Published: Jul 27, 2023 by Daning Huang
We hosted Dr. Sicheng He from University of Tennessee, Knoxville, for a seminar in our department. Dr. He’s expertise lies in the high-fidelity optimization of multi-disciplinary dynamical systems.
Abstract Multidisciplinary design optimization (MDO) has been widely used in the aerospace industry. Problems with millions of state variables and hundreds of design variables are solved. However, from the perspective of dynamical systems, most of the previous research focused on equilibrium point problems. To enable MDO for general dynamical systems, we develop efficient computation methods based on Jacobian-free Newton-Krylov (JFNK), algorithmic differentiation (AD), and adjoint methods. Applications in wind turbine design and aeroelastic limit cycle oscillation (LCO) (or flutter) suppression are discussed. Besides, control co-design (CCD) is an emerging field that aims to incorporate optimal control as one discipline into MDO to substitute the traditional sequential design workflow used in the industry. CCD is a powerful idea that can exploit the complex interaction between the plant and the control variables. However, most of the previous research focused on low-fidelity models. To enable CCD for high-fidelity models, we propose an efficient sensitivity computation method based on the coupled adjoint method. A quadrotor blade CCD problem is presented where the control cost is reduced by about 50% from a sequentially optimized design. Finally, we discuss the data-driven method in aerodynamic shape optimization. Using our neural network model, we can conduct real-time airfoil shape optimization with accurate lift, drag, and moment coefficients (<1% relative error) for both subsonic and transonic regimes.