Audio-based shape recognition is often used in robotics and navigation. With the popularity of in-ear earphones and tiny robots and drones with low-power microphones, we see an opportunity where acoustic sensing can be used to monitor small enclosed spaces, e.g., ear canals, pipes, and machinery, for human, infrastructure, and machine health monitoring. This paper studies the opportunity of using acoustic sensing to identify and monitor various small hollow objects from the inside. We propose an algorithm that fuses signal processing and machine learning techniques to distinguish between three shapes and assess the health of the shapes. The results show that under ideal conditions, audio can distinguish between three shapes with 90% accuracy, and distinguish between three shapes under non-ideal conditions(different heights, background noises, and setups) with 70% accuracy. We further determine shape deformation with 100% accuracy and identify the side of the deformation with 90% accuracy. Our results demonstrate that as individual features acoustic Loudness, power Spectrogram, mel Spectrogram, and spectral Entropy have great results. However the best combination of features is zero crossing rate, Acoustic Loudness, and mel Spectrogram. Additionally decreasing the number of microphones decreases the accuracy of the experiments. Furthermore, attempting localization techniques in small hollow areas to determine the height of the shape is difficult due to sound reflection and refractions. Localization through triangulation yields results with a high variance, while localization through a gradient boosted machine requires a large dataset.

• This report represents the work of one or more WPI undergraduate students submitted to the faculty as evidence of completion of a degree requirement. WPI routinely publishes these reports on its website without editorial or peer review.

Creator

• Rothstein, Michael
• Liu, Zhuolin

Subject

• Health
• Computing
• Innovation
• Project-Based Learning
• Infrastructure
• Industry

Publisher

• Worcester Polytechnic Institute

Identifier

• E-project-042423-153701
• 104666

Keyword

• Maintenance
• Signal Processing
• Machine Learning
• Health care
• Audio
• Object Recognition

Advisor

• Islam, Bashima

Year

• 2023

UN Sustainable Development Goals

• 3 - Good Health and Well-being
• 9 - Industry, Innovation and Infrastructure

Date created

• 2023-04-24

Resource type

• Major Qualifying Project

Major

• Electrical & Computer Engineering

Source

• E-project-042423-153701

Rights statement

• In Copyright

Last modified

• 2023-06-22