Telemetry from industrial assets like PLCs, sensors and more is ingested by AWS IoT Greengrass leveraging pre-fabricated connectors.

Hot anomaly detection originates at the edge performing regular data stream analytics and machine learning inference.

Edge to Cloud ingestion can be performed in real-time when network connectivity is available. AWS IoT Core provides an interface
to ingest data via MQTT to further processing, where as IoT SiteWise provides aggregation, assets and model management specifically geared towards IoT assemblies.

Scheduled AWS IoT Jobs perform detection on the edge to identify machine
learning (ML) result drift.

Edge and close-to-edge compute and storage is used to store raw device
telemetry as well as ML inference results, but is also used to retrain models and
perform data analysis.

Terrabytes of data can be ingested into the data lake more effectively to perform
anomaly detection in the Cloud and retrain or train new ML models.

An AWS S3 data lake architecture stores raw and process telemetry data,
trained machine learning (ML) models and ML inference results.

Amazon Kinesis Data Analytics runs queries to determine anomalistic behavior in datasets.

Time Stream database stores time series data including dimensions of any choice.

Kinesis data analytics can be used to filter or process telemetric data
before storing time series records.

Perform cold anomaly detection leveraging cold data stored in the data lake structured
in AWS S3. After drifts are detected – either in the cloud, or at the edge based on hot data –
Lambdas and step functions play important roles to orchestrate ML training and analysis.

Push re-trained ML model to the IoT device in order to perform hot data anomaly detection.

Operational technology teams consume alerts from SNS as emails, text messages, or
Integration into ticketing systems.

Assess real-time and
historical machine performance.