Manufacturing Vision Systems for OEE: Measuring the Losses Sensors Miss

Key takeaway: Computer Vision OEE adds cameras to the machine signal. The camera counts output where no signal counts well, and you open its video for any stop on the shift timeline. The machine signal still times the stops, and the operator adds the reason. Much of the lost capacity sits in short stops that never reach a paper log, and at stations with no usable machine signal.
Today the camera counts output, so short gaps in the count show where a station keeps stopping, and the video for that minute shows what happened. A camera that detects each jam by itself is on our roadmap.
Plan on about one month to install the devices and connect the machines. Cameras are an add-on, quoted per plant once a review of your stations shows how many you need, and a camera station needs no PLC work. See how Fabrico is priced.
The "black hole" on your factory floor
Quick answer: Computer Vision OEE puts AI cameras over the stations a machine signal cannot count well, such as hand-pack tables, transfers and older machines. Today the camera counts output there, and you open its video for any stop on the shift timeline, next to the machine data and the operator's reason. Camera-detected stops and suggested stop reasons are on our roadmap. A camera station counts output without PLC integration or extra sensors. It suits older lines, closed controllers, manual stations and lines where a PLC tap costs more than it returns. Where a machine already gives a clean signal, use the signal.
Machines are connected in about one month, and the full rollout, with data import, setup and on-site training, is complete in 3 to 4 months.
When to use it: older equipment without OPC-UA, mixed-vendor lines, multi-tenant plants, or any line where the cost of PLC retrofit exceeds the value of automated OEE for 12 months.
For OEE basics, see the OEE Complete Guide. For benchmark context, what a good OEE score is. For the data layer, OEE data collection methods compared. For the visual verification angle, 5 best CV OEE platforms.
Every plant has a black hole: the time the line is not running, but nobody can tell you why. The PLC reports cycle complete. The operator log says nothing happened. The OEE dashboard records 87 minutes of "minor stops" with no cause attached.
Events under 5 minutes often never get logged. On a line running 6,000 units a shift, each 1% of capacity lost that way is 60 units, every shift.
Operators cannot write down a 30-second jam. PLCs only know "stopped" or "running", not why. The black hole exists because the data layer was never built to see it.
Computer vision OEE narrows that gap. The camera counts what the station really made, and the video for that minute lets the operator pick the right reason instead of a guess. See related: OEE Monitoring Without PLC and OEE Data Collection Methods Compared.
Related deep-dives: OEE without PLC integration · OEE complete guide · Best CV downtime-verification software 2026 · OEE data collection methods.
How a computer vision OEE tracking system works, end to end
A computer vision OEE tracking system is cameras, the machine signal and the operator, feeding one OEE calculation. The camera counts output and gives you video. The machine signal times run and stop. The operator adds the reason. Here is the path from a unit on the conveyor to an OEE figure, as it works in Fabrico today.
- Line review. You agree which stations get a camera, which connect through the PLC and which get a Fabrico IoT device. A camera goes where there is no usable signal: a hand-pack table, a transfer, an older machine. See camera or sensor, machine by machine and how many cameras a line needs.
- Count. The camera over the station counts units as they pass, every minute of the shift. Camera counts and PLC counts feed one OEE calculation, so lines connected in different ways compare side by side. More on AI visual counting.
- Time the stops. The machine signal gives each stop its start and end, short stops included. At a camera station with no signal, gaps in the count show on the shift timeline when the station made nothing, and the operator logs the stop.
- Add the reason. The operator picks the reason from your own list on a tablet or phone. Time with no reason shows as its own figure, so missing reasons stay visible.
- Look at the minute. Click a stop on the shift timeline and open the camera video for that minute, next to the count and the reason. In the standard setup the video is streamed when someone opens it and is not stored in Fabrico. Access is role-based.
- OEE and money. Availability, Performance and Quality come out by line and shift, with the losses ranked in money. Data leaves through a REST API, webhooks and Excel or CSV export.
What runs today, and what is next, job by job:
| Job | Who does it today | Next (roadmap) |
|---|---|---|
| Count output at a station | The camera, or the PLC where it counts well | Already live |
| Time each stop | The machine signal, or the operator where there is none | The camera detects the stop |
| Name the reason | The operator, from your reason list | The camera suggests, the operator confirms |
| Show what happened | Camera video for that minute, on request | Jam and blockage zones on the camera view |
| Check each pack | An inspection system, not the OEE camera | Stays with inspection vendors |
Computer vision OEE by FMCG sector
Each sector page shows where cameras go on its lines, what stays with the PLC or the inspection system, and what to ask vendors.
- Food: food manufacturing, bakeries, confectionery, dairy, meat and poultry, seafood, ready meals, fresh produce, eggs, pasta, canned food, coffee
- Drinks: beverage and bottling lines, breweries
- Personal care, home and health: cosmetics, household products, supplements
- Across sectors: FMCG manufacturing, co-packers, packaging converters
Which OEE losses a vision system can actually resolve
Manufacturing vision systems are often sold against OEE in general. It is more useful to ask which specific loss they resolve, because the answer is not all six.
- Small stops. The strongest case. Events too short to register as a downtime reason in controller data are visible on video.
- Reduced speed. Cameras can show the human or material cause behind a slow cycle, which a rate figure alone does not explain.
- Setup and adjustment. Footage shows how long a changeover really took and where the time went, rather than when someone remembered to log it.
- Defects and rework. Useful where the defect is visible at the point it is created, so the cause and the reject sit on one timeline.
- Breakdowns and startup losses. Usually better served by machine signals. Vision adds context, not the primary reading.
What is Computer Vision (in simple terms)?
Computer Vision is a camera plus AI software that watches a station and recognises the units passing through it. In Fabrico today, that means it counts output, every minute of the shift, and shows the video for any minute you open from the timeline.
The hardware: an industrial-grade camera over each station that has no usable machine signal, with a clear view of the product flow. How many cameras a line needs, where they mount and what the network must carry are agreed after a review of your stations. Camera cost: quoted per plant after that review.
The software: a vision model set up for your line that recognises good output coming through and counts it. Recognising a jam or a stopped machine from the camera is on our roadmap. Checking parts against a spec is a job for an inspection system, not for an OEE camera. Before you scale, compare the camera count with a hand count on your own pilot line: see how accurate a camera count is. If you are weighing a model your own data team would build and maintain, our guide to the build or buy decision for line cameras sets out who owns what.
What it is NOT: it is not generic surveillance. It is not facial recognition. It is not a black box. Cameras point at the product flow, and in the standard setup the video is streamed when someone opens it and is not stored in Fabrico. See line cameras without surveillance.
See Fabrico's implementation: Computer Vision OEE feature page.
Use Case 1: monitoring legacy machine uptime (Availability)
A 1995-era hydraulic press has no PLC interface you can tap. No Modbus, no OPC UA, no Ethernet at all.
The control cabinet is sealed by the OEM. Traditional answer: install IoT vibration sensors and a current clamp on the motor.
That means an electrician and integration work on every machine.
Computer Vision answer: one camera with a view of the platen. Today the camera counts the parts the press makes, so the press gets a count and a Performance figure without opening the cabinet. Gaps in that count show on the shift timeline when the press made nothing. Each stop is timed by a machine signal, from a Fabrico IoT device if there is no PLC to read, or logged by the operator. Reading the press states (idle, closing, hold, opening) straight from the camera is on our roadmap.
Open the video for any gap or stop to see what happened. Timing and cost: agreed after a review of the station.
No PLC work for the camera.
To compare the two routes on your own presses, ask each vendor for a written quote per machine, sensor retrofit and camera, with the install time, and pilot one press each way. Then check both counts against a hand count for the same shift.
This is why tracking OEE without PLC is practical when a camera counts the output and a simple machine signal or the operator times the stops.
Sensor-first OEE tools reach the same stops by another route. Our Fabrico vs TeepTrak comparison shows where cameras fit and where plug-and-play sensors fit.
Use Case 2: automating part counts (Performance)
Performance loss is the silent killer. The line is running. The PLC reports normal operation. But you are doing 92 parts a minute on a 100 part-per-minute design rate. Where did the 8 parts go?
Computer Vision answers this in three ways:
- Direct count. The camera counts every unit passing a marker on the conveyor. Cross-check with PLC output exposes the gap.
- Short stops. A machine signal times each stop, short ones included. Where there is no signal, gaps in the camera count show where short stops cluster in the shift, and you open the video for those minutes. Camera-detected micro-stops are on our roadmap. Paper logs miss many of them, because nobody writes down a 30-second jam. See micro stops and OEE.
- Cycle time drift. The live count per station shows the running rate against the rated speed. A 0.4 second drift over a week is invisible to the eye but visible in the data. By the time the operator notices, you have lost a shift of output.
Worked example, illustrative numbers: at 92 parts a minute on a 100 part-a-minute rate, 8 hours of running lose 3,840 parts, or 8% of rated output on every shift at that rate. The camera count shows when in the shift the rate drops, so the team knows where to look first. Price it with your own margin per part, and count it only at a station that limits output.
Compare options: Best Real-time OEE Monitoring Systems.
Use Case 3: feeding quality data into OEE (Quality)
Most quality control happens after the fact. End of line, off-line, sample-based. A defect is found, but 200 units of the same defect have already shipped.
Inline inspection of every unit is a job for inspection vision systems from vendors such as Cognex, Keyence or Mettler-Toledo. Fabrico does not sell inspection. Those systems check labels, fill levels, lids and colour, and reject bad units in real time. Their reject count, or a reject signal from the line, feeds the Quality part of OEE in Fabrico, next to the count of what reached the end of the line. See label and code vision vs an OEE camera.
The bigger value is closing the OEE-CMMS loop. When rejects rise at one station, OEE shows it by line and shift, and the team can open the camera video for those minutes. If a machine needs a fix, the stop becomes a work order for the technician. No "the line was producing scrap and we did not know" stories at the morning meeting.
This is the "OEE Diagnoses, CMMS Cures" pattern we describe in OEE benchmarks guide. Quality data without action is a museum exhibit.
Frequently asked questions
Do we need a clean room for cameras?
No. Industrial-grade cameras work in most production halls, from packaging to automotive. Only severe washdown zones (full sanitation cycles with 80°C water) need enclosures rated for washdown, such as IP69K. Fabrico's camera setup for a washdown zone is agreed per station, so ask any vendor, us included, for the IP rating of each enclosure in writing.
What about lighting changes?
Models are trained on your actual lighting (shift-to-shift variation, sun through skylights, fluorescent flicker). Check it on your own line: on the pilot, compare the camera count with a hand count for the same shift, a night shift included, before you scale. See camera placement and lighting.
Is the video stored on the cloud or on-prem?
Neither, in the standard setup. Video is streamed when someone opens a stop on the timeline and is not stored in Fabrico. Access is role-based. How long video can be opened, and anything beyond the standard setup, is agreed per plant in writing. See GDPR and line cameras.
What is the ROI?
Measure the payback on your first line before you plan a wider rollout.
See how Fabrico is priced and how a camera pilot works. For the quote logic, see How a CMMS Quote is Built.
The real power isn't the camera. It's the connection.
A camera that just produces a dashboard is a more expensive version of the same problem. Computer vision matters because of what the team does with the count and the video.
The right architecture: the machine signal logs a stop on Line 3 and the operator adds the reason. Anyone with access opens the camera video for that minute from the shift timeline. If the reason is a breakdown, the stop becomes a work order for the right technician. Next on our roadmap: the camera detects the stop and suggests the reason, and the operator confirms.
The work order lands on the technician's tablet. From the stop, the technician opens the camera video for that minute. Measure your own response time on the pilot line, before and after.
That is the Fabrico approach: Computer Vision OEE wired natively to our CMMS. Not two systems pretending to talk. One system from sensor to fix.
To test it on one of your hardest lines, read how a computer vision OEE pilot works. For lines with no PLC, see OEE monitoring without a PLC.
To see this on your own lines, look at Fabrico's camera-based OEE software, which puts camera counts, camera video and machine data on one OEE timeline.
Computer vision is one pillar of a broader AI OEE software stack that also covers micro-stop detection and an AI assistant you can question.
See Fabrico live. Book a demo.
Request a demoRelated articles

Real Fabrico tiles, sample data.











