Session: 06-01: Machine Learning and Statistical Methods in NDE I

Paper Number: 147800

147800 - Enhancing Liquid Characterization in Sealed Containers: Integrating Swept-Frequency Acoustic Interferometry With Convolutional Neural Networks 

Abstract: 

Swept-Frequency Acoustic Interferometry (SFAI) is a nonintrusive liquid characterization technique developed specifically for detecting and identifying liquids inside sealed containers. The SFAI technique uses analytical expressions to determine sound speed, density and sound attenuation of a liquid inside a container from the frequency-domain obtained over a wide frequency range.  However, depending on the acoustic characteristics resulted from the sensor/containers/liquid system configuration, retrieving accurate parameters (in particular, density and attenuation) may be challenging. In this study, we investigate a novel SFAI approach leveraging the Convolutional Neural Networks (CNNs) framework as an alternative method for retrieving acoustics properties of the liquid from frequency-domain signal. A multi-layered analytical model was used to generate a vast amount of labeled signal using various sensor/containers/liquid configurations. The CNN model was trained using two strategies: 1) trained with the synthetic data and tested on experimental data. 2) trained with a combination of synthetic and experimental data and tested on experimental data. The results showed the CNN model as useful approach for retrieving physicochemical properties of fluids. The CNN model trained on the combined experimental and synthetic dataset has shown to effectively enlarge the dataset and enhance the robustness and generalization of the CNN model. Superior performance was achieved by the CNN model in density and attenuation estimation compared to traditional techniques. This innovative approach not only improves the accuracy of some liquid properties but also reduces reliance on manual signal processing and a priori knowledge. By integrating synthetic data generation with advanced machine learning techniques, this study represents a significant advancement in noninvasive fluid characterization inside sealed containers, promising broader applications across various industries.

Presenting Author: Daniel Pereira Los Alamos National Laboratory

Presenting Author Biography: Daniel Pereira was born in Brazil, in 1986. He received the B.Sc. degree in materials engineering and the M.Sc. degree in materials and metallurgy from the Federal University of Rio Grande do Sul, Porto Alegre, Brazil, in 2011 and 2014, respectively, and the Ph.D. degree in mechanical engineering from the École de Technologie Supérieure, Montreal, QC, Canada, in 2019. He is currently a postdoctoral researcher at Los Alamos National Laboratory. His current research interests include non-destructive testing, numerical simulations, signal processing and machine learning.

Authors:

Daniel Pereira Los Alamos National Laboratory
Eric Sean Davis Los Alamos National Laboratory
Pavel Vakhlamov Los Alamos National Laboratory
Abhishek Saini Los Alamos National Laboratory
John Greenhall Los Alamos National Laboratory
Gonzalo Seisdedos Los Alamos National Laboratory
Cristian Pantea Los Alamos National Laboratory