|
Title |
A Study of Impact-Echo Experiments and Deep Learning Models Applied for Void Investigation within Plastic Ducts
|
|
Authors |
이성호(Seong-Ho Lee) ; 김기현(Ki-Hyun Kim) ; 윤석구(Seok-Goo Youn) |
|
DOI |
https://doi.org/10.4334/JKCI.2022.34.6.579 |
|
Keywords |
충격공진법; 딥러닝; 합성곱 신경망; 장단기 기억 신경망; 플라스틱 덕트 Impact-Echo; deep learning; CNN; LSTM; plastic duct |
|
Abstract |
A PSC bridge is a structure in which prestress is introduced into the concrete in advance. In a PSC bridge, it is important to investigate voids in the ducts because they cause corrosion of strands. Recent studies have been conducted which applied deep learning models to Impact-Echo (IE) which is a non-destructive testing method, to investigate voids in PSC bridges. However, few studies have been conducted using the LSTM model, and the one-dimensional CNN model, to find the voids located inside a circular plastic duct. Therefore, this study evaluated the accuracy of void detection using the LSTM model and CNN model, and a combined CNN and LSTM model, for data collected during the IE experiments. Based on the test results, it was determined that the CNN-LSTM model was the most accurate deep learning model, with 93 % accuracy, among the three tested models.
|