Machine Learning-Based Surrogate Modeling for Aerodynamic Loads Predictions

Abstract

The rapid prediction of loads for engineering applications is of high interest in the aerodynamic community. An approach for obtaining these rapid predictions is through the use of surrogate modeling. Surrogate models enable a faster turn-around time for various engineering needs that high-fidelity computational models cannot accommodate. A machine learning (ML) framework to support surrogate modeling of integrated aerodynamic loads predictions for aircraft is investigated in this effort. The ML framework includes core Deep Neural Network (DNN) components built to support surrogate models for both steady and unsteady aerodynamics. A vital aspect of a successful surrogate model is the prediction accuracy when non-linear flow phenomena such as flow separation and transonic effects impact the flowfield. For highly separated flows – such as dynamic stall – a two-step, novel physics-state predictor approach is laid out. The dual framework employs an intermediate physics-state predictor that enables the surrogate model to more accurately model dynamic separated and non-linear flow patterns. Parameter studies are conducted to investigate the impact of different input feature sets on the surrogate model's predictive capabilities. A discussion on the potential benefits of incorporating Convolutional Neural Network (CNN) based DNN architectures for the surrogate model is also included. Initial 2D/3D verification results for steady and unsteady aerodynamic problems are presented. The eventual goal is to develop trusted surrogate models for real-time system assessments supporting Digital Engineering initiatives.