Multi-level Decoupled Optimization of Wind Turbine Structures Using Coefficients of Approximating Functions as Design Variable

This dissertation proposes a multi-level optimization method for slender structures such as blades or towers of wind turbine structures. This method is suited performing structural optimizations of slender structures with a large number of design variables (DVs). The proposed method uses a two-level optimization process: a high-level for a global optimization of a structure and a low-level for optimizations of sectioned computational stations of the structure.;The high-level optimization uses approximating functions to define target structural properties along the length of a structure, such as stiffness. The approximating functions are functions of the distance from the root of the structure that are defined using basis functions such as polynomials or exponential functions. The high-level DVs are the coefficients of the functions. Thus, the number of the high-level DVs is independent of the number of sections. Moreover, selecting smooth approximating functions help to obtain alternative designs with smooth shapes.;The low-level optimization finds an optimum parametric design, such as laminate layups, that matches with the target structural properties defined at the high-level optimization. At the low-level optimization, the proposed method uses an optimizer in each section. Each optimizer is independent of the optimizers in the other sections, thereby decomposing a large optimization problem into several small ones. This approach reduces the number of DVs per optimizer at the low-level optimization which reduces the design space of each section and eliminates the design space of coupling between sections. Once optimum designs are found from all sections at the low-level, the high-level solvers evaluate them for the entire structure.;The advantage of the proposed method is that it reduces the number of iterations of the high-level optimization because it considers a small number of high-level DVs. Computational efficiency increases because the computationally extensive high-level solvers need to be run less frequently to obtain an optimum solution. An additional advantage of the proposed method is that it produces many feasible alternatives.;Using example problems, the paper demonstrates that the proposed method converges faster in the early iterations, and generates more alternative designs with smooth geometry than traditional single-level methods.