Our algorithm
A team of experts for industry-tested AI
BENEFITS
.
PERFORMANCE IN INDUSTRIAL ENVIRONMENTS
A leader in made in France cosmetics
Customer goal
- Technological breakthrough
- Move from sampling to 100% control.
- Ensure 100% conformity of exported parts Minimize handling of heavy parts.
Results
- 100% of defects detected and eliminated.
- 3% false positives.
- 98 of ROC AUC.
A workshop specializing in the manufacture of silver and gold links
Customer goal
- Objectivize inspection, which varies from one inspector to another (10% difference).
- Relieve experienced operators of large production runs, allowing them to concentrate on exceptional parts.
- Provide decision-making support for less-skilled FTEs.
Results
- 90 to 100% of defects detected and eliminated.
- 3 to 19% false positives.
- 89 to 98 ROC AUC.
Leading manufacturer for automotive components
Customer goal
- Delete a control station needed in 3/8.
Results
- 100% of defects detected and eliminated.
- 0% false positives.
- 100 AUC ROC.
A workshop specializing in crimping for the luxury industry
Customer goal
- Add a checkup stage to avoid sending a part with a missing stone.
Results
- 100% of defects detected and eliminated.
- 0% false positives.
- 100 AUC ROC.
MARKET-STANDARD PERFORMANCE
MVTec's databases (MVTec Anomaly Detection) are widely used in the industry to evaluate artificial intelligence models. Here is AnotherBrain's algorithms’ evaluation of these data sets.
Our algorithms stand out in the industry for their ability to outperform existing solutions by 1% to 10%. This result is achieved with remarkable efficiency, requiring here only 20 compliant parts for training.
99.7
ROC AUC
100
ROC AUC
100
ROC AUC
100
ROC AUC
100
ROC AUC
100
ROC AUC
97.02
ROC AUC
100
ROC AUC
99.98
ROC AUC
96
ROC AUC
95
ROC AUC
100
ROC AUC
100
ROC AUC
98.5
ROC AUC
OUR METHODOLOGY
We based our early research on our ability to detect any defect on all types of materials, surfaces and aspects.
To meet industrial needs when defects are rare or can take many forms, we mainly base our algorithms' learning on your parts judged as OK, and then deduce those that are KO (as a human being would do).
Our algorithm will form a representation of what a perfect part is, and compare it with the parts to be checked. This enables it to deduce whether the part is OK or KO, and to pinpoint the exact location of the fault. It can also discover new, unspecified defects.
OUR EXPERTS
AnotherBrain is a team of researchers dedicated to the creation of true artificial intelligence, aiming for general and frugal AI.
Having seen our ability to respond to many industrial use cases, particularly quality control, we formed a specialized team in 2019. We co-developed our first AI-enhanced visual quality control solution with an automotive leader.
Our refined expertise in defect detection has led to the development of :
- Blue Phosphor®: an automated quality control solution, that combines precision and objectivity.
- FA-VAE: our technology is protected by 2 published patents.