Page 22 - Kỷ yếu hội thảo khoa học lần thứ 12 - Công nghệ thông tin và Ứng dụng trong các lĩnh vực (CITA 2023)

6

3 Experiment and Results

3.1 Experiment Setup

In this experiment, a real concrete block with dimensions of 1300 1300 1300 mm is
used, consisting of four planes labeled as A, B, C, and D. Each plane contains 36
damage locations that are artificially generated by a pencil-lead breaks [15] and serve
as the proposed damage sites, as illustrated in Figure 4. To collect the artificial AE
events generated from each crack source, a network of eight acoustic emission sensors
labeled from S1 to S8 is mounted at the surface of the concrete block. The
experiments are conducted multiple times to ensure a significant dataset is collected
for all cases.

Fig. 4. Experiment setup for 3D crack localization

The crack sources were repeated roughly 20 times (+/-) for each of the 144 crack
sources on the four planes, resulting in approximately 2,898 experiments. A total of
23,184 signals were obtained from the eight sensors utilized to detect the crack
sources. The acquired data was then divided into a training dataset of 80%, consisting
of 2,322 experiments, and a testing dataset of 20%, consisting of 576 experiments.
The data was collected by Korea Atomic Energy Research Institute (KAERI) in 2022.
When a crack occurs, it can be detected using AE sensors. AE sensors can detect
the sound waves generated by the crack's propagation. These sensors are placed on
the surface of the concrete and can detect even the smallest cracks, making them
particularly useful for early detection and monitoring of crack growth. The signals

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