Wind Turbine Optimization Under Uncertainty with High Performance Computing

The widespread proliferation of wind turbines for power generation is driving the de-velopment of devices that are both ecient and minimally disruptive to surrounding com-munities. Optimized congurations can become ineective in the presence of unexpected operating scenarios, for example in the presence of insect or dust contamination, thus requiring recalibration and adjustments. Robust design procedures aim at limiting the po-tential impact of uncertainties on performance and are an eective risk-mitigation strategy. In this work a wind turbine multi-physics simulation tool EOLO that represents the aerodynamic loading, realistic turbulent wind scenarios, uid-structural interactions and acoustics is presented. The geometrical conguration of a realistic turbine is optimized un-der insect contamination (input uncertainty) using a novel Simplex Stochastic Collocation (SSC) method. This methodology provides a wealth of information regarding the overall performance statistics (output uncertainties) and forms the basis for the exploration of several robust design options. We introduce a new process that extends the usual concept of variance minimization. The proposed robust design strategy is coupled to a customized multi-objective genetic algorithm. The overall process requires a large number of compu-tations and therefore we developed a simulation environment LELAND to carry out these analysis on large scale parallel clusters. Leland is a dynamic scheduler that starting from a small initial set of computations automatically selects the new candidate simulations to be performed to increase the eciency of both the optimization procedure and the uncertainty quantication method. I. Introduction and motivations Wind turbine design continues to receive considerable attention due to the desire to increase energy pro-duction while limiting the environmental impact. Large scale horizontal turbines have benetted considerably from the use of simulation tools and computational optimization