Reliability-based control co-design of horizontal axis wind turbines

Wind energy is one of the fastest-growing energy sources due to its cleanness, sustainability, and cost-effectiveness. In the past, wind turbine design studies focused primarily on a sub-system or single-discipline design and analysis, including control, structural, aerodynamic, and electro-mechanical studies, for example. More recent studies formulated wind turbine design problems using multidisciplinary design optimization (MDO) strategies, with either static or dynamic system models, providing the potential for identifying system-level optimal designs. On the other hand, efforts have also been made to increase the reliability and robustness of wind turbines by accounting for various sources of uncertainty explicitly in the design process. In the presented study, the MDO formulation of wind turbine design problem has been extended to include both control system co-design and reliability considerations in an integrated manner. As a result, the optimal wind turbine design that has an optimal control solution and is robust to uncertainties can be obtained at an early design stage, which would benefit the controller design and maintenance design at latter phases. In this paper, the design of a horizontal axis wind turbine (HAWT) supported by a tubular tower is considered and formulated as a multi-objective control co-design problem with design parameter uncertainties and stochastic wind load. A physics-based multidisciplinary dynamic model of tubular-tower-supported pitch-controlled HAWT that captures the main design conflicts under extreme wind is provided and implemented, along with the necessary modifications to make nested control co-design comply with modern reliability-based design optimization structures, forming a new class of reliability-based co-design (RBCD) problems. In particular, we provide detailed discussions about RBCD problem formulations and implementation strategies, and with the HAWT design problem, we demonstrate the results and computational costs with integrated double-loop, single-loop, as well as decoupled methods.