A platform for research: civil engineering, architecture and urbanism
A platform for research: civil engineering, architecture and urbanism
High-accuracy rebar position detection using deep learning–based frequency-difference electrical resistance tomography
Abstract Rebar corrosion is one of the most critical mechanisms causing structural deterioration in reinforced concrete structures. However, rebar corrosion assessment is difficult in that typical non-destructive testing methods have limitations in accurately detecting rebar positioning in concrete. This study presents high-accuracy rebar position detection using a deep learning–based electrical resistance tomography (ERT) technique. Two data sets were prepared as input data: (1) the original circular ERT images in a Cartesian coordinate system and (2) the transformed rectangular ERT images in a polar coordinate system. The proposed convolutional neural network (CNN) model successfully distinguished rebar position from ERT images. Most of the radial and angular positions of the rebar were accurately identified by the model, despite rebar's wide distribution of high conductivity in the raw ERT images. Notably, the detection performance clearly depended on the coordinate types in the ERT images, whether they were Cartesian or polar coordinates.
Highlights The combination of ERT and CNN model successfully improved detection accuracy. Raw ERT images were coordinate-transformed to suit the target information well. The detection accuracies were 82.6% (Cartesian coordinates in ERT images) and 95.6% (polar coordinates), respectively. Validation using X-ray CT demonstrated good agreement of the detection results.
High-accuracy rebar position detection using deep learning–based frequency-difference electrical resistance tomography
Abstract Rebar corrosion is one of the most critical mechanisms causing structural deterioration in reinforced concrete structures. However, rebar corrosion assessment is difficult in that typical non-destructive testing methods have limitations in accurately detecting rebar positioning in concrete. This study presents high-accuracy rebar position detection using a deep learning–based electrical resistance tomography (ERT) technique. Two data sets were prepared as input data: (1) the original circular ERT images in a Cartesian coordinate system and (2) the transformed rectangular ERT images in a polar coordinate system. The proposed convolutional neural network (CNN) model successfully distinguished rebar position from ERT images. Most of the radial and angular positions of the rebar were accurately identified by the model, despite rebar's wide distribution of high conductivity in the raw ERT images. Notably, the detection performance clearly depended on the coordinate types in the ERT images, whether they were Cartesian or polar coordinates.
Highlights The combination of ERT and CNN model successfully improved detection accuracy. Raw ERT images were coordinate-transformed to suit the target information well. The detection accuracies were 82.6% (Cartesian coordinates in ERT images) and 95.6% (polar coordinates), respectively. Validation using X-ray CT demonstrated good agreement of the detection results.
High-accuracy rebar position detection using deep learning–based frequency-difference electrical resistance tomography
Jeon, Dongho (author) / Kim, Min Kyoung (author) / Jeong, Yeounung (author) / Oh, Jae Eun (author) / Moon, Juhyuk (author) / Kim, Dong Joo (author) / Yoon, Seyoon (author)
2021-12-23
Article (Journal)
Electronic Resource
English
Bond in high-strength concrete - influence of the rebar position
| Tema Archive | 2002
Monitoring of Rebar Corrosion by Means of the Electrical Resistance Probes
| British Library Conference Proceedings | 2007
Effects on corrosion resistance of rebar subjected to deep cryogenic treatment
| British Library Online Contents | 2017