Ramon Perez

The Ægir Predictive Maintenance Solution is designed to enhance the reliability of hydropower generator equipment by leveraging advanced data analytics, machine learning, and sensor data processing. Developed by Elder Research, Inc. (USA) and Sira-Kvina kraftselskap (Norway), Ægir helps prevent costly failures in large industrial generators by predicting potential issues before they become critical. The Problem Unexpected failures in hydropower generators can result in millions of dollars in lost revenue. The rise of new sensor technologies has provided power providers with vast amounts of data, but extracting actionable insights from this data remains a challenge. Additionally, high-quality failure data is scarce, making predictive maintenance difficult. The central challenge is how to combine human expertise with machine-driven analysis to anticipate and prevent equipment faults. The Ægir Solution Ægir processes large amounts of sensor data, incorporates historical maintenance records, and integrates subject matter expertise with machine learning models. The solution identifies anomalous activity in hydropower generators and turbines, provides early warnings for potential failures, and continuously updates the condition grade of equipment. It creates cases for engineers to review, allowing maintenance teams to take timely corrective actions. Over time, the system refines its predictions by learning from expert feedback. Key Components of Ægir:

1. Data Sources: Ægir ingests data from multiple sources, including:

• SCADA systems for generator, turbine, and transformer sensor data.
• Local control systems that provide indicator data for power cycle actions.
• Protection system databases that contain high-resolution pressure and vibration sensor readings.
• Turbine governor systems for additional fluid flow management readings.
• CMMS maintenance ticketing systems, which store historical fault records and repair actions. • Visual inspection test results.

2. CBETT (Condition-Based Error Type Taxonomy):

• A structured hierarchy that classifies equipment failures based on expert knowledge from Sira-Kvina, Norconsult, and IRIS Power.
• Mapped to IEC/ISO 81346 standard RDS codes, ensuring a systematic approach to diagnosing and categorizing issues.

3. Machine Learning & Predictive Models:

• Survival Models: Predict major generator and turbine faults up to 12 weeks in advance.
• Deep Learning Models: Identify faults within a 4-week window with higher accuracy.
• Anomaly Detection Models: Monitor startup sequences and sensor activity to flag unusual behavior.
• Models are trained across all units and then refined for specific equipment, improving precision.
• If multiple models agree on an anomaly, a case is generated for engineer review.

4. RADR (Risk Assessment Data Repository):

• A case management system that allows engineers to review flagged issues, log actions, and investigate further using visual analysis tools. Sira-Kvina’s Role in the Project Sira-Kvina, a Norwegian hydropower company, has been a key partner in implementing Ægir. Before using Ægir, the company relied on traditional monitoring and manual inspections, leading to reactive maintenance strategies. With Ægir, Sira-Kvina now benefits from a proactive approach, allowing them to detect issues early, schedule repairs when power prices are low, and minimize unplanned downtime. By integrating Ægir, Sira-Kvina has improved its ability to prevent failures, optimize maintenance schedules, and reduce operational risks, making it a valuable case study for predictive maintenance in hydropower.

Keywords: AI, Machine Learning, Predictive Maintenance, Condition Monitoring