Hybrid Experimental—Numerical Methodology For NACA 0012 Trailing Edge Noise with Auxiliary Machine Learning Based Modeling

Date of Award

5-2026

Degree Name

Master of Science

Department

Mechanical and Aerospace Engineering

First Advisor

Shiva Om Bade Shrestha, Ph.D.

Second Advisor

Tianshu Liu, Ph.D.

Third Advisor

Alvis Fong, Ph.D.

Keywords

Large eddy simuation (LES), machine learning, NACA 0012 airfoil, particle image velocimetry (PIV), passive noise control, trailing edge noise

Access Setting

Masters Thesis-Abstract Only

Restricted to Campus until

5-1-2036

Abstract

Understanding the physical mechanisms underlying trailing edge aeroacoustic noise is essential for the development of effective noise mitigation technologies in applications such as wind turbines and aircrafts. Accurate characterization of the coupled acoustic and aerodynamic fields is therefore important to properly discern dominant noise generation mechanisms and evaluate probable design modifications. In this study, a hybrid experimental–numerical method-ology was utilized to investigate trailing edge noise from a NACA 0012 airfoil at a low Reynolds number of approximately 1.1×105. Experiments were conducted in a low-speed anechoic wind tunnel using three-dimensional (3D) printed airfoils. Acoustic pressure fluctuations were ac-quired using microphones and a data acquisition system, and the measured voltage signal was converted for frequency domain analysis. To assess noise mitigation potential, channel based trailing edge configurations with varying diameters and exit angles were compared against the baseline geometry.

Direct acoustic simulations were performed using a pressure-based solver with Large Eddy Simulation (LES) and a Smagorinsky-Lilly subgrid scale model to capture the aeroacoustic behavior of the baseline and channel-based configurations. Particle image velocimetry (PIV) was utilized to visualize the wake near the trailing edge and extract spatially resolved turbulence statistics to relate flow behavior to the acoustic spectra. To reduce reliance on time consuming experimental and numerical methods, an auxiliary machine learning strategy was introduced. CatBoost, Deep Neural Network (DNN) and Support Vector Regression (SVR) models incorporating frequency domain feature engineering and metaheuristic hyperparameter optimization were trained using the experimentally obtained acoustic spectra to predict airfoil self-noise characteristics. The results demonstrated that CatBoost achieved the strongest predictive performance, with Fourier enhanced feature representations yielding consistent improvements for both broadband and tonal components.

Overall, this work provides a combined methodology for experimental characterization, numerical validation, and selective machine learning based modeling to study trailing edge noise and supports noise mitigation strategies for low Reynolds number airfoils.

This document is currently not available here.

Share

COinS