Machine Vision Lighting for EV Battery Manufacturing: A Technical Guide

Which lighting technique reveals what your camera needs to see? This guide breaks down six proven approaches — from high-speed electrode web inspection to 3D weld profiling — with the illumination requirements for each.

Why Lighting Determines Whether Vision Systems Succeed or Fail

A machine vision system can only analyze what the light reveals. Cameras capture reflected light, and the illumination — not the sensor, not the algorithm — sets the ceiling on what the system can detect. If a defect doesn’t produce contrast at the sensor, more resolution or better AI cannot recover it. Lighting technique is what creates that contrast.

Battery manufacturing makes this unusually hard. Mirror-finish copper and aluminum foils throw specular glare that saturates the image, while near-black cathode coatings absorb nearly everything — and the two often appear in the same frame. Defects are sub-millimeter at full line speed: electrode webs run at meters per second, and cell assembly runs continuously, requiring strobe-grade intensity and microsecond timing. Every cell needs a traceable inspection record from raw foil to finished pack. And dry-room and cleanroom constraints add further limits on how the lighting can be built and cooled.

Poor lighting cannot be compensated for by better algorithms or AI. Different inspection tasks require fundamentally different lighting approaches — and choosing the right one starts with understanding what you need to see.

Choosing the Right Technique

Line Lighting: Electrode and Coating Inspection at Web Speed

Electrode inspection is where battery quality begins — and where the speed problem is most severe. Line-scan cameras image the coated electrode web continuously as it runs, catching pinholes, streaks, agglomerates, uncoated zones, and edge defects in real time, before defective material is wound into a cell.

The system has to handle dark cathode coatings that absorb most of the incident light, mirror-finish copper and aluminum foil on the uncoated margins, sub-millimeter defects at meters-per-second web speed, and coating-edge geometry that has to hold tight tolerance.

Lighting requirements: A line-scan camera captures a single row of pixels at very high rates, so the illumination must concentrate extreme intensity into a narrow, perfectly uniform strip across the full web width. Exposure times shrink as web speed rises, and dark coatings return only a small fraction of the incident light — so intensity is the gating factor. Uniformity along the line matters just as much: any brightness variation across the web shows up as a false gradient in every scanned line.

SVL Products for This Application

Dark-Field: Finding Particles Before They Become Failures

A metal particle inside a cell can cause a latent short circuit — a defect that may not appear until the pack is in the field. Foreign particle detection is therefore one of the highest-stakes inspections in cell assembly, and it is fundamentally a lighting problem: the particles are small, and the electrode surfaces they land on are nearly black.

Lighting requirements: In a dark-field configuration, light grazes the surface at a low angle, so flat surfaces reflect it away from the camera and appear dark. Only raised or recessed features — particles, scratches, pinholes, debris — scatter light up into the lens, appearing bright against a dark background. The geometry does the work: contrast comes from the angle of illumination, not from the brightness of the defect itself.

SVL Products for This Application

Diffuse and Coaxial Illumination: Reflective Welds Without Glare

A modern pack contains hundreds of welds — tab welds, busbar joints, can seams, terminal connections — and each one is critical to cell integrity. The inspection problem: these are specular surfaces with specular defects. Under standard directional lighting, the weld glares, the image saturates, and cracks, porosity, and spatter disappear into the bloom.

Lighting requirements: The solution is to remove directionality. Diffuse dome illumination surrounds the part with soft, even light from every angle — the machine vision equivalent of a cloudy day — so mirror-finish surfaces image flat and defects produce contrast instead of hotspots. Coaxial illumination sends softened light down the camera’s own optical axis, ideal for flat reflective surfaces like foils and busbars. The same glare-free imaging that reveals weld defects also makes codes readable on dark, curved, or laser-marked cell surfaces — the foundation of cell-level traceability.

SVL Products for This Application

Backlighting: Separator Overhang and Dimensional Gauging

Some of the most consequential checks in cell assembly are edge measurements. The separator must overhang the electrode on every layer of every cell — and once the cell is sealed, that feature is no longer accessible to inspection. Stack height, layer count, winding concentricity, cell height, and terminal position are all dimensional checks that must happen inline, without contact, at takt time.

Lighting requirements: Backlighting places the light source behind the part, so the camera sees a silhouette. Edges render with maximum contrast and no interference from surface texture, color, or reflectivity — the measurement is pure geometry. For fast-moving stacking and winding operations, the backlight must strobe at high intensity so each frame freezes the edge cleanly.

SVL Products for This Application

SWIR: Past the Visible Band

Some battery inspection problems produce no visible contrast at all. Shortwave infrared imaging, from 1,050 to 1,650 nm, captures how materials interact with light at wavelengths where the physics of absorption are fundamentally different.

Water and electrolyte absorb strongly in SWIR, so wetting, fill verification, and moisture contamination are clearly visible where visible light shows little contrast. Materials that look identical in visible color can separate sharply in SWIR, making coating-chemistry differences, contamination, and residue visible. And some packaging and separator materials become translucent in SWIR, allowing inspection of features hidden in visible light.

What’s changed: SWIR sensor costs have dropped past the old 1-micrometer barrier, and matched SWIR LEDs are now available at peak wavelengths of 1,050, 1,200, 1,450, 1,550, and 1,650 nm in standard machine vision form factors, with the same strobe and continuous capabilities as their visible-spectrum counterparts. But the camera is only half of the system. Without matched SWIR illumination, the sensor sits in the dark — the lighting finishes the job.

SWIR applications vary widely — different wavelengths target different materials, working distances depend on line geometry, and environments range from electrode inspection to fill verification. Smart Vision Lights offers SWIR-enabled configurations across multiple product families, including the L300G2 linear light platform, and works directly with customers to specify the right solution.

3D Structured Light: Measuring Geometry Directly

A projected pattern lets the camera measure height directly — surface topography is measured rather than inferred from brightness. On the battery line, that capability closes the gaps 2D imaging leaves open.

Weld bead profiling quantifies height, width, undercut, and missing material on every weld, regardless of surface reflectivity — turning a subjective visual check into a dimensional measurement. Coating and surface flatness inspection catches thickness variation, lumps, and depressions on the electrode that 2D imaging cannot distinguish from shading. Stack height and cell dimensional checks complete the picture.

3D works alongside 2D rather than replacing it: 2D locates candidate defects quickly, and 3D confirms the geometry.

SVL Products for This Application

Where This Is All Heading

New cell formats are changing the inspection task itself. Large-format cylindrical cells, dry-coated electrodes, and cell-to-pack construction each change what has to be inspected and how fast. Dry electrode coating removes the solvent line but adds new surface textures to characterize. Larger cells mean fewer cells per pack — and less tolerance for a defect in any single one. Line speeds continue to increase. In each case, the lighting approach has to be reworked before the camera or algorithm can be selected.

Meanwhile, SWIR and 3D are moving from specialty add-ons to standard stations on new battery lines. AI keeps improving at the decision — but it still cannot see through glare, or recover contrast that was never captured. Cell volumes and line speeds are not slowing down, and neither is the cost of a missed defect.

The choice of technique — line, dark-field, coaxial, backlight, SWIR, 3D — will continue to determine whether a defect is caught. The lighting remains the foundation that decides whether a camera captures useful data or noise, and its role in battery manufacturing will only become more central.