100% Accuracy on Tiny Deviation of Laser Welding on Battery
Real-time laser weld inspection for battery manufacturing, ensuring critical electrical connections in EV battery packs
An electronics manufacturer of portable power bank batteries needed to improve the detection of laser welds. In around 2% of parts, laser welds would deviate from the center, which was unacceptable for battery quality. Given the accuracy required (~3 pixels), parts had to be inspected and measured by hand, limiting throughput.
With the OV80i, we built a segmentation recipe using just 12 images in under 1.5 hours. We achieved 100% accuracy on 40 test samples. After moving into production, the OV80i removed the need for manual inspectors and allowed for increased production volume.
Application Details
- Industry: Battery Manufacturing
- Application: Laser Welding
- Key Challenge: Small deviation of welding defects
- Time to Train: <1.5h setup/training
- Accuracy: 100% accuracy
Customer Quote
"I honestly didn't expect this camera could catch deviations so small, but it did. And I didn't expect the setup and training to be this fast and straightforward either. This investment has definitely increased our output quality."
— Head of Production
The Problem
Battery tabs had to be laser welded onto batteries and this required very accurate placements, within microns. The deviations could be very minor and difficult for humans to detect.
Human inspectors were slow: quality stations had microscopes and teams would manually measure the deviation, taking significant time and limiting throughput.
The Solution
We implemented a segmentation algorithm on the OV80i, enabling the system to calculate the precise measurement of interest—in this case, whether the weld center lies within the acceptable range of the battery center. Using a 12mm lens paired with an on-axis ring light, we captured high-quality images with excellent contrast.
Setup Process
Step 1: Hardware Setup
- • Camera: OV80i, 385mm above the base
- • Lens: 12mm, helps zoom into the area of interest with clarity
- • Light: Ring light, 225mm above the base, delivers even illumination directly along the optical axis, reducing shadows to make welding defects more visible
Step 2: Alignment
Selected unique edges for alignment, enabling inspection even when parts are rotated.
Step 3: Inspection Setup
Selected area of interest for inspection.
Step 4: Labeling for Segmentation
We labeled and trained on 6 good samples and 6 bad samples. Yellow labels the outer circle, blue labels the center point—allowing the system to calculate whether the blue center point is positioned in the middle of the yellow outer circle.
Step 5: Custom Logic with Node-Red
Goal: Decide Pass or Fail by checking if the center point lies within an acceptable distance from the outer circle's center. Specifically, it checks if the distance between them is less than a set tolerance (here, 15 pixels).
Testing & Results
Tested on 20 good samples and 20 defective samples, achieving 100% accuracy with zero misses and zero overkills. Each test sample was correctly labeled, and the program automatically calculated the number of pixels by which the welding deviated from the center.
| Inspection | Total | Overkill | Missed | Accuracy |
|---|---|---|---|---|
| Manual | 325 | 1% | 2% | 97% |
| OV80i | 1,876 | 0% | 0% | 100% |
OV80i increased inspection accuracy from 97% (manual) to 100%, reduced missed defects from 2% to 0%, and lowered overkill from 1% to 0%, while boosting throughput nearly 6× (325 parts → 1,876 parts).
Take the Next Step Toward Zero Defects
See how Overview AI can transform your battery manufacturing quality inspection.
Request a Demo