Machine vision trigger timing: sequence camera, light and reject gate

One part-present pulse can time an entire inspection station. From that single edge, the camera exposure opens at a computed delay, each strobe channel fires microseconds around it, and the reject gate waits at its own distance-derived delay with the part's pass or fail result already latched. That chain, the machine vision trigger timing sequence, is what makes production-line images repeatable part after part. We walk the chain end to end, give the delay formulas, work a 600 mm/s conveyor example with the arithmetic shown, and close with the decision for each device: sensor, camera and controller outputs.

One pulse, one clock

A free-running camera grabs frames whenever its internal clock says to, so part position drifts frame to frame and auto exposure lets brightness wander. A timed sequence replaces that clock with one external event: the pulse from a part-present sensor. Camera, lights and reject gate all count their delays from that same edge, in hardware. Sequencing light controllers pulse, delay and release outputs at the microsecond level, and a hardware camera trigger fires on its input pin, so the chain repeats far more tightly than a software loop that reacts only after an image arrives. Software still runs the inspection; it just never owns the timing.

The chain from part-present sensor to reject gate

The devices are the same on almost every production line: one photoelectric sensor as the part-present input, then timed outputs to the camera trigger, one strobe channel per light, and the reject gate.

1. The part breaks the sensor beam; the input card debounces the edge and fixes its polarity.
2. After the camera delay, the camera trigger output pulses and the camera exposes for its programmed time.
3. Around that instant, each light channel fires its strobe with its own delay and pulse width.
4. The vision system stores a pass or fail result against that trigger's part number.
5. After the gate delay, counted from the original sensor edge, the gate output fires only if the stored result is a fail.

Set the camera and lights into hardware trigger modes first, or a free-run default silently ignores the whole chain; the hardware trigger configuration per camera guide covers that step. In a one-camera, one-light cell a smart camera strobe output can drive the light directly and removes one device from the chain.

The four delays you set

Every delay in the chain comes from the same two quantities: distance from the sensor along the belt, and line speed v at the inspection point.

Here d_sensor-to-FOV is the distance from sensor beam to field-of-view centre, d_sensor-to-gate the distance to the gate, and p_pixel the pixel footprint on the part, equal to field-of-view width divided by horizontal pixel count. The exposure formula is the blur budget: exposure time times line speed is the smear length, and it must stay near one pixel or edges soften. The exposure time versus shutter speed guide for moving parts works the same budget in detail, and a short flash shrinks the effective exposure further, which is the mechanism behind how strobe pulse width freezes motion blur and why trigger pulse width sets your camera exposure time.

The light channel needs a small lead over the camera, not an exact match. LED driver stages settle in microseconds to a few tens of microseconds, so program the strobe 20-50 µs ahead and keep the exposure inside the flat top of the flash; the LED strobe rise time guide shows what that leading edge costs. When the sensor sits upstream of the field-of-view centre, the camera's own programmable trigger delay can absorb part of the offset instead of a controller channel, as the camera trigger delay timing guide for offset inspection sensors explains.

Worked example: a 600 mm/s conveyor

Start from the mechanical facts: belt at 600 mm/s, sensor 180 mm upstream of the field-of-view centre, sensor 500 mm upstream of the gate, field of view 96 mm across a 1920-pixel sensor, parts on a 100 mm pitch, flash flat within 40 µs, image transfer plus inspection 120 ms.

Then the arithmetic:

• Pixel footprint: 96 mm / 1920 px = 0.05 mm/px.
• Exposure cap: 0.05 mm / 600 mm/s = 0.0000833 s ≈ 83 µs. Set 80 µs; smear = 600 mm/s × 80 µs = 0.048 mm, about 0.96 px.
• Camera delay: 180 mm / 600 mm/s = 0.300 s = 300 ms.
• Light delay: 300 ms - 40 µs ≈ 299.96 ms, strobe width 200 µs, so the 80 µs exposure sits inside the flat top.
• Gate delay: 500 mm / 600 mm/s = 0.833 s = 833 ms.
• Decision ready: exposure closes at 300.08 ms, plus 120 ms inspection ≈ 420 ms, leaving 833 - 420 = 413 ms of slack before the gate.
• Line rate: 600 mm/s / 100 mm = 6 parts per second, one every 167 ms.

At 167 ms per part and 300 ms of travel, up to two parts sit between sensor and camera at once. Each sensor edge must start its own timed chain; a chain that cannot overlap drops or double-fires parts, which appears as sequenced exposure trigger timing overtrigger errors.

What a speed error does to every delay

Fixed delays are only as good as the speed they assume. Position error at the exposure equals sensor offset times fractional speed error. At the example's 180 mm offset, every 1% of speed error moves the part 180 mm × 0.01 = 1.8 mm, which is 36 px at 0.05 mm/px. A belt running 5% slow catches the part at 570 mm/s × 0.3 s = 171 mm instead of 180 mm: a 9 mm, 180 px miss that breaks any fixed region of interest.

Read the tolerance backwards from the pixels. If your locate routine must hold ±20 px, that is ±1.0 mm, so total speed variation must stay under 1.0 mm / 180 mm = 0.56%. Regulate the belt to half a percent, or abandon clock time and trigger off an encoder: with a 1 mm-pitch encoder, the camera fires 180 counts after the sensor and the gate fires 500 counts after, at any speed.

The reject gate follows the sensor, not the vision result

The gate must never wait for the inspection to finish, because its position depends on the part, not on the answer. Time the gate output at its fixed delay from the original sensor edge and store each part's result in order; when the gate pulse falls due, the controller releases it only for a stored fail. In the worked example the decision lands 413 ms before the gate, so store-and-release is comfortable. The same sensor-locked pattern runs a full colour-sorter circuit with vision, control and reject wired as one chain.

When processing is slower than the gate, buy distance: each extra 100 mm of belt between sensor and gate adds 100 mm / 600 mm/s = 167 ms of decision time at the example speed. Cutting the region of interest or resolution shortens processing too. Keep the gate delay itself fixed and mechanical.

Choosing the sensor, camera and controller

Part-present sensor

A diffuse-reflective sensor suits belt-fed parts whose height varies, because its trip point tunes to the part rather than the belt behind it; through-beam types give the sharpest edge detection when the trigger point must repeat within a millimetre. The trigger sensor finder filters by sensing range, output type and beam type. The MV-PE5502 is a diffuse-reflective photoelectric sensor with a 10-350 mm settable sensing range and an NPN output, which fits short stands over a belt.

Camera for a strobed moving part

Use a global shutter so every row integrates the same flash instant; rolling shutter rows would sample different parts of the pulse and band the image. The MV-CU020-90GC is a 2.5 MP GigE camera with a global shutter and a C-mount, a common pick for belt cells. The camera finder filters by shutter type, resolution and interface when your field of view needs more pixels.

Controller channels

Count channels before buying: one strobe channel per light, and a home for the camera trigger and reject outputs in the cell controller or PLC. The LS-CTL-0070 is a 4-channel digital control box rated 200 W with RS232, a strobe feature and constant-voltage 24 V outputs. A four-channel box covers a two-light strobe cell with capacity to spare.

For two larger strobes, the LS-CTL-0264 is a 2-channel high-power digital controller rated 400 W with a network option. Match output type to the light: constant-current channels drive current-driven lights, constant-voltage 24 V channels drive the rest, and every channel you plan to pulse needs the strobe feature. The light controllers category lists all channel counts and output types, and the controller configurator narrows the field by channels, power and control interface.

Set-up order, and proving the timing repeats

1. Put the camera and lights into hardware trigger modes and switch off every auto exposure and auto gain.
2. Wire the sensor to the sequencer input, and confirm one clean edge per part on an oscilloscope or input counter.
3. Load the four delays from the formulas, keeping the strobe lead at 20-50 µs.
4. First light: run the belt with the light on continuous, centre the field of view, then switch to strobe and re-balance exposure and strobe current.
5. Trigger 100 parts and overlay the first and last frames. In the worked example's scale, hold total part-position scatter within ±20 px, which is ±1.0 mm at 0.05 mm/px; more scatter points to belt ripple, sensor bounce or a missing debounce.
6. Mark one failing part, run 30 parts, and confirm that exact part lands in the reject bin.
7. Save the delay set into the recipe with the line speed written beside it, so changeover recomputes distance divided by speed instead of guessing.

If the camera still misfires, compare its configured minimum frame interval against the sensor pulse rate; the camera frame rate with external trigger pulses guide works that arithmetic.

When the fixed-delay answer is wrong

Four cases break the pattern. Speed that varies product to product or ramps at start-up: switch to encoder counts, as the tolerance arithmetic above dictates. Multi-station lines where one part crosses several cameras: give each station its own input and channel set sequenced independently; multi-channel light controller sequencing for photometric stereo pushes channel-by-channel timing further. Ambient light leaking into long exposures: strobe shorter and brighter, and handle the residue with the tactics in ambient light variation in pick and place vision. A vision decision that lands after the gate: move the gate downstream or cut processing time; never slow the gate to wait for software.

When the sequencer is a PC or PLC program rather than a dedicated box, structure the chain as a fixed order of timed steps driven by the input edge, not as scattered timers inside the vision code; the flat-sequence state-machine timing pattern shows that structure.

Start from your own numbers: measure the two distances, read the belt speed, compute the four delays, then size the chain in the configurator or send the distances and speed through the contact page and we will check the delay set with you. The controllers topic page gathers the rest of the timing guides.