Short shots in injection-moulded closures: what a camera can detect

What does a short shot look like on a closure?

Depending on the geometry, an incomplete moulding may leave a missing rim segment, an unfinished feature or an interrupted outline. An image must show the relevant region with sufficient contrast. Do not label every broken bridge or distorted edge a short shot: damage after moulding can produce a similar appearance.

Have the moulding and quality teams label the sample set by the agreed defect definition. Keep acceptable moulding variation in that set, including colour and resin changes that affect contrast.

Can vision identify why it happened?

An incomplete fill is a process issue for the moulding team to investigate. The inspection result should identify the visible symptom, time and available production context. It does not by itself establish whether filling conditions, material flow, the tool or another process factor caused that symptom. Avoid automatically changing process settings on the basis of an unvalidated image classification.

Which views are needed?

The exact number of cameras follows the coverage study. A catalogue photograph of a cap is not proof that all surfaces can be inspected in the proposed conveyor arrangement.

How should the station be accepted?

Use actual incomplete caps, damaged caps that might be confused with them, and acceptable parts from the normal production range. Separate setup samples from the final test set. Count false rejects and missed defects at the agreed production conditions, then verify tracking and physical rejection. Where cavity identifiers are visible and reliably readable, they can be logged for process investigation; do not infer a cavity number from timing alone without validation.

Our closure inspection system describes a HIKROBOT-based station and its budget assumptions. Retain torque, leak, material and mechanical checks wherever the acceptance specification requires them.

Why incomplete filling produces different optical signatures

A short shot is incomplete filling of the mould cavity. The visible result depends on where the flow stopped: an outer rim may be interrupted, a thin bridge may be missing, or an internal feature may be incompletely formed. The camera sees the resulting geometry, not the underlying melt pressure, vent condition or material history. Process engineers investigate those causes using the moulding process records; inspection provides defect evidence and, where available, the cavity association.

Separate a short shot from flash and from permitted design openings. A silhouette can reveal missing material at an exposed boundary, but internal ribs require a view into the closure. Flash adds material and may create the opposite contour deviation. A moulded symbol or an intentionally interrupted tamper feature must not be taught as a defect merely because its appearance differs from a smooth reference.

South Africa’s Plastics SA describes industry training and skills development. For a local closure project, include the moulding and toolroom teams in defining the reference samples; the camera recipe should preserve their product judgement in a repeatable inspection method. (Plastics SA: industry representation and skills development) This is particularly useful where cavities or colour changes produce acceptable variation that the downstream filler has already approved.

Choose the view around the missing feature

HIKROBOT describes locating, measurement and defect-detection tools, including learning-based options. Use those capabilities only after confirming the missing feature is visible. (HIKROBOT VisionMaster product page) A locating tool should tie each inspection region to the closure geometry. Comparing an unregistered image to a good template can mistake normal translation for missing material, or average away a local gap when the permitted position range is too broad.

Lighting changes the evidence. Backlighting highlights an outside boundary, while reflected light may reveal a shallow incomplete rib. The HIKROBOT light-source glossary explains these different geometries. (HIKROBOT Light Source Catalog, 2025 Q3, glossary) Test all scoped resin colours and translucencies. A light setting that gives a clean silhouette on an opaque closure may transmit through a translucent wall and change the apparent edge. Lock the image settings after the trial and record the material variant in the recipe.

A hypothetical trial that separates shape from pose

Hypothetical example: the requirement is a complete exposed rim and a present internal liner. The sample set contains a good closure, a closure with a missing rim segment, a closure with no liner and a good closure rotated within the allowed range. Use the silhouette view for the rim and an open-side view for the liner. Requiring both checks prevents an intact outline from masking an absent internal component.

Challenge a reversed closure as well. If the open side is not visible, the correct output is an invalid presentation or incomplete inspection, according to the handling design. It is not a liner pass inferred from the outer shape. Follow the failed closure to its physical reject location and ensure that its neighbour is not removed instead. The example specifies a test method; it does not state a defect size, cycle rate or achieved detection accuracy.

Keep natural production failures separate from hand-modified challenge parts. A cut rim can be useful for testing the logic and reject path, but its sharp edge may image differently from an actual moulded short shot. Label that distinction in the trial record and include real incomplete-fill samples before accepting a claim about production coverage. Retain images with cavity and material information where those data are available.

Validate across cavities and maintain the defect boundary

For a trained detector, group images by physical closure and production lot before splitting development and acceptance data. Consecutive frames of the same defective closure are strongly related. Google’s dataset guidance explains why duplicated or overlapping examples make evaluation misleading. (Google Machine Learning Crash Course: dividing datasets) Use the validation set for tuning and reserve a final test set that has not influenced the threshold or model. If a failed test image is used for retraining, it is no longer independent acceptance evidence.

Report the outcome by defect type and presentation condition. A large population of easy good parts can make overall accuracy look excellent while the important internal feature is never inspected successfully. Count acceptable closures rejected, incomplete rims accepted, missing liners accepted and invalid presentations separately, with their own tested populations. NIST’s interval guidance explains why a finite sample cannot establish that the missed-defect rate is exactly zero. (NIST e-Handbook, 7.2.4.1 Confidence intervals for a proportion)

During production, trend faults by cavity or moulding condition when that association is reliable. A recurring pattern can guide process investigation, but the vision result alone does not prove the root cause. After a mould repair, resin change, new colour or fixture adjustment, re-run the affected reference samples. The closure-inspection solution page covers station handling and integration; this article explains the defect mechanism, image coverage and evidence needed to support that project.

From short-shot samples to a closure inspection station

Bring good parts from the relevant cavities together with incomplete and borderline closures. The visible defect and part presentation determine the views the station needs.