Cold Plate: Four 2D Cameras Stitch One Plate in 42 s
The cold plate in a new-energy battery box is a wide workpiece, and base and cover plates come in multiple variants. Inspection must cover missing or wrong parts, defects and presence across several variants in one system. The previous manual method fell behind the cycle on wide, multi-variant parts. This cell uses nearly four 2D industrial cameras to capture by zone, applies internal correction and stitches one complete workpiece image, and completes plate inspection in about 42 seconds, with over 15,000 units inspected per month.
A single camera faces two impossible requirements
Field of view and resolution are in direct conflict. Widen the view to cover the plate and each feature gets too few pixels; keep the pixels and the view no longer covers the plate. This is imaging geometry, not parameter tuning. The question therefore shifts from which resolution to buy to whether the plate should be captured in zones.
Multi-camera is not just more cameras; it is more correction
Nearly four cameras cover the width in zones, which adds only a few mounting positions in hardware. The real work is in the internal algorithm: each camera image must be corrected to remove position error from multi-view parallax and mounting tolerance before stitching into one complete image. Skip that and the images fail to line up at the seams, leaving nothing to judge.
Correction brings a secondary benefit: the workpiece does not need precise positioning. Placement variation is absorbed by software correction, so a cart can deliver the part and the station needs no fine positioning mechanism.
Seams, brightness and perspective decide whether judgment is possible
The most underrated difficulty in a multi-camera system is that joining images is not the same as judging them. Seams, brightness differences and perspective differences between cameras make the same feature look different from zone to zone. Templates applied per zone then no longer match the actual image, and false calls start near the seams. Full-plate stitching with unified correction keeps the image readable and consistent, which is what allows stable mixed-variant inspection.
Variant differences go to software; hardware stays fixed
Base plates and cover plates come in several models, with different outlines, holes and feature layouts. If every variant required new teaching or camera adjustment, inspection capacity would fall behind the line. Here variant differences sit in software: the matching inspection template is called per variant to compare missing parts and defects, and changeover switches only the program while camera and lighting configuration stay fixed. The more variants, the more this matters, because hardware changeover time rises linearly with variant count while software switching does not.
Four things inside 42 seconds
Capture, correction, stitching and judgment all complete in about 42 seconds, subject to site acceptance results. Synchronized zone capture is the precondition; serial capture alone would eat most of the cycle. Correction and stitching are automatic, and judgment compares templates for missing parts and defects with compute kept under control, or the algorithm would stretch the takt.
What 15,000 units a month really means
The cell raises inspection throughput several times over traditional measurement and inspects more than 15,000 units a month, matching the line's inline inspection need. The significance is not peak capacity but repeating the same criteria every shift. After inspection, the system generates a record and a complete workpiece image, anomalies are intercepted before transfer, and data is archived by batch so assembly quality can be traced.
Selection questions for a multi-camera cell
Why not one line-scan camera for the whole width?
A line scan needs stable relative motion between part and camera. Here parts arrive by cart and stand still, so zone-based area cameras fit the delivery method better.
Can stitching error cause wrong verdicts?
Yes, which is why correction and seam consistency are mandatory. Put seam consistency and software correction capability on the evaluation list, not just camera count.
Does a new variant need a new station?
No. Variant differences are carried by software templates, hardware stays fixed, and changeover is a program switch.
Can this cell measure dimensions?
Its role is missing-part and defect error proofing, not precision dimensional measurement. Items requiring a number need a separate solution.