Date of Award

Spring 2016

Project Type


Program or Major

Mechanical Engineering

Degree Name

Doctor of Philosophy

First Advisor

Martin Wosnik

Second Advisor

Kenneth Baldwin

Third Advisor

Diane Foster


Cross-flow (often vertical-axis) turbines (CFTs), despite being thoroughly investigated and subsequently abandoned for large scale wind energy, are seeing renewed interest for smaller scale wind turbine arrays, offshore wind, and marine hydrokinetic (MHK) energy applications. Though they are similar to the large scale Darrieus wind turbines, today's CFT rotors are often designed with higher solidity, or blade chord-to-radius ratios, which makes their behavior more difficult to predict with numerical models. Furthermore, most experimental datasets used for numerical model validation were acquired with low solidity rotors.

An experimental campaign was undertaken to produce high quality open datasets for the performance and near-wake flow dynamics of CFTs. An automated experimental setup was developed using the University of New Hampshire's towing tank. The tank's linear motion, control, and data acquisition systems were redesigned and rebuilt to facilitate automated cross-flow turbine testing at large laboratory (on the order of 1 meter) scale.

Two turbines were designed and built---one high solidity (dubbed the UNH Reference Vertical-Axis Turbine or UNH-RVAT) and one medium-to-low solidity, which was a scaled model of the US Department of Energy and Sandia National Labs' Reference Model 2 (RM2) cross-flow MHK turbine. A baseline performance and near-wake dataset was acquired for the UNH-RVAT, which revealed that the relatively fast wake recovery observed in vertical-axis wind turbine arrays could be attributed to the mean vertical advection of momentum and energy, caused by the unique interaction of vorticity shed from the blade tips.

The Reynolds number dependence of the UNH-RVAT was investigated by varying turbine tow speeds, indicating that the baseline data had essentially achieved a Reynolds number independent state at a turbine diameter Reynolds number $Re_D \sim 10^6$ or chord based Reynolds number $Re_c \sim 10^5$. A similar study was undertaken for the RM2, with similar results. An additional dataset was acquired for the RM2 to investigate the effects of blade support strut drag on overall performance, which showed that these effects can be quite significant---on the order of percentage points of the power coefficient---especially for lower solidity rotors, which operate at higher tip speed ratio. The wake of the RM2 also showed the significance of mean vertical advection on wake recovery, though the lower solidity made these effects weaker than for the UNH-RVAT.

Blade-resolved Reynolds-averaged Navier--Stokes (RANS) computational fluid dynamics (CFD) simulations were performed to assess their ability to model performance and near-wake of the UNH-RVAT baseline case at optimal tip speed ratio. In agreement with previous studies, the 2-D simulations were a poor predictor of both the performance and near-wake. 3-D simulations faired much better, but the choice of an appropriate turbulence model remains uncertain. Furthermore, 3-D blade-resolved RANS modeling is computationally expensive, requiring high performance computing (HPC), which may preclude its use for array analysis.

Finally, an actuator line model (ALM) was developed to attempt to drive down the cost of 3-D CFD simulations of cross-flow turbines, since previously, the ALM had only been investigated for a very low Reynolds number 2-D CFT. Despite retaining some of the disadvantages of the lower fidelity blade element momentum and vortex methods, the ALM, when coupled with dynamic stall, flow curvature, added mass, and end effects models, was able to predict the performance of cross-flow turbines reasonably well. Near-wake predictions were able to match some of the important qualitative flow features, which warrants further validation farther downstream and with multiple turbines. Ultimately, the ALM provides an attractive alternative to blade-resolved CFD, with computational savings of two to four orders of magnitude for large eddy simulation and RANS, respectively.