Production decisions
How to design a control that does not let the defect through
Detecting the fault is the easy part. The difficult part is deciding what the system does when it finds one, and getting that right.
A small defect slipping through a fast line does not stay small. It travels, multiplies and appears three stages later as a stoppage, scrap or an unhappy customer.
The obvious response is to add inspection. The less obvious point is that most controls installed do not solve the problem, because they are designed around the sensor alone.
Detection is usually straightforward. Four other things determine whether the control helps or gets in the way.
The case used as an example
A six-unit printing press with a rotary die. The die makes small circular perforations that act as references for downstream automatic equipment.
When the cut is incomplete or the cut-out does not detach, that reference disappears. Downstream, the final equipment can jam.
The line speed made reliable visual inspection impossible. The reel could be inspected afterwards and material recovered, but that does not scale: it means checking everything to find the small amount that fails.
The brief was to detect missing perforations in all six lanes while running.
Decision 1 · What to look at, and how many times
Six parallel lanes means six independent channels. Not one scanning sensor, but six fibre-optic barriers, one per lane, each aligned with its perforation path.
It sounds obvious put like that, but the temptation to save on sensors is constant, and whenever lanes are grouped you lose the ability to say which one failed. Identifying which one is exactly what makes the warning useful.
The mechanics exist for that purpose: four passive rollers guide the web and six adjustable mounts on four bars align each barrier with its lane. Adjustable, because jobs change, and an inspection system that has to be dismantled to readjust it eventually gets dismantled altogether.
Decision 2 · What to compare against
This is the part most often underestimated.
The distance between perforations is constant. The time between them is not. The line starts, accelerates, decelerates and stops. With a fixed timing reference, the control sees missing holes during each ramp when there is only a speed change.
A system that gives false alarms at every start gets disabled within two weeks. Not because it fails, but because nobody tolerates a beacon shouting for no reason.
The solution is a live reference: the control updates its timing reference with each valid reading, following the ramps instead of mistaking them for a defect. Detect, measure the interval, adapt the reference and record the absence.
That ability to follow the process separates an inspection system that stays installed from one that ends up bypassed.
Decision 3 · What happens when something is found
This is the most counterintuitive decision of all: the system does not stop the line.
It could. Technically, it is trivial. But a direct stop for every incident can affect the process more than it helps — and on many lines, stopping at the wrong moment creates more waste than the defect itself.
So it does something else: briefly activates the beacon, identifies which lane failed on screen, increments the counter and saves the event against the reel and lane.
The result is that the decision remains with the person. The system provides precise information; production decides what to do with it.
A control that informs well is worth more than one that acts badly.
Decision 4 · What it records
A warning is useful at the time. A record is useful afterwards.
Every incident is saved with reel identification, date and time, and quality can export the history as CSV over USB. That makes the control more than an alarm: it provides traceability to look back and answer questions that previously had no answer.
And it completely changes inspection work. Before, reels were inspected to find the defect. Now, what has been flagged is inspected.
A fifth decision, less visible
The equipment works without depending on signals from the press's original control.
That installation decision seems minor, but is not: it means the control is mounted with four screws and a power supply, without touching the host machine, changing its program or creating a dependency between two systems that age at different rates.
It also means the same architecture can be adapted to other rotary-die lines, with the appropriate integration modifications.
When an inspection system has to be married to the machine's PLC, every machine update becomes an inspection-system problem. Keeping them separate costs some engineering initially and pays it back for years.
The summary
If you are adding in-line quality control, the sensor is the least of it. Ask:
- Can it say exactly where the failure occurred, or only that something failed?
- Does it adapt to the real process — starts, stops and speed changes — or will false alarms continue until someone disconnects it?
- What does it do on detection? Is that better than what happened before, or just more elaborate?
- Does it leave a record? Without that, there is no improvement, only scares.
- Does it depend on the host machine to work?
A control that fails any of these five ends up bypassed. And a bypassed control is worse than none, because it also creates false reassurance.
At 7Axis Industrial we design and integrate inspection and error-control systems for running processes. Mechanics, sensors and control working as one system.
View the full RV-510 case → Quality, inspection and poka-yoke →