A weld seam is the one feature on the part that has to be right every metre, and it is often inspected while the part is still moving: on a tube mill, behind a welding carriage, or on the post-weld conveyor. This application guide sizes a line scan weld seam inspection from four numbers — travel speed, pixel size on the part, line rate and exposure. You will match the line rate to the weld travel, choose between a 2K and a 4K camera, fix the lens magnification and working distance, pick dark-field or bright-field line lighting for the defect class, and clock every line from an encoder so the image stays square to the part. The worked example follows a 250 mm/s weld carriage, and the same arithmetic sets any speed from a bench pass to a tube mill.
Why choose a line scan camera for weld seam inspection?
Choose a line scan camera when the seam travels continuously past the optics — tube-mill welding, carriage-welded long joints, post-weld conveyor inspection — because it builds one blur-free image of the whole seam at whatever speed the encoder reports. A line sensor exposes one row at a time and stitches rows into a record of unlimited length, so a 60 m weld becomes a single scrollable image instead of thousands of overlapping frames.
Motion blur is speed times exposure. A 1 ms area-scan frame at 250 mm/s smears the bead 0.25 mm, about 9 pixels at the 0.029 mm scale this build lands on; a 100 µs line exposure smears 0.025 mm, under one pixel. That difference is the whole argument at production speed.
The decision flips when the part stops. If the seam is short, or travel stays below roughly 50 mm/s with a dwell at the end, a resting area-scan snapshot needs simpler optics, lighting and software; the area scan camera line scan mode for moving-web inspection method covers the slow in-between case with hardware you may already own. For anything faster or longer, start from the line scan cameras category or the camera finder.
How do you match the line rate to the weld travel speed?
Set the line rate to travel speed divided by pixel size on the part: a weld carriage at 250 mm/s with 0.05 mm pixels needs 250 ÷ 0.05 = 5,000 lines per second, so one acquired line equals one pixel step of travel.
Run the rate low and every line covers more travel than the pixel width, so the seam stretches along its length; run it high and the image compresses. Hold the mismatch inside 10% and defect tools stay content — a 7% error turns a round pore slightly oval without hiding it. The line rate also caps the exposure, which section 6 budgets.
| Seam and process | Typical travel | Line rate at 0.05 mm pixels |
|---|---|---|
| Post-weld bench pass | 50 mm/s | 1,000 Hz |
| Laser-hybrid seam at 5 m/min | 83 mm/s | 1,660 Hz |
| Tube mill or post-weld conveyor | 250 mm/s (15 m/min) | 5,000 Hz |
Those speeds are typical for the process named, not limits: put your own carriage or mill speed into the formula. The camera-side settings are walked through in how to set the line rate of a line scan camera, and the calculator below runs the arithmetic on your numbers.
How many pixels do you need across the weld bead?
Give the smallest defect you must catch at least 4 pixels: a 0.5 mm undercut notch then demands 0.05 mm pixels or finer, and a 2,048-pixel sensor meets that over a 60 mm field with 0.029 mm pixels — 17 pixels across the same notch.
Undercut, porosity and spatter on a single-pass bead run 0.5-2 mm across, so the 4-pixel detection rule puts pixel size at 0.05-0.1 mm, and measurement work wants 0.03 mm or finer. Field of view is the other half of the choice: a bead 6-12 mm wide, plus heat-affected zones and ±5 mm of tracking wander, lands near 60 mm.
A 2,048-pixel sensor over that 60 mm resolves 0.029 mm. Joints that must include both plate edges push the field to about 100 mm, where a 2K sensor coarsens to 0.049 mm — 10 pixels per notch, detection intact but measurement margin gone. A 4K sensor holds 0.024 mm over the same width. The line scan camera field of view guide shows the geometry end to end.
Both example cameras are GigE with 7 µm pixels and M42 mounts, so one lens family serves either. The LS-LCM-0016 is a 2K GigE line scan camera with 7 µm pixels, a 58 kHz line rate and an M42 mount. The MV-CL042-91GC-V2 is a 4K colour GigE line scan camera with 7 µm pixels and an M42 mount, the pick when the defect is a tint change rather than geometry.
- Lens mount
- M42
- Interface
- GigE
- Lens mount
- M42
- Interface
- GigE
With the optics settled below, this build's pixel works out at 0.029 mm, so the 250 mm/s carriage actually needs 250 ÷ 0.0292 ≈ 8,600 Hz. At that pixel size the 58 kHz ceiling corresponds to about 1.7 m/s of travel, so the line rate is never the binding limit in this design.
Which lens magnification and working distance cover a 60 mm seam view?
Magnification is sensor width divided by field of view — 14.34 mm ÷ 60 mm ≈ 0.24x — and a 25 mm lens at 0.24x focuses at 25 × 1.24 ÷ 0.24 ≈ 129 mm object distance, low enough for a compact bracket on the carriage.
Two cautions on that 129 mm: it is measured to the lens principal plane, so front-of-lens clearance is slightly less, and depth of field is thin at 0.24x. The estimate uses the 7 µm pixel as the circle of confusion, so treat it as a planning number. A bead crown sitting 1-3 mm above the plate drifts out of focus at f/4, so stop to f/8, focus at mid-crown height and plan around ±1.2 mm of usable depth.
Fixed-magnification line scan lenses turn 0.24x into an order code rather than a tuning exercise; matching lens resolution to the sensor is covered in line scan lens selection for 4K, 8K and 16K cameras. The LS-LSL-0215 is a 25 mm, 0.24x, M42-mount line scan lens with a 30-45 mm image circle, and the mount matches both example cameras.
- Sensor format
- APS-H
- Mount
- M42
- Image circle
- 30-45 mm
- Focal length
- 25-50 mm
Dark field or bright field: which light shows the weld defect?
Light the seam in dark field — a line light on each side, angled low across the plate — and undercut lips, overlap, spatter and porosity pits read as bright edges on dark metal; switch to bright field only for colour and finish defects such as HAZ tint.
Dark field works because the illumination axis sits far from the imaging axis: the light skims at about 60° from vertical, flat plate reflects it away and stays dark, and any raised or sunken edge throws light back into the lens and reads bright. Bright field floods the scene from above and renders reflectance and colour instead: HAZ tint, mill scale, oil, coating skips.
Weld geometry lives in dark field; surface condition lives in bright field. Band choice for tint work is covered in which LED colour gives the best contrast in machine vision. Both light families sit in the line scan lights category, and the light finder filters them by length, colour and voltage.
Size each light to span the field plus margin and to sit clear of the bead crown. The LS-LSC-0046 is a double angle oblique light 500 x 53 x 121.5 mm that runs on 24 V — the dark-field pick. The LS-LSC-0074 is a 484 mm high-uniformity, fan-cooled line scan light on 24 V — the bright-field pick for tint and finish work.
- Size
- 400 mm +
- Voltage
- 24 V
- Size
- 400 mm +
- Voltage
- 24 V
How much light does a 100 µs line exposure need?
At 8,600 Hz the line period is 116 µs, so the exposure budget is about 100 µs — and at 86% duty the line light runs essentially continuously, which means you buy a genuinely bright fan-cooled light rather than relying on strobe overdrive.
Brightness needs scale with the inverse of exposure: halve the exposure and the light must double its intensity on the part to hold the same signal level. Dark field compounds the problem because most of the emitted light skims past the lens. At 100 µs on against a 116 µs period the duty cycle is 86%, so the LED stack sits in its continuous thermal regime and burst overdrive is unavailable — the honest fix is a fan-cooled, high-brightness light sized with the method in line scan light: how much power a fast line really needs.
A controller still earns its slot for gating, sync and brightness trim; what a lighting controller fixes on a line scan camera lists those jobs. Gate the exposure to the light and fit a bandpass filter matched to its wavelength: nearby arc flashes and shop lamps are broadband, and they will wash out exactly the low-angle contrast this design depends on. The stray light diagnosis guide covers the usual culprits and fixes.
How do you mount and trigger the camera on the weld machine?
Mount the optics square to the travel direction, clock every line from a rotary encoder instead of the camera's free-running timer, and size the encoder so at least one count arrives per pixel step of travel.
Square means the scan line sits at 90° to travel; a few degrees of skew shears the image and tilts every measurement made across the bead. Keep the front glass outside the spatter cone, run a sacrificial replaceable window at the working distance, and keep the bracket stiff — motion during a 100 µs exposure only freezes while the image moves less than a pixel.
Clock lines from a quadrature encoder on a measuring roller. The build needs 1 ÷ 0.0292 ≈ 34.3 lines per mm. A 2,500 P/R encoder decoded at ×4 delivers 10,000 counts per revolution; on a 50 mm roller (157 mm circumference) that is 63.7 counts per mm, and with the camera's trigger divider set to 2 the camera fires 31.8 lines per mm — a 7% stretch along travel.
Swap the roller for a 47 mm one (147.7 mm circumference) and the same parts deliver 33.9 lines per mm, within about 1% of the pixel grid. Then calibrate each axis separately: the moving checkerboard calibration guide shows the per-axis scale check, and wiring quadrature encoder inputs covers the electrical side.
A mill that holds speed within a couple of percent can free-run and accept distortion equal to the speed error. When the weld head is a robot and results must reach the cell controller, the integration steps in a robot weld-inspection vision setup guide cover that side of the install.
What proves the system out before production?
Prove the system on the machine, not the bench: run welded, greasy and ground coupons at full travel speed and confirm that focus, contrast and trigger hold — vibration, material hue changes and grease or spatter on the window break more weld inspections than optics choice does.
- Run good, undercut and porosity coupons at full travel speed: the camera must separate a notch from bead ripples at production rate, not bench rate.
- Add greasy and ground coupons: oil and rework change the surface and flip dark-field contrast, so re-reference the grey levels whenever mill finish changes.
- Command a speed ramp — low, nominal, 20% overspeed — and confirm the trigger divider keeps the image square at each step.
- Check the sacrificial window after each shift: a spatter film shows up as gradual contrast loss, not a sudden fault.
- Measure contrast margin on the critical defect: the undercut lip should sit well clear of the noise spread in the image histogram.
Three decisions fall out. A part that stops, or travel under about 50 mm/s: skip line scan and take an area snapshot. Mechanized travel up to about 100 mm/s: the 2K mono camera with 0.24x optics over 60 mm, an oblique dark-field line light, encoder-clocked at speed ÷ 0.029 and roughly 100 µs exposures. Tube-mill or conveyor duty from 250 mm/s: the same optics at 8,600 Hz and up, with the fan-cooled light, bandpass filter and sacrificial window doing the survivability work.
Size your own build in the configurator below, starting at the camera step, or send a joint drawing and travel profile through the contact page and we will check the numbers with you.
The configurator matches a camera, lens, light, light controller and cables to each other, step by step, and lists only parts that fit what you already picked.
Questions
Can a 2D line scan camera also measure bead height and penetration?
Width, undercut length and defect position all come from the calibrated image in millimetres per pixel. Height and penetration are depth, which a single 2D view infers only from shading. When you need real bead height, add a second view or move to a 3D laser line profiler; triangulation returns a height profile of the crown directly, and undercut depth becomes a measurement instead of a judgement.
What happens to the image if the carriage speeds up mid-seam?
With encoder triggering, nothing: every line still covers one pixel step of travel, so the image stays geometrically true at any speed. In free-run mode the seam stretches or compresses by the speed error — 10% overspeed makes the image 10% too long along the travel axis. That is why weld inspection should be clocked from a shaft or wheel encoder, not an internal timer.
Do I need a colour line scan camera for weld inspection?
Only when the defect itself is a colour change: HAZ tint on stainless, temper colours, coating faults or burn-through discolouration. Geometry defects — undercut, overlap, porosity, spatter — show in monochrome under dark-field lighting, and a mono sensor puts its whole quantum efficiency into one channel instead of splitting it across three. Colour triples the data rate and usually costs sensitivity, so choose it deliberately.
How do I protect the optics from weld spatter and heat?
Run the lens at its working distance behind a sacrificial, replaceable window in a quick-change holder, keep that window outside the spatter cone, and add an air wipe if spatter lands on it every shift. Put window condition on the maintenance schedule: a film of spatter or grease dims the image gradually and reads like a lighting fault, so teams chase the wrong problem.
Is GigE bandwidth enough for a 2K line scan camera at these line rates?
Yes, easily. A 2K line at 8 bits and 8,600 Hz moves 2048 × 8,600 ≈ 17.6 MB/s, well inside GigE's roughly 118 MB/s usable rate. Bandwidth starts to matter at 4K and 16K lengths or 10-bit depth; that is where 10GigE, Camera Link and XoFLink interfaces earn their place.




