Classifying bubble cavitation with machine learning trained on physical models
Loading...
Date
2024
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
When an air filled bubble inside a liquid is excited by an acoustic signal the oscillations may cause cavitation where the bubble collapses and disintegrates into multiple smaller bubbles. This is a crucial phenomenon in various engineering applications, such as underwater noise and biomedical engineering. Accurately modeling cavitation is essential for optimal design of acoustical systems. In this study, our objective is to develop a machine learning model capable of classifying bubble oscillation as stable or transient cavitation. We implemented numerical solvers for the Rayleigh-Plesset, Keller-Miksis, and Gilmore differential equations to describe bubble oscillations with different mathematical models. These models takes the bubble radius, acoustic pressure, frequency, and temperature as input variables and provide a time series of the bubble radius. We used multiple thresholds to distinguish between stable and transient cavitation, based on variables such as the maximum radius, maximum velocity, acoustic emission, inertial and pressure functions for each model. We created a training dataset on a range of relevant physical scenarios. The machine learning algorithm combines different thresholds and models to predict the expected cavitation type. Our machine learning approach allows for fast predictions and uncertainty estimates of cavitation type on a wide range of scenarios. This approach offers greater robustness compared to numerical solution methods, leveraging four thresholds and three equations for enhanced accuracy.
Description
Tesis (Master of Science in Engineering)--Pontificia Universidad Católica de Chile, 2024
Keywords
Cavitation, Machine-learning, Differential equations, Bubbles, Classification