Read anomaly model results through OPC UA
Access model trends and alarm levels alongside the sensor data.
Data & interfaces
Documented anomaly-service models and Sensor Dashboard node structure. Available models and node names depend on the installation.
Find the model
In the documented OPC UA address space, expand the sensor’s tensorFlow attribute and models node. Model names correspond to those shown in the Sensor Dashboard.
Choose the output
The source lists lossMAE for raw prediction error, Expectile_05pct, Expectile_25pct, Expectile_50pct, Expectile_75pct and Expectile_95pct for trend statistics, and alarmLevel for the configured threshold.
| OPC UA attribute | Meaning in the original guide |
|---|---|
| lossMAE | Raw prediction error, shown as raw. |
| Expectile_05pct | 5% expectile, shown as LO 5%. |
| Expectile_25pct | 25% expectile. |
| Expectile_50pct | 50% expectile, labeled median in that dashboard. |
| Expectile_75pct | 75% expectile. |
| Expectile_95pct | 95% expectile, shown as HI 95%. |
| alarmLevel | The alarm level set in the Sensor Dashboard. |
Check model availability
These nodes require an available anomaly model. The original service description separates model training from local monitoring; model provisioning is not a standard measurement-export step.