Machine Learning Framework for Predicting Shock Locations on Rotorcraft Airfoils

The rotor blade tip can operate at the transonic regime and location of transonic shock is important. Most mid/lower fidelity comprehensive codes for rotor blade analysis uses airfoil tables. The airfoil tables are usually generated using Computational Fluid Dynamics (CFD) codes, and the it's crucial to capture the tansonic shock to get accurate integrated quantities. This study focuses on predicting the location of transonic shocks on rotorcraft airfoils using a machine-learning-based surrogate model. A comprehensive methodology is developed to parameterize airfoil geometries, generate a design of experiments, and train neural network models for predicting shock locations. CFD simulations are performed on 560 unique airfoils using an in-house solver (HAM2D) over a range of angles of attack at Mach 0.7 to obtain a large sample of shock data. A combination of classification and regression models are employed to detect the presence of shocks and predict the location of shocks. The models demonstrate high accuracy, effectively predicting shock occurrence and locations, with regression errors well within acceptable tolerances for localized mesh refinement. The proposed framework provides a robust and efficient approach for predicting shock locations, which is essential for optimizing rotorcraft performance under transonic conditions.