Session: New & Innovative Floating Designs I

Paper Number: 98334

98334 - Control Co-Design of Floating Offshore Wind Turbine Blade via Reinforcement Learning 

Offshore wind energy is a promising emission-free source of power, as its full technical potential can satisfy the entire U.S. energy demand. However, most of the energy lies in the water deeper than 60m, thus harvesting offshore wind energy has been challenging. Floating offshore wind turbines (FOWTs) have been developed by integrating the technologies involving onshore wind turbine towers and floating foundations, but their high levelized cost of energy, about three times higher than that of onshore wind turbines, has been the major bottleneck of widespread of the technology. To address the challenge, it is necessary to optimize FOWTs design as well as its controllers (e.g., blade pitch controller) to maintain its power generation while minimizing the fatigue loading due to the nonlinear dynamics involving unbalanced nonstationary wind/wave loading. In this paper, a control co-design framework is proposed, which utilizes reinforcement learning to learn a general design policy for FOWT blades and matching blade pitch controllers. Specifically, founded upon open-source toolsets for FOWTs design and control, including OpenFAST, a virtual environment is developed where an AI agent can design a turbine blade and matching pitch controller and obtain rewards based on the performance of the design. Based on the rewards provided to the agent, the agent gradually improves its design policy to improve the rewards. As the reward is designed to be high when the power generation relative to fatigue loading is high, the agent can learn how to design a good turbine blade design as well as a matching blade pitch controller. The feasibility of the proposed method is demonstrated in representative design load cases by providing an effective turbine blade design as well as a corresponding blade pitch controller reducing the fatigue loading to the turbine blade. The comparison between the proposed control co-design and the baseline (i.e., only blade pitch controller is optimized) is also presented. The proposed method provides a general control co-design for any application domain, given a realistic simulation environment and practical reward functions. Therefore, the proposed method can be applied to similar problems where the simultaneous design of the system and its controller is beneficial: unmanned aerial vehicle (UAV) design and heating, ventilation, and air conditioning (HVAC) system design.

Presenting Author: SungKu Kang Northeastern University