Mid-Fidelity Performance Analysis for Fixed-Pitch Speed-Controlled Multirotor Aircraft
Restricted (Penn State Only)
- Author:
- Cornelius, Jason
- Graduate Program:
- Aerospace Engineering
- Degree:
- Doctor of Philosophy
- Document Type:
- Dissertation
- Date of Defense:
- May 10, 2023
- Committee Members:
- Amy Pritchett, Program Head/Chair
Daniel Haworth, Outside Unit & Field Member
Sven Schmitz, Chair & Dissertation Advisor
Jacob Langelaan, Major Field Member
Jose Palacios, Major Field Member
Edward Smith, Major Field Member
Larry Young, Special Signatory - Keywords:
- Helicopter
Multirotor
CFD
Rotorcraft
High Performance Computing
GPU
Coaxial
Coaxial Rotor
Machine Learning
Surrogate Models
Dragonfly
Drone
UAS
Urban Air Mobility
Unmanned Aerial Systems
BEMT
Aerodynamics
Aircraft
C81 Tables
OVERFLOW
ARC2D
Airfoil Performance
NASA
Langley TDT
Langley 14-by 22-ft. Subsonic Tunnel Facility
Langley Transonic Dynamics Tunnel
Hybrid BEMT-URANS CFD
Linux Bash Scripting
GPU CFD Acceleration
Gaussian Process Regression
Neural Networks
Conceptual Design
Preliminary Design - Abstract:
- This dissertation presents new methodology towards increasing the capabilities for performance analysis of fixed-pitch speed-controlled multirotor aircraft with an emphasis on coaxial rotor systems. This vehicle configuration is seeing much interest across many sectors of commercial industry, public transport, military applications, and even extra-terrestrial space exploration. Although these aircraft can be analyzed with conventional approaches, the specific details of the rotor system involved have created opportunities for increases in computational efficiency and accuracy. This work starts with the presentation of an improved BEMT rotor modeling methodology for variable-speed rotors. A novel approach for developing the C81 rotor input decks as a function of not only the blade radial location but also the rotor RPM is described. Thirteen distinct C81 tables are used for the most discretized rotor model, each with seven distinct rotor speeds captured in the tables. More than 3,000 two-dimensional OVERFLOW simulations are used to create the rotor model. Validation data was used from experimental testing of a coaxial rotor system at the NASA Langley Transonic Dynamics Tunnel and the 14- by 22-ft. Subsonic Tunnel Facility. The novel rotor modeling approach inside a hybrid BEMT-URANS flow solver was validated to within 5-10% accuracy of steady rotor thrust and torque measurements. The next major portion of this dissertation focuses on mass GPU parallelization of a commercial off-the-shelf CFD solver. Specialized GPU accelerated machines developed through this dissertation were combined with a custom Linux bash script to achieve a 41 times speedup over the off-the-shelf program. The last major contribution of this work involves leveraging machine learning to develop rotor performance surrogate models. Metrics for the training and testing of the resulting models is reported, and the best models achieved R-squared values of 0.99 using a 5-fold cross validation strategy. Gaussian process regression was found to be the most effective, and the resulting model can be used to estimate the rotor performance of a new condition in millionths of a second. The surrogate models are fast enough for real time simulation and capture the highly non-linear and high dimensionality of the underlying data. The methodologies presented in this work have obtained large increases in both accuracy and computational efficiency for predicting the steady rotor performance of multirotor configurations using hybrid BEMT methods. Accuracy on the order of 5-10% has been achieved with several order of magnitude reduction in computational cost per simulation. These developments enable rapid conceptual and preliminary design studies to be conducted and can also be used to build large rotor performance lookup tables for many applications relevant to rotorcraft design and testing.