Simulation-Based Optimization of Lattice Support Structures for Offshore Wind Energy

Today, design of wind energy support structures is to a large extent a manual process. It requires a lot of experience, and the design tools are often based on simplified methods. As larger structures are being developed and installations move to larger water-depths, the need for efficient and accurate design tools increases. Simulation-based design is a promising technique that can help automate this process. In this study, Spall’s simultaneous perturbation stochastic approximation (SPSA) method was implemented to automatically optimize thickness and diameter of the members in offshore lattice tower support structures. The method utilizes a pseudo-gradient based on only two function evaluations per iteration, which allows for a computationally efficient process. Each evaluation of the design consists of time-domain simulations of the complete wind turbine in FEDEM Windpower, subsequent rainflow counting and calculation of joint lifetimes with stress concentration factors. The utilization of both ultimate and fatigue limit states is reported for each joint. Tower weight was chosen as an indicator of cost, and an objective function comprising variables for weight and joint lifetimes was defined. Joint lifetime was ignored whenever its value was above the design lifetime of the tower, allowing the algorithm to search solely for the lightest design, as long as the design lifetime constraint was sustained. The method has shown promising results, and is able to successfully find viable designs, even when starting from highly unacceptable starting points. Some of the major challenges when using SPSA for lattice support structures are to find a good objective function, as well as appropriate values for the parameters controlling perturbation and step size. Existing guidelines were followed when doing this calibration, but for an efficient search the parameters had to be adapted. Results for both appropriate parameters and the optimization itself are reported for the 10MW NOWITECH reference turbine on a full-height lattice tower. These results show that superior results can be achieved, but at a high cost in terms of computational time. Recommendation is given to use alternative methods to come up with a partially optimized staring point, from which the SPSA method can optimize further.