Optimization of a low speed wind turbine using support vector regression

Wise, John Nathaniel (2009-03)

Thesis (MScEng (Mechanical and Mechatronic Engineering))--University of Stellenbosch, 2009.


NUMERICAL design optimization provides a powerful tool that assists designers in improving their products. Design optimization automatically modifies important design parameters to obtain the best product that satisfies all the design requirements. This thesis explores the use of Support Vector Regression (SVR) and demonstrates its usefulness in the numerical optimization of a low-speed wind turbine for the power coe cient, Cp. The optimization design problem is the three-dimensional optimization of a wind turbine blade by making use of four two-dimensional radial stations. The candidate airfoils at these stations are selected from the 4-digit NACA range. A metamodel of the lift and drag coe cients of the NACA 4-digit series is created with SVR by using training points evaluated with XFOIL software. These SVR approximations are used in conjunction with the Blade Element Momentum theory to calculate and optimize the Cp value for the entire blade. The high accuracy attained with the SVR metamodels makes it a viable alternative to using XFOIL directly, as it has the advantages of being faster and easier to couple with the optimizer. The technique developed allows the optimization procedure the freedom to select profiles, angles of attack and chord length from the 4-digit NACA series to find an optimal Cp value. As a result of every radial blade station consisting of a NACA 4-digit series, the same lift and drag metamodels are used for each station. This technique also makes it simple to evaluate the entire blade as one set of design variables. The thesis contains a detailed description of the design and optimization problem, the implementation of the SVR algorithm, the creation of the lift and drag metamodels with SVR and an alternative methodology, the BEM theory and a summary of the results.

Please refer to this item in SUNScholar by using the following persistent URL: http://hdl.handle.net/10019.1/2737
This item appears in the following collections: