This article is for plant managers, production managers, quality, continuous improvement and maintenance leaders who already run, or are about to install, machine vision inspection on their lines. The camera detects and ejects defective parts, but the reject count often stays inside the inspection system. This article shows how to feed that count into the OEE quality factor and, above all, how to cross it with machine state to learn when and why rejects appear.
- Machine vision gives a timestamped verdict on every part.
- That count can feed the OEE quality factor directly.
- Crossed with machine state, it shows when rejects appear.
- Inspecting and measuring are two different, complementary jobs.
What machine vision does on the line
ATRIA Innovation, an engineering and technology innovation company based in Zaragoza, Spain, describes how an inspection station works in Control de calidad por visión artificial (Machine vision quality control, in Spanish): the product passes through the station, one or more cameras capture images, software processes them in real time and compares what it finds with defined parameters. Then, in the article’s words: “El sistema determina si la pieza es válida o presenta defectos.” (The system determines whether the part is good or has defects.) An automatic action follows: accept, reject or sort the product.
The ATRIA Innovation article also covers the usual applications: surface defects such as scratches, cracks, stains and deformations, dimensional checks, assembly verification, labeling and packaging, and code reading for traceability. It points out that the information generated can be stored for later analysis. That is the starting point here: inspection produces reliable quality data, part by part. The question is what the plant does with that data once the part has been rejected.
From reject to OEE quality factor
OEE is the product of three factors: availability, performance and quality. The quality factor is the share of good parts in the total parts produced. On paper it is the simplest of the three. In practice it is often the least reliable, because it is fed by manual declarations at the end of the shift, scrap weighed in a bin, or figures that arrive days later from the lab.
A vision station changes this: every inspected part gets a verdict, with a timestamp. If that count reaches the OEE calculation, the quality factor stops being an estimate. The real value appears when the reject is no longer an isolated number and sits on the same timeline as stops, short stops and machine speed.
Two precautions before you start:
- Define the denominator. Parts produced must be counted at the same point for every product: before the inspection station, not after it.
- Handle re-inspection. If rejected parts get a second check and some go back to the line, decide whether they count as good first time. Classic OEE only counts parts that are right first time as good.
Crossing rejects with machine state
The camera knows a part is defective. It does not know what state the machine was in when it made that part. Put the two data sets side by side and patterns appear that neither shows alone. The table below lists typical situations; they are examples of how to read the data, not results from a specific plant.
| Situation | What vision inspection sees | What machine state shows | First lead |
|---|---|---|---|
| Start-up after a changeover | Rejects concentrated on the first parts | End of a format change, speed not yet stable | Start-up settings, changeover standard |
| Rejects after short stops | Defects appearing in bursts | Repeated short stops just before | Restarts that disturb temperature, pressure or positioning |
| Running at reduced speed | Reject rate differs by period | Speed below the rated rate | Operators slow the machine to limit defects: a hidden performance loss |
| Steady rejects, unrelated to stops | Constant reject rate | Normal running | Likely cause in material or process, not in restarts |
| Stop caused by the inspection station itself | Station alarm, doubtful rejects | Stop recorded on the line | Optics cleaning, lighting, threshold settings: separate false rejects |
For example, a packaging line can show an acceptable quality factor over the shift average and still concentrate most of its rejects in the minutes after each short stop. In that case the useful action is not at the vision station, which is doing its job, but on the causes of the short stops.
Step by step: feeding the quality factor
- Pick one line with vision inspection. Start where rejects weigh most on cost or delivery.
- Set the counting rules. Where parts produced are counted, what counts as a reject, how re-inspection and false rejects are handled. Write it down and share it with quality and production.
- Get a timestamped reject count. Depending on the setup, it can come from the eject signal, a counter or the inspection system log. What matters is the time of each reject, not just the shift total.
- Measure machine states automatically. Running, stopped, short stop and actual speed, without relying on end of shift recall.
- Analyze rejects by state. Separate start-ups, periods after a stop and steady running. Compare the reject rate in each.
- Review weekly with quality and production. One table, one OEE calculation, and actions assigned by cause: process, maintenance, changeover or inspection settings.
Inspecting and measuring: two different jobs
It helps to keep the roles clear. Designing vision inspection, choosing cameras and lighting, training the algorithms and setting acceptance criteria is the work of machine vision specialists such as ATRIA Innovation. Measuring stops, short stops and machine performance, and relating them to quality, is a different job.
TEEPTRAK handles that second part and does not perform quality inspection. A module is wired in the electrical cabinet and a sensor sits at the edge of the line; installation takes about an hour per machine, with no connection to the PLC. OEE, stops and short stops are available within 48 hours, and rejects can be recorded against that same timeline to calculate the quality factor. TEEPTRAK is used in more than 700 plants in over 30 countries.
Further reading
- Control de calidad por visión artificial (ATRIA Innovation, in Spanish)
- Quality Rate OEE: How to Calculate, Measure and Improve This Critical Manufacturing Metric
- Nonconformance reports: linking each scrap part to machine history
FAQ
Is machine vision enough to calculate the OEE quality factor? It provides the reliable numerator, the rejected parts. You still need a consistent count of parts produced, taken before inspection, and clear rules for re-inspection.
Why cross rejects with machine stops? Because many defects cluster around start-ups, changeovers and restarts after short stops. Without machine state you see how many parts are rejected, but not when or why.
How should false rejects from the inspection station be handled? Identify them during re-inspection and analyze them separately, as an inspection tuning issue. Mixed with real defects, they distort the quality factor and point actions in the wrong direction.
0 Comments