A platform for research: civil engineering, architecture and urbanism
A platform for research: civil engineering, architecture and urbanism
Machine Learning-Based Structural Damage Identification Within Three-Dimensional Point Clouds
https://orcid.org/http://orcid.org/0000-0002-2368-3962
https://orcid.org/http://orcid.org/0000-0002-8642-2217
Damage identification via remotely sensed data amid routine structural inspections or assessments in the aftermath of extreme events has been the subject of intensive research in recent years. By analyzing remotely sensed data, these tasks can be performed rapidly, safely, and economically while maintaining high accuracy and objectivity. Therefore, many methods have been proposed to detect damage from remotely sensed point clouds of the civil structures. These methods use various feature extraction techniques based on geometry and other spectral information and classification based on supervised or unsupervised learning algorithms. However, the proposed solutions are typically optimized to detect particular types of damage or group various types of damage as a single class, limiting the application of these methods. This article proposes an advanced workflow to identify damage from point clouds by extracting semantic information using supervised and unsupervised learning methods. The supervised learning method is comprised of a deep learning model developed based on both voxel- and index-based data structures to segment the point cloud data into various objects semantically. An unsupervised learning algorithm classifies the segmented scene into instances for damage detection into various damage states and handles rare cases of damage instances.
Machine Learning-Based Structural Damage Identification Within Three-Dimensional Point Clouds
https://orcid.org/http://orcid.org/0000-0002-2368-3962
https://orcid.org/http://orcid.org/0000-0002-8642-2217
Damage identification via remotely sensed data amid routine structural inspections or assessments in the aftermath of extreme events has been the subject of intensive research in recent years. By analyzing remotely sensed data, these tasks can be performed rapidly, safely, and economically while maintaining high accuracy and objectivity. Therefore, many methods have been proposed to detect damage from remotely sensed point clouds of the civil structures. These methods use various feature extraction techniques based on geometry and other spectral information and classification based on supervised or unsupervised learning algorithms. However, the proposed solutions are typically optimized to detect particular types of damage or group various types of damage as a single class, limiting the application of these methods. This article proposes an advanced workflow to identify damage from point clouds by extracting semantic information using supervised and unsupervised learning methods. The supervised learning method is comprised of a deep learning model developed based on both voxel- and index-based data structures to segment the point cloud data into various objects semantically. An unsupervised learning algorithm classifies the segmented scene into instances for damage detection into various damage states and handles rare cases of damage instances.
Machine Learning-Based Structural Damage Identification Within Three-Dimensional Point Clouds
https://orcid.org/http://orcid.org/0000-0002-2368-3962
https://orcid.org/http://orcid.org/0000-0002-8642-2217
Damage identification via remotely sensed data amid routine structural inspections or assessments in the aftermath of extreme events has been the subject of intensive research in recent years. By analyzing remotely sensed data, these tasks can be performed rapidly, safely, and economically while maintaining high accuracy and objectivity. Therefore, many methods have been proposed to detect damage from remotely sensed point clouds of the civil structures. These methods use various feature extraction techniques based on geometry and other spectral information and classification based on supervised or unsupervised learning algorithms. However, the proposed solutions are typically optimized to detect particular types of damage or group various types of damage as a single class, limiting the application of these methods. This article proposes an advanced workflow to identify damage from point clouds by extracting semantic information using supervised and unsupervised learning methods. The supervised learning method is comprised of a deep learning model developed based on both voxel- and index-based data structures to segment the point cloud data into various objects semantically. An unsupervised learning algorithm classifies the segmented scene into instances for damage detection into various damage states and handles rare cases of damage instances.
Machine Learning-Based Structural Damage Identification Within Three-Dimensional Point Clouds
Structural Integrity
Cury, Alexandre (editor) / Ribeiro, Diogo (editor) / Ubertini, Filippo (editor) / Todd, Michael D. (editor) / Mohammadi, Mohammad Ebrahim (author) / Wood, Richard L. (author)
Structural Health Monitoring Based on Data Science Techniques ; Chapter: 21 ; 437-456
Structural Integrity ; 21
2021-10-24
20 pages
Article/Chapter (Book)
Electronic Resource
English
| Elsevier | 2025