Cosmetics and consumer chemicals
Filling, sealing and label, with visual checks on the line.
Teal Neural
How many parts came out. And how many fell off.
The line counted from the camera: units, rate and the exact moment the line rate broke. Without touching the controller.
The controller knows it is switched on.
The PLC counts machine cycles. That is not the same as good product coming out: a cycle can finish with the part badly placed, fallen or with no label.
So the real production figure is known at the end of the shift, counting boxes, or the next day when the report is reconciled. By then, whatever was done wrong has already been done.
And nobody knows why: the report says 4,200 instead of 4,600, and from there the questions start.
You count what comes out, not what turns.
You mark on the image the point the product passes and what has to be counted. Out comes the number of units, the rate per minute and the curve of the shift.
When the rate drops, the video of that moment is there. There is nothing to reconstruct: you look at it.
You do not have to stop the line to install it or touch the controller, and the camera can measure something else next month without any cabling being run again.
What it counts.
- Units passing a point, and the real rate per minute and per hour.
- Cycle time, and how it varies over the shift.
- Micro-stoppages: the ones under a minute, which are written down in no report.
- Material jams and build-ups upstream.
- Product fallen, tipped over or out of format.
- Visual quality checks: colour, shape, seal, label present.
- Temperature changes in processes that need it, with a thermal camera.
- Apparent consumption and machine hours, to break down the energy cost.
- Raw material level in the hopper and at the feed point.
Lo que Teal Neural entiende
5 detecciones grabadas.
Producción y línea de fabricación.
Teal
Teal
Teal
Teal
Teal
Cada una es una grabación del sistema funcionando. Pasa el ratón por encima —o tabula hasta ella— para verla en movimiento.
A real case.
A packaging line was giving 4,200 units a shift when the nominal figure was 4,800. The report logged only two long stoppages.
Counting from the camera turned up 34 micro-stoppages of between 20 and 50 seconds, all at the same point: a bend where the packs jammed one at a time. They added up to 22 minutes of the shift, which is almost exactly the 600 units that were missing.
Where it is used.
Meat and prepared foods
Format, process temperature and line rate.
Glass and ceramics
Pieces per cycle, breakages and kiln temperature.
Plastics and processing
Injection cycle, visible flash and consumption per unit.
Drinks and packaging
Units per minute on the filler and the case packer, packs tipped over and gaps on the outfeed line.
Food
Product coming out of an oven or a tunnel, correct format and pieces coming off the belt.
Pharmaceuticals
Vials present in the tray, complete blisters and position checks before sealing.
Automotive and components
Pieces per cycle, assembly sequence and part present before the next station.
What cannot be promised.
Nobody can tell you how much production will go up, because that depends on what gets done with what you see. The system shows you where you are losing it; getting it back is a plant decision.
What can be promised is the measurement: real units, real rate and the breakdown of the stoppages. And that nobody writes that number by hand.
If with that in front of you there is no margin to be found, then there is none — and knowing that is worth something too.
How you start.
- One counting point where the finished product comes out.
- Two weeks comparing with the production report.
- Lowering the micro-stoppage threshold until the twenty-second ones show up.
- After that, visual quality and temperature if they are needed.
Cycle time is the missing figure.
A line's average rate hides almost everything. A line doing 300 units an hour could be doing 5 a minute steadily, or 12 for twenty seconds and none for forty.
Measuring the time between one unit and the next gives you the real distribution: how many good cycles, how many slow ones and where the tail of bad ones sits. That is where the margin is, and not in raising the nominal speed.
And because the figure is per unit and not per shift, when something degrades you see it the same day and not in the monthly report.
Temperature and energy.
With a thermal camera, the same installation measures temperature: an oven that does not get there, a sealing zone that cools down, a motor that heats up before it breaks.
And from the real hours of machine running you get a breakdown of consumption by line and by product that today is estimated. It is not an electricity meter, but it answers the question almost nobody can answer: how much energy each unit made is carrying.
Quality checks, and where they stop.
The camera tells apart what can be seen: colour out of range, wrong shape, seal open, label missing or crooked, pack deformed. That covers a large share of the defects that today are caught at the end, once a thousand have already been made.
What it does not cover is what cannot be seen: weight, real seal integrity, composition. There are scales and inspection machines for that, and this does not replace them.
The combination that works is visual checks on the line for the obvious and sampling for the rest — with the difference that now the sampling is done knowing where to look.
Casos grabados
Funcionando, en sitios de verdad.
Botellas atascadas - Optimiza tu línea de producción con Infinity Neural
Optimización de la Producción Frutícola con Teal Neural
Monitorización de manómetros con Teal Neural
Teal Neural: Optimizando la producción industrial
Teal Neural: Optimización y control en cintas de transporte
What it does not do.
It does not measure what cannot be seen. If the product is inside a closed box or moving at a speed the camera cannot resolve, this is a sensor and not a camera.
And it needs steady light. A line where the lighting changes every hour means fixing the light before anything else.
And what sits alongside it.
Other things the same family understands, on the same cameras.
What people ask us.
▸Do we have to replace the cameras we already have?
Almost never all of them, and sometimes none. What decides it is not the brand or the megapixels: it is the angle, the distance and the light. A modest camera well placed works better than a good one looking from the wrong spot.
The first thing we ask for is an image from the camera exactly as it is. With that you can see whether it works, whether it has to be moved or whether another one is needed. And if another one is needed, we say so then and not after signing.
▸Where is the video processed?
Wherever it needs to be. It can be on a machine on your own site, with the image never leaving it, or on a server. On sites where the video cannot leave for reasons of policy or regulation, it is processed inside, full stop.
What goes out in that case is not image: it is the alerts and the figures.
▸And data protection?
None of this identifies people. It tells apart figures, postures, objects and vehicles, not faces or names.
Even so, a system that processes images of a workplace has its obligations: informing people, defining what it is used for and how long things are kept. That gets sorted at the start, with whoever handles data protection at the company, and it is not paperwork left to the end.
▸Does it need to connect to the PLC?
No. It measures from the image, and that is why it works on lines where the controller gives no access or does not exist.
If there is a PLC and it can be read, so much the better: you cross the two figures, and the difference between cycles and good product is a piece of data in itself.
▸How long does it take to give results?
The count, from day one. What takes time is having something to compare against: until there are two or three weeks of history, the number is there but it does not tell you whether it is good or bad.
That is why the first thing to look at is not the average, it is the micro-stoppages: you see those on the first shift.
▸Does it tell good product from bad?
What can be seen: shape, colour, seal, label, deformation. That is a large share of the defects.
What cannot be seen — weight, seal integrity, composition — still needs its own equipment. And saying so beforehand avoids the awkward conversation later.
▸Can it be used on several lines at once?
Yes, and that is the normal way. Each line has its own configuration and thresholds, and the data gets compared between them: the same machine on two different lines performing differently is an interesting question.
▸Do we have to stop the line to install it?
No. You work with the camera that is already there, or a new one goes up during a planned stoppage. What does not happen is touching the controller or getting involved in the process.
The set-up — marking the counting point and saying what gets counted — is done with the line running, and it is checked against the report for a few days before it is accepted.
▸Does it match what the PLC says?
Almost never, and that is the point. The controller counts cycles; this counts product coming out. The difference between the two figures is exactly what is being lost.
The usual thing is to leave both counters side by side: the controller's for the machine and this one for real production.
▸At what speed does it stop working?
It depends on the format and on the camera, there is no universal number. With large, well-spaced product you can count very fast; with small product, packed together and edge-on, much sooner.
It is the first thing we look at with a video of your line. If it does not work, we say so then and not after installing it.
▸And if we change format every week?
That is the usual thing and it is taken into account when it is set up: you configure one profile per format and change it with the production changeover, the same way you change a recipe.
Get started
Do you want this on your site?
This is already recorded working. What is left is to see whether your camera allows it: where it is fitted, what angle it has and what can be seen from there. If it will not work, we will tell you.