Abstract
Cross-flow turbines, also known as vertical axis turbines, use blades that rotate about an axis perpendicular to the incoming flow to convert the kinetic energy in moving fluid to mechanical energy. In this work, the performance of a two-turbine array in a recirculating water
channel is modeled using Gaussian process regression. In prior experiments, we optimized “coordinated control” set points (equal tip-speed ratios with an azimuthal phase offset between turbines) to maximize the power output 64 unique geometric configurations with the turbines counterrotating. While this approach identified promising configurations where turbine pair out-performed geometrically identical turbines in isolation, the experiments were timeconsuming to conduct. In this work, a Gaussian process regression model is initialized with a subset of random points from the geometric configuration space and returns confidence intervals for the full parameter space. Subsequently tested points are then chosen based on the regions of the parameter space model with the highest uncertainty. This is repeated until the model converges (i.e., the model is unchanged with additional points tested). Results are benchmarked against the experimental “truth”, but, in
future work, would actively guide experimental exploration of high-dimensional spaces.