Page 18 - Kỷ yếu hội thảo khoa học lần thứ 12 - Công nghệ thông tin và Ứng dụng trong các lĩnh vực (CITA 2023)

2

AECWT-3DR-Net: Damage Localization Network for

Concrete Structures Using Acoustic Emission Sensors

Van Vy 1[0009-0006-3087-7462] and Hyungchul Yoon 2[0000-0003-2558-6303]

1 Ph.D. candidate, Chungbuk National University, South Korea
2 Corresponding author, Associate Professor, Chungbuk National University, South Korea
hyoon@cbnu.ac.kr

Abstract. In the construction industry, the deterioration of structures is a
significant concern. To detect cracks in concrete structures, acoustic emission
sensors are commonly used. The traditional approach relies on measuring the
time of arrival, time difference of arrival, and received signal strength indicator.
However, conventional methods are prone to error in the presence of
inhomogeneous materials. In this research, we introduce a new method that
employs deep learning techniques to detect cracks using acoustic emission
sensors. The aim of this approach is to enhance the accuracy of crack detection
while automating the process. The proposed method entails the following steps:
capturing signals from acoustic emission sensors and then converting them into
a time-frequency representation using continuous wavelet transform. These
representations are fed into a convolutional neural network that is specifically
designed to locate the crack. Finally, the convolutional neural network is trained
to predict the coordinates of the crack. The proposed method's effectiveness and
advancements were confirmed through experiments conducted on a concrete
block that had a crack artificially created by pencil-lead breaks.

Keywords: Damage Detection, Acoustic Emission Sensor, Deep Learning.

1 Introduction

Ensuring safety is a vital priority in the construction industry, particularly due to
the potential for structures to deteriorate with time. The conventional approach for
identifying cracks in solid materials is through the application of acoustic emission
(AE) sensors. Initially, AE signals were analyzed based on parameters, and
mathematical calculations were used to determine crack locations. However, with the
advent of deep learning networks, AE sensor signals can now be integrated and
analyzed to estimate the location of cracks more accurately [1-5].
The traditional approach to identifying cracks with AE sensors relies on
measuring time of arrival (ToA) [6] and time difference of arrival (TDoA) [7]. These
techniques are based on precise calculations of the time it takes for a signal to travel
from the location of the crack to the sensor, and the speed at which the signal travels.
However, the accuracy of these methods can be limited due to the impact of

CITA 2023 ISBN: 978-604-80-8083-9