· 3 min read
AI Vision for Battery Cell Inspection: Quality at Every Step, from Electrode to Pack
A single defective cell can stop a battery pack, trigger a recall or, in the worst case, start a fire. Battery manufacturers need to inspect every electrode, every weld and every seal, at line speed and on materials that are hard for cameras: reflective foils, dark coatings, transparent separators.
Logirobotix brings AI vision to battery lines. Our models find the defects that rule-based systems miss, adapt to new cell formats without reprogramming, and record every decision for full traceability.
Why battery inspection is different
- Safety is at stake. Misaligned electrodes, burrs and particles can cause internal short circuits and thermal runaway.
- Defects are small and materials are difficult. Metal foils reflect light, coatings are dark and uniform, separators are thin and transparent. Fixed thresholds struggle with this variability.
- Volumes are high and margins are thin. Inspection must keep up with production without generating false rejects that waste good cells.
- Traceability is mandatory. Automotive customers expect every cell to be linked to its process data, from electrode batch to final pack.
Cell formats we inspect
- Pouch cells: seal integrity, wrinkles and folds, tab position, pouch surface damage.
- Prismatic cells: can and lid welds, surface dents and scratches, venting features, terminal position.
- Cylindrical cells: crimp and cap quality, can surface defects, insulation rings, end-face inspection.
Inspection along the production line
| Step | What we inspect | Why it matters |
|---|---|---|
| 1. Electrode coating and calendering | Defects: pinholes, scratches, agglomerates, uncoated areas, coating edge defects, foreign particles. | Coating defects reduce capacity and can become starting points for internal shorts. |
| 2. Slitting and notching | Defects: burrs, edge cracks, tab shape and dimension errors, delamination at the cut edge. | A burr that pierces the separator is one of the main causes of internal short circuits. |
| 3. Stacking and winding | Checks: alignment and overhang of anode, cathode and separator; layer count; folds or wrinkles in the separator. | Misalignment between electrodes is a direct safety risk. Vision can also guide pick-and-place robots for faster, more accurate stacking. |
| 4. Tab welding | Checks: weld position and size, burns, spatter, bumps, cracks, incomplete welds on dissimilar materials (copper and aluminium). | A weak weld means high resistance, heat and early failure. |
| 5. Electrolyte filling and sealing | Checks: seal width and continuity, wrinkles, contamination on the seal area, electrolyte residue, fill-port closure. | A leaking seal lets moisture in and electrolyte out, degrading the cell over time. |
| 6. Formation, grading and final cell inspection | Checks: swelling and dimensional deviation, surface defects, label and marking accuracy, barcode and DataMatrix reading. | Only cells within specification move on to module assembly, and each one stays traceable. |
| 7. Module and pack assembly | Checks: cell orientation and polarity, busbar welds, connector presence and position, screws and fasteners, cable routing, labels. | An assembly error at pack level can make an entire pack unusable or unsafe. |
What AI adds over traditional machine vision
- Handles reflective and textured surfaces. AI learns what a real defect looks like on foils, welds and coatings, instead of relying on fixed contrast thresholds.
- Fewer false rejects. Normal variation in weld appearance or coating texture is not flagged as a defect, so good cells keep moving.
- Classifies defects, not just finds them. Each detection is labelled (burr, spatter, wrinkle, misalignment) with a confidence score, so engineers can act on root causes.
- Fast to adapt. New cell formats, chemistries or suppliers are handled by retraining on a small set of images, not by rewriting inspection rules.
- 2D and 3D. Surface defects are checked with 2D imaging; height, flatness, swelling and weld profiles with 3D data where needed.
How Logirobotix works
- Works with your cameras. We use standard industrial cameras and the images your line already produces. New cameras are added only where coverage is missing.
- Trained on your cells. Models learn from your own good and defective samples and keep improving as production runs.
- Connected to your line. Results go to the PLC for automatic reject and to your MES for traceability, linked to each cell's serial number.
- Full audit trail. Every image, decision and confidence score is stored, ready for customer audits (for example IATF 16949) and failure analysis.
- Subscription model. No large upfront investment: inspection as a service, scaled station by station.