Imaging hot forged parts when the part itself glows

A forging above roughly 500 °C emits its own light, and at pressing temperature that glow can outshine every LED on your panel. Lighting the part the usual way stops working: your exposure budget is spent rejecting light, not adding it. This guide is for engineers putting a camera on a forging line, quench line or press station. It gives the physics in one table, a worked field-of-view and exposure example you can rescale to your part, the blue-light trick for parts still hot enough to glow, and the point at which SWIR or thermal becomes the better camera.

Why a hot forging lights itself

Every object above absolute zero radiates, and the radiation shifts to shorter wavelengths as the object gets hotter. Two numbers set up the whole imaging problem.

The first is where glow becomes visible: at around 525 °C a metal surface emits enough visible light to be seen, first against a dark background (the thermal radiation article covers the underlying physics). Below that line the part is an ordinary lit subject. Above it, the part is the light source and your LED panel becomes a rounding error.

The second is where the emission peaks. Wien's displacement law gives:

At 1000 °C (1273 K) the peak sits near 2.3 μm, deep in the infrared and far past a silicon sensor. The visible glow is the steep short-wavelength tail of that peak, so it lives almost entirely in red. A tail that steep means a small temperature change moves brightness enormously. Planck's law at 650 nm gives the ratios a camera actually sees:

These are blackbody ratios. Oxidized steel emits less than a blackbody (emissivity typically 0.7–0.9 in the visible band), but close to the same fraction at every visible wavelength, so the ratios stand. The consequence for design: no single exposure serves a line where parts cool from 1250 °C toward 650 °C, because the signal itself falls by more than four orders of magnitude on the way down.

The rule runs backwards: exposure first, lights last

Sort the job into four temperature bands and the decisions make themselves:

• Below about 500 °C: no visible glow. Treat it as a cold part and light it normally.
• About 500–800 °C: dull to cherry-red glow. The glow is the illumination: expose in the millisecond range, add no light, and fit a red bandpass filter if the shop lights pollute the frame.
• Above about 800 °C: the glow saturates the sensor. Cut it: exposure down to tens of microseconds, iris stopped down, ND stacked on top if needed.
• When temperature itself is the measurement: move to SWIR or LWIR hardware, covered below.

Then work the three controls you have in order of range. Exposure spans the most (2 ms to 10 μs is 200×, and faster cameras go lower still). The iris adds 5–8 stops, which is 32–256×. ND sheets add as many decades as you stack. In series, they cover any ratio in the table.

Worked example: a 140 mm forging at 500 mm working distance

Task: find laps and surface cracks across a 140 mm square face. A 0.15 mm crack should span at least 3 pixels, so the target is 0.05 mm/px or better.

The 12 MP USB3 camera used in this example delivers 4000 × 3000 pixels on a 1/1.7-inch sensor. A 1/1.7-inch optical format works out to about 7.5 mm of active width (the convention treats 16 mm as one inch, so the diagonal is 16/1.7 ≈ 9.4 mm and a 4:3 sensor is roughly 7.5 mm across). Even if the true width is 7.7 mm, the numbers below shift by a few percent and the final pick still covers the part. Choose 500 mm working distance to keep the camera away from the radiated heat, and aim for a 180 mm field of view around the 140 mm face.

m = 7.5 / 180 = 0.0417, and f = 500 × 0.0417 / 1.0417 = 20 mm. Common focal lengths bracket 20 mm, so take the nearest lens that meets the resolution target. A 25 mm lens gives FOV = 7.5 × 500 / 25 = 150 mm, and 150 mm / 4000 px = 0.0375 mm/px — finer than the target, covering the face with 5 mm of margin per side. That is the pick: 25 mm at 500 mm working distance.

Check the pixel against line speed before you trust it: at 1 m/s, blur stays within one pixel only if exposure stays under 0.0375 mm ÷ 1 m/s = 37 μs.

Camera used in this example:

Setting exposure: one test shot, then scale

Anchor everything with a single test exposure at the coldest part you must inspect. Adjust exposure until the glow reads mid-gray and write the number down; every hotter part then scales by the table above.

Example chain. The anchor shot on an 800 °C part lands at 2 ms, f/2.8. Billets then arrive from the press at 1250 °C, radiating 12,700× more red light:

1. Drop exposure to the camera floor, say 10 μs: 2 ms / 10 μs = 200×.
2. Stop the iris from f/2.8 to f/16: five stops, 2^5 = 32×. Running total 6400×.
3. The remaining factor is 12,700 / 6400 ≈ 2×, so add an ND 0.3 sheet (a 2× filter). Total 12,800×, just past the requirement.

A fresh 1250 °C billet therefore shoots at about 10 μs, f/16 with ND 0.3 — derived, not guessed. Any ratio converts to a filter value the same way (optical density to aperture conversion). As that billet cools toward 800 °C, brightness falls about 450-fold (12,700 / 28), so set exposure per part: trigger on part-present, then scale. When you write those limits through the API, query the units first — exposure nodes are commonly in microseconds, and a 10 written as milliseconds is a blank frame (auto exposure limit units in microseconds).

Ambient light barely matters on this route: at 10 μs the sensor collects 1/200 of what the 2 ms anchor shot collected, so ceiling lights and their flicker hardly register (camera flicker under artificial lighting).

Killing the glow when you need reflected detail

Some features carry their information in reflected light, not emitted light: a die number stamped beside the flash, an engraved grade mark, an oil sheen. On parts below roughly 700 °C the glow is weak enough that one lit exposure carries both signals. On hotter parts, suppress the glow instead of fighting it, and the physics hands you a shortcut, because blackbody emission collapses toward short wavelengths.

At 1000 °C, Planck's law puts the part's own 450 nm emission about 2,300× below its 650 nm emission. A strobed blue light, by contrast, reflects off the surface at whatever drive current you feed it. So strobe a 450 nm blue bar light hard, put a narrow 450 nm bandpass on a mono camera, and the reflected signal beats the glow by that margin before the filter rejects the red flood. Expect the working window to run from about 500 °C, where the glow starts, to around 1000 °C, where only a strong strobe wins.

Two practical points: a light that switches too slowly wastes its flash at these exposure times (LED strobe rise time), and the filter logic is exactly the one used with red LEDs against ambient (bandpass filter with red LED light to block ambient light). Where you need only less glow, not reflected contrast, plain ND does the job (picking an ND value by stops).

When the heat is the signal: SWIR and thermal

Two jobs want a different camera entirely.

Temperature pattern as the inspection — scale coverage, hot spots, confirmation that a part left the press in range — is a thermal job. A 640 × 512 LWIR camera reading 8–14 μm over GigE at 50 Hz returns calibrated temperature per pixel; set emissivity to match the surface or every reading shifts, and trend relative differences when it varies. Standard glass is opaque in this band, so thermal cameras pair with their own infrared-transmitting lenses rather than ordinary machine vision optics. The thermal option, C-mount, 9–24 V DC:

Glow imaging with more signal is a SWIR job. At 1000 °C a blackbody radiates about 90× more at 1.1 μm than at 650 nm, and more still closer to its 2.3 μm peak, so a SWIR sensor collects the glow with headroom — useful through scale and oxide, and on the hottest parts. The SWIR option, 1.3 MP on the IMX990 sensor, 1/2-inch format, C-mount, fanless, GigE:

How the sensor families divide the spectrum, from silicon through SWIR to microbolometers, is laid out in camera sensor technologies by spectral range. In both bands, emissivity contrast is your defect contrast: scale, oxide and fresh fracture radiate with different efficiency, so laps and cracks modulate the image before any detection runs. When the analysis needs numbers rather than pictures, pull the frames out at full depth (export high bitdepth grayscale images from thermal cameras). More hardware along these lines sits in the infrared cameras range.

Camera, trigger and mount details

• Mono beats color. The glow scene is monochrome orange. A Bayer color sensor throws away sensitivity, and its IR-cut filter blocks the deep-red and near-IR band where emission is strongest — what that filter does and when to image past it is covered in color camera near infrared imaging: IR-cut filter options.
• Trigger per part, expose per part. A cooling line walks through a 12,700× brightness range; a fixed exposure serves exactly one temperature. Trigger on part-present and scale from the anchor shot.
• Keep the camera out of the radiation. A 140 mm billet at 1250 °C radiates about 6 kW in total (σT^4 = 5.67e-8 × 1523^4 ≈ 305 kW/m² for a blackbody, across 0.0196 m² of face). Most of that is infrared landing on whatever faces it, housing included. Keep 500 mm or more between a >1000 °C surface and the lens, fit a hood, and know what the camera's rating actually promises (camera operating temperature range: what happens outside it).
• Rolling shutter is usually fine here. At 10 μs each row samples essentially the same instant, so a die station works fine; on a fast conveyor prefer a global shutter model from the area scan cameras range.

For the visible-band route, a concrete starting point: 20 MP mono on USB3, IMX183 sensor, 1-inch format, C-mount:

When the usual answer is wrong

Below roughly 500 °C, forget the glow logic entirely: the part is a lit subject again. A diffused white bar such as the 372 × 28 mm LS-BAR-1276 (24 V) spreads even light across scale and fresh metal:

• Never add red light to a glowing part. Red LEDs pour photons into the band the part already floods. If you add light at all, add blue.
• Treat thermal-camera temperatures as functions of surface condition. Emissivity moves the reading; set it per surface state (scale, oxide, fresh metal) or trend relative differences.
• Do not carry exposure settings between stations. They hold at one temperature only; rescale by the ratios.

Cooled forgings that still need metal-on-metal contrast raise the older, familiar problems: metal part on metal background lighting handles the contrast fight, and partial bright field lighting explains why matte surfaces want directed light.

Checks and parts list before you order

• Part temperature at the imaging point — measure it; the whole design hangs on the band.
• Camera: mono, exposure floor at 10 μs or below, hardware trigger input — or SWIR/LWIR when temperature is the signal.
• Lens: focal length from the worked arithmetic (25 mm at 500 mm WD in the example), with a locking iris.
• Filters: narrow 450 nm bandpass for the blue-strobe route, red bandpass for 500–800 °C glow work, ND 0.3 and up for glow suppression.
• Light: none above about 800 °C, a strobe-capable blue bar for 500–1000 °C reflected work, a diffused white bar below about 500 °C; the bar lights range covers lengths and diffusers.
• Mount: 500 mm or more working distance, hood or radiant shield between billet and lens.
• Controller: strobe output with microsecond timing for the blue route.

Send the part size, the temperature at the imaging point and the line speed, and we will size the chain with you: the camera finder and light finder narrow the candidates, the configurator assembles the set, and a quote is the fast route for multi-station lines. The lighting topic page collects the neighboring techniques.