Inverse design of a cyclorotor using Bayesian inference and CFD simulations

Abstract

Within the context of the new wave of eVTOL aircraft development, Cyclotech GmbH has been working on the development of cyclorotors, a type of propeller consisting of rotating blades along a common longitudinal axis, where the pitch angle of the blades is actively controlled through what is known as a pitch curve. This thesis proposes a methodology to solve an inverse design of this pitch curve, which results in a certain target lift distribution. A framework using Bayesian inference is proposed, altogether with CFD simulations for the aerodynamic solution of the cyclorotor. To reduce the number of CFD evaluations and improve the framework efficiency, gradient descent methods are used altogether with semi-transient adjoint CFD gradients. Furthermore, low-fidelity information is added to the framework through the definition of a prior model to help speed up the inference. Using this methodology, three different test cases, from lower to higher complexity, are defined and successfully solved.