Spatiotemporal deep learning approach on estimation of diaphragm wall deformation induced by excavation

This paper proposes a convolution neural network (CNN) based prediction method for concrete diaphragm wall (CDW) deflections. CNN algorithm is modified for processing the CDW deformation data collected from in-situ measurement in both time and space dimensions, and capable of making dynamic predicti...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Veröffentlicht in:Acta geotechnica Jg. 16; H. 11; S. 3631 - 3645
Hauptverfasser: Zhao, Hua-jing, Liu, Wei, Shi, Pei-xin, Du, Jiang-tao, Chen, Xiu-ming
Format: Journal Article
Sprache:Englisch
Veröffentlicht: Berlin/Heidelberg Springer Berlin Heidelberg 01.11.2021
Springer Nature B.V
Schlagworte:
Algorithms
Artificial neural networks
Back propagation networks
Complex Fluids and Microfluidics
Construction accidents & safety
Construction industry
Convolution
Deep learning
Deflection
Deformation
Diaphragm wall
Dimensions
Engineering
Excavation
Feature extraction
Foundations
Geoengineering
Geotechnical Engineering & Applied Earth Sciences
Hydraulics
Long short-term memory
Machine learning
Measurement
Neural networks
Occupational safety
Predictions
Research Paper
Safety management
Soft and Granular Matter
Soil Science & Conservation
Solid Mechanics
Diaphragm wall
Deformation
Deep excavation
Deep learning prediction
ISSN:1861-1125, 1861-1133
Online-Zugang:Volltext
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Beschreibung
Zusammenfassung:This paper proposes a convolution neural network (CNN) based prediction method for concrete diaphragm wall (CDW) deflections. CNN algorithm is modified for processing the CDW deformation data collected from in-situ measurement in both time and space dimensions, and capable of making dynamic prediction based on the extracted spatiotemporal features of wall deflection. The proposed method is validated through investigating a project of deep excavation in Suzhou, China. The predicted results show excellent agreement with field measurement and yield mean absolute errors of 0.86 mm and 1.55 mm for nowcasting and forecasting tasks, respectively. Three prevailing algorithms in time series prediction, namely, back propagation neural network, long short-term memory and autoregressive integrated moving average, are conducted for comparison. The results illustrate that the CNN outperforms the other algorithms in terms of accuracy and execution time. Therefore, the proposed CNN model is the most suitable for CDW deflection prediction, and can provide reasonable references for construction safety management on site.
Bibliographie:ObjectType-Article-1
SourceType-Scholarly Journals-1
ObjectType-Feature-2
content type line 14
ISSN:1861-1125
1861-1133
DOI:10.1007/s11440-021-01264-z