Operational Modes Detection in Industrial Gas Turbines Using an Ensemble of Clustering Methods.

Area of Science:

• Industrial Process Monitoring
• Machine Learning Applications
• Data-Driven Modeling

Background:

• Industrial systems generate vast sensor data, revealing dynamic relationships and operational states.
• Traditional gas turbine mode identification often relies on limited variables or predefined states.
• Detecting unknown operational modes is crucial for effective gas turbine management.

Purpose of the Study:

• To develop a data-driven methodology for discovering a priori unknown operational modes in industrial processes.
• To implement unsupervised machine learning for automatic identification and characterization of these modes.
• To enhance real-time monitoring systems for gas turbine management.

Main Methods:

• Utilized an ensemble of clustering techniques to group similar sensor data points.
• Generated automatic descriptions of discovered clusters for expert interpretation.
• Applied filtering thresholds to refine cluster quality and exclude outliers.
• Tested the methodology on Siemens gas turbine sensor data.

Main Results:

• Successfully discovered and characterized previously unknown operational modes and sub-modes.
• Automatic cluster description generation improved partition quality.
• Expert interpretation of cluster descriptions validated the identified operational modes.
• The approach demonstrated effectiveness in a real-world gas turbine case study.

Conclusions:

• The proposed methodology enables the discovery of unknown operational modes through data-driven models.
• Unsupervised machine learning offers a powerful tool for enhancing industrial process monitoring.
• Expert validation confirms the practical utility and positive impact of the identified modes.