Steel billet surface inspection methods: vision or NDT?

If you run quality on a billet mill or a continuous casting line, the question usually arrives as an acronym fight: ECT, MLF, MPI or a camera on the 160 × 160 mm face. The right split is by defect depth and by the record you must keep. This comparison states what each method actually finds, then works the vision side in full numbers — field of view, focal length, working distance, line rate, exposure and depth of field — so you can commit per face, per grade and per line speed.

Which methods can inspect the surface of a 160 × 160 mm billet?

Four methods do real work on a 160 × 160 mm billet face: eddy current testing (ECT) and magnetic leakage flux (MLF, generally MFL) sense cracks electrically, magnetic particle inspection (MPI) shows them with particles, and a line-scan camera images every visible defect face by face.

The two jobs differ. ECT, MLF and MPI read a crack's electrical or magnetic signature, so they respond to discontinuities the eye cannot grade and, within limits, to ones it cannot see. The camera turns each face into an image, measures and locates every open-to-view defect, and files the evidence. Most billet surface specs name defects that show after descaling — cracks, laps, seams, rolled-in scale, pits, roller marks — and that list is exactly what optical imaging covers at line speed. Keep the probes for depth sensitivity and the camera for coverage and records.

What can eddy current, flux leakage and MPI actually find on steel?

ECT and MLF respond to cracks electrically: ECT needs its probe riding close to the surface because the signal decays steeply with lift-off, MLF needs the steel magnetized toward saturation, and MPI is a particle test that belongs offline, not on the roller table.

Eddy current testing induces circulating currents in the billet and reads the disturbance a crack makes to them. It works on any conductive steel, senses cracks open to the surface and slightly beneath, and is standard on bar and rod lines with rotating or encircling probes. On a square billet the geometry is the catch: corners and oxide scale move the probe's stand-off, so the carrier must follow the section rather than ride a fixed radius.

Magnetic leakage flux (MFL) magnetizes the ferromagnetic section toward saturation and reads the flux leaking out at a discontinuity. It forgives scale more readily than ECT's stand-off, but each sensor head covers a strip, so a 160 mm face needs an array spanning the full width, and the magnetizer adds hardware to the frame.

Magnetic particle inspection is the offline one: magnetize the part, apply dry powder or wet fluorescent particles under UV light, and particles build up along crack indications, including some slightly subsurface ones. Cleaning, magnetizing and demagnetizing per piece make it an audit tool, not an inline one.

When does a line-scan camera beat the probe on a billet line?

A line-scan camera wins when the deliverable is a full image of every face: a 4K camera covers one 160 mm face in a single pass at 41.5 μm per pixel, archives the picture against the billet ID, and reads the corners as well as the flats.

What the probe senses, the camera shows. Vision holds no lift-off, no coupling and no magnetization: the billet passes a frame with one line scan camera per face, the line rate follows the rollers, and inspection costs no cycle time. Every metre is stored against an encoder position, so a defect becomes a millimetre coordinate on an identified billet — the record a grinding crew and a customer claim both need. Multi-camera frames on steel sections are routine; the steel coil end face multi-camera layout shows the pattern.

The limit is physics: the camera sees only what light can reach. A crack that has not opened to the surface, porosity under the descaled skin, a seam closed over scale — none of it lands in the image, and that is exactly why NDT keeps its place on the same line. Contrast is the other discipline: scale, water and oil all compete with the defect, so lighting, the next question, decides more than pixel count.

Which camera and lens cover one 160 mm face in a single pass?

Size the field of view at 170 mm — the 160 mm face plus 5 mm of margin each side — and a 4K camera with 7 μm pixels resolves 41.5 μm per pixel, which takes a 50 mm lens working at about 346 mm.

Worked example: one face, one pass

1. Sensor line: 4096 px × 7 μm = 28.7 mm.
2. Field of view: 160 mm face + 2 × 5 mm margin = 170 mm.
3. Magnification: m = 28.7 / 170 = 0.169.
4. Focal length at a 300 mm working distance: f = 300 × 0.169 / 1.169 = 43 mm, so take the next standard focal length, 50 mm.
5. Working distance with the 50 mm lens: WD = 50 × 1.169 / 0.169 = 346 mm. Check: m = 50 / (346 − 50) = 0.169, so FOV = 28.7 / 0.169 = 170 mm.
6. Resolution: 170 mm / 4096 px = 41.5 μm per pixel, so a 0.1 mm crack spans 2.4 pixels — enough to detect and grade it.

The camera that carries the arithmetic is the MV-CL042-91GC-V2: 4K resolution, 7 μm pixels, a 29 mm image circle and an M42 mount for line-scan lenses. The 29 mm circle covers the 28.7 mm sensor line with margin, and line scan lens selection for 4K, 8K and 16K explains why the lens must cover the full line length, not just the focal length. Colour helps grade oxide stains; mono variants in the same resolution class give the strongest crack contrast.

Line rate and exposure

Line rate = travel speed / pixel footprint. At a 100 mm/s roller speed: 100 / 0.0415 mm per pixel = 2409 Hz, so the camera shoots about 2.4 kHz with a 0.42 ms exposure — comfortable territory for LED line lights and a GigE link alike.

Bandwidth at 4K: 4096 px × 2400 lines/s × 24 bit ≈ 236 Mbit/s for colour at 2.4 kHz, well inside GigE. At 24 kHz, mono 8-bit needs about 790 Mbit/s and still fits, while colour needs 2.36 Gbit/s and does not — faster billets go mono, or move to 10GigE and Camera Link. The line rate set-up guide walks the camera-side settings, and an 8K mono camera on 10GigE such as the LS-LCM-0099 buys back headroom at those speeds.

Depth of field

With one pixel as the circle of confusion, DOF ≈ 2 × N × c × (1 + m) / m². At f/8: 2 × 8 × 0.0415 × 1.169 / 0.0286 = 27 mm total, so focus mid-face and everything within about ±13 mm stays sharp — generous for a cast face. Stopping to f/5.6 tightens it to about ±9.5 mm.

Which light shows cracks on cold steel, and what changes when the billet is hot?

On cold steel, cracks read best in dark field: an oblique line light grazing the face at roughly 45-60° turns an open crack into a bright streak on a dark background, while flat, clean steel stays dark.

Dark field works because grazing light reflects off the flat face and away from the camera; only edges that stand up — crack lips, laps, seams, scale spikes — scatter light back. The background lands near black, small cracks carry large contrast, and a modest LED does the job. Line lights come in lengths from under 50 mm to beyond 400 mm, so a 300 mm-class unit covers the 170 mm field with mounting margin.

Bright field flips the image: a straight line light near the camera axis makes pits, rolled-in scale and marks go dark on a bright face — the right mode when the defect list is surface texture rather than standing lips. The bright-field-to-backlight illumination guide maps modes to defect types.

Both lights below fit a 160 mm face — the oblique unit for dark field, the straight unit for bright field.

What changes when the billet is red-hot?

At casting discharge the billet runs around 900-1,000 °C and glows on its own: by Wien's law its emission peaks near 2.3 μm, so LED lighting adds almost nothing and you expose for the glow, with all hardware mounted clear of the radiant zone.

Self-glow imaging is the hot forged parts playbook: short exposures freeze the moving face, descaling water and flaking scale cost contrast, and the imaging band matters — SWIR versus visible changes what scale hides and shows. Inspect cold, after the cooling bed, and every LED number above holds unchanged.

A thermal camera does not find cracks but earns its place beside them: it reads face temperature, flags short cuts, and logs the heat pattern per billet for process control. The infrared cameras category holds the LWIR choices; the LS-IRC-0044 reads 640 × 512 frames at 50 Hz over GigE, sensitive across 8-14 μm and powered from 9-24 V DC.

How do the methods compare side by side?

Read the table by the row your acceptance spec forces on you: if the spec demands subsurface sensitivity only ECT, MLF or MPI qualify, if it demands images and full-face coverage at line speed the line-scan row wins, and thermal rows serve temperature and cut length, never crack detection.

No single row covers a plant. Critical grades run vision for full-face mapping plus ECT or MFL probes for subsurface assurance, with MPI as the periodic audit that re-checks sampled indications — a combination that costs less than one missed customer claim.

When is the camera the wrong tool?

The camera is the wrong tool when the defect of record sits below the surface: tight subsurface cracks, porosity under the descaled skin, or a customer spec written around ECT or MFL acceptance criteria — no lighting trick will make those visible.

MPI keeps a separate, useful job: on a bench it confirms, piece by piece, that the inline methods are calling real defects rather than scale shadows. Internal soundness stays with ultrasonic testing entirely — no camera and no surface probe replaces it. Where the grade is critical, run both stacks: the camera grades and locates every visible indication for the grinding plan, the probe carries the subsurface go/no-go. Better lighting never turns a camera into an eddy current instrument; depth sensitivity is physics, not pixels.

What do you check before ordering?

Before ordering, tie every number in this comparison to your own line: the smallest defect in the spec sets the pixel size, the fastest roller speed sets the line rate, and the billet temperature where the camera sits decides the whole mounting plan.

• Smallest defect named in the spec: size the pixel at 2-3 across it. A 0.1 mm crack needs 41.5 μm pixels or finer; tighter defects push the face to 8K.
• Fastest roller speed: line rate = speed / pixel footprint — 24 kHz at 1 m/s here. GigE caps 4K colour near 9 kHz and 4K mono 8-bit near 29 kHz.
• Face coverage: field of view = face + 10 mm total margin, and the lens image circle must cover the sensor line — 29 mm versus 28.7 mm in the worked example.
• Light mode per defect: oblique dark field at a 45-60° grazing angle for cracks, laps and seams; straight bright field for pits, scale and marks.
• Temperature at the camera: hot means self-glow exposure and heat shielding; cold means LED length sized to the 170 mm field.
• Sync: encoder ticks per pixel, one line per increment, defect positions in millimetres from the billet head — set with the encoder ticks per pixel guide.
• Records: one image per billet with its ID and defect coordinates; the archive is part of the delivered product.

Build the frame from the line scan camera finder and the line light finder, or start at the camera configurator. Send the billet section, defect list and line speed through the quote desk and the four-face layout comes back sized to your line.