Convolutional neural networks for automated damage recognition and damage type identification
Artículo
Open/ Download
Publication date
2018Metadata
Show full item record
Cómo citar
Modarres, Ceena
Cómo citar
Convolutional neural networks for automated damage recognition and damage type identification
Abstract
Recurring expenses associated with preventative maintenance and inspectionproduce operational inefficiencies and unnecessary spending. Human inspec-tors may submit inaccurate damage assessments and physically inaccessiblelocations, like underground mining structures, and pose additional logisticalchallenges. Automated systems and computer vision can significantly reducethese challenges and streamline preventative maintenance and inspection.The authors propose a convolutional neural network (CNN)‐based approachto identify the presence and type of structural damage. CNN is a deep feed‐for-ward artificial neural network that utilizes learnable convolutional filters toidentify distinguishing patterns present in images. CNN is invariant to imagescale, location, and noise, which makes it robust to classify damage of differentsizes or shapes. The proposed approach is validated with synthetic data of acomposite sandwich panel with debonding damage, and crack damage recogni-tion is demonstrated on real concrete bridge crack images. CNN outperformsseveral other machine learning algorithms in completing the same task. Theauthors conclude that CNN is an effective tool for the detection and typeidentification of damage.
Indexation
Artículo de publicación SCOPUS
Identifier
URI: https://repositorio.uchile.cl/handle/2250/169441
DOI: 10.1002/stc.2230
ISSN: 15452263
15452255
Quote Item
Structural Control and Health Monitoring, Volumen 25, Issue 10, 2018, Pages 1-17.
Collections