Cameras for continuous steel casting monitoring: heat and distance

Your process department wants cameras watching a continuous caster, and wireless looks like the fastest route. It is not. An uncompressed camera stream needs far more bandwidth than plant Wi-Fi sustains between steel structures, and the camera end of any link sits in radiant heat, scale dust and interference from drives. This guide commits to a design — a wired GigE thermal camera near the strand, in a shielded, air-purged enclosure, with fiber carrying the images to the control room — and gives you the numbers behind it: bandwidth, cable reach, radiant load and pixels on target.

The decision: wired, thermal, shielded

Wire the cameras. A compressed viewing stream for operators can go wireless at 2–5 Mbit/s, but the full image path cannot: a 16-bit thermal stream runs at hundreds of Mbit/s, and GigE Vision camera selection — bandwidth, cables and software — is built for exactly that rate over copper and fiber.

Choose an LWIR thermal camera for the strand itself, because glowing steel saturates a silicon sensor. Put it 1–3 m away, behind a reflective visor with a purged window slot, and keep its body inside the −20 °C to +60 °C ambient band most industrial cameras are specified for. Back it up with a SWIR camera where you need surface detail, and an ordinary area scan camera further back for logistics views. Run copper up to 100 m and fiber beyond it, with the camera powered from a field cabinet. If you wire several stations, steel coil end face inspection with multi-camera imaging shows the same discipline on another steel line.

Why wireless loses on a caster

Do the bandwidth arithmetic first. A 640 × 512 thermal sensor streaming 16-bit data at 50 Hz produces:

That payload is a quarter of a GigE link — trivial on wire, hopeless on radio. Caster halls are steel cladding that reflects 2.4 and 5 GHz instead of passing it, with arc loads, large drives and welding gear lifting the noise floor. A nominal 300 Mbit/s access point in that environment sustains a small fraction of its rating, and every dropped packet on a vision stream is a torn or lost frame, not just slower video.

The decision: the image path is wired, end to end. Wireless earns one job — carrying a compressed sub-stream of a few Mbit/s to tablets and office screens for people who only want to look.

Which camera: LWIR for heat, SWIR for detail

The strand leaves the mould with a surface in the 800–1,100 °C range, and the steel in the mould is near 1,500 °C. Point a visible camera at that and you get a saturated white blob; heavy neutral-density filtering only makes the blob dimmer. Match the band to the job.

LWIR, at 8–14 μm, maps the temperature field directly: mould level, uneven secondary cooling, sticker and breakout hot spots. For the wide view, the LS-IRC-0044 prints 640 × 512 resolution, 8–14 μm, 50 Hz, GigE, a 88.5° × 73.2° field of view and 9–24 V DC power. For a tight crop of the mould from the same spot, the MV-CI003-GL-N25 is a 0.3 MP LWIR camera with GigE and a 25 mm EFL on its M34×0.75 mount. Sensitivity and lens trade-offs for exactly this kind of monitoring are worked through in thermal camera NETD and lens choice for process monitoring.

Compare the wide, tight-crop and compact thermal options:

All 3 share: Pixel size 17 μm. Values as printed in each product's data sheet; open a part for its drawing and full specifications.

SWIR, roughly 0.9–1.7 μm, is the band for surface detail on hot steel: scale pattern, tears, water streaks from the spray zones. The MV-CI013-GS-NN is a 1.3 MP SWIR camera with the IMX990 sensor, GigE and a fanless build, taking standard C-mount lenses.

Pixel size sets the detail you resolve. The LS-IRC-0029 prints 15 μm × 15 μm pixels behind a C-mount. Divide pixel size by focal length for the instantaneous field of view: 15 μm behind a 25 mm lens is 0.6 mrad, or about 1.2 mm of strand at 2 m. The band-by-band trade-offs are in SWIR vs NIR vs visible imaging, and what changes outside visible light is in SWIR lens and UV lens basics.

To size a lens and standoff before you buy, run the geometry in the calculator:

More infrared cameras are in the finder if neither band fits the view you need.

Standoff, shielding and cooling

Heat, not bandwidth, is what actually kills cameras on casters, and it arrives by four paths: radiation from the strand, hot ambient air, conduction along the mounting arm, and scale dust on the window.

Radiation is the big one. A surface at 1,000 °C (1,273 K) radiates σT⁴ ≈ 149 kW/m²; at an emissivity of 0.85 for oxidized steel, that is about 127 kW/m². Distance helps less than you hope — flux falls with the square of distance for a point source, and the strand is a broad plane. Reflection, not distance, does the work: a polished visor that lets the window see the strand only through a narrow slot throws back about 90 % of the load, and filtered purge air across the window strips the rest while keeping scale off.

Conduction is the sneaky one. Your mounting arm is a steel bar with one end near the strand and the other at the enclosure — calculating heat transfer distance in steel pipe shows how to estimate how far along the steel the heat actually reaches, so you can mount the enclosure past it, off the radiant side, with a thermal break.

Ambient: out of direct sight, the air near the strand still runs hot, so hold the enclosure under the camera's rating with a filtered fan or a vortex cooler. Expect focus to drift as the housing warms — infrared lens athermalization covers the lens designs that hold focus through the swing.

The link: 100 m of copper, kilometres of fiber

Copper Ethernet reaches 100 m per segment — an IEEE 802.3 physics limit, not a camera choice. Inside that radius, wire straight to the switch with industrial cable: the LS-CBL-0117 is a 15 m RJ45 camera data cable with locking studs, and the LS-CBL-0199 is a 3 m high-flex GigE cable with screw locks for drag-chain duty where the camera rides a moving carriage.

Past 100 m, put the media conversion in a cabinet beside the caster and pull fiber: about 2 km on multimode, 10 km and more on single-mode, which covers any plant layout. Camera Link systems reach the same distances with dedicated fiber extenders — frame grabber image acquisition for Camera Link and CoaXPress covers that route.

Fiber buys a second benefit: galvanic isolation. The caster structure and the control room sit at different ground potentials while the drives run, and copper between them becomes a conductor for fault currents. Power the camera from the field cabinet — the LS-CBL-0300 is a 7 m M8 8-pin power and I/O cable rated for drag-chain use — and send only light the long way. Connector, flex and shielding choices are covered in machine vision camera cable selection for reliable links, and the camera cable range lists the standard lengths.

For the cable side:

Worked example: LWIR camera at 2 m, control room 350 m away

Inputs:

• Strand surface: 1,000 °C, typical just below the mould; the mould itself holds steel near 1,500 °C.
• Camera: the wide LWIR model from the comparison above, streaming 16 bits per pixel at 50 Hz. The 16-bit depth is the one assumption; 14-bit output trims the rate by an eighth.
• Geometry: 2 m standoff, a 100 mm × 100 mm window, and a slot that limits the strand to about 40 % of the window's view (view factor ≈ 0.4).
• Route: 350 m of cable path to the control room.

Radiant load on the window:

A visor reflecting 90 % leaves roughly 50 W entering the enclosure — a small filtered fan or vortex cooler handles that with margin under the camera's rating. Decision points: at this standoff and shielding, air cooling is enough; above about 1 kW absorbed, or closer than 1 m, specify a water-cooled housing.

Pixels on target: the covered width at 2 m is 2 × 2.0 m × tan(88.5° / 2) ≈ 4.0 × 0.974 ≈ 3.9 m. Across 640 pixels that is 3.9 / 640 ≈ 6 mm per pixel — mould level and breakout hot spots stand out; a 3 mm defect does not. A lens four times longer narrows the view to about 0.8 m and about 1.2 mm per pixel at the same distance, so make that the mould camera and keep the wide one for context.

Bandwidth and route: the payload is 640 × 512 × 2 bytes × 50 Hz ≈ 33 MB/s ≈ 262 Mbit/s, comfortable on one GigE link with headroom for control traffic. At 350 m the copper budget is long gone, so the decision is a fiber pair from the field cabinet — 350 m sits far inside the 2 km multimode budget.

Storage: 33 MB/s written continuously is 33 × 86,400 ≈ 2.8 TB per day per camera. Record a 1 Hz radiometric trend (0.66 MB per frame ≈ 57 GB per day) plus full-rate clips around alarms, not 24 hours of full-rate video.

Recording without drowning in data

Record on the wired side, close to where the bandwidth lands, and push only alarms and short clips upstream. Long distance camera recording for triggered infrared capture covers the trigger loop that turns full-rate recording on around an event. Keep the data at full depth while you do: exporting high-bitdepth grayscale images from thermal cameras shows how 16-bit radiometric frames survive software that wants 8-bit. When process engineers ask for degrees instead of counts, start from raw thermal counts to temperature calibration — each camera carries its own calibration curve.

Checks before ordering

• Window material. Ordinary glass goes opaque past about 3 μm, so an LWIR camera behind glass sees a wall. Order germanium, zinc selenide or chalcogenide for 8–14 μm; fused quartz passes SWIR.
• Purge air. Dry, filtered plant air flowing outward at the window. Wet air fogs the window; unfiltered air plates it with scale in days.
• Temperature logging. Put a sensor in the enclosure and alarm at 50 °C, leaving margin under the 60 °C rating most cameras carry.
• Local power. Feed the camera from the field cabinet — 9–24 V DC covers the LS-IRC-0044 — never across the long cable.
• Spare fiber. Pull one spare pair with the working pair and label both ends; caster ducts are not fun to re-pull.
• A spare camera on the shelf. Caster downtime costs dwarf a camera's cost, and thermal models are often built to order.
• EMC routing. Keep signal fiber away from drive cables and cross power runs at right angles.

When the usual answer is wrong

• Operators only want a look. Then a compressed sub-stream over Wi-Fi is the right answer; still buy the camera for analysis, and let the wired recorder serve the viewers.
• The camera must sit closer than about 1 m, or in full direct sight. Air cooling is out; use a water-cooled furnace-camera housing with a replaceable window tip.
• You only need events, not a live stream. Record locally at the camera and upload clips on trigger; store-and-forward tolerates a wireless hop that continuous streaming cannot.
• Private LTE or 60 GHz links are already on site. They carry compressed multi-camera views to a control room well; uncompressed radiometric streams still belong on fiber.
• You expect internal defects, not surface conditions. Cameras cannot see into the product; how normal and angle probes detect steel casting flaws is the ultrasonic side of the quality question.

Next step: filter the camera finder by GigE and LWIR or SWIR, or open the configurator at the camera step and work toward cables. Bring the standoff, ambient temperature and cable-route length to the contact page and we will size the link and the cooling with you. More process-monitoring set-ups sit under the applications topic.