Prediction of Laboratory Peak Shear Stress Along the Cohesive Soil–Geosynthetic Interface Using Artificial Neural Network
详细信息 查看全文
- 作者：Prasenjit Debnath ; Ashim Kanti Dey
- 关键词：Artificial neural network ; Peak shear stress ; Sensitivity analysis ; Multi ; variable regression ; Statistical analysis
- 刊名：Geotechnical and Geological Engineering
- 出版年：2017
- 出版时间：February 2017
- 年：2017
- 卷：35
- 期：1
- 页码：445-461
- 全文大小：
- 刊物类别：Earth and Environmental Science
- 刊物主题：Geotechnical Engineering & Applied Earth Sciences; Hydrogeology; Terrestrial Pollution; Waste Management/Waste Technology; Civil Engineering;
- 出版者：Springer International Publishing
- ISSN：1573-1529
- 卷排序：35
文摘
In general, soil–geosynthetic interface behaviour is modeled by interface element which involves the assumption of stiffness values which are difficult to determine experimentally. Most of the geosynthetic-reinforced earth structures fail at the interface of the geosynthetic and the soil due to slip or plastic yielding of the reinforced soil. Hence, for a proper design of the soil–geosynthetic interface, an artificial neural network (ANN) model can be used as an alternative approach for the prediction of the soil–geosynthetic interface behavior. The present study uses an ANN model to predict the peak shear stress along the cohesive soil–geosynthetic interface. Three-layer feed-forward back-propagation neural networks with 4, 10 and 15 hidden nodes using three different learning algorithms are examined. Out of three learning algorithms, Bayesian regularization learning algorithm with four hidden nodes is used for its highest coefficient of determination (R2 = 0.988) for the testing set and all of the predicted data falling within the 99% prediction interval. The prediction performance of the ANN model with Bayesian regularization learning algorithm with four hidden nodes is compared with the multi-variable regression analysis. Different sensitivity analyses to quantify the most importance input parameters are also discussed. A neural interpretation diagram to visualize the effect of input parameters on the output is presented. Finally, a predicted model equation is obtained based on the neural network parameters.