Online quality monitoring for EV battery manufacturing: three-channel real-time acquisition of power/amplitude/frequency with three-way fusion judgment — closing the gap left by sampling inspection.
The Pain Point: Quality Cannot Rely on Post-Hoc Sampling
For EV batteries, tab and module welding are critical processes — one cold weld or over-weld can scrap an entire batch, or worse, cause a safety incident. Yet most factories rely on sampling inspection plus destructive testing. Sampling rates are low, misses are common, and defects are often discovered only after shipment.
Why not inspect 100%? Because welding is fast (50-200ms per weld) and manual visual checks cannot keep up with takt; post-weld methods (pull testing, metallography) are destructive and unsuitable for full inspection.
Solution: Real-Time In-Process Monitoring — Three-Channel Capture + Three-Way Fusion
This solution samples while welding happens — no added takt, no product damage, every weld point intelligently judged:
- Three-channel real-time capture: power, amplitude and frequency signals received from the welder via UDP (10kHz sampling), recording the full energy curve of every weld
- CNN fault classification: a lightweight 1D-CNN (25KB, 0.3ms) infers in real time at the edge, recognizing known fault modes (cold weld, over-weld, insufficient power)
- VAE anomaly detection: a variational auto-encoder learns the distribution of normal weld curves and catches novel anomalies the CNN has never seen — real faults are too diverse to be fully labeled
- Rule engine explainability: power offset, amplitude offset and first-ultrasonic offset translate 'something is wrong' into concrete physical parameters an operator understands — solving the human-trust problem in industrial settings
Three signals fuse into a judgment; high confidence auto-passes, low confidence goes to human review — forming a self-improving loop of label → train → re-infer.
Value: From Sampling to Full Inspection, From Reactive to Proactive
| Aspect | Traditional | After Deployment |
|---|---|---|
| Coverage | 5%-10% sampling | 100% of weld points |
| Defect discovery | Customer complaint after shipment | Instant alarm at weld time |
| Traceability | Destructive tests, no records | Full energy curve per point, replayable |
| Miss risk | Manual visual misses | CNN + VAE dual-model backup |
Reuse with Light Customization
Built on the standard online-monitoring product with per-welder customization (protocol integration, thresholds, model fine-tuning). Adapts to ultrasonic welding, resistance spot welding and laser welding.
Suitable for: EV batteries, 3C batteries, energy-storage batteries and any scenario demanding strict weld quality.
Every Weld Point on Record
Weld quality cannot be left to chance — the missed cold weld in sampling may become the complaint and recall after shipment.
In-process monitoring judges every weld point intelligently, replays the full energy curve, and intercepts anomalies on the spot. Want to know which approach fits your welding line? Contact us for an assessment.
