01
Defined the sorting rules with the processor.
We set the defect classes (good, empty, minor, small, medium and large cracks, leakers) and what happens to each egg: package, redirect to liquid egg, or discard.
Case study · Eggs
We built non-contact crack detection for commercial egg grading.
- 216K EGGS/HR
- 0 EGGS TOUCHED
- 99.6% ACCURACY
Egg graders found cracks by tapping every egg.
The industry standard was acoustic: small hammers tap each egg, and the pitch of each tap signals a crack. On average, those systems catch 75 to 85% of cracks.
Tapping every egg also means touching every egg. If one shell carries salmonella, fecal matter or leaking yolk, the hammers pick it up and pass it to every egg they tap after it.
The hardware is hard to keep running, too. The sensors have to be washed daily, water gets in, and they break or degrade, so crews spend their time cleaning, repairing and replacing parts. Optical detection was the obvious alternative, but it had been tried for years without a system that worked in production.
What we built
A non-contact vision system that finds cracks and leaks, from hairline fractures to leakers, and sorts every egg by the processor’s own rules.
02
Engineered the imaging.
12 high-speed cameras at two opposing 45° angles capture 16 images of every egg as it rotates, covering the entire shell, at about 1,000 images per second. We tested polarized backlighting, 395nm UV, and monochrome imaging across UV, white and infrared light to find what makes cracks visible.
03
Trained the models.
Thousands of labeled images per class, trained and retrained until the system could tell a real crack from a natural mark on the shell. A human inspector or an acoustic sensor can confuse the two.
04
Integrated with the line.
The processor’s hardware team built the vision booth to our specs, and we connected our software to their controls and ejection points for automatic sorting. Every egg gets a record: ID, lane and confidence score.
The result
- 99.6%
Accuracy, against the 75 to 85% typical of acoustic tapping.
- 0
Eggs touched, so inspection can’t spread contamination from egg to egg.
- 0
Moving parts to wash, repair or replace.
SANOVO licensed the technology worldwide.
At the test site, the acoustic detectors were switched off once our system was running. In 2020, SANOVO Technology Group, one of the world’s leading makers of egg grading and processing lines, took a worldwide exclusive license to it.
“This new and special designed artificial neural network algorithm … has made a revolutionary change to the future of egg crack detection.”
Have an inspection problem nobody’s solved yet?
We’ll tell you whether we can solve it and what it would take.