be installed?

In most cases directly after the labeler, where containers are still separated, a reject can be removed cleanly and the feedback is close enough to the labeler to correct a drifting head or a bad reel during the same run. The exception is any process that changes the label after application, such as a shrink tunnel, after which the check has to move downstream because the label has not reached its final geometry. A lighter confirmation check near the case packer is worth adding only where a specific downstream operation threatens the label.

Can one station inspect paper labels, film labels and shrink sleeves?

One physical station can usually handle all three, but not with one setting. Paper, matt film, metallised film, clear film and shrink sleeve present different optical problems, and each format needs its own recipe covering illumination, exposure, the defect mask and the reference image. A station offered without a format-specific setup procedure will be tuned once and will then pass progressively less of what it was bought for. The enquiry should therefore state every label family that will run on the line, not only the one that is running today.

How should the reject threshold be decided for a label check?

By writing the two costs down and comparing them. A false reject costs one container, one label and the contents if the pack is already filled. An escape costs a chargeback, a complaint, a recall that cannot be limited if the batch code was wrong, or a regulatory conversation for a controlled product. Because the second cost is far larger and because the packer is the last point at which a pack is still individually visible, the threshold is normally set to over-reject rather than to under-reject, and the resulting reject volume is accepted as a known cost rather than fought.

How do I set a position tolerance for a wrap-around label?

Express it against a datum the station can actually find, such as a mould seam, an embossed mark or the container outline, and state the offset in millimetres and the rotation in degrees as two separate limits rather than as one. Both limits should be agreed with the customer where the customer has a specification, and both should be reviewed when the label material, the container or the lighting changes. A tolerance that exists only as a number inside the machine software cannot be argued about with a supplier or audited later.

Does a label inspection station replace manual checks at start-up?

No. A trained inspector under controlled lighting will catch defects that the station was never told to look for, which makes a defined manual check of the first containers of every run valuable whether or not a station is installed. What a manual check cannot do is examine every container at production speed for a full shift. The practical division is that the automatic station takes the repeatable, specified checks on every pack, and the manual check covers new formats, changeovers and defect classes that are not yet in the station’s specification.

Send the Label Type, the Mandatory Checks and the Line Speed

To receive a station proposal rather than a camera description, send three things together rather than one at a time. The first is the label family: whether the pack uses paper, matt or gloss film, a metallised or clear no-label-look film, or a full-body shrink sleeve, and whether the container is flint, amber, frosted or decorated. This single input changes the illumination and sometimes the sensing principle more than any other. The second is the list of checks that are mandatory rather than desirable, separated into checks that must have zero escapes and checks where a stated escape rate is tolerable, because that separation is what sets the threshold.

The third is the line: the rated speed the station has to keep up with, whether containers arrive in single file or in mass flow at the inspection point, the container height and base diameter, whether the presentation can be steadied, and how much conveyor length is available downstream for a reject. Add the condition of the container at the station, especially whether it is dry or carrying condensation, and add the reject handling that already exists on the line so that the removal interface can be matched rather than invented.

With those inputs, the reply can state which checks a vision function will cover, which require code reading or verification, which cannot be covered at the proposed position and have to be handled by a manual routine, and what the false reject volume is expected to be at the stated tolerance. It can also state what a format change requires, which parts are format specific, and which consumables are replaced on an interval. A proposal that names only a camera resolution and a rate has not addressed any of the three interfaces, and it will have to be rebuilt once the first production run exposes the gap.