Using artificial neural network (ANN) in prediction of collapse settlements of sandy gravels

H. Hasanzadehshooiili, Reza Mahinroosta, Ali Lakirouhani, Vahid Oshtaghi

Research output: Contribution to journalArticlepeer-review

30 Citations (Scopus)


Collapse settlement is one of the main geotechnical hazards, which should be controlled during first impoundment stage in embankment dams. Imposing large deformations and significant damages to dams makes it an important phenomenon, which should be checked during design phases. Also, existence of a variety of contributing parameters in this phenomenon makes it difficult and complicated to well predict the potential of collapse settlement. Thus, artificial neural networks, which are commonly applied by majority of geotechnical engineers in predicting various perplexing problems, can be efficiently used to calculate the value of collapse settlement. In this paper, feed forward back propagation neural networks are considered. And three-layered FFBPNNs with the architectures of 4–6–2 and 4–9–2 accurately predicted the coefficient of stress release and collapse settlement value, respectively. These networks were trained using 180 datasets gained from large-scale direct shear test, which were carried out on gravel materials. High correlation between measured and predicted values for both collapse settlement and coefficient of stress release can be easily understood from the coefficient of determination and root mean square error. It is shown that sand content and normal stress applied to the specimens, respectively, are most effective parameters on the collapse settlement value and coefficient of stress release.
Original languageEnglish
Pages (from-to)2303-2314
Number of pages12
JournalArabian Journal of Geosciences
Issue number6
Early online date10 Feb 2013
Publication statusPublished - 2014


Dive into the research topics of 'Using artificial neural network (ANN) in prediction of collapse settlements of sandy gravels'. Together they form a unique fingerprint.

Cite this