A Machine Learning Approach to Predicting Unsteady Transonic Flow of Pitching Airfoils

Abstract

Abstract A machine learning methodology has been proposed in this paper to study the unsteady transonic aerodynamics in the flutter regime. The methodology is based on a well-established regularization technique, and it compares very well with the data modeling approach proposed recently by the author in the prediction of the lift coefficient, cl, of NACA00 series airfoils over a range of reduced frequency. The present methodology has been extended to the prediction of the airfoil pitching moment coefficient, cpm, also. Just as in the case of the data model proposed earlier, the regularization-based machine learning model is trained on a subset of the considered reduced frequency range and a subset of the NACA00 series airfoils. The model predictions are in good agreement with the computational fluid dynamics (CFD) results, in the reduced frequency range for the selected test NACA00 profiles including those with a thickness typical of supercritical wing sections. The machine learning methodology presented here represents a new technology that can be used in the prediction of transonic flutter aerodynamics of wings using a strip theory approach. This new approach can be coupled with a simple finite element model such as a beam element model offering a rapidly implementable aeroelastic framework for the design of new transonic wings.