BIM for existing infrastructure
Transcript of BIM for existing infrastructure
![Page 1: BIM for existing infrastructure](https://reader031.fdocuments.net/reader031/viewer/2022022206/621372c57c691e18703ab1ad/html5/thumbnails/1.jpg)
BIM for existing infrastructure
Viorica Pătrăucean, Ph.D. Prof. Campbell Middleton Prof. Roberto Cipolla Dr. Ioannis Brilakis
![Page 2: BIM for existing infrastructure](https://reader031.fdocuments.net/reader031/viewer/2022022206/621372c57c691e18703ab1ad/html5/thumbnails/2.jpg)
BIM costs vs. benefits
Peter Cholakis’s blog
![Page 3: BIM for existing infrastructure](https://reader031.fdocuments.net/reader031/viewer/2022022206/621372c57c691e18703ab1ad/html5/thumbnails/3.jpg)
BIM for future constructions – 3D design
Design software: Autodesk, SolidWorks, CloudCompare, Blender
Staffordshire case study Designed IFC model
As-Built Point Cloud
![Page 4: BIM for existing infrastructure](https://reader031.fdocuments.net/reader031/viewer/2022022206/621372c57c691e18703ab1ad/html5/thumbnails/4.jpg)
BIM for future constructions – Geometry checker
Staffordshire case study
Progress monitoring – use BIM’s temporal dimension
![Page 5: BIM for existing infrastructure](https://reader031.fdocuments.net/reader031/viewer/2022022206/621372c57c691e18703ab1ad/html5/thumbnails/5.jpg)
BIM for existing infrastructure
![Page 6: BIM for existing infrastructure](https://reader031.fdocuments.net/reader031/viewer/2022022206/621372c57c691e18703ab1ad/html5/thumbnails/6.jpg)
BIM for existing infrastructure
Paper
![Page 7: BIM for existing infrastructure](https://reader031.fdocuments.net/reader031/viewer/2022022206/621372c57c691e18703ab1ad/html5/thumbnails/7.jpg)
Reverse-engineer 3D design
Images courtesy of 3D ATA, Slovenia
![Page 8: BIM for existing infrastructure](https://reader031.fdocuments.net/reader031/viewer/2022022206/621372c57c691e18703ab1ad/html5/thumbnails/8.jpg)
Reverse-engineer 3D design
Images courtesy of 3D ATA, Slovenia
![Page 9: BIM for existing infrastructure](https://reader031.fdocuments.net/reader031/viewer/2022022206/621372c57c691e18703ab1ad/html5/thumbnails/9.jpg)
Reverse-engineer 3D design
Manual work
3D point cloud
2D CAD drawings 3D BIM
![Page 10: BIM for existing infrastructure](https://reader031.fdocuments.net/reader031/viewer/2022022206/621372c57c691e18703ab1ad/html5/thumbnails/10.jpg)
Reverse-engineer 3D design
Time consuming and error prone
![Page 11: BIM for existing infrastructure](https://reader031.fdocuments.net/reader031/viewer/2022022206/621372c57c691e18703ab1ad/html5/thumbnails/11.jpg)
Reverse-engineer 3D design Automated approach: Machine learning
3D point cloud
BIM
![Page 12: BIM for existing infrastructure](https://reader031.fdocuments.net/reader031/viewer/2022022206/621372c57c691e18703ab1ad/html5/thumbnails/12.jpg)
Reverse-engineer 3D design Automated approach: Machine learning
3D point cloud
BIM
traditional
modern
![Page 13: BIM for existing infrastructure](https://reader031.fdocuments.net/reader031/viewer/2022022206/621372c57c691e18703ab1ad/html5/thumbnails/13.jpg)
Traditional machine learning approach
Input point cloud Feature extraction Voxel-wise classification/segmentation
Smoothing: Dense CRF Part/model fitting: non-rigid ICP Output
User-defined features
![Page 14: BIM for existing infrastructure](https://reader031.fdocuments.net/reader031/viewer/2022022206/621372c57c691e18703ab1ad/html5/thumbnails/14.jpg)
Traditional machine learning approach
Input point cloud Feature extraction Voxel-wise classification/segmentation
Smoothing: Dense CRF Part/model fitting: non-rigid ICP Output
Library of parts User-defined features
![Page 15: BIM for existing infrastructure](https://reader031.fdocuments.net/reader031/viewer/2022022206/621372c57c691e18703ab1ad/html5/thumbnails/15.jpg)
Machine learning = Training
Library of parts Classification & segmentation
Training (labelled) set Parts in context
![Page 16: BIM for existing infrastructure](https://reader031.fdocuments.net/reader031/viewer/2022022206/621372c57c691e18703ab1ad/html5/thumbnails/16.jpg)
Training data: Labelled real point clouds M11 (11 scans)
10 classes • deck • column • pier • abutment • wing-wall • parapet • handrail • road • vegetation • noise
![Page 17: BIM for existing infrastructure](https://reader031.fdocuments.net/reader031/viewer/2022022206/621372c57c691e18703ab1ad/html5/thumbnails/17.jpg)
Training data: Labelled real point clouds
Addenbrooke’s bridge (14 scans)
![Page 18: BIM for existing infrastructure](https://reader031.fdocuments.net/reader031/viewer/2022022206/621372c57c691e18703ab1ad/html5/thumbnails/18.jpg)
Training data: Synthetic models (3D Warehouse)
![Page 19: BIM for existing infrastructure](https://reader031.fdocuments.net/reader031/viewer/2022022206/621372c57c691e18703ab1ad/html5/thumbnails/19.jpg)
Input point cloud Feature extraction Voxel-wise classification/segmentation
Smoothing: Dense CRF Part/model fitting: non-rigid ICP Output
User-defined features
The future of data modelling: Deep learning
![Page 20: BIM for existing infrastructure](https://reader031.fdocuments.net/reader031/viewer/2022022206/621372c57c691e18703ab1ad/html5/thumbnails/20.jpg)
Input point cloud Feature extraction Voxel-wise classification/segmentation
Smoothing: Dense CRF Part/model fitting: non-rigid ICP Output
User-defined features
The future of data modelling: Deep learning
![Page 21: BIM for existing infrastructure](https://reader031.fdocuments.net/reader031/viewer/2022022206/621372c57c691e18703ab1ad/html5/thumbnails/21.jpg)
The future of data modelling: Deep learning Feature learning Voxel-wise classification/segmentation
End-to-end optimisation
Deep artificial neural networks
![Page 22: BIM for existing infrastructure](https://reader031.fdocuments.net/reader031/viewer/2022022206/621372c57c691e18703ab1ad/html5/thumbnails/22.jpg)
The future of data modelling: Deep learning Feature learning Voxel-wise classification/segmentation
End-to-end optimisation
Deep artificial neural networks
• ~ 20% accuracy increase • sometimes better than humans • large number of training examples
![Page 23: BIM for existing infrastructure](https://reader031.fdocuments.net/reader031/viewer/2022022206/621372c57c691e18703ab1ad/html5/thumbnails/23.jpg)
Conclusion
Deep learning
As-Built BIM
• BIM adoption – highly dependent on its implementation for existing infrastructure
• Current manual modelling methods are overly expensive; costs vs. benefits
• Object recognition systems based on deep learning surpass humans
• Need large amount of training data
• Joint efforts to collect data (point clouds, 3D CAD models)
![Page 24: BIM for existing infrastructure](https://reader031.fdocuments.net/reader031/viewer/2022022206/621372c57c691e18703ab1ad/html5/thumbnails/24.jpg)
Parallel projects
Mobile system for fast scanning (Prof. Kenichi Soga)
As-built bridge modelling and change detection (Dr. Ioannis Brilakis)
IFC converter and dedicated tools for bridge design (Prof. Campbell Middleton)
Condition monitoring (Dr. Ioannis Brilakis)