Refine and improve a skill

A skill can be retrained and constantly tuned to improve the results. To better prepare for the deployment of the skill in a production environment, we recommend you establish acceptable limits for the skill results. An important consideration is that a skill cannot deliver 100% accuracy for all datasets. There will be a trade off between accuracy and usability. The skill also improves over time when provided with new images or retrained.

Determine the level of acceptable tolerance. As AVEVA Vision AI Assistant can integrate with AVEVA Operations Management Interface (OMI) and AVEVA Insight, the skills results can be used to trigger alarms or workflows in those systems. There could be a scenario where every anomaly triggers an alarm, resulting in significant noise and unnecessary burden on the users. Depending on the severity and impact of the anomaly, configure alarms to respond to only the most important anomalies. Refine these configurations continuously to suit new results from the skill.

What about false alarms?

False alarms can occur when the input image differs from the training by a very small margin. If you notice an image incorrectly classified, use the steps described in Retrain a skill to correct the classification. The skill can be retrained by providing more images or creating a new skill with new and old images.