Machine learning based aerodynamic models for rotor blades

Computational tools that are both efficient and accurate can significantly improve and speed up the design of future vertical lift configurations. Current higher-fidelity CFD-based approaches that use discrete blades are accurate, but remain prohibitively expensive for routine evaluations in design environments. On the other hand, lower fidelity approaches that are based on lifting line comprehensive analysis, while efficient, are not sufficiently accurate for detailed design purposes. In this work, we develop a mid-fidelity approach using a Machine Learning (ML) surrogate model that can learn the aerodynamic behavior of rotor blades from data-rich higher fidelity CFD simulations. The ML model consists of a Deep Convolutional Neural Network (DCNN) trained with supervised learning techniques on a rich content data set. The data utilized for training is obtained from high fidelity simulation of the UH-60 Utility Tactical Transport Aircraft System (UTTAS) Maneuver, which encompasses a large range of operating environments for rotor blades. The model achieves an accuracy of 99.6% on training data and is further evaluated with similar but unseen inputs for validation against full CFD simulations and flight test data. The current mid-fidelity implementation is one order of magnitude faster compared to an engineering quality full CFD simulation and provides superior predictions than traditional lower fidelity approaches.