Lithium-ion battery electrode web inspection with line scan lighting

Battery electrode webs punish a vision system twice over. The graphite and ceramic coatings swallow most of the light you send, the copper or aluminium foil underneath mirrors the rest, and the defects that matter — particles, pinholes, coating voids, slitting burrs — measure tens to hundreds of microns on a web moving a metre per second. The arithmetic below sizes an inspection station for a coater, calender or slitter: pixels per defect, line rate, exposure and duty cycle, TDI gain, light length and lens magnification. Everything is calculated for a 600 mm web at 1 m/s; swap in your width and speed and the same formulas give your parts list.

What pixel size catches a 300 µm defect on an electrode web?

Set the pixel scale at a third of the smallest defect you must flag. A 300 µm particle, pinhole or coating void needs 100 µm per pixel or finer, because below about three pixels across the mark blends into coating texture and there is nothing for the classifier to measure.

Three pixels is a detection floor, not a classification standard. Sorting a pit from a dirt particle from a drying crack wants five to ten pixels across, and measuring a defect's length wants more still. Pick the smallest defect from your own defect catalogue, weighted by consequence. Pinholes and embedded metal particles are the ones that resurface later as cell failures, long after the coater that let them through, as the root causes of lithium-ion battery degradation make clear. For wide dimensional gauges the rule relaxes: edge-fitting algorithms measure coating width sub-pixel, so the pixel rule guards defects and small features, not width.

How many pixels does a 600 mm web need?

Divide web width by pixel scale: 600 mm ÷ 0.1 mm = 6000 pixels. An 8K camera covers the full web with about 27% of its line still in reserve for web wander, mounting tolerance and uncoated edge margins.

The reserve is not padding. Webs meander a few millimetres and slitters drift; a camera sized to exactly 6000 pixels clips the edge the day the coating runs wide. An 8K line with 7 µm pixels is 8192 × 7 µm = 57.3 mm at the sensor, which sets the minimum lens image circle at 57 mm. Splitting the web across two 4K cameras also works, at the cost of a second capture channel and a seam to calibrate — the field of view calculation for line scan cameras shows the cross-check.

Two cameras fit the full-width station:

The MV-CL081-41CM is specified exactly for these optics: 8K, 7 µm pixels, a 57 mm image circle and an M72 mount. The LS-LCM-0099 is the 10GigE variant: 8K mono with an M72 mount. Shortlist the rest of the line scan cameras by line length, pixel size and interface; the camera finder filters the same way.

What line rate does a 1 m/s web force on the camera?

Line rate is web speed divided by pixel scale, so the example web demands about 13.7 kHz. The interface must carry roughly 0.9 Gbit/s of 8-bit pixel data at that rate, every line, for the whole shift.

1,000,000 µm/s ÷ 73 µm = 13,700 Hz ≈ 13.7 kHz, which makes the line period 1 ÷ 13.7 kHz = 73 µs. Data follows: 8192 pixels × 13,700 lines/s × 1 byte = 112 MB/s, about 0.90 Gbit/s on the wire. A GigE link carries roughly 950 Mbit/s of usable payload, so mono 8-bit squeezes in with almost no headroom; 10GigE or Camera Link carries it comfortably and leaves room for 10-bit output later. Camera Link buyers should settle the grabber pairing early — the Camera Link frame grabber compatibility guide maps camera families to grabber configurations.

Gate the acquisition from the machine encoder rather than an internal clock: that keeps pixels square through accel and decel. The calculator below turns your own speed and pixel scale into the line rate to configure; setting the line rate of a line scan camera shows where the number lives in the camera, and making line rate follow machine speed at runtime covers the encoder side.

Why does a graphite coating push you to TDI?

Graphite and ceramic coatings absorb most incident light, and the 73 µs available before one pixel of smear returns a weak signal. Time delay integration (TDI) adds signal instead of time, summing N sensor lines over the same point for N times the charge and a √N gain in signal-to-noise.

A 4-stage TDI camera therefore collects four times the charge and roughly twice the shot-noise-limited SNR of the same sensor without TDI — equivalent to 4× the light without an extra LED. The price is motion discipline. TDI integrates along the direction of travel, so speed variation across the stages smears the image, and encoder sync stops being optional. Where no TDI camera fits the budget, 2 × 2 pixel binning trades cross-web resolution for light — a trade to make consciously, because the pixel rule above fixes that resolution.

The TDI camera to grow into:

The MV-CL166-91F2C pairs 16K resolution with 5 µm pixels on RGB/4TDI lines, an 82 mm image circle and an F-mount. On that sensor the same 600 mm web resolves 600 mm ÷ 16,384 = 36.6 µm per pixel: a 300 µm defect spans 8 pixels, and the line rate at 1 m/s rises to 1,000,000 µm/s ÷ 36.6 µm ≈ 27.3 kHz — a pinhole-hunting geometry if the capture budget allows.

How much light does a 73 µs line exposure need?

At 13.7 kHz the line period and the smear-limited exposure are both 73 µs, so the duty cycle is exactly 100 percent. The light runs continuously at the irradiance the histogram demands, and heat rather than strobing becomes the design constraint.

That 100% figure is the dividing line in light selection. Below it, a strobed controller overdrives an LED row briefly for extra peak light; at 100% there is nothing to overdrive, so you need a light rated for continuous output with active cooling. Aim the camera at a histogram filled to roughly half or two-thirds of the 8-bit range at the working aperture: enough SNR in the coating, with headroom so a specular flash off the foil does not clip. The smear arithmetic sits in exposure time versus shutter speed for moving parts, constant-current stability in what lighting controllers fix for line scan, and the irradiance budget in how much power a fast line really needs.

Order light longer than the web; the line scan lights category sorts by length:

The LS-LSC-0513 is a 750 mm fan-cooled line light on 48 V. The LS-LSC-0415 is a 750 mm dual-angle body on the same 48 V supply, so one housing carries both a straight row and an oblique row. Fan cooling is not a nicety at 100% duty; it is what keeps the output stable across a shift.

Which lighting geometry separates foil glare from coating defects?

Run two lighting passes rather than one compromise. Diffuse bright-field light directly over the camera reads contrast inside the dark coating, while oblique dark-field light skims the bare foil so scratches, dents and slitting burrs glow bright against a black background.

The physics is reflection geometry. Smooth foil mirrors: bright-field illumination bounces straight back into the lens and whites out the image, while a dark-field angle sends the specular reflection away from the camera so only roughness scatters light back in. Graphite does the opposite, scattering diffusely, so bright field reads coating texture, bubbles and missing coat cleanly. That is why slitting stations lean on the oblique pass — the fresh-cut edge is where burrs, chips and powder loss live. Orient dual-angle rows across the web so the oblique sweep crosses both slit edges.

For the dark-field row:

The LS-LSC-0441 is a 493 × 60 mm oblique line light with natural cooling on 24 V, sized for the strip and edge passes at the slitter.

Which lens magnification matches the pixel scale?

Magnification is sensor pixel over object pixel: 7 µm ÷ 73 µm = 0.0955. At a 900 mm working distance the lens equation gives f = 900 mm × 0.0955 ÷ 1.0955 ≈ 78 mm, so an 80 mm-class large-format lens with a 57 mm-plus image circle fits the camera's mount.

Check depth of field before fixing the aperture. At f/4 with c = 7 µm (one sensor pixel): 2 × 4 × 0.007 mm × 1.0955 ÷ 0.00912 ≈ 6.7 mm total, about ±3.4 mm of web flutter; at f/5.6 the same arithmetic gives ±4.7 mm. Diffraction sets the other wall: the Airy diameter is 2.44 × 0.55 µm × N, so f/4 gives 5.4 µm, inside the 7 µm pixel, while f/8 gives 10.7 µm and visibly soft edges. Settle at f/4 to f/5.6 and buy depth of field from geometry, not aperture.

Select line scan lenses from printed magnification rather than focal length alone. They are designed at fixed magnification points, so pick one within roughly 10% of 0.0955 with an image circle over 57.3 mm and trim the working distance at the mount. The lens selection guide for 4K, 8K and 16K line scan cameras walks the image-circle and mount checks, and the line scan lenses category filters by both.

Large-format line scan lenses for the two stations:

The LS-LSL-0023 is built for 16K 5 µm sensors: 116 mm focal length, 60-90 mm image circle, 127.2 mm working distance, V-mount. If the TDI camera is F-mount, filter the lens finder for F-mount large-format lenses in the same image-circle class — that 16K sensor is 16,384 × 5 µm ≈ 82 mm wide. The LS-LSL-0241 prints 146 mm and 0.75x in M42. At 0.75x a 7 µm sensor resolves 7 µm ÷ 0.75 = 9.3 µm per pixel across a 57.3 mm ÷ 0.75 ≈ 76 mm field — the slitter's fine-strip station, where 1 m/s demands 1,000,000 µm/s ÷ 9.3 µm ≈ 107 kHz and about 7 Gbit/s of data, a job for a 10GigE camera or a slower strip.

What do you check before ordering the parts?

Four numbers close the design before any purchase order: encoder resolution against pixel scale, working distance against machine clearance, light length against web width plus wander, and the camera line-rate ceiling against your fastest speed plus a quarter for acceleration transients.

• Encoder: at least one clean quadrature tick per pixel; the exact scale is a software ratio, and setting encoder ticks per pixel walks the count.
• Clearance: the 900 mm working distance must exist physically beside the machine, plus the light's standoff; changing a lens is cheaper than moving a frame.
• Light length: web width plus the machine's documented edge wander, so end-of-line falloff never enters the field.
• Line rate: fastest web speed ÷ pixel scale × 1.25 against the camera's rated maximum, so a speed ramp never outruns the acquisition.
• Calibration: verify scale and squareness with a moving checkerboard target per the line scan calibration guide before trusting any width number.

When is one bright-field 8K camera the wrong answer?

Three layouts break the single-camera plan. Slitted strips narrower than about 80 mm want a fine camera per strip, separator film wants transmitted light because pinholes are holes rather than surface marks, and webs that stop and start run better on an area camera driving its line-scan mode.

After slitting, a strip is tens of millimetres wide and the defects shrink with it, so magnification moves to the 0.75x station and the full-web 8K no longer applies. Separator base film is a translucent polymer: backlight it and a pinhole flares bright against a dark field, a contrast reflection cannot give; once opaque foil sits under a full coating, backlighting stops working. On winders that index stop-start, an area camera run in line scan mode for moving-web inspection avoids TDI's motion-matching demands altogether, at a speed cost the machine usually accepts.

Set the two inputs of the worked example to your own line — web width and web speed — and every other number here follows: pixels, line rate, exposure, duty, magnification. Shortlist hardware in the camera finder or start the system configurator with those two numbers; more build guides sit on the applications topic page.