Multi-camera AI inspection that finds black spots, thread damage, colour drift and deformation on blow-moulded bottles, then rejects them online before they reach your filler.
Six to sixteen industrial cameras give 360-degree coverage of cap top, sidewall and thread. Defective closures leave the line through high-speed air valves; good ones are counted and boxed.
Our Good Product Training Model learns from OK images alone, so you are not blocked waiting for rare defect samples. Defect Synthesis generates the rest, annotations included.
Light source, industrial camera, edge computing unit, AI software and mechanical frame are all developed in house on a 29,000 m2 campus. That is why support does not stall at a supplier boundary.
KeyeTech designs and builds the machines that sit at the end of a plastic packaging line and decide, part by part, what ships and what does not. A feeder presents the part, a customised light source and industrial cameras image it from every angle, the KVIS AI system compares what it sees against your accepted standard, and an air valve removes anything that fails.
We are not an integrator reselling someone else's cameras. The optics, the strobe light sources, the embedded computing unit and the inspection software are developed by our own team in Hefei, which is why a detection problem on your line can be solved by the people who wrote the algorithm.
A person checking caps on a light table gets slower and less consistent after two hours. Line speed goes up, headcount goes up, and the escape rate stays where it was.
A black spot or a broken tamper ring that reaches the filler is no longer a scrap cost. It is a rejected lot, a claim and a supplier audit.
Traditional vision needs labelled examples of every defect. On a high-yield line those examples take months to collect, so the project stalls.
Cap top, bottle sidewall, preform gate and a printed label need different lighting, different camera counts and different tolerances. One standard machine does not cover them.
Every configuration on the old catalogue is here, with its own specification table and FAQ. Choose by the part you make.
The largest family in our range. These systems inspect plastic closures for water, beverage, dairy, daily chemical and pharmaceutical packaging at production speed, and remove the failures without stopping the line.
Explore the range →
For pharmaceutical, beverage and dairy bottle producers who cannot pass a black spot or a damaged neck finish to the filler.
Explore the range →
The core machine in an automated preform quality line, sitting between the moulder and the blower where a bad preform is still cheap to remove.
Explore the range →
In-mould labelling puts the print inside the wall, so a label fault is a scrapped container, not a re-labelling job. These systems catch it at the moulder.
Explore the range →
A visual inspection machine built for the printing defects that get past a press operator watching a moving web.
Explore the range →
Visual inspection machines for paper and plastic cup quality inspection, covering the rim, the seam, the print and the body in one pass.
Explore the range →
For parts that do not fit any standard machine. We follow the principle that everything can be inspected, and design the optics from your samples.
Explore the range →Optics, electronics, mechanics, algorithm and software sit in the same R&D building. Core members of the team are PhDs from USTC. When the imaging is wrong, the person who can change the light source is upstairs.
The Good Product Training Model performs pixel-level anomaly detection using OK images only. Major defects need no labelling, changeover to a new part is fast, and unknown defects are still caught.
Forward and backward diffusion algorithms generate realistic defect images from a handful of samples, with annotation information attached, so a model can be built without waiting for the line to produce failures.
Connected machines are updated online through the KeyeTech AI cloud platform. Inspection standards can be re-tuned from Hefei, 7x24, without flying an engineer to your plant.
A modular embedded unit with a 32 TOPS NPU chip replaces a traditional PLC control system, tolerates harsher temperature and humidity, and protects data through a sudden power loss.
An engineer is on site within 48 hours of the machine arriving, and stays until it runs stably. Your technicians can also train at our factory free of charge.
Different sectors fail in different ways. These pages describe the defect that actually costs you money in each one.
Water and soft drink lines run fast and audit hard. A closure defect that reaches the filler stops a line that costs more per hour than the inspection machine.
Pharma packaging buyers audit their suppliers. An inspection machine with a defined, adjustable acceptance standard is easier to defend in an audit than a table of operators.
Short shelf life and cold chain mean dairy packaging faults are discovered late and cost a full batch. The inspection has to happen at the converter.
Personal care packaging is a visual product. A scuff, a colour shift or a crooked label is a customer complaint even when the pack functions perfectly.
The same anomaly-detection approach that grades a moulded cap can grade a natural product, where no two accepted items are identical.
Outside packaging, our systems inspect electronic components and new energy parts, where the defect is small and the tolerance is not negotiable.
Put the quality gate where the defect is made, so nothing downstream ever sees it.
A standalone station with its own feeder, for plants that cannot stop or rebuild the production line to add inspection.
Our AI 2.0 core technology. Train on good parts, catch the defects nobody has catalogued yet.
When the defect you need to catch is too rare to collect, synthesise it.
We need the part itself, photographs or drawings, the defect types that matter, the precision you need, your line speed and whether inspection is inline or offline. Physical defect samples are required before a firm quote.
The optical R&D team designs the imaging: light source type, camera count, lens and resolution matched to your smallest defect. We image your real samples before quoting, not after.
Hardware reliability testing, software testing, joint debugging of light, camera, valve and statistics, then model testing against your accepted standard. Nothing ships until it meets your test criteria.
On-site commissioning within 48 hours of arrival, operator training, then one year machine warranty, three years on camera, lens and sensor, and long-term remote maintenance through the cloud platform.
A capacitor manufacturer with the world's largest market share in aluminium electrolytic capacitors recognised KeyeTech's inspection work on automotive-grade product.
A global automation group reviewed KeyeTech's self-developed edge computing, cameras, algorithms and cloud training platform on site.
Recognition of KeyeTech's inspection equipment by international closure manufacturers, formalised in a cooperation agreement signing.
Injection blow moulding and AI inspection running as one connected line at the 91st China International Pharmaceutical exhibition in Xi'an.
We answer within 24 hours. For a firm quotation we need physical defect samples, the required speed and precision, and whether the machine runs inline or offline.
