[A-CG47] Promotion of climate and earth system sciences using manned/unmanned aircrafts

[ACG47-P03] Airflow Estimation using airborne Doppler lidar and Machine Learning

*Ryota Kikuchi¹, Masayuki Sato², Masahide Onji¹, Yoshiro Hamada¹, Nobuhiro Takahashi³ (1.JAXA Aviation Technology Directorate, 2.Kumamoto University, 3.Nagoya University)

Keywords:Aviation Safety, Atomospheric Turbulence, Airborne Doppler Lidar

Gust-alleviation systems that use airborne Doppler lidar technology are expected to enhance aviation safety by significantly reducing the risk of turbulence-related accidents. Accurate measurement and estimation of vertical wind velocity are crucial for the successful implementation of such systems. In this study, an airflow vector estimation algorithm based on data from airborne lidars is proposed and investigated for preview control to prevent turbulence-induced aircraft accidents in flight. Existing techniques assume that the wind field between lidars is homogeneous or can be functionally approximated. However, this assumption leads to performance degradation when dealing with complex turbulence fields. To address the issues, a machine learning-based method for estimating airflow, specifically a neural network-based method, is proposed, utilizing a methodology for generating turbulent flow data as the learning data. With the proposed method, airflow estimation can be extended to complex turbulent flow fields that were previously problematic for existing methods. This paper presents case studies and statistical evaluations of the proposed machine learning-based airflow estimation method and an existing method, with a focus on differences in predicted airflow field characteristics and their statistical prediction performance.