The rapid advancement of wind turbine technology has enabled the deployment of increasingly powerful wind turbine generators (WTGs). This study focuses on a Siemens Gamesa Renewable Energy (SGRE) WTG with a rotor diameter of 236 m and a hub height of 154 m, installed in a wind farm in the seismically active Taiwan Strait. Seismic ground motions can impose significant loads on WTGs, particularly during the idling state, where the absence of aerodynamic damping can lead to amplified structural responses and potentially challenge project feasibility.

Current deterministic seismic load calculation methods do not account for the probability of such combined scenarios and lack a clear connection with the safety philosophy, where seismic design is associated with a target annual probability of failure that balances the consequences of failure against the cost of the necessary safety measures. Without accounting for these considerations in the simulation, the resulting loads may become either overly conservative or insufficient. This study applies a probabilistic approach to seismic load calculation to address these limitations and provide a clearer understanding of the results

The probabilistic approach models dependencies among four key environmental variables: wind speed, wind direction, wind shear exponent, and ocean wave direction, using a Non-Parametric Bayesian Network (NPBN). In the NPBN, dependencies are captured via a rank correlation matrix describing both conditional and unconditional monotonic relationships. Given wind speed (defining turbine operating conditions), conditional sampling of the remaining variables is performed through Monte Carlo simulation (MCS), preserving these dependencies. Independent variables such as azimuth angle, seismic wave direction, and seismic seeds are also sampled using MCS.

Seismic loads are computed for four seismic return periods (100, 475, 1300, and 4500 years) using SGRE’s aeroelastic simulation tool, BHawC, under both deterministic and probabilistic approaches. The annual probability of failure associated with deterministic seismic load limits is evaluated and compared with optimal load limits derived through iterative optimization to meet the target recommended by the IEC standard and supported by literature. Results show that deterministic approach results in a seismic load limit on the blade that can be up to 26 % higher than what is required, whereas for the tower component, it can be approximately 8 % lower than the optimal level.

The probabilistic framework not only improves the computational efficiency of seismic load simulations by requiring significantly fewer simulation groups but also produces more realistic, site-specific load distributions. Moreover, it enables the seismic load limits to be defined based on a target probability of failure, allowing the wind turbine designs to optimally balance the consequences of failure against economic benefits, such as the material costs of safety measures. This approach is particularly valuable to SGRE, contractor, and client, as even a small percentage reduction in load limit can lead to substantial cost savings for the entire project.

Offshore wind energy is one of the emerging renewable technologies toward a zero-carbon emission society. Historically, offshore wind farms have been developed in the North Sea, but nowadays wind farms are expanded to other regions such as North America and Asia, which are seismic prone.
In the design of Rotor-Nacelle-Assembly (RNA) and support structure for offshore wind farms, the number of design positions is reduced by using representative design positions (i.e. clustering). The number of designs is directly related to the efficiency of the design work, so it is worthwhile to minimize it by wisely selecting the representative position. However, such a clustering methodology has not been systematically developed in seismic design due to the limited number of projects and researches in seismic regions.
As for seismic analysis, there are several challenges to develop an efficient clustering strategy. First, higher (and multiple) vibration modes are excited in the system consisting of the wind turbine itself in interaction with the soil. Second, blade vibration (and blade-tower coupling effect) has a significant role in the output forces. Third, energy intake from the seismic ground motion is dependent on the position due to soil amplification. Considering the problem statement above, this research develops a clustering strategy for seismic analysis to optimize the design process on RNA and support structure for offshore wind farms. To achieve this goal, as a first step, a simplified calculation model is created to efficiently carry out the sensitivity studies. A combination of point masses with Euler-Bernoulli beam is used for RNA model while lumped mass at the tower top is conventionally used in past researches, which allows to properly represent the blade-tower coupling effect contrary to most of past researches. Considering modal participation mass, higher vibration modes are truncated, and principle vibration modes are detected, which improves the computational efficiency. Both the frequency domain method and the response spectrum method are developed and compared with proven software, BhawC, to verify their accuracy. Then, a sensitivity study is carried out by using the developed model. Through the sensitivity study, the system property is approximately quantified as a form of ”scaling factor”. Combining the scaling factor with acceleration response spectrum, the maximum bending moment for an arbitrary position (i.e. arbitrary water depth, soil condition and bedrock depth) can be quickly estimated. Subsequently, the positions in a wind farm are sorted into groups with similar load trends based on the estimated bending moment; this is the basis of the clustering strategy proposed in this thesis. Finally, the strategy is applied to virtual wind farms and its applicability is verified. Although it shows some errors in the estimation of the maximum bending moment itself, the accuracy is considered satisfactory to detect the most loaded position (i.e. design representative position) and categorize the positions which have similar load trend, which are one of the main requirements for a reliable clustering methodology. Also, the strategy is applied to more realistic soil condition, such as inhomogeneous soil and its applicability and limitation are investigated.