Start with visual inspection.
Put AI to work on your line.
Support verification review and defect classification using images from existing automated inspection equipment. Run inference on edge PCs and manage model performance with MLOps after deployment.
Reduce the manual review left after automated inspection.
Automated inspection can flag candidates that require further review. AI helps distinguish likely false positives from defects and supports defect classification within existing inspection workflows.
Automated optical or final visual inspection identifies candidate defects.
Review flagged images and identify likely false positives and defects.
Support defect classification and final quality review.
The diagram illustrates one possible inspection workflow. Equipment, review procedures and quality decision points vary by manufacturing site.
Six model types
Different model types support visual inspection, anomaly monitoring and process analysis. The appropriate combination depends on the target task, available data and deployment environment.
Defect detection
Identify defect presence and location in inspection images.
Defect classification
Classify detected defects into defined categories.
Anomaly detection
Detect deviations from expected patterns in inspection or operating data.
Segmentation
Identify regions of interest or defects at the pixel level.
Causal inference
Analyze possible cause-and-effect relationships among process variables.
Anomaly prediction
Predict potential anomalies from relevant operating data.
Measured in BFAI PoCs
PoC in back-end visual inspection: verification review, final inspection and classification.
Back-end visual inspection PoC, compared with manual visual review.
Based on BFAI PoC results. Performance and evaluation conditions may differ by process, equipment and inspection task. Contact us for detailed evaluation conditions.
Deploy on your line. Keep accuracy over time.
After the PoC, models run on your inspection data in your environment, and are monitored and updated with MLOps.
Data & outputs
Visual inspection models use images from existing inspection equipment and output defect presence, type and location. Other model types use data appropriate to their application.
Deployment
Run visual inspection inference on an on-site edge PC. Cloud and on-premises options are also available depending on the application and deployment requirements.
Maintain accuracy after deployment
Enterprise MLOps supports model monitoring, evaluation and controlled updates after deployment.