Bemærk
Adgang til denne side kræver godkendelse. Du kan prøve at logge på eller ændre mapper.
Adgang til denne side kræver godkendelse. Du kan prøve at ændre mapper.
Choose a pattern based on the input you have and the output you need. All example values are synthetic.
| Goal | Operation | Implementation |
|---|---|---|
| Add context to readings | Look up the latest record by asset ID | Built-in enrichment; an application maintains the state store |
| Align and correlate records | Match readings by timestamp or item ID | Custom stateful sample or an external service |
| Noise reduction | Average a nonoverlapping batch | Built-in tumbling window |
| Route readings by severity | Filter impossible readings and route by severity | Built-in filter, branch, map, and concatenate |
| Route more than two message kinds | Select a path for each message kind, then merge paths | Built-in branch and concatenate |
| Route invalid messages to a dead-letter queue | Separate rejected payloads for a consumer to retain | Built-in branch, map, and concatenate |
| Model prediction and updates | Evaluate a model, manage its parameters, or deliver an update | Choose an independent pattern below |
Model prediction and updates
Choose an independent pattern for prediction, fitting, or delivery. Combine them as needed.
| Pattern | Use when |
|---|---|
| Calculate a prediction with known coefficients | You need to evaluate an existing formula, using parameters in the message or retrieved through enrichment |
| Fit and activate an approved model | Your application needs to fit and approve new parameters, then persist the complete active-model record |
| Deliver approved model updates through MQTT | You need to send approved coefficients to a custom predictor for later readings |
Prerequisites
Start with an existing Azure IoT Operations deployment, a data flow profile, and an MQTT endpoint with broker access. Follow Create a data flow graph to configure them.
The examples use profile default and MQTT endpoint factory-mqtt in namespace azure-iot-operations. Built-in transforms use the default registry endpoint.
For custom patterns, use the ready-to-apply manifests or Helm chart in the Azure Samples deployment instructions, including registry setup and single-instance deployment. To check for sensor silence, configure the clock publisher against your existing broker.