Aegir Predictive Maintenance Solution

Aegir ingests sensor data and maintenance records, integrating with AVEVA, IBM Maximo, Cognite, and others to power AI models with real-time insights on equipment health and performance.

Aegir’s event processor, powered by Elder Research’s CEPtor, filters high-frequency sensor data to highlight only key events, reducing noise and improving model accuracy.

The Aegir platform is developed utilizing cloud services by default yet portable to premise as needed. The cloud infrastructure allows for “as needed” scaling without a separate engineering exercise.

Our data scientists are not bogged down in software or infrastructure setup when developing new models, as the model development environment allows them to focus on the best-fit models for the targeted pursuit.

This same environment can be offered to customers for internal development.

This is a snapshot of some of the models employed as part of Aegir, but the list continues to grow.

Predictive Models

• Survival Models: predict major faults of generator and turbine within 12 weeks
• Deep Learning Models: predict major faults of generator and turbine within 4 weeks
• Trained across all units, then refined for a given unit

Anomaly Models

• Startup timing: identify unusual startup sequence behavior
• Subsequence model: identify unusual sensor activity in a subsequence
• Shutdown timing: identify unusual shutdown sequence behavior

Aegir provides a series of visuals organized around cases and evidence to support a prediction deemed worthy of investigation. The visuals support the model scores and allow for establishing context around the event(s) that support the case using the data collected (sensor or maintenance records). Each case and its evidence can then be labeled by operations personnel to further inform the models.