TL;DR
This paper introduces a novel Bayesian Spectral Decomposition (BSD) method for operational modal analysis (OMA) of offshore wind turbine (OWT) towers. By applying BSD to real-world vibration data from a 5.5 MW OWT, the study demonstrates its ability to accurately identify modal parameters (frequencies and damping ratios) and crucially, quantify their uncertainties, offering a more robust approach to structural health monitoring.
What Problem Does This Paper Address?
Offshore wind turbines (OWTs) are critical for renewable energy but face harsh marine environments that can lead to structural fatigue and failure. Accurate understanding of their dynamic behavior, specifically their modal parameters (natural frequencies, damping ratios, and mode shapes), is essential for ensuring structural integrity, predicting responses, and implementing effective vibration control. Traditional OMA methods often make simplifying assumptions (e.g., Gaussian excitation) that may not hold true for OWTs operating under complex, real-world conditions. Furthermore, quantifying the uncertainty associated with these identified parameters is crucial for reliable decision-making in structural health monitoring and design.
Methods
The core of this paper is the application of a Bayesian Spectral Decomposition (BSD) method for OMA. This approach combines frequency-domain decomposition techniques with Bayesian inference. Key steps include:
- Data Acquisition: A vibration monitoring system with uniaxial accelerometers was installed on a 5.5 MW monopile OWT.
- Data Preprocessing: This involved outlier removal, low-pass filtering, and importantly, a coordinate transformation to project sensor data onto the structural fore-aft (FA) and side-side (SS) directions, accounting for the turbine's nacelle yaw.
- BSD Application: The BSD method decomposes the power spectral density (PSD) matrix using singular value decomposition (SVD). It then formulates a probabilistic model to estimate modal parameters and their uncertainties. This is a novel aspect, offering a more comprehensive view than methods that only provide point estimates.
- Validation: The results from the BSD method were compared against those obtained using the widely used Stochastic Subspace Identification (SSI) technique to verify its reliability.
Datasets
The study utilized vibration and SCADA data collected from a single 5.5 MW monopile offshore wind turbine (Turbine No. 29) located at the Biqing Bay Offshore Wind Farm in the South China Sea. Data were collected over two days, representing two distinct operational conditions:
- Rated Rotor Speed: Data collected on November 14, 2024, under typical operating conditions.
- Idle State: Data collected on November 12, 2024, specifically focusing on a one-hour period when the turbine was idling (wind speed below cut-in value).
The data were preprocessed into 20-minute segments for analysis.
Key Insights and Ideas
- BSD's Uncertainty Quantification Advantage: The study highlights BSD's strength in providing not just modal parameter estimates but also their associated uncertainties (quantified by coefficients of variation, CV). This is a significant advantage over methods like SSI, which primarily offer point estimates.
- Damping Ratio Sensitivity: Damping ratios were found to be more uncertain than frequencies, a common challenge in OMA. The BSD method's ability to quantify this uncertainty is crucial for interpreting results.
- Idle vs. Rated State Dynamics: The second bending mode was only identifiable under idle conditions, suggesting that rotor-induced harmonics at rated speed can mask this mode. This underscores the importance of analyzing OWTs under various operational states.
- Wind Speed's Influence on Damping: A clear increasing trend of damping ratio with wind speed was observed in the FA direction, particularly at rated rotor speed. This is attributed to aerodynamic damping, which plays a dominant role in the FA response. The SS direction showed less variation.
- "Soft-Stiff" Design Confirmation: The first bending modal frequency falling between the 1P and 3P frequencies confirms the typical "soft-stiff" design of the wind turbine tower.
Key Results
- Frequency Stability: Identified modal frequencies remained largely stable across both rated and idle conditions, with the first-order FA and SS frequencies averaging around 0.284 Hz.
- Damping Ratio Trends:
- At rated rotor speed, the mean damping ratio was 1.253% (FA) and 1.120% (SS).
- At idle state, the mean damping ratios decreased to 0.944% (FA) and 0.960% (SS).
- The damping ratio in the FA direction showed a significant increase with wind speed.
- Uncertainty Levels: Coefficients of variation (CV) for frequencies were very low (below 0.002), indicating high precision. CVs for damping ratios were higher but generally within acceptable ranges (below 0.09), with the BSD method providing these quantified uncertainties.
- Mode Shape Consistency: Modal Assurance Criterion (MAC) values between identified mode shapes were consistently high (exceeding 0.98), confirming the reliability of the BSD method's mode shape identification.
- BSD vs. SSI: The BSD and SSI methods yielded generally consistent frequency and damping ratio results, with BSD offering the added benefit of uncertainty quantification.
Limitations and Future Work
- Rotor Harmonics Interference: The accuracy of BSD may decrease when rotor-induced harmonic loads are close to structural modal frequencies.
- Rapid Environmental Variations: The method's accuracy might be affected by rapid variations in wind speed, rotor speed, and yaw angle, which violate the quasi-stationary assumption.
- Tuned Mass Dampers (TMDs): The current BSD formulation is not suitable for damping identification after a TMD is installed due to potential double peaks in the frequency domain.
Future work suggested includes:
- Improving the BSD framework to mitigate these limitations.
- Enhancing its applicability to more complex operating conditions.
- Further investigation into damping identification after TMD installation.
- Leveraging the quantified uncertainties for more objective tower design, alarm threshold setting, and long-term structural damage detection.
Summary updated