Measuring OEE with Cameras: Sensorless Production Tracking
Cameras instead of PLCs and sensors for OEE: machine downtime, cycle time and part counts are extracted from video. The advantages of sensorless OEE tracking.
Translated from the Turkish original · Türkçe aslı
Measuring OEE with cameras means automatically extracting the three components of OEE from the footage of a fixed camera that sees the machine and the station: availability from machine stops, cycle time and performance from product passes, and quality from defective parts removed from the line. The PLC is not touched, and no new sensors are fitted to the line; it also works on older machines without PLC data. The operator labels the downtime reason with a single tap, and the data flows to the MES or ERP.
How is OEE calculated, and why is it under-measured in most plants?
OEE (Overall Equipment Effectiveness) shows how much of the planned production time actually turns into good parts. Availability measures for how much of the planned time the machine runs, performance how close it comes to the ideal cycle speed while running, and quality how many of the parts produced are good the first time. OEE is the product of these three ratios; the formula is explained step by step in the how to calculate OEE article. The formula is simple; the hard part is collecting the input for each component reliably.
The classic approach requires PLC signals, sensors and manual data entry. Older machines often have no readable PLC data; what remains is a form filled in by hand at the end of the shift. Every hand-filled form is a rounded version of reality. Short stops never make it onto the form: waiting for material, clearing a jammed part by hand, the operator going to help at another station. The reason on the form is often the remembered reason, and the start time of the stop is an estimate. As a result, manually tracked OEE shows availability higher than it really is, and the improvement meeting is built on the wrong item.
Because OEE is a product, a small error in one component is reflected in the whole score; and a single score does not tell you in which component the loss occurs. Of two lines with the same OEE value, one may be losing through long breakdowns and the other through frequent, short stops. For one the discussion is the maintenance plan, for the other material flow and setup times. This is where the value of measuring the components separately and automatically comes from.
How does measuring OEE with cameras work?
The system looks for answers to three questions. The first is whether the machine is running: the motion signature and the indicator lamp are evaluated together. Is the press ram coming down, is the machine spindle turning, what colour is the stack light: when these signals confirm each other, the state is established. The second is the cycle: product passes are counted, and cycle time and its deviation are calculated. The third is line balance: stations are compared and the point slowing the line down is flagged. Because the operator's interaction at the station is also tracked, the moment the machine is waiting can be distinguished from the moment the operator is waiting.
A single frame does not produce an event; a state change is recorded only after it has been confirmed over consecutive frames. This way, a forklift passing in front of the machine or an operator briefly blocking the frame is not recorded as downtime. When a stop starts, the timer opens automatically, and the operator labels the reason with a single tap on the shop-floor screen. The start and end times come from the system, the reason from the shop floor. The resulting data has minute-level precision and typically reveals 15-20% more downtime than manual records; lost time stops being an estimate and becomes an item you can budget for.
How are availability, performance and quality extracted from video?
For availability, what matters is classifying downtime correctly. The shift calendar, break times and planned maintenance windows are defined in the system; stops within these periods count as planned, and the rest are recorded as losses. On the performance side, the ideal cycle time is defined according to the product recipe; when the measured cycle is compared with it, speed losses and short stops that never get recorded become visible separately.
For the quality component, the reject bin or reject chute where defective parts are placed is monitored; a part is counted when it leaves the line. However, detecting the defect itself, such as a scratch or a dimensional deviation, is not the job of a general-purpose CCTV camera. That is a separate machine vision application, such as surface and defect inspection, which uses triggered industrial cameras and controlled lighting; the decision it produces can feed into the quality component of OEE.
How is cycle time measured at manual stations?
Stations that depend on the operator's pace, such as sewing, manual assembly and packing, mostly remain invisible to PLC-based systems. Here the camera measures active working time at the station, losses such as waiting and searching for material, and fluctuation between cycles; on one pilot line, active working time at a station was measured at 46.8 s. This measurement replaces the short stopwatch study with continuous data and feeds directly into line balancing work. The focus of the measurement is the station, not the person; we cover efficiency on the human side separately in the personnel performance analytics (PPA) article.
What is the difference between camera-based and PLC-based OEE?
PLC-based measurement reads the machine's own signal, so it gives the running state and the part count directly; if the signal is reliable and accessible, it is a strong source. Its limit is that the signal does not explain the reason: you learn that the machine stopped, but not whether it was due to waiting for material or a setup. Older machines have no signal at all, and manual stations are out of scope. The camera closes these gaps and records every stop with visual evidence.
The camera has its limits too. It needs a line of sight; it cannot see motion inside a closed cabin or behind the machine body. On very high-speed lines, a product recognition and counting application using encoder-synchronised, high-frame-rate cameras is the more accurate choice for part counting. In practice the two approaches are not rivals: if PLC data exists, it is combined with the state information from video and the two sources verify each other. We explained the general framework of MES in the what is MES article.
Breakdown of bottlenecks and downtime reasons
The value of OEE lies not in the score itself but in its breakdown. In the sample dashboard on our solution page, 38% of total downtime comes from die and setup changes, 21% from waiting for material, 17% from breakdowns, 13% from break handovers and 11% from quality interventions; in this distribution, halving setup change time alone adds about 6 points to OEE. A table like this settles, with data, the debate over whether the improvement budget should go to a die change (SMED) project or to the material feeding arrangement.
OEE gains usually come not from equipment investment but from planning the downtime that has become visible; the impact at your site is measured during the pilot. Because measurement is automatic, comparisons between shifts and lines stay fair. We discussed why shift comparisons should measure the process, not the person, in the human performance in shift-based production article.
How is OEE data transferred to MES and ERP systems?
The OEE data produced is fed into your existing MES or ERP system via a standard API; no separate software island is created. Events are published over REST API and MQTT with a timestamp, line and station ID and event type; this is how they connect to MES, ERP/SAP and Power BI. For the shop floor there is a shift board view: the operator sees the open stop and the shift's current availability at the head of the line. The downtime reason code list should be kept identical to the list in the MES; otherwise the two systems report the same stop under different names. When downtime data is matched with energy consumption, the cost of a machine running idle also becomes visible; we covered this in the energy efficiency analytics article.
Where does the camera go, and what hardware is needed?
What you need is a fixed camera that clearly sees the machine and the station; the moving area, the stack light if there is one, and the product exit should all be visible in the frame. In most plants, security cameras see the corridor and the general area, not the machine's moving area. That is why suitability is checked camera by camera for each station during the site survey: if the existing camera's angle is sufficient, it is used via its RTSP stream; if not, an additional camera facing the station is recommended. Continuously patrolling PTZ cameras are not suitable; as the frame changes, the reference is lost. For angle, lighting and areas blocked by the operator's body, see the camera placement guide.
For processing, a single GPU server typically handles 8–16 cameras in real time; if heavy models such as pose analysis will also run on the same server, this drops to 6–10 cameras.
Data privacy and KVKK
In OEE measurement the goal is to measure the machine and the process, not the person. In the default setup, footage is processed on an edge server inside the facility and raw video is not sent to the cloud (processing can also be done in the cloud if preferred). Facial recognition is not used; analytics outputs are anonymous, reports are aggregated at station and shift level, and event clips are automatically deleted at the end of the defined retention period. Because the camera also captures employees, a written definition of the purpose of processing and informing employees are required under Law No. 6698 on the Protection of Personal Data, KVKK (Türkiye's Personal Data Protection Law). The form this takes is determined by the facility's legal department; the explanations here are general information and do not replace legal advice.
Common mistakes when measuring OEE with cameras
- Taking the ideal cycle time from the machine label: if the label is not realistic, performance always comes out low; the ideal cycle should be defined per product, as a joint decision of production, maintenance and planning.
- Starting without clarifying the definition of planned downtime: if breaks, shift handovers and planned maintenance are not entered from the start, the score drifts in the wrong direction.
- Keeping the downtime reason list long: with a long list, the operator will still press “other”.
- Not adjusting the short stop threshold to the site: if the threshold is too low, every wait counts as downtime; if it is too high, micro-stops disappear.
- Turning OEE into a performance review of people: once the score becomes a target, people start playing with the definitions.
How is the pilot set up?
The pilot starts with a single line. During the site survey, we jointly assess which station is most likely to be the bottleneck, which machine has no PLC data and which downtime reason is written most often on the form. Installation for a single line is completed within a week; in the first week of the 30-day pilot that follows, you start to see your real availability figure. In the first days, operators get used to labelling reasons; then the thresholds are calibrated to the rhythm of the site: how long an idle period counts as downtime, when short stops are reported separately, and how planned breaks are separated. Setting the criteria in writing at the start of the pilot makes it possible to base the rollout decision on numbers; for this you can use the approach in the pilot design article. The camera suitability assessment before installation is free of charge.
Frequently asked questions
Does measuring OEE with cameras require a PLC connection?
No. Whether the machine is running is extracted from the motion signature and indicator lamp, cycle time from product passes, and the quality component from parts placed in the reject bin, all from video. This is why it can also be used on older machines without PLC data. If PLC data exists, it can be combined with the state information from video; the two sources verify each other, and when they disagree you can see which one is wrong.
How is OEE calculated?
OEE is the product of the availability, performance and quality ratios. Availability shows for how much of the planned time the machine runs, performance how close it comes to the ideal cycle speed while running, and quality how many of the parts are good the first time. A camera-based system automatically collects the inputs for these components from video: downtime durations, cycle times and the number of defective parts removed.
Can our existing security cameras be used to measure OEE?
They can if the camera clearly sees the machine and the station. Because security cameras are usually placed for corridors and general areas, they may not see the machine's moving area or the stack light. Before installation we assess your camera inventory free of charge and report which stations can be measured with the existing camera and where additional cameras are needed.
How long does installation take, and when are the first results visible?
Installation for a single line is completed within a week; in the first week of the 30-day pilot that follows, the real availability figure starts to become visible. Downtime thresholds, the separation of short stops and the definition of planned breaks are calibrated to the rhythm of the site during the pilot. The rollout decision is made by comparison against criteria set in writing at the start of the pilot.
Which systems can OEE data be transferred to?
OEE data, broken down by availability, performance and quality, can be transferred to MES, ERP/SAP and Power BI; events are published over REST API and MQTT. There is also a shift board view for the shop floor. We recommend keeping the downtime reason code list identical to the list in the MES, so that both systems report the same stop under the same name.
In which cases is camera-based OEE not sufficient on its own?
If the machine's running state or product output cannot be seen from any angle, for example if all the motion takes place inside a closed cabin, the camera alone is not enough. On very high-speed lines, part counting requires an encoder-synchronised industrial camera; detecting the defect itself is also a separate machine vision application. In these cases the camera is used together with PLC data or in-line inspection.
Doesn't this mean employees are monitored by camera? What about KVKK?
The target of the measurement is the machine and the process. In the default setup, footage is processed on an edge server inside the facility, raw video is not sent to the cloud (processing can also be done in the cloud if preferred), facial recognition is not used and analytics outputs are anonymous. Event clips are automatically deleted at the end of the defined retention period. Informing employees and defining the purpose under Law No. 6698 should be prepared together with the facility's legal department; this explanation is not legal advice.