In this paper, the use of XGBoost™ gradient boosted trees for the prediction of on-disk velocity in a coaxial rotor helicopter is analyzed for higher state data with extremely sparse data sets. In particular, the use of these machine learning algorithms was evaluated for their prediction capabilities when intermediary state data was both reduced and withheld. This analysis showed a distinct tradeoff between model characteristics in order to produce the best performing models, as has been consistent with previous work. Additionally, it was found that these models can perform sufficiently well to predict higher harmonic solutions across the rotor disk when only trained on lower state data. This result indicates that application to finite-state inflow modeling, and in particular, higher harmonic solutions, could help to significantly reduce the associated computational cost of higher harmonic solutions.

Document Type

Final Report

Author's School

McKelvey School of Engineering

Author's Department

Mechanical Engineering and Materials Science

Class Name

Mechanical Engineering and Material Sciences Independent Study

Date of Submission