The growing demand for wind energy necessitates efficient health monitoring strategies to ensure the long- term reliability of wind turbines. Monitoring critical loads, such as flapwise blade root moments and tower base fore- aft moments, is crucial for preventing turbine fatigue and failure. However, direct measurements through physical sensors are costly, time- consuming, and limited to specific locations. This study introduces a probabilistic data- driven virtual sensing framework that uses multi- hidden Gauss- Markov model (Multi- HGMM) to estimate these loads by capturing the relationship between measurable quantities and key structural metrics, without requiring extensive physical sensors. An expectation–maximization algorithm is used to determine the HGMM parameters from a comprehensive dataset. This dataset includes routinely recorded SCADA data, such as wind speed, rotor speed, and pitch angles, along with additional key features that were carefully selected for their relevance to load estimation. In a subsequent stage that includes operational measurement data, the probabilistic HGMM can be used to estimate loads. We validate our approach on a 5- MW wind turbine model developed by the National Renewable Energy Laboratory (NREL), for above- rated wind speeds where turbines face heightened loads due to increased aerodynamic forces, critical for structural integrity. The results demonstrated that the multi- HGMM approach achieved a mean absolute error of approximately 6% for estimating both the tower base moment and flapwise moment when incorporating tower top accelerations and shaft bending moments alongside baseline features. By reducing reliance on physical sensors, this virtual sensing methodology offers a scalable, cost- effective solution for wind turbine monitoring.
Loading....