Data-Driven Approach For Acoustic Source Localization

Authors

  • Arnav Joshi Indian Institute of Technology Indore, India and University of Waterloo, Canada
  • Hamid Daryan University of Waterloo, Canada
  • Jean-Pierre Hickey University of Waterloo, Canada

Abstract

The spatial and temporal characteristics of the dominant hydrodynamic contributors to far-field aeroacoustic noise are identified through a data-driven approach. This approach is proposed as an alternative to traditional acoustic source localization methods such as beamforming which suffer from poor resolution at lower source frequencies. A training database was developed using analytically-defined monopoles as static and moving acoustic sources. These sources were randomly distributed over a scanning grid of fixed size and a microphone array was simulated to gather information about the sources in the form of the Cross-Spectral Matrix (CSM). The CSM was used as an input feature to train a Convolutional Neural Network (CNN) to identify position, strength, and velocity of the sources with far greater accuracy than traditional methods. Citing a particular result, the network could detect all the 6 sources at a frequency of 8000 Hz spread randomly over a 12x12 grid with just a handful of training (100 epochs). The performance was equally good for lower frequencies. Networks built for more realistic cases such as detecting sources with different strengths, or, detecting a random number of sources also yielded promising results. This approach is then used to identify acoustic noise sources in simplified vortex pairing.

Author Biographies

Arnav Joshi, Indian Institute of Technology Indore, India and University of Waterloo, Canada

4th year undergraduate student at Department of Mechanical Engineering, Indian Institute of Technology Indore interning at the University of Waterloo, Canada.

Hamid Daryan, University of Waterloo, Canada

PhD student at the Department of Mechanical and Mechatronics Engineering, University of Waterloo, Canada

Jean-Pierre Hickey, University of Waterloo, Canada

Professor at the Deparment of Mechanical and Mechatronics Engineering, University of Waterloo, Canada

Additional Files

Published

2021-08-17

How to Cite

1.
Joshi A, Daryan H, Hickey J-P. Data-Driven Approach For Acoustic Source Localization. Canadian Acoustics [Internet]. 2021 Aug. 17 [cited 2024 May 27];49(3):64-5. Available from: https://jcaa.caa-aca.ca/index.php/jcaa/article/view/3905

Issue

Section

Proceedings of the Acoustics Week in Canada