Abstract
Noise, vibration and harshness (NVH) have become important design drivers for wind turbine drivetrains. Detailed knowledge about the modal model is essential in this regard. Given the size of the wind turbine system and necessity for rotation of the drivetrain systems to be representative for normal operation, operational modal analysis (OMA) approaches are increasingly explored to assess these modal properties. The main challenge then are the harmonics
originating from the gears. These excitations are important since they are the sources of tonal noise. Amplification by structural resonances plays a potential important role. Therefore, the resonances close to the harmonic excitations are in focus. To be able to extract them, the influence of the coinciding harmonics needs to be reduced using a cepstrum lifter. Moreover, to allow the processing of a large amount of data, automated OMA techniques are used. Once the frequency and damping can be characterized for the important resonances, it becomes possible to gain insights in their changes. Particularly unveiling the links between these modal parameters and the operational parameters of the machine helps in further understanding turbine NVH behaviour. Correlation investigation and signal modelling by means of machine learning approaches allow for this. This paper will focus on processing experimental data of an offshore wind turbine and will investigate the changes in resonance frequency and damping over time. Modal parameters are extracted using automatic OMA. The changes in frequency and damping are investigated during the different measurements to understand how they link with the operational parameters of the machine (e.g. rpm). This is done in order to gain insights in their natural changes due to the non-linearity of the system loading. Better understanding this can help in defining statistical boundaries around expected resonance and
damping values. These can then be used in as probabilistic boundaries in the turbine design process for noise mitigation.
originating from the gears. These excitations are important since they are the sources of tonal noise. Amplification by structural resonances plays a potential important role. Therefore, the resonances close to the harmonic excitations are in focus. To be able to extract them, the influence of the coinciding harmonics needs to be reduced using a cepstrum lifter. Moreover, to allow the processing of a large amount of data, automated OMA techniques are used. Once the frequency and damping can be characterized for the important resonances, it becomes possible to gain insights in their changes. Particularly unveiling the links between these modal parameters and the operational parameters of the machine helps in further understanding turbine NVH behaviour. Correlation investigation and signal modelling by means of machine learning approaches allow for this. This paper will focus on processing experimental data of an offshore wind turbine and will investigate the changes in resonance frequency and damping over time. Modal parameters are extracted using automatic OMA. The changes in frequency and damping are investigated during the different measurements to understand how they link with the operational parameters of the machine (e.g. rpm). This is done in order to gain insights in their natural changes due to the non-linearity of the system loading. Better understanding this can help in defining statistical boundaries around expected resonance and
damping values. These can then be used in as probabilistic boundaries in the turbine design process for noise mitigation.
Original language | English |
---|---|
Title of host publication | IMAC 2019 Conference proceedings |
Publication status | Accepted/In press - 2019 |
Keywords
- Automatic Operational Modal Analysis
- Wind Turbines
- Harmonics
- Machine learning