Recently, Offshore Wind Turbines (OWT) have attracted great attention in an effort to make a shift from fossil-based energy sources towards an enhanced sustainable and renewable energy production. In order to achieve the renewables targets and reduce the cost of wind energy, OWTs are consistently increasing in size. Therefore, research has targeted the optimization of OWT design. For many years, System Identification has played a central role in obtaining the actual modal properties of existing structures. Operational Modal Analysis (OMA) is a subset of these technique that applies on measurement data obtained from a structure loaded by ambient excitation. In the case of an OWT, such methods would be highly important in validating and/or updating the design and monitoring the structural health of the structure, which would potentially lead into lifetime extension. In practice, using OMA techniques on operating OWT is not a straightforward procedure. Most of these techniques assume that the excitation is a white-noise process, which is not the case when waves and operational loads (e.g. rotational sampling) are present. Additionally, the system itself is considered to be Linear Time-Invariant. Unfortunately, the modal properties of the system are highly affected by the varying rotor speed and in general do violate the LTI assumptions. Given these challenges, the use-ability of existing methods need to be further investigated through application on OWTs under different operational conditions. Also, it is vital to asses and if possible eliminate the impact of the limitations related to loading on the identification. Thus, a benchmark study of OMA algorithms has been performed on simulated data obtained from two models. The first model is a simplified OWT numerical model in Matlab, which can be used to validate the algorithms, and the second is the NREL 5-MW baseline offshore wind turbine in FAST that was used to simulate multiple different operational conditions. Using the simulated responses, Eigensystem Realization Algorithm (ERA), Stochastic Subspace Identification (SSI), Frequency Domain Decomposition (FDD) Least-Squares Complex Frequency-domain (LSCF) estimator and Transmissibility-based Operational Modal Analysis (TOMA) were examined. Through this study the robustness of the algorithms against harmonic excitation and measurement noise were investigated, providing the user with guidelines. In the application on simulated data obtained from FAST, the results showed that all the algorithms were able to derive several stable modes, even when theoretically fundamental assumptions are violated. In general, the algorithms performed better for low wind speeds, while at high wind speeds they leaded in poorer identification. The greatest deviation compared to analytically obtained modal properties was observed in the damping ratios of the flapwise blade bending modes, where none of the algorithms was able to obtain such large damping ratios (>30%). However, most of them were still able to obtain two fore-aft tower modes and an accurate damping estimation. TOMA did remove the influence of external loading from the identification, but faced difficulties in obtaining blade modes. Finally, this benchmark study revealed the strengths and weaknesses of each technique when the core assumptions are violated.

Vibration-based structural health monitoring of civil engineering structures is receiving increasing attention in recent years. This is due to the development of more robust system identification techniques as well as improvements with regard to the practicality of installing the necessary instrumentation. Health monitoring systems are more often replacing static deflection tests and detailed visual inspections, where continuous monitoring of vibration data aims for early damage detection.

This thesis can be summarized as the development of a structural health monitoring method for bridges. It forms a part of the Zwartewaterbrug project, related to the Zwartewater bridge located in the city of Hasselt, in the Dutch province of Overijssel. The Zwartewater bridge served as a case study for validation of the proposed early damage detection algorithm.

The wider purpose of this work was to address the problem of having a large number of bridges in Europe that are facing the end of their service life and should either be retrofitted or decommissioned.. The chosen approach consisted of a low-cost vibration monitoring method, that is suitable for continuous structural health monitoring.

The initial challenge was to introduce a small change to the structure, representative of early damage, without actually damaging the structure. For this an added mass approach was taken, where weights of respectively 25, 50, 75, and 100 kg were added below the bridge deck. The vibration data (accelerations) measured on the ”damaged” structure were then used to solve an inverse problem where the aim was to detect the induced damage without closing the bridge to traffic.

Since the data was gathered with only ambient excitation by wind and traffic, a new challenge arose, which is considered to be the core of this thesis. The structural differences between vibration measurements related to the ”healthy” and ”damaged” structure were concluded to be smaller than the structural changes due to the differences in unknown traffic loading. The traffic loads were namely varying from 200kg motorcycles to 50t trucks.

In this thesis, the method of Probabilistic Filtering was developed to face the aforementioned challenge. The method involves mainly a data pre-processing step, and the probabilistically filtered signals are subsequently processed with two output-only identification methods: Frequency Domain Decomposition and data-driven Stochastic Subspace Identification. The full process was integrated into an automatic analysis algorithm, performed in MatLab.

It was concluded that both detection and localization of the induced damage is possible with the proposed methodology. Further work was then performed to quantify the detected added mass in terms of structural damage. A quantitative finite element analysis was performed with the substructure approach (super-element, static boundary conditions to the substructure of interest). The given analysis concluded that the developed structural health monitoring method is able to detect anomalies comparable to a 35cm crack in the welds securing the longitudinal stiffeners to the transverse beams.

Finally, recommendations were made regarding an expansion of the method to structure service life predictions and more computationally efficient calculation set-ups.

In the race for cost reduction in the offshore wind industry, support structure optimization leading to weight reduction plays a prominent role. The fatigue limit state is often the driving consideration for support structure design. Monitoring the monopile loads can offer an accurate knowledge of its consumed and remaining fatigue lifetime, which in turn has the potential benefits of lifetime extension and feedback on current design practices among others. Since within an offshore wind farm not all turbines are subjected to the same loading, a turbine or cluster-specific internal load monitoring scheme is investigated using data-driven approaches and specifically feedforward artificial neural networks (ANNs) and linear regression (LR).

The purpose of this thesis is to examine whether it is possible to accurately estimate the actual loading of the offshore wind turbine at the monopile mudline fatigue sensitive location by utilizing standard signals and/or sensor measurements and machine learning techniques instead of a structural model.

Data simulated with the Siemens Gamesa Renewable Energy in-house aeroelastic code - Bonus Horizontal axis wind turbine Code (BHawC) – is used, including operation and idling fatigue load cases. Ten minute time and frequency-domain statistics of signals that are collected in turbines are used as inputs to the data-driven models, whereas 10-minute damage equivalent loads (DELs) of monopile moments are used as targets. By using statistics of rotor rotational velocity, electrical power output, blade pitch angle, hub wind speed and nacelle accelerations, a mean absolute error of DEL estimation smaller than 4% in both operation and idling load cases is achieved, under the condition that the training set properly reflects the entire variation in environmental conditions. Furthermore, errors average out over multiple 10-minute intervals, resulting in accurate long-term estimation with residual errors between -1% and +1%. The accuracy of the estimation can be further improved by including additional sensors; accelerometers placed at the tower bottom (TB) can help reduce ANN mean absolute error by up to 40% in operational load cases.

In addition, if inclinometers are installed at TB they can allow 10-minute equivalent ranges of rotation signals at TB to be used as inputs to a LR scheme to accurately estimate DELs. When the 10-minute intervals are binned based on whether the mean wind speed is below, around, or above the turbine rated wind speed, this method gives mean absolute errors below 2.5%. As a result, TB inclinometers are recommended as extra sensors for load monitoring, provided that factors such as their accuracy level and durability compared to alternatives are considered.

The data-driven models examined in this thesis show potential not only for online internal load monitoring, but also for fast estimation of past load histories using recorded data.