Spectral mass gauging of unsettled liquids is a technique that
could aid in human space exploration. Researchers have shown that through the use of Weyl's Law, it is
possible to predict the amount of liquid that exists in a container in a low-gravity environment where
normal, dry-wet sensors are not adequate. This is done through the counting of peaks caused by acoustic
resonances in the liquid when observed over a specific frequency range.
Currently, this peak counting is being done manually by a human. In this project, we explore the
automation of this peak counting by using neural networks. We explore many model architectures including
convolutional neural networks (CNN), multiple recurrent neural networks (RNN) such as LSTM and GRU, and
ensemble models. We show that in a small spectral window, neural networks can count the number of peaks
with an accuracy greater than 90%.
Read the Paper
See Presentation Slides
This project was driven through the Cal Poly Data Science 2020 program in cooperation with NASA. Three students: Nathan Philliber, Mateo Ibarguen, and Steven Bradley worked with Dr. Michael Khasen and Dr. Matteo Corbetta to explore and develop machine learning methods to classify spectral window data — the ultimate goal being to develop a tank architecture capable of measuring fluid volume in low-gravity environments.