Inline measurement does not replace the laboratory. It maps machine-direction, cross-web, and defect distributions continuously, then uses offline samples for correlation so every anomaly returns to a roll and web position.
Separate three signal types
Loading, thickness, and surface defects answer different questions
A loading gauge estimates coating mass per area, a thickness gauge measures foil-plus-coating thickness or profile, and vision uses reflectance, transmission, or surface appearance to find streaks, pinholes, bare foil, particles, cracks, and edge defects. They support but do not replace each other: a local particle may barely affect scan-averaged loading yet become critical after calendering or slitting.
Sodium-ion cathodes span broad composition and density ranges, while hard-carbon electrodes can differ in roughness and optical absorption. A change in chemistry, foil thickness, coating width, or target loading requires renewed gauge correlation and vision recipes rather than inherited coefficients from the previous product.
| System | Primary output | Cannot answer alone |
|---|---|---|
| Inline loading | Machine/cross-web mass distribution | Particles, cracks, local appearance |
| Inline thickness/profile | Thickness, heavy edge, profile | Composition, adhesion, residual moisture |
| Vision inspection | Defect class, size, and location | True coating mass and internal pores |
Calibrate before coating
Bare-foil baseline, reference samples, and offline weighing are all required
Run stable uncoated foil through the gauge first to establish the bare-foil baseline and check left-right drift. Then use reference pieces or qualified sections spanning the operating range to check zero, span, and cross-web scan. Correlate online values with fixed-area, fixed-position offline weighing and thickness; sample positions must be reproducible on the roll.
Passing calibration does not guarantee an entire shift. Verify after foil, slurry, or product changes, gauge maintenance, material temperature shifts, or widening online-offline differences. Store reference ID, time, operator, deviation, and correction state instead of recording only a “calibrated” flag.
- —Zero with bare foil
- —Span the operating range with references
- —Correlate with fixed-position offline samples
Reading production histories
Beyond mean loading, watch cross-web profile, periodicity, and transients
The operator screen should show the current cross-web profile, recent machine-direction history, vision alarms, and web position together. A left-right bias points first to die, feed distribution, backing roll, or scanner. A streak fixed in cross-web position points to die lip, agglomerate, or backing roll. For periodic variation, compare frequency with pump stroke, roll rotation, tension, and filter pressure.
Coating starts, stops, splices, acceleration, deceleration, and slurry transitions must not be mixed with steady-state statistics. Tag events automatically or manually and write affected start/end positions into the roll record. Slitting can then avoid precise sections instead of scrapping the roll or sending the defect into cells.
| History/image feature | Check first | Section action |
|---|---|---|
| Persistent cross-web slope | Die, backing roll, feed distribution, scanner | Hold until profile is stable again |
| Fixed-position machine streak | Lip debris, agglomerate, roll damage | Remove by defect map and web position |
| Periodic loading variation | Pump, roll, tension, filter-pressure frequency | Mark full cycles and verify post-fix sections |
| Scattered pinholes/particles | Deaeration, foil, filtration, debris | Grade by size, density, and consequence |
Turn data into disposition
The defect map must follow roll, web position, slitting, and cell identity
A polished inline chart is not enough. Each alarm should retain defect class, size, cross-web coordinate, machine-direction position, image, and process signals, linked to master roll, daughter roll, knife position, and electrode lot. Cell linkage then allows performance data to improve defect grading.
Disposition should separate safety-critical defects, removable sections, and trend shifts requiring sample verification. Evaluate false calls and misses through periodic human review, known-defect samples, and offline inspection. Simply reducing alarm count over time can suppress real defects as well.
Inline inspection finds and locates problems faster; final release still combines chemistry, downstream process, defect consequence, and offline validation.
Bibliography
References
- 01Coating Defects of Lithium-Ion Battery Electrodes and Their Inline Detection and Tracking
- 02In-line electrode quality assurance in Li-ion battery manufacturing
- 03Wet and Dry Electrode Manufacturing and Thin-Film Technology
- 04Detection and Identification of Coating Defects in Lithium Battery Electrodes Based on Improved BT-SVM
Updated: 2026-08-26