Trigger a camera from a PLC output to log rare failures

Every so often the process makes a bad part when nobody is standing there, and the shift note never mentions it. The machine's PLC already knows the moment it happens — it drives an output for exactly that condition. Fire an industrial camera's opto-isolated input from that output and every event leaves a timestamped image behind, ready to review when the shift ends. This guide walks the wiring, the trigger settings, the camera choice and the storage math, so a line chasing a 0.1% rejection rate collects its own evidence without anyone babysitting it.

The decision: camera I/O, not a shutter-button relay

Do not de-solder a shutter button and wire a relay across it. A consumer camera powers itself off between shots, wants a half-press before the full press, hides its images on a memory card and stamps nothing with a time. An industrial area scan camera is built for this job: its opto-isolated input accepts the 24 V logic your panel already uses, the camera stays powered and armed for weeks, and each trigger returns one complete frame to a PC with a time stamp, captured while the failure is still on the table.

This is the same hardware trigger a photoeye inspection station uses; the only change is that the PLC's failure bit fires it instead of a part-present sensor. The sections below wire the loop, set the trigger, pick the camera and size the evidence locker.

What you need

• A camera with at least one opto-isolated input. Filter the camera finder by interface and mount; the input lives on the camera's power and I/O connector.
• A power and I/O cable with bare leads at the PLC end. One multi-core lead usually carries the supply and the trigger together.
• A spare discrete output. A 24 V transistor output is ideal; a dry relay contact also closes the loop, with a bounce caveat covered in the FAQ.
• A data path to a PC. GigE runs up to 100 m to a switch; USB3 keeps its runs to a few meters.
• Power for the camera. PoE from the switch, or a DIN-rail supply matched to the voltage the camera's I/O pinout lists — the power supplies page filters enclosed supplies.
• Acquisition software that saves on trigger, plus disk space sized by the worked example below.

Wire the trigger loop

The input is not a logic pin. It is an LED behind an isolator, and the PLC output closes a small current loop through it. Three wiring decisions decide whether the first pulse works:

1. Direction. Work out which side sources and which side sinks from the pinout before landing wires — the opto-isolated GPIO direction walkthrough shows both topologies.
2. Loop current. An industrial 24 V output drives the loop directly with the series resistance the pinout calls for; a 5 V controller output drives the same loop with a smaller resistor, and the how to trigger a camera from a 5 V controller output walkthrough works that arithmetic.
3. Common reference. Tie the PLC output's common to the camera input's reference, or the loop never conducts. The classic symptom is a camera that ignores every pulse; the trigger common reference guide diagnoses it step by step.

One cable handles the field wiring. The LS-CBL-0300 carries power and trigger together in a single 7 m, 8-core M8 lead with open ends for the terminals.

For the data run on a fixed mount, the LS-CBL-0199 is a 3 m GigE camera cable with screw locks and a high-flex, drag-chain-rated jacket.

Set the trigger mode and the pulse

Configure the capture once; the events then run themselves. Using the standard GenICam feature names in your acquisition app:

1. Turn the trigger on and point the trigger source at the input line you wired. If a forced pulse produces nothing, the GenICam line settings checklist resolves the usual mismatches.
2. Set the exposure mode. Timed gives a fixed exposure you choose; TriggerWidth lets the PLC pulse itself gate the exposure — the trigger pulse width guide explains when each wins.
3. Fix exposure, gain and focus once for the worst-lit failure state and leave them there. A documentation camera must not hunt mid-event.
4. Force the output from the PLC for a full second and confirm exactly one frame arrives.
5. Save every frame under a timestamp filename and log the trigger count alongside the PLC's event list.

Pulse length, in numbers. Inputs of this kind register pulses a few microseconds wide, so a 1 ms pulse carries a 100-to-1,000-fold margin (1 ms = 1,000 µs). Hold it longer than one PLC scan — a small controller scans in roughly 1-10 ms — and 10 ms is a comfortable default. Frame rate sets the only hard ceiling: each capture needs its exposure plus readout before the next trigger is accepted, the arithmetic in maximum frame rate on a hardware trigger, so a chattering sensor yields tens of captures per second, never thousands.

Choose the camera for the scene

If the part keeps moving when the pulse fires, use a global shutter: a rolling sensor scans line by line and shears fast motion into a skewed image that hides the defect. If the machine stops or the scene barely moves, rolling shutter carries no penalty and resolution buys documentation detail.

The MV-CU020-90GC is a 2.5 MP color GigE camera with a global shutter, a 1/1.7" sensor and a C-mount.

For a stopped scene, 12 megapixels lets a reviewer read a date code or count thread turns in the evidence image. The LS-CMR-0166 records 4000 × 3000 frames at 30 fps over USB3 with a 12 MP rolling-shutter sensor.

Freeze the moment with light

A failure in a dim corner of the cell needs light for exactly one exposure. Wire the camera's strobe output to the light's strobe input: the line's timing against exposure and timestamp is covered in the camera strobe output timing guide. An LED pulsed during a 1-5 ms exposure freezes residual motion and marks the event in the image itself. Size the light's duty cycle for a low event rate and it will run for years; the LED strobe duty cycle guide shows the matching math.

Worked example: a day of evidence at 60 parts/min and a 0.1% failure rate

Inputs, and where each comes from:

• 60 parts/min — an illustrative mid-rate for a small monitored cell; scale the result linearly to your own rate.
• 0.1% failure rate — the rejection target being watched.
• 8 h shift — the end-of-shift review cadence.
• 4000 × 3000 pixels at 8-bit mono — the 12 MP camera's frame from the section above, saved lossless; color triples the bytes.

The rule, in words: events per shift equals parts per shift times the failure fraction, and log size equals events times bytes per frame.

Arithmetic:

• Parts per shift: 60 parts/min × 60 min/h × 8 h = 28,800 parts.
• Events: 28,800 × 0.001 = 28.8, so plan for 29 images.
• Log size: 29 images × 12 MB = 348 MB ≈ 350 MB per shift.

Decision, with the threshold. At the target rate a shift's evidence is about 350 MB, so a 16 GB buffer holds 16,000 MB ÷ 350 MB ≈ 45 shifts — roughly nine working weeks. The picture flips if the process degrades: at 1%, 288 images × 12 MB ≈ 3.5 GB per shift, and the same buffer covers about 4.5 shifts. Keep lossless saves while a shift's log stays under about 2 GB; past that, switch to JPEG (about a tenth of the bytes) or cap retention at two weeks and alert when the disk passes 80% full.

Get the images off and review them

The camera stays armed; a small acquisition loop on the PC waits for each trigger, grabs the frame and writes it out — the Python hardware-trigger grab loop is about thirty lines and runs unattended. Put the wall-clock time and a sequence number in each filename, keep the PC's clock synced to the plant's time source, and each image pairs with the PLC's own event rows by timestamp. At review time, sort by hour: failures cluster, and a cluster at the same time every day points at a shift change, a warm motor or a raw-material batch faster than any statistics package.

Mistakes that empty the evidence locker

• Wiring into the wrong pin. Strobe outputs, power pins and trigger inputs share one connector; a swapped wire captures nothing. Settle direction from the pinout before energizing.
• Relay chatter counted as three events. A bouncing contact fires multiple captures for one event; if the count matters, drive the loop from a transistor output.
• Rolling shutter on a moving part. The image shears and the defect hides in the skew — that is the global-shutter case above.
• Exposure too long for residual motion. Even a stopped line vibrates; shorten the exposure and let the strobe carry the brightness.
• Autofocus or auto-exposure left on. The one frame that matters gets hunted mid-capture; lock everything once.
• A disk that fills silently. 350 MB per shift sounds endless until the failure rate runs to 1%; set the retention cap and the 80% alert from the worked example.

Before you order

• Confirm the output card: voltage, transistor or contact, and one spare channel.
• Match the cable to the camera's connector family and core count before choosing lengths; the power and I/O cables page filters by both.
• Decide raw or JPEG now — it sets storage and the retention cap.
• Trial the chain end to end: force one event, confirm one timestamped image on the PC, then walk away and let it collect.

If the wiring or the trigger settings stall, the camera configurator sizes the whole chain, and support is free with every order. More trigger and sync patterns live on the systems topic page.