Every inspection you specify today stands on forty years of lighting development. The first industrial vision systems of the late 1970s ran on illumination borrowed from other applications, and their results showed it. The lights, sensors and controllers in a modern station exist because of the specific limits those early systems hit. Machine vision lighting moved from borrowed lamps to high-power, multi-wavelength LEDs, and faster sensors turned lighting into something you can sequence: several strobed sources captured frame by frame through one camera. By the end you will know when to commit to a multi-light sequence, how to size the camera rate it needs, and which controller features hold the sequence together.
The decision this history leads to
Forty years of development ends in one purchasing decision. When a part needs more than one lighting geometry or wavelength to show its whole defect list, sequence several strobed lights through one camera instead of buying more cameras. A multi-channel digital controller fires light 1 during frame 1, light 2 during frame 2, light 3 during frame 3, and the software grades each frame for the defects that light reveals.
You pay in camera speed: an area-scan camera must deliver one frame per light per part, and a line-scan camera must run at the number of lights times the line rate each image needs. Commit to the sequence when the camera has that rate in hand and one light genuinely cannot cover the defect list. Stay with one light when it can, and the history below explains why that choice even exists.
Borrowed lamps: how vision lighting began
Machine vision is a product of the late 1970s and 1980s, and automotive was first through the door. One large car maker already used cameras to confirm that components were assembled correctly, years before a complete vision system could be bought as a product. A system was an illumination device, a camera, a lens and some digital image analysis, with parts sourced from whatever other applications offered. The illumination, in particular, was borrowed: incandescent and fluorescent fixtures built to light a workspace, not to inspect a part.
Those lamps set hard limits. Most of their input power left as heat, so enclosing a lamp with a part warmed everything nearby. Bulb life was short, so brightness drifted as the source aged. A single wavelength meant filtering a broadband lamp and discarding most of its output. And the source ran continuously: it could not be gated per image, so the camera shutter alone froze motion. The same lamp-versus-LED trade-offs still show up wherever CFL and LED machine lighting are compared, and the LED wins on efficiency and life every time.
Through the 1980s the technology spread from automotive into electronics, pharmaceutical and packaging plants, and vision grew into a billion-dollar industry on constant technical change. Lighting was the component holding it back.
What the LED changed
The LED was the breakthrough the industry needed: lighting with much higher power and wider wavelengths, in a source small enough to build into any geometry. Almost every machine-vision light sold today is an LED, because an LED is compact, efficient and long-lived, and because wavelength became a manufacturing choice rather than a filter loss. Across today's bar lights and ring lights the range runs from 365 nm in the ultraviolet to 940 nm in the infrared, with white, red, green and blue alongside.
Power came first. High-output LEDs overdriven for microseconds deliver far more light during a strobe than a continuous source of the same size, freezing motion without any shutter at all, and strobe overdrive, duty cycle and timing covers how far you can push that. Wavelength followed: a 365 nm source reveals fluorescence that white light never will, and narrow-band red or infrared cuts through glare and some plastics. Which LED colour gives the best contrast and monochromatic lighting in machine vision turn that into a selection method.
Geometry came free with the package. Because the source is small, one LED family builds ring lights, bar lights and domes, so a bright-field view and a dark-field view of the same part are one lighting budget rather than two systems. Lighting geometry for image contrast maps the camera, light and part positions behind each view.
Strobing also changed maintenance. An LED pulsed for microseconds per frame runs cool, and why strobe mode makes lights last longer quantifies the lifetime gain, while matching lights and controllers on duty cycle keeps the drive inside the light's rating.
Sensors and processing caught up
Faster sensors finished what the LED started. New camera technology delivered smaller bodies with higher resolution and higher frame and line rates, which moved vision onto faster lines, and made multi-light inspection possible, because successive frames or lines could now be captured from different light sources while one part stayed in view. Processing followed: processor speed rose as cost fell, and image-processing software grew a versatile toolbox of measurement tools, so grading three images per part became as affordable as grading one. The light, the sensor and the processor each removed a different limit, and multi-light capture needed all three.
How multi-light capture works
The controller drives the lights in sequence and the camera records the result: images from each light are acquired on consecutive frames or lines with a single camera, then separated in software. Two patterns cover nearly every installation.
Area scan, one frame per light. The controller fires light 1 inside frame 1's exposure, light 2 inside frame 2, and so on, while the part holds position or indexes one station between frames. Sequencing several colours through one camera this way is standard practice for colour-versus-geometry checks. The budget is frames per second:
The 1.2 factor leaves twenty percent for trigger jitter and readout gaps.
Line scan, one line group per light. The lights alternate line by line, bright field, dark field, UV, repeat, and the software de-interleaves three full images at one third of the camera's line rate. The budget is line rate:
Line period shrinks as rate rises, so each strobe pulse must finish inside the exposure window of its own line, and why microseconds of light delay blur your image sets the LED rise-time budget. The sequence itself lives in the controller, not in host software, so the light order and its timing stay fixed; stabilizing lighting changes at the vision trigger covers the trigger chain that starts it.
Worked example: three lights, one line scan camera, 500 mm/s
Inputs for a mid-speed surface-inspection line, each an example value you would replace with your own:
- Web speed: 500 mm/s
- Resolution required along travel, in each lit image: 0.1 mm per line
- Lights: 3, a bright-field bar, a low-angle dark-field bar, and a 365 nm UV bar
Rule: lines per second per image equals web speed divided by line spacing, and the camera must deliver that rate times the number of lights.
Arithmetic:
- Lines per second per image = 500 mm/s ÷ 0.1 mm/line = 5,000 lines/s
- Camera line rate = 3 × 5,000 lines/s = 15,000 lines/s = 15 kHz
- Line period = 1 ÷ 15,000 s ≈ 66.7 µs, the window each strobe pulse must fit inside
- Each light fires one line in three: 5,000 pulses per second, about 33 percent duty while the web runs
Decision: commit to the three-light sequence when the camera's maximum line rate reaches 18 kHz, the 15 kHz requirement plus the same twenty percent margin. At a 10 kHz ceiling, drop to two lights at full resolution and move the UV check downstream, or keep three lights and accept 0.15 mm per line per image (500 ÷ (10,000 ÷ 3) ≈ 0.15 mm), which passes only if that still resolves your smallest defect.
Choosing the lights and the controller for multi-light work
Lights follow the defect list. Ring lights give even illumination around the lens axis for surfaces, codes and turned parts; bar lights line up with line-scan sensors and long fields. Wavelength follows material, and the machine vision lighting color wheel picks the LED that suppresses the background you do not want to see.
Two building blocks from the current range. The LS-RNG-0100 is a 120 mm ring light at a 60° angle, 620 nm red, 24 V, an axis-mounted even source for the area-scan version of a station. The LS-BAR-1047 is a 290 x 30 mm bar light, 24 V, sized for a line-scan field of similar width.
- Size
- 100-200 mm
- Type
- Standard
- Angle
- 60°
- Voltage
- 24 V
- Size
- 200-400 mm
- Type
- Standard
- Voltage
- 24 V
The controller decides whether sequencing is easy. You need one channel per independently strobed light, a stored sequence the controller steps through from a single trigger input, and an interface to program it. Current control matters as much: constant-current channels hold LED output steady as the sources warm, which keeps every frame in the sequence matched in brightness. Two four-channel digital units show the split. The LS-CTL-0180 is a constant-current controller, 4 channels, 120 W; the LS-CTL-0070 is a digital control box, 4 channels, 200 W, RS232, constant-voltage 24 V output with strobe support.
| Specification | LS-CTL-0180 | LS-CTL-0070 |
|---|---|---|
| Baud rate | 9600-115200bps optional | 9600 bps |
| Control | Digital (4-digit display, buttons), RS232 | Manual encoder knob adjustment, PC remote adjustment |
| Working mode | 0: constant off, 1: constant on, 2: strobe, 3: trigger delay shutdown | Constant on / Trigger / Strobe |
| Dimensions (mm) | 89 x 147 x 175 | W 100 x H 156 x D 134 (from the drawing; body 149 high, 110 deep) |
| Price | $359.00 | $549.00 |
All 2 share: Number of channels 4. Values as printed in each product's data sheet; open a part for its drawing and full specifications.
Choose constant current when brightness match across frames decides pass or fail; choose constant voltage when the lights carry their own regulation and you need switched strobe power. Stations beyond four lights move to eight-channel digital controllers in the same light controllers category, so filter the controller finder by channel count and output type before ordering.
When one light is still the right answer
Multi-light is a tool, not a default. Keep one light when a single geometry and wavelength already show every defect: you keep the full frame rate, one exposure to tune and the simplest trigger chain. Keep one light when the camera cannot pay the rate, since a 60 kHz camera spends only 20 kHz per image on three lights, and some lines cannot spare that. Keep strobing regardless: strobed lighting with ambient rejection suppresses factory light far better than continuous operation at the same camera settings.
The deliberate exception is multi-light science: photometric stereo with a sequenced controller fires three or more directions per frame set and computes surface shape from the shading, giving a height-like map that no single image contains. That is a defect class one light can never show, and it is the modern endpoint of the same history.
Checks before you commit to a multi-light sequence
- Camera rate: maximum frame or line rate at least 1.2 × lights × the per-image rate you sized. The margin keeps trigger jitter out of the sequence.
- Channels: one controller channel per independently strobed light, with current and pulse width set per channel.
- Pulse window: each strobe pulse, including LED rise time, fits inside its own frame or line exposure.
- Duty cycle: each light's duty stays inside its rating, and overdrive current only within the duty limit the light itself specifies.
- Stored sequence: the controller, not the host, times the light order from one trigger input.
- Fixed camera settings: every frame in the sequence uses the same exposure and gain, so turn every auto mode off before trusting the sequence.
- Safety: UV and infrared sources shrouded or interlocked, because they look dark at full output.
When the defect list and the rate budget agree, start at the controller step of the system configurator or browse the lighting finder to match channels and sources to the sequence you have sized, and bring the sequence timing to the contact page if a station needs review.
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
When did LED lighting take over machine vision?
The first industrial systems of the late 1970s and 1980s ran on incandescent and fluorescent lamps borrowed from other applications. LEDs displaced them once high-power, multi-wavelength devices arrived, because an LED is compact, efficient, long-lived and can be strobed in microseconds, while a lamp can do none of that. Almost all modern machine-vision lights now use LEDs, so the takeover is effectively complete.
How many lights can one camera sequence?
As many as you have controller channels and camera speed for. Four-channel digital controllers are the common choice and eight-channel units are available; the real ceiling is the rate budget, because each added light costs one more frame per part or one more multiple on line rate. Most stations settle on two or three lights, which covers a bright-field, dark-field and wavelength set.
Do I need a separate controller channel for each light?
Yes, whenever the lights must strobe independently: one channel per light gives each its own current, pulse width and place in the stored sequence. Lights that always fire together can share a channel if their combined current stays inside that channel's rating, so add up the light currents against the channel and total output figures before paralleling anything.
Why did early vision systems struggle with lighting?
Their lamps came from other applications: fixtures built to light a workplace, not to inspect a part. Most input power became heat, brightness drifted as bulbs aged, a single wavelength meant filtering away most of the output, and the source could not be gated per image. Early teams spent as much effort taming the lamp as writing the image analysis.
Does strobing really make LED lights last longer?
Yes, and the measure is L70: the operating hours at which output falls to 70 percent of its initial value. Strobing cuts junction heating and lowers duty cycle, which extends L70 well beyond continuous running at the same current. Overdriven pulses trade some of that back, so keep overdrive current inside the duty limit the light itself specifies.
Is multi-light capture the same as photometric stereo?
No, though both sequence lights through one camera. Multi-light capture switches geometry or wavelength to expose different defect classes, and each image is graded on its own. Photometric stereo fires three or more directions per frame set and computes surface shape from the shading, producing a height-like map no single image contains. Same controller technique, different output.