Home // SMART 2025, The Fourteenth International Conference on Smart Cities, Systems, Devices and Technologies // View article

Deep Learning-based Failure Detection for Safety Diagnostics of Hydrogen Storage Vessels

Authors:
Dongju Kim
Da-Hyun Kim
Hyo-Jin Kim
Young-Joo Suh

Keywords: Hydrogen Storage Vessels; Acoustic Emission; Multimodal; Deep-learning.

Abstract:
Hydrogen is a clean energy source that is essential for responding to climate change and ensuring energy security. Typically, hydrogen storage vessels are exposed to high pressure environments, which can pose an immediate risk of explosion in the event of failure. Therefore, technologies are needed to detect and resolve failures early through diagnostics of hydrogen storage vessels. In this paper, we propose a deep learning-based multimodal failure detection technique to ensure the safety of hydrogen storage vessels. To develop the failure detection technique, we first performed tensile tests on the storage vessel material to collect Acoustic Emission (AE) signals, and also collected failure and normal data based on tensile load graphs. The Synthetic Minority Over-sampling Technique (SMOTE) method was applied to solve the data imbalance. Finally, we developed a multimodal deep learning model using time-domain waveforms and frequency spectra for failure detection, and the proposed method achieved an accuracy of 99.19% and an F1 score of 0.9733, demonstrating excellent failure detection performance. Furthermore, we confirmed that the proposed method shows better performance than using only time-domain waveforms or frequency spectra, and we expect that this research will contribute to the safety diagnosis and maintenance of hydrogen storage vessels.

Pages: 10 to 15

Copyright: Copyright (c) IARIA, 2025

Publication date: April 6, 2025

Published in: conference

ISSN: 2308-3727

ISBN: 978-1-68558-251-7

Location: Valencia, Spain

Dates: from April 6, 2025 to April 10, 2025