Machine Learning Methods for Sound Power Minimization
Loading...
Authors
Dossett, Wesley C.
Date
Type
thesis
Language
eng
Keyword
Mechanical Engineering , Design Optimization , Acoustics , Machine Learning
Alternative Title
Abstract
Design analysis and optimization for acoustics is a challenging task for aerospace engineers. Elevated noise levels can lead to negative effects on passenger and crew health indices. Radiated sound power is a broad measure of acoustic performance of a design. Existing sound power optimization methods are limited to tonal frequencies and low frequency bands. It is computationally difficult to integrate broadband sound power minimization into numerical optimization algorithms. Machine learning is a promising avenue for sound power minimization as it is effective at finding and exploiting patterns in complex data. This work proposes two machine learning methods for performing broadband sound power design optimization.
First, a residual neural network was trained using supervised learning to estimate the broadband average sound power of a specified design. The first method implemented the residual network into a traditional topology optimization framework to perform the evaluation and sensitivity analysis steps. The second method implemented the residual network into a reinforcement learning framework. Here, an agent network was trained to synthesize designs with minimal radiated sound power. Both methods were applied to design the thickness distribution of a flat aluminum panel and stiffened panel. These machine learning methodologies were compared to a conventional acoustics optimization methodology which used equivalent radiated power as a proxy for sound power.
In both studies, the reinforcement learning agent consistently generated designs with lower sound power than all designs in the dataset used to train the residual network, the designs generated using the first machine learning method, and designs generated using the conventional optimization approach. Future work may include improving generalization of the machine learning model and expanding its capabilities to more complex problems.
Description
Citation
Publisher
License
Queen's University's Thesis/Dissertation Non-Exclusive License for Deposit to QSpace and Library and Archives Canada
ProQuest PhD and Master's Theses International Dissemination Agreement
Intellectual Property Guidelines at Queen's University
Copying and Preserving Your Thesis
This publication is made available by the authority of the copyright owner solely for the purpose of private study and research and may not be copied or reproduced except as permitted by the copyright laws without written authority from the copyright owner.
Attribution 4.0 International
ProQuest PhD and Master's Theses International Dissemination Agreement
Intellectual Property Guidelines at Queen's University
Copying and Preserving Your Thesis
This publication is made available by the authority of the copyright owner solely for the purpose of private study and research and may not be copied or reproduced except as permitted by the copyright laws without written authority from the copyright owner.
Attribution 4.0 International