ESTIMATION OF RESIDUAL FRICTION ANGLE OF CLAY SOILS USING ARTIFICIAL NEURAL NETWORKS MODELLING
Mostafa Abolfazl Zadeh, Amin Falamaki
In the proceedings of: GeoQuébec 2015: 68th Canadian Geotechnical Conference & 7th Canadian Permafrost ConferenceSession: Physical and Numerical Modelling / Modélisation physique et numérique
ABSTRACT: Accurate estimation of site-specific soil strength parameters (e.g., the internal friction angle and cohesion) is challenging in geotechnical engineering due to the limitations and complexities associated with obtaining undisturbed soil samples and laboratory shear test analysis. The residual friction angle (r) of clay soils is particularly important parameter in slope stability analysis, especially in case of pre-existing slip surfaces and large deformations, and is commonly approximated from Atterberg limits and grain size distribution using traditional regression analysis. In this study, we tested the reliability of Artificial Neural Networks (ANNs) in predicting the residual friction angle degrees of different soil types based on their Atterberg Limits, clay size fraction and normal stress. The main objective was to find a satisfactory relationship between input and actual measured values using artificial neural network models. The effect of the network geometry on the performance of the models was also assessed. Strong correlation factors (e.g., 0.99) for training and testing data sets in model MLP741 demonstrate that ANNs are powerful tools for predicting soil strength parameters.
RÉSUMÉ: Dans cette étude, nous avons déterminé la fiabilité des réseaux de neurones artificiels (RNA), pour prédire la valeur de rargile at la contrainte normale. L'objectif principal était de trouver une relation satisfaisante entre l'entrée et les valeurs mesurées en utilisant les modèles RNA. L'effet de la géométrie du RNA sur la performance des modèles est également évalué. Nous avons trouvé que, pour tous les modèles RNA testés, les facteurs de la corrélation sont supérieurs à 0,99. Cela montre que les RNA sont des outils puissants pour prédire des valeurs du r. Il a été également observé que parmi les modèles RNA, le MLP741 (perceptron multicouche) mène aux meilleurs résultats.
Access this article:
Canadian Geotechnical Society members can access to this article, along with all other Canadian Geotechnical Conference proceedings, in the Member Area. Conference proceedings are also available in many libraries.
Cite this article:
Mostafa Abolfazl Zadeh ; Amin Falamaki (2015) ESTIMATION OF RESIDUAL FRICTION ANGLE OF CLAY SOILS USING ARTIFICIAL NEURAL NETWORKS MODELLING in GEO2015. Ottawa, Ontario: Canadian Geotechnical Society.
@article{575,author = Mostafa Abolfazl Zadeh ; Amin Falamaki,title = ESTIMATION OF RESIDUAL FRICTION ANGLE OF CLAY SOILS USING ARTIFICIAL NEURAL NETWORKS MODELLING,year = 2015}