Fleet-Wide Anomaly Detection for Wind Turbines Using Amazon SageMaker Edge Manager
Energy companies managing large fleets of wind turbines face challenges in efficiently monitoring and maintaining these assets to reduce downtime and operational costs. AWS solution uses Amazon SageMaker Edge Manager to deploy and manage ML anomaly detection models on edge devices connected to each turbine, enabling real-time anomaly detection. Sensor data from turbines is collected, streamed to edge devices, and analyzed locally using ML models optimized and compiled with SageMaker Neo. AWS IoT Core manages the device fleet and over-the-air updates, while Amazon OpenSearch Service visualizes metrics and anomalies on dashboards. The solution supports scalable lifecycle management of ML models across possibly thousands of devices with centralized monitoring and alerts.
- Organization
- Energy companies managing wind turbines
- Industry
- Energy & Utilities
- Location
- United States
- Published
- April 2021
Reported outcomes
Strategic outcomes
Primary read
Use case focus
Showing 2 of 2
- 1Edge AI
- 2Anomaly Detection
- Energy companies needed to reduce costs and downtime by detecting anomalies in wind turbine performance in real time across large geographically-distributed fleets.
- Traditional approaches could not easily scale to manage and update ML models running on edge devices at this scale.
- AWS provided a platform leveraging Amazon SageMaker Edge Manager to optimize, secure, deploy, and manage ML models on Jetson Nano edge devices attached to each turbine.
- Data scientists trained PyTorch autoencoder models on sensor data using SageMaker, then optimized and compiled the models for Jetson Nano using SageMaker Neo.
- The Edge Manager agent runs on each device, exposing APIs to invoke models and capture metrics in real time.
- Over-the-air deployment of model and software updates is handled through AWS IoT Core and managed through SageMaker IoT integration.
- Collected data and model metrics flow into Amazon OpenSearch Service where customizable dashboards visualize turbine status and anomaly alerts.
- The solution enabled real-time anomaly detection on thousands of wind turbines, reducing downtime costs by up to $1,600 per turbine per day.
- Predictive maintenance efficiency improved, helping extend the equipment life and maximizing productive uptime.
- Automated lifecycle management of ML models at the edge reduced operational overhead and scale-up complexity.
- The open-source mini wind turbine testbed demonstrated feasibility and ease of replication for similar fleets.
Architecture
The solution architecture features Jetson Nano devices connected to wind turbine sensors streaming data to the local device software which invokes ML models via SageMaker Edge Manager agent APIs. The models are optimized and compiled using SageMaker Neo. The device fleet is managed via AWS IoT Core and over-the-air updates enable seamless deployment. Data and model metrics are pushed to Amazon OpenSearch Service for monitoring dashboards and alerts.
Sources & evidence1
AI-generated summary. Verify important details with the linked sources before relying on this case.
Explore related AI use cases
Was this useful?