Updated July 2026 · 8 min read · Adente Vision Engineering Team
Why does a fail result only matter when it moves the part?
An inspection unit that flags a bad part has done half the job. The other half is physical: a defective part has to leave the line, and nothing leaves the line until an actuator pushes, blows or diverts it. A fail flag sitting on an output terminal, with no wiring to a reject device, lets the same defect ship. On a live cap-inspection line, an Adente Vision unit rejects broken, unclosed and hinge-damaged caps at a 99.65% F1-score and a 0.69% false-negative rate, deciding in about 30 ms per part. That result is only worth anything because a reject actuator downstream acts on it every time a cap fails.
So the wiring question is narrow and practical. How do you turn a fail decision into a movement, on the correct part, without stopping the line or ejecting a good one? The answer is a signal path from a 24V output, a tracking delay, a latch, and a small set of interlocks. None of it is exotic, but each step has a way to go wrong that only shows up on the moving line, not on the bench.
How do you wire a reject actuator to a 24V inspection output?
The unit exposes four inputs and four outputs at 24V, and one output is assigned to the fail or reject signal. A typical assignment uses the remaining outputs for pass, ready and fault, so the reject line has company that tells the cell what state the inspection is in. Adente Vision is an edge-AI visual inspection unit built by ADENTE Advanced Engineering Technologies, part of the Aden Group, sold through automation system integrators, so the reject signal leaves the same enclosure that made the decision, with no PC or cloud sitting between the judgement and the actuator.
That 24V fail output then drives the reject device. The device itself is ordinary cell hardware: a solenoid valve feeding a pneumatic pusher, an air-blast nozzle for small light parts, or a diverter gate for larger ones. Which one fits is an actuator-selection question set by the part mass, the line speed and the space you have, not an inspection question. What matters for wiring is the current the device draws. A standard industrial 24V digital output, as defined in IEC 61131-2, sources a limited current; a solenoid or contactor that draws more than that is switched through an interposing relay or a solid-state relay rather than directly. Sizing that relay to the coil is basic electrical practice, but skipping it is a common way to overload an output.
Two more details decide whether the reject works reliably. First, an inductive load such as a solenoid needs a freewheeling diode or an RC snubber across it, so the coil's collapsing field does not stress the switching output. Second, sinking versus sourcing, PNP versus NPN, has to match between the output and whatever it drives. A correctly triggered reject that does nothing on the bench is almost always a sinking-versus-sourcing mismatch.
Reject signal path: timing and interlock at a glance
| Stage | What happens | Design point |
|---|---|---|
| Inspect | Unit decides pass or fail, about 30 ms per part on the cap line | Fail asserted on a 24V output |
| Track | Part travels from the inspection point to the reject point | Delay = distance divided by line speed, measured on the line |
| Actuate | Output drives the solenoid, air-blast or diverter | Interposing relay when the current exceeds the output rating |
| Hold | Fail stays latched through the reject window | Latch and clear sized to the actuator stroke and part spacing |
| Interlock | Ready and fault outputs gate the reject | No ejection when the unit is not inspecting |
The table compresses the whole chain, but two rows carry most of the risk on a real line: the tracking delay and the latch. Both are worth their own paragraph.
How do you time the reject to the correct part?
The unit decides at the inspection point, but the part is rejected somewhere downstream, so the fail signal has to be delayed to line up with the moment the failed part reaches the actuator. That tracking delay is the distance from inspection to reject divided by the line speed, and it is specific to your cell: the same about-30-ms decision pairs with a reject 200 mm away in one cell and two metres away in another. On an encoder-triggered line the delay is counted in encoder pulses, so it stays correct when the line speeds up or slows down; on a fixed-speed conveyor a timer is enough. Get the delay wrong and you eject the part in front of, or behind, the defective one, which reads as a random escape and a random good-part loss at the same time.
The fail then has to be latched. A 24V output pulse that lasts only as long as the decision can be too short for a pneumatic stroke, and if the signal drops while the part is still in the reject window, the part is not ejected. Latching the fail until the part has cleared the reject point, then clearing it for the next cycle, is what makes the ejection dependable. The latch-and-clear window is sized to the actuator stroke and the part spacing, so like the tracking delay it is a per-cell measurement rather than a fixed number. Because the decision itself is fast, about 30 ms per part on the cap line, the inference stage is rarely the timing constraint; the mechanical reject and its tracking delay are. Treat the about-30-ms figure as the first link only, and measure the full trigger-to-ejection time on your own parts and conveyor.
How do fault and ready outputs keep the reject safe?
A reject should fire only when the unit is actually inspecting, and that is what the ready and fault outputs are for. A ready output tells the cell the unit is powered, triggered and running its model; a fault output flags a problem such as a lost trigger or a failed capture. Wired into the reject logic, these inhibit the actuator when the unit is not ready, so an unmonitored gap does not let uninspected parts pass as good, and a genuine fault stops the line or diverts to a manual check instead of shipping unjudged product.
That is the line between a reject that is merely wired and one that is safe. The fail output ejects bad parts; the ready and fault outputs make sure the cell knows when the inspection itself cannot be trusted. Adding an acknowledge handshake, where the PLC confirms it saw the fail before the unit clears it, closes the loop on a fast line so no reject is silently missed. For the metric that tells you how many defects still slip through even a correctly wired reject, and why a low false-negative rate is the number to hold a vendor to, see the false-negative rate guide.
This wiring note is a spoke of the pillar guide on AI visual inspection; to see the inspection tasks that feed a reject on a real line, browse the real applications.