AI-based visual inspection for pharmaceuticals
Overcome the most challenging visual inspection use cases with AVIS, the first AI software built specifically for difficult-to-inspect products.
60 minutes
Time to train a new inspection recipe from scratch
No defect sets
Defects aren’t required to train AVIS - only 500 units of compliant product are required for training
1.5 FTE
Replaced on average with every deployed AVIS system
99% Accuracy
Detect defects with up to 100% accuracy and as low as 1% false ejects
Increase quality and productivity with AI-driven inspection
Increase quality and productivity with AI-driven inspection
Enhance quality, consistency, and overall throughput of your operations. AVIS automatically detects defective products with heightened accuracy compared to human inspectors, giving you confidence in the quality of each unit going out the door.
Detecting anomalies is straightforward once you capture all variations that are considered normal. AVIS trains using compliant product instead of labeled defect images. It’s the only practical way to automate visual inspection for the world’s toughest-to-inspect products.
Four steps to create a recipe
Record
Create a recording of 500 compliant units of product.
Configure
Select regions of interest (ROIs) for all cameras. AVIS creates an individual model for each ROI.
Train
AVIS creates a recipe by autonomously training to recognize all normal variation within each ROI. This typically is complete in less than 15 minutes.
Run
AVIS uses its reference models of normal variation to detect defects which are displayed on the AVIS HMI and archived in the audit trail.
Record
Create a recording of 500 compliant units of product.
Configure
Select regions of interest (ROIs) for all cameras. AVIS creates an individual model for each ROI.
Train
AVIS creates a recipe by autonomously training to recognize all normal variation within each ROI. This typically is complete in less than 15 minutes.
Run
AVIS uses its reference models of normal variation to detect defects which are displayed on the AVIS HMI and archived in the audit trail.
Record
Create a recording of 500 compliant units of product.
Configure
Select regions of interest (ROIs) for all cameras. AVIS creates an individual model for each ROI.
Train
AVIS creates a recipe by autonomously training to recognize all normal variation within each ROI. This typically is complete in less than 15 minutes.
Run
AVIS uses its reference models of normal variation to detect defects which are displayed on the AVIS HMI and archived in the audit trail.
Record
Create a recording of 500 compliant units of product.
Configure
Select regions of interest (ROIs) for all cameras. AVIS creates an individual model for each ROI.
Train
AVIS creates a recipe by autonomously training to recognize all normal variation within each ROI. This typically is complete in less than 15 minutes.
Run
AVIS uses its reference models of normal variation to detect defects which are displayed on the AVIS HMI and archived in the audit trail.
Record
Create a recording of 500 compliant units of product.
Configure
Select regions of interest (ROIs) for all cameras. AVIS creates an individual model for each ROI.
Train
AVIS creates a recipe by autonomously training to recognize all normal variation within each ROI. This typically is complete in less than 15 minutes.
Run
AVIS uses its reference models of normal variation to detect defects which are displayed on the AVIS HMI and archived in the audit trail.
Automated inspection powered by AVIS
DAI-50
Built by Dabrico powered by AVIS
- Compact 36″x36″ automated visual inspection machine
- Three camera, AI-based inspection at up to 50 vials per minute
- Train a recipe for new products in one day
- No AI or computer vision expertise needed
- Trains from compliant units (no defect sets needed for training)
- Integrated material handling system and ejection
Advisum-AI
Built by Genesis Packaging Technologies powered by AVIS
- Mounts into operator inspection cabin turning SAVI into AVI
- Three camera, AI-based inspection at up to 50 vials per minute
- Train a recipe for new products in one day
- No AI or computer vision expertise needed
- Trains from compliant units (no defect sets needed for training)
- Currently available for the Seidenader V90 (ask about other SAVI machines)
- -integrates with SAVI conveyor control and ejection system
Success Stories
Replacing human inspection of lyophilized powder with AVIS
- Challenge: A pharma manufacturer needed to meet regulations for 100% visual inspection of 3-ml glass vials with lyophilized cakes, but manual inspection was challenged by high variability in compliant units.
- Solution: Our compact automated system uses unsupervised machine learning to create an inspection recipe from the normal variation in 500 pre-inspected vials.
- Outcome: A feasibility study showed 90% detection accuracy and a 6% false eject rate, outperforming human inspectors.
Replacing human inspection of molded glass with AVIS
- Challenge: A CMO recently failed numerous AQLs from semi-automated visual inspection and was looking for a solution to inspect their powder-filled molded glass container.
- Solution: Our compact automated system uses unsupervised machine learning to create an inspection recipe from the normal variation in 500 pre-inspected vials.
- Outcome: A feasibility study showed over 98% accuracy and a 3% false eject rate for defined defects, far surpassing human inspectors.
