Opsio - Cloud and AI Solutions
Injection moulding machine with a take-out robot, its black moulded housings lined up on a belt conveyor that passes under a camera with a white ring light

AI Visual Inspection for Plastics & Molded Parts

Camera-based inspection for short shots and flash on technical parts, sink marks and splay on cosmetic surfaces, and thin walls on blow-moulded bottles, with every reject traced to its mould cavity.

Book a walkthroughSee use cases

What is AI visual inspection for plastics and molded parts?

AI visual inspection for plastics and moulded parts is Opsio's camera and edge-AI service that checks every shot for injection molding defects such as short shots, flash and warpage, cosmetic defects such as sink marks, splay and black specks, and blow-moulded container defects such as thin walls and neck flash, and links each reject to its mould cavity.

Key facts

  • Covers short shots, flash, warpage, sink marks, splay, weld lines, black specks and bottle wall-thickness variation.
  • Imaging uses diffuse dome light, low-angle dark-field, laser triangulation and backlight transmission.
  • Inference runs on NVIDIA Jetson or Intel OpenVINO hardware at the press in under 50 ms.
  • Reads mould, cavity number and shot counter from the moulding machine over Euromap 77 (OPC UA).
  • Sends pass, reject or flag to the cell PLC over OPC UA, Modbus or PROFINET.
  • Stores images and results on the edge, on-premises or in an EU cloud region.
  • A feasibility study on your sample parts typically takes 1 to 2 weeks.

Main defect families

Fill and dimensional defects
Short shots, flash, warpage, dimensional deviations, gate vestige and ejector pin marks that affect fit and function.
Cosmetic surface defects
Sink marks, splay, weld lines, colour variation, black specks, burn marks and scratches on visible surfaces.
Container and closure defects
Wall-thickness variation, neck and thread flash, parting-line defects, contamination and colour in blow-moulded parts.

Last reviewed by Praveena Shenoy, Head of AI & Data, Opsio, a Swedish cloud and AI services company. Part of Opsio's AI Visual Inspection service.

Why are injection molding defects hard to catch at the press?

Every cavity in a mould can drift on its own, and the press keeps producing while it does. A 32-cavity closure tool on a 6-second cycle, for example, releases 320 parts a minute, far more than an operator at the conveyor can turn over and examine. Defects also arrive in bursts: a blocked vent or a cold hot-runner tip affects one cavity until someone notices.

The material makes it harder. Black and dark-grey polymers swallow light, so flash and shallow sink marks show only at the right lighting angle, while grained and high-gloss surfaces hide or exaggerate the same defect. Glass-filled grades keep warping after ejection, so a part that looks right at the press can be out of tolerance an hour later.

Manual inspection breaks down on judgement as much as on volume. Two inspectors grade the same faint weld line or colour shift differently, and boundary samples fade or go missing. When a short-shot clip or a thin-walled bottle escapes, the cost is a line stop or sorting action at your customer, a containment charge on the next scorecard and, for automotive and medical parts, a formal complaint against the whole lot.

Plastic part inspection: three use cases

Each use case follows one part family through situation, defects, inspection setup and outcome, so you can match it to your own presses and blow-moulding lines.

Technical injection-moulded parts

Black glass-filled moulded bracket on a granite plate under projected light stripes; one cylindrical boss is incompletely filled with a ragged rim, and a thin flash fin runs along the base edge.
Glass-filled nylon bracket with a short shot on one boss and parting-line flash, under structured light.

Situation

You mould functional parts: glass-filled PA66 brackets, clips, connector housings and medical device components, typically on 4- to 32-cavity tools with cycles of 10 to 40 seconds. A robot takes each shot off the tool and sets it on a conveyor or tray. Today an operator checks a few parts an hour with a go/no-go gauge and a CMM measures a sample per shift, so a cavity that starts short-shotting can run for hours unseen.

Defects

  • Short shotan incompletely filled partMelt freezes before it reaches the end of flow and leaves a missing boss, rib or clip tip, usually from a blocked vent, a cold hot-runner tip or too little shot volume.
  • Flasha thin fin of excess plastic at the parting line or around insertsMelt escapes between worn or under-clamped mould halves, and the fin can stop a clip seating or a seal closing.
  • WarpageUneven cooling and fibre orientation make the part bow or twist after ejection, so flat faces and mounting points move out of tolerance.
  • Dimensional deviationsShrinkage that varies with melt temperature, packing pressure and cavity wear moves holes, bosses and snap features outside the drawing tolerance.
  • Gate vestigethe stub left where the part was separated from the runnerA worn gate insert or poor degating leaves a stub that stands too high and interferes with assembly or sealing.
  • Ejector pin marks out of specPins set too proud or too deep, or pushing on a part that is still soft, leave imprints deeper than the drawing allows or white stress marks around them.

Inspection setup

A 12-megapixel GigE Vision area camera with a 25 mm C-mount lens looks down through a diffuse dome light for the 2D outline, and low-angle dark-field bar lights trace flash along the parting line. A laser triangulation sensor (a 3D method that reads height from a projected laser line) scans each part as the robot passes it over, measuring flatness, boss height and hole positions against the CAD nominal. Our dimensional inspection Q&A covers the measurement side in more depth.

The station sits directly after robot take-out, before parts reach the box. The robot passes the cavity number with each part, so the system rejects the part into a quarantine chute, flags the cavity and logs the measurements for SPC. The same 3D method checks gap and flush on automotive body and powertrain lines.

Outcome

Every shot is checked, not one sample per shift. A short shot or flash fin is caught on the cycle it occurs and the cavity is named, so your setter clears the vent or hot-runner zone instead of sorting a full box. Warpage and dimensional trends show per cavity on the SPC chart, typically warning you before a feature crosses its tolerance limit.

Cosmetic and consumer parts

Close-up inside a grey moulded housing: a circular ejector pin area sunk below the surface and ringed with flash, a thin flash line along a rib, and fine scratches on the wall.
Grey moulded housing with a sunken ejector pin mark, flash along a rib and surface scratches.

Situation

You mould visible parts: appliance housings, automotive interior trim, cosmetics packaging and electronics covers in grained, high-gloss or coloured finishes. Parts come off 1- to 8-cavity tools at 20 to 60 seconds per cycle and inspectors grade them in a light booth against boundary samples. That grading changes with the inspector, the shift and the viewing angle.

Defects

  • Sink marksshallow depressions over thick sectionsThick sections keep shrinking after the skin has frozen and pull the surface inward, typically above ribs, bosses and screw towers.
  • Splaysilver streaksMoisture in under-dried resin or volatiles from overheated material turn to gas at the flow front and leave silvery streaks fanning out from the gate.
  • Weld linesalso called knit linesTwo flow fronts meet around a hole or core without fully fusing, leaving a visible line and a weaker zone.
  • Colour variationMasterbatch dosing, regrind ratio or thermal degradation shifts shade or gloss from part to part or batch to batch.
  • Black specks and contaminationDegraded material flaking off the screw and barrel, or foreign particles in the resin, show as dark specks and streaks in the surface.
  • Burn marksdieselingAir trapped at the end of fill is compressed until it ignites, scorching the last area to fill black or brown.
  • ScratchesContact with conveyors, grippers or other parts after ejection drags fine lines across gloss and textured faces.

Inspection setup

A colour GigE Vision camera with a 16 to 35 mm lens views the part under a diffuse dome light, which removes glare on gloss surfaces and gives an even field for colour values and black specks. A second image under low-angle dark-field light (light grazing across the surface) makes sink marks, weld lines and scratches cast a shadow. A robot presents each face of large housings in turn; small parts pass on a conveyor.

Resolution is set so the smallest speck in your boundary sample covers at least 3 by 3 pixels, typically 0.05 to 0.1 mm per pixel. Colour is compared with a stored reference in a calibrated colour space. The system rejects clear failures, flags borderline parts for an inspector and logs every image, a method described in our guide to surface defect detection.

Outcome

Your cosmetic standard is applied the same way on every shift. The grading lives in labelled images instead of boundary samples in a drawer, and parts near the limit go to a person for the final call. Splay and black-speck trends point to drying or purging problems before a whole batch is affected, and colour drift is caught at the masterbatch change rather than at your customer.

Blow-moulded containers and closures

Row of white HDPE bottles in front of a backlight panel: one bottle shows a brighter translucent band where the wall is thin and a flash fin on the neck thread, ring-light camera above
Thin-wall band and neck flash on a blow-moulded HDPE bottle in transmitted light

Situation

You extrusion- or stretch-blow-mould HDPE, PP and PET bottles and jerry cans, and injection-mould caps and closures alongside them. Container lines typically run at 3,000 to 20,000 units per hour, and closures come off high-cavitation tools. Inspection is often a leak tester plus a person checking a few containers an hour, so a thin wall or damaged neck is found at the filler, not in your plant.

Defects

  • Wall-thickness variationthin or thick bands in the container wallParison programming errors, die-gap wear or uneven preform heating stretch the material unevenly, and thin zones buckle under top load or split when dropped.
  • Neck and thread flashExcess material or poorly trimmed tails on the neck finish stop the closure seating and cause leaks.
  • Parting-line defectsMismatched mould halves or a poor pinch-off leave ridges, notches or weak seams along the parting line and the base.
  • ContaminationBlack specks, gels and foreign particles from degraded or recycled resin show as dark spots or unmelted lumps in the wall.
  • ColourMasterbatch dosing errors or variable recycled content shift the shade between containers or leave streaks.

Inspection setup

Containers pass a backlight panel in front of a monochrome area camera with a 12 mm lens; light transmission through translucent HDPE and PP maps wall thickness variation, because the amount of light that passes changes with wall gauge. Above the neck, a camera with a ring light and a mirror arrangement captures a 360° view of the neck finish and threads in one image, as in our bottle defect detection guide. A colour camera under diffuse light checks shade and contamination.

The station sits after the trimmer and before bagging or palletising, and closures get the same top and side views on a separate conveyor. Failed containers are blown off by an air-jet reject, and the mould station number is read from the blow-moulding machine and logged. Backlight and neck checks on filled bottles appear on our food and beverage page.

Outcome

Every container is checked for wall distribution, neck finish and contamination before it leaves your plant. Thin-wall trends show per mould station, so you adjust the parison profile instead of waiting for a top-load failure. Neck flash is rejected before it becomes a leaking closure at your customer's filler, and contamination counts tell you when a recycled-resin lot needs attention.

How does injection moulding inspection work on your line?

A moulding inspection station runs the same six steps on every line; what changes is how parts are presented and which machine signals the system reads.

Black dome light with an upright industrial camera on top, clamped to an aluminium post a few centimetres above a belt conveyor of grey moulded housings, with an air-jet reject nozzle and reject bin beside it
Dome light and camera close over the belt, with an air-jet reject into the bin for failed parts.
  1. 1

    Trigger and part presentation

    Robot take-out places each shot at a fixed pose and sends a digital trigger with the cavity number; conveyors use a photoelectric sensor and an encoder. Blow-moulded containers are spaced by a timing screw or star wheel so each one is imaged on its own.

  2. 2

    Image acquisition

    GigE Vision or USB3 Vision cameras with C-mount lenses sit in an enclosure that blocks ambient light. Dome, dark-field, backlight and laser sources are strobed in sequence, so one trigger produces several images of the same part.

  3. 3

    Edge inference and latency budget

    Models run on NVIDIA Jetson or Intel OpenVINO hardware beside the press, with inference under 50 ms per image. That fits well inside a 5 to 60 second moulding cycle and leaves time for multi-view checks on fast blow-moulding lines.

  4. 4

    Decision and actuation

    Pass, reject or flag goes to the cell PLC over OPC UA, Modbus or PROFINET. The PLC drives a reject chute, flap or air-jet, and can stop the press or block one cavity after a set number of consecutive rejects.

  5. 5

    Traceability and data retention

    Each result is stored with time stamp, machine, mould, cavity number and shot counter read over Euromap 77. Images stay on the edge for an agreed period, typically 30 to 90 days, and summaries go to MES and SPC.

  6. 6

    Active learning for new defects and moulds

    Uncertain images are queued for operator review. Confirmed labels feed retraining on SageMaker, so a new mould, colour or defect class is added without rebuilding the station.

Which imaging suits which moulding defect?

Each defect family has a lighting geometry that makes it visible; the station combines two or three of them in one enclosure.

Defect familyBest-suited imagingWhy
Short shots and gate vestigeBright-field 2D under a diffuse dome, plus laser triangulationMissing features show in the 2D outline, and stub or boss height shows as a 3D deviation.
Flash, sink marks and ejector pin marksLow-angle dark-field lightGrazing light makes thin fins and shallow depressions cast shadows that diffuse light hides.
Warpage and dimensional deviations3D laser triangulation or structured lightFlatness and feature position need height data; a 2D image cannot see bow or twist.
Splay, weld lines and scratchesDark-field plus multi-angle bright-fieldSilver streaks and fine lines change reflectance with angle, and two angles separate them from grain.
Colour variation, black specks and burn marksColour camera under a diffuse dome lightEven, glare-free light gives stable colour values and high contrast for dark specks.
Voids and residual stress in clear partsBacklight and crossed-polariser imagingVoids scatter transmitted light, and stress shows as coloured fringes between crossed polarisers.
Wall-thickness variation in containersBacklight transmissionLight passing through translucent HDPE or PP changes with wall gauge.
Neck finish, thread flash and parting-line defects360° neck imaging with ring light and mirrors, plus side backlightThread profile and flash show in silhouette; a single overhead view misses the back of the neck.

Integration with the moulding cell, MES and quality system

The station reads from the moulding machine, writes to your quality systems, and keeps data where you decide.

Moulding machine (Euromap 77 / OPC UA)
Reads mould ID, cavity number, shot counter and process values over Euromap 77, the OPC UA interface for injection moulding machines. Can send a reject or stop request back to the machine or cell PLC.
Cavity-level SPC
Supports defect rates and dimensional measurements per cavity on control charts, with alarms when a cavity trends towards its limit. Results export to your SPC software or MES.
MES, ERP and QMS
Writes good, reject and quarantine counts per order to MES and ERP, sends nonconformance data to your QMS, and attaches images to an 8D or complaint record.
Where data lives
Inference and short-term images on the edge device at the press; long-term images and model training on-premises or in an EU cloud region you choose.
Automotive end customers
Designed to support the IATF 16949 traceability and PPAP evidence your automotive customers ask for, with results linked to lot and cavity. The control plan and its approval stay with you.
Medical device end customers
Opsio supplies audit trail, user roles and validation documentation support for ISO 13485 and FDA 21 CFR Part 11 environments; the validated state and SOPs remain your responsibility.

From first test to your line in five steps

Moving from sample parts to a station at your press takes five steps, and you decide after step 2 whether to continue.

  1. 1

    Opsio runs tests in the AI Lab

    You ship a sample set, typically 50 to 100 good parts and 20 to 50 defective parts per defect class, taken from as many cavities as possible: short shots, flash, sink marks, splay, black specks and warpage, or thin-wall and neck-flash containers. Opsio images them under dome, dark-field, backlight and laser light, trains a first model, and measures detection rate and false-reject rate per defect class.

    You provide
    Good and defective parts marked with cavity number, drawings, cosmetic specification and boundary samples.
    Opsio delivers
    Imaging trials, a first model and per-class detection and false-reject results.
  2. 2

    Present results and pricing

    Opsio reviews the results per defect class with your quality and production engineers, using example images of your own parts, including borderline cosmetic cases. You receive the recommended cameras, lighting and 3D sensor, the cycle-time fit for your press or blow-moulding line, and a fixed quote for the production station and the managed service.

    You provide
    Cycle times, reject criteria and the engineers who will own the station.
    Opsio delivers
    Results report, bill of materials, cycle-time analysis and fixed quote.
  3. 3

    Decision and PO/MSA signing

    You decide whether to go ahead. The purchase order and master services agreement set the scope (presses, moulds and part families), the acceptance criteria per defect class, ownership of images and models, and the support level once the station is in production, including how new moulds are added to the model.

    You provide
    Purchase order, signed MSA and agreed acceptance criteria.
    Opsio delivers
    Signed MSA, project plan and named contacts.
  4. 4

    Installation in the Opsio AI Lab

    Opsio builds the production station in the AI Lab on the real cameras, lighting, enclosure, edge computer and reject interface, and tests it with your parts against the agreed acceptance criteria. For medical device components, the validation documentation that supports your IQ and OQ is prepared at this stage, ready for your quality team to review.

    You provide
    Further sample parts, Euromap 77 and PLC interface details, and the cell layout.
    Opsio delivers
    Pre-commissioned station, acceptance test record and validation documentation where needed.
  5. 5

    Installation at the customer location

    The station is installed after robot take-out or the trimmer, connected to the moulding machine and PLC, and run in shadow mode next to your current inspection. Once its results meet the acceptance criteria it switches to live rejection, your operators are trained, and the active-learning loop starts.

    You provide
    Installation window, utilities, network access and operators for training.
    Opsio delivers
    Installation, integration, shadow-mode report, training and the managed service.

Pricing for the feasibility study, production stations and the managed service is on the AI Visual Inspection service page.

AI visual inspection for Plastics & Molded Parts: questions

What are some common defects in injection moulding?

The most common are short shots, flash, sink marks, warpage, weld lines, splay, burn marks, black specks, colour variation, flow lines and jetting. Almost all are visible to a camera with the right lighting: dark-field for flash and sink marks, a diffuse dome for colour and specks, and 3D measurement for warpage. Internal voids in opaque parts are the exception, because they need X-ray or a cut section.

What is splay in injection moulding?

Splay is silvery streaking on the part surface that fans out from the gate in the direction of flow. It is usually caused by moisture in under-dried resin or by volatiles from overheated material. A camera sees splay as a change in reflectance under dark-field or multi-angle light, and a splay rate rising on several cavities at once typically points to the dryer rather than the mould.

What is a sink mark defect in injection moulding?

A sink mark is a shallow depression in the surface above a thick section such as a rib, boss or screw tower. The core keeps shrinking after the skin has frozen and pulls it inward. Sink marks are often only tenths of a millimetre deep, so they need low-angle dark-field light or 3D measurement; diffuse light alone tends to hide them on dark or textured parts.

What are weld lines in injection moulding?

Weld lines, also called knit lines, form where two flow fronts meet and fuse incompletely, typically around holes, cores or between two gates. They show as a fine line or slight notch and can weaken the part. Camera inspection uses grazing light to make the line visible and tracks its position and length, because a weld line that moves often signals a change in melt or mould temperature.

What is a short shot in injection moulding?

A short shot is a part that is not completely filled, so a rib, boss, clip tip or edge is missing. Causes include a blocked vent, a cold hot-runner tip, too little shot volume or low injection pressure. A camera finds short shots by comparing the outline and feature heights with a reference, and cavity tracking shows whether one cavity or the whole tool is affected.

What are the common blow molding defects?

Common blow-moulding defects are wall-thickness variation, neck and thread flash, poor pinch-off and other parting-line defects, contamination and black specks, colour streaks, and deformed or ovalised containers. A backlight shows wall distribution through translucent HDPE and PP, a 360° view of the neck finds thread flash and trimming faults, and a colour camera under diffuse light checks shade and specks.

How does warpage inspection work on moulded parts?

Warpage inspection measures the part in 3D, because bow and twist are height changes that a 2D image cannot see. A laser triangulation or structured-light sensor scans the part on a flat reference or fixture and compares flatness and feature positions with the CAD nominal. Glass-filled parts keep moving after ejection, so agree with your customer whether the press-side or a conditioned measurement is the reference.

How does flash inspection work?

Flash inspection looks for thin fins of plastic along the parting line, around inserts and at ejector pins and vents. Low-angle dark-field light makes the fin edge catch the light against the part, and a backlit silhouette shows flash on the outline. The model learns where a small fin is acceptable and where it is not, for example on a sealing face or a snap-fit edge.

What does plastic birefringence show in moulded parts?

Birefringence shows residual stress and molecular orientation in transparent parts. Between crossed polarisers, stressed zones appear as coloured fringes, typically around gates, weld lines and inserts. A camera with a polarised backlight can image these patterns on clear lenses, covers and medical housings and flag parts whose pattern differs from the approved reference, a sign of a moulding change that can lead to cracking later.

How do you get rid of flow lines in injection moulding?

Flow lines are removed on the process side, typically by raising melt or mould temperature, increasing injection speed or moving the gate. Camera inspection does not remove them, but it shows which cavities and which shots are affected and correlates them with process values read over Euromap 77, so your process engineer can confirm whether a setting change worked.

What should an injection moulding inspection checklist include?

A checklist typically covers fill, flash, sink marks, weld lines, splay, burn marks, black specks, colour, gate vestige, ejector pin marks, critical dimensions and flatness, plus packing and labelling. AI inspection automates the visual and dimensional items on every shot and records results per cavity. Destructive tests such as burst, drop and material checks stay in your laboratory plan.

Is there a visual inspection standard for plastic parts?

There is no single universal standard; cosmetic acceptance criteria usually come from your customer's drawing or appearance specification. These typically set viewing distance, lighting, viewing time, surface zones and allowed defect sizes, often backed by boundary samples. An AI system turns that specification into labelled image classes and thresholds, so the same criteria apply on every shift and every change is documented.

What are the quality standards for plastic injection moulding?

Most moulders work to ISO 9001, with IATF 16949 for automotive parts and ISO 13485 for medical device components, plus customer-specific requirements. These standards require process control and traceability rather than a specific inspection method. Camera inspection supports them by recording a result per part and cavity, feeding SPC and keeping images as evidence; the quality system and its approval stay with you.

Book a 20-minute walkthrough for plastics & molded parts

Pick the defects that cost you most, and we will show you live how a vision model finds them on a line like yours.

  • Live on your defect type, no slides
  • A reply within one business day
  • No obligation
Your call is with
  • Vaishnavi ShreeDirector & MLOps Lead
  • Praveena ShenoyCountry Manager, India

We use your details to answer your request, as described in our Privacy Policy.

Other industries