Keep production AI performing.
A model management foundation for continuous operations, with AI quality managed throughout its lifecycle.
Manufacturing changes. AI quality needs continuous management.
Changes in products, fixtures, lighting and cameras can affect AI accuracy. Semiconductor production requires ongoing monitoring, evaluation and controlled updates to maintain AI quality.
Detect the change.
Maintain AI performance.
Statistical monitoring, including control-chart-based rules, helps detect early signs of drift. Investigate model behavior, changes in the inspected product and environmental conditions before deciding whether corrective action or retraining is needed.
- 01 Monitor & detect drift
- 02 Analyze factors
- 03 Retrain
- 04 Evaluate & approve
- 05 Deploy models
Return to monitoring the deployed model and continue the cycle. Model adoption and production release require the applicable human approvals.
Decide where people need to be involved.
Automate monitoring and evaluation while keeping decisions about corrective action, retraining and production release under the site’s defined approval procedures.
Platform
Monitor data and model behavior, detect abnormal patterns and evaluate candidate models against configured criteria.
Site engineers
Investigate whether changes arise from the model, product or operating environment. Determine corrective action and whether retraining is needed.
Quality assurance team
Review evaluation evidence and approve model adoption and production release according to the site’s requirements.
A detected drift does not automatically trigger an approved production model change. Model adoption and release follow the required review and approval process.
The platform and AI models
AI models, including visual inspection models, operate under MLOps management. AI Models
Data & outputs
Input: model inference results, input data distributions and quality indicators. Output: drift alerts, evaluation results, candidate models and deployment status.
Deployment & related products
Cloud, on-premises and hybrid. Related products: AI Models and Causal AI Platform.
Start with your manufacturing challenge.
Talk to us about your process, data and deployment environment.