RANS-based design optimization of dual-rotor wind turbines

• Published in: Engineering computations Vol. 35; no. 1; pp. 35 - 52
• Main Authors: Thelen, Andrew, Leifsson, Leifur, Sharma, Anupam, Koziel, Slawomir
• Format: Journal Article
• Language: English
• Published: Bradford Emerald Publishing Limited 01.01.2018; Emerald Group Publishing Limited
• Subjects: Accuracy; Algorithms; Approximation; Case 2; Computational efficiency; Computational fluid dynamics; Computational grids; Computer simulation; Design optimization; Energy industry; Flow velocity; Fluid flow; Investigations; Kriging interpolation; Mathematical models; Navier-Stokes equations; Optimization techniques; Partial differential equations; Pattern search; Reynolds averaged Navier-Stokes method; Rotors; Simulation; Trends; Turbines; Turbulence models; Two-equation turbulence model; Variables; Wind turbines; Wind power; Dual-rotor wind turbines; RANS simulations; Rotor design; Design optimization
• ISSN: 0264-4401; 1758-7077
• DOI: 10.1108/EC-10-2016-0354

Purpose An improvement in the energy efficiency of wind turbines can be achieved using dual rotors. Because of complex flow physics, the design of dual-rotor wind turbines (DRWTs) requires repetitive evaluations of computationally expensive partial differential equation (PDE) simulation models. Approaches for solving design optimization of DRWTs constrained by PDE simulations are investigated. The purpose of this study is to determine design optimization algorithms which can find optimal designs at a low computational cost. Design/methodology/approach Several optimization approaches and algorithms are compared and contrasted for the design of DRWTs. More specifically, parametric sweeps, direct optimization using pattern search, surrogate-based optimization (SBO) using approximation-based models and SBO using kriging interpolation models with infill criteria are investigated for the DRWT design problem. Findings The approaches are applied to two example design cases where the DRWT fluid flow is simulated using the Reynolds-averaged Navier−Stokes (RANS) equations with a two-equation turbulence model on an axisymmetric computational grid. The main rotor geometry is kept fixed and the secondary rotor characteristics, using up to three variables, are optimized. The results show that the automated numerical optimization techniques were able to accurately find the optimal designs at a low cost. In particular, SBO algorithm with infill criteria configured for design space exploitation required the least computational cost. The widely adopted parametric sweep approach required more model evaluations than the optimization algorithms, as well as not being able to accurately find the optimal designs. Originality/value For low-dimensional PDE-constrained design of DRWTs, automated optimization algorithms are essential to find accurately and efficiently the optimal designs. More specifically, surrogate-based approaches seem to offer a computationally efficient way of solving such problems.