Machine Learning

Realistic wind data are essential in developing, testing, and ensuring the safety of small unmanned aerial systems in operation. We present a non-intrusive reduced order modeling (NIROM) approach to replicate realistic wind data and predict wind fields.

The project aims to validate the hypothesis that ‘in-time’ gust awareness by a pilot or an autopilot, can enhance safety, efficiency and robustness of future autonomous aircraft operations in low altitude airspace.

We use Large-Eddy Simulation to understand the unsteady and highly coherent turbulent flow structures produced by buildings in neutral atmospheric boundary layer flow. Furthermore, we demonstrate a non-intrusive machine learning methodology to predict flow fields to augment safe wind-aware navigation systems for Unmanned Aerial Vehicles as a first step toward safely integrating UAS into existing aerial infrastructure.

We attempt to create accurate wind data for an urban environment using high-fidelity CFD data from Large Eddy Simulations (LES) and Convolutional Auto-Encoders (CAE) for non-linear surrogate modeling. The non-linear surrogate model extracts underlying non-linear modes from the high-resolution data snapshots, and the LSTM network trains on these specific modes.

A single building setup in neutral atmospheric conditions is considered a test case to demonstrate the method. The method relies on using Large Eddy Simulation data from a computational fluid dynamics simulation and a non-intrusive Reduced Order Modeling approach (ROM) coupled with Recurrent Neural Networks like Long Short Term Memory (LSTM).