Making offshore wind energy more cost competitive in comparison to fossil-fuel based production, is vital to maintain the direction the European Union has taken in renewable energy. Increasing the lifetime of a turbine can play a big role in driving down the overall costs of ener
...
Making offshore wind energy more cost competitive in comparison to fossil-fuel based production, is vital to maintain the direction the European Union has taken in renewable energy. Increasing the lifetime of a turbine can play a big role in driving down the overall costs of energy. The more energy the turbine is able to produce in its lifetime, the lower the costs per MWh will be. Fatigue damage is one of the limiting factors in a turbine’s lifetime. These damages are inversely proportional to the damping ratios and as such, estimating these ratios accurately, allows for optimisation of the structural design as well for improving control algorithms.
Modelling software is used to estimate the modal properties, such as damping ratios, in the
design phase of a turbine. However, these modelled properties often have a mismatch with
reality due to differences in material properties, soil characteristic and others. Design based
on these mismatched properties can lead to suboptimal control and a decrease in lifetime of
the turbine. In order to eliminate this mismatch, it is of importance to accurately obtain the
modal properties from the real turbine.
System identification can play an important role in this. Using measurement data obtained
during idling and operation of the turbine, the modal properties can be identified. When
using measurement data, the danger of over-fitting is always present however. Often, a tradeoff
needs to be made between the variance and bias of the estimation. To protect against
ill-conditioned data matrices, as well to better deal with the variance-bias trade-off, regularisation
can be added to the identification algorithm.
The first goal of this thesis is to successfully identify the modal properties of the first tower
modes and first coupled drive-train mode of a real turbine. Secondly, the effect of adding
regularisation will be examined on the estimation of these modal properties. The optimised
Predictor Based Subspace Identification algorithm will be used for identification. This will
be extended to include Tikhonov regularisation, truncated SVD regularisation and nuclear
norm regularisation. The performance of these techniques are compared in two case studies,
after which one is selected to be used on the measurement data from the turbine.
What follows is the estimation of modal properties from the turbine and evaluation of the
results for both with and without added regularisation.