Volume 15, Issue 5 (2015)
MCEJ 2015, 15(5): 181-190 |
Back to browse issues page
Download citation:
BibTeX | RIS | EndNote | Medlars | ProCite | Reference Manager | RefWorks
Send citation to:
BibTeX | RIS | EndNote | Medlars | ProCite | Reference Manager | RefWorks
Send citation to:
Komakpanah1 H, Yasrebi , S S, Golshani A. Application of Artificial Neural Networks in Prediction of Expansive Behavior of Clay Soils. MCEJ 2015; 15 (5) :181-190
URL: http://mcej.modares.ac.ir/article-16-8121-en.html
URL: http://mcej.modares.ac.ir/article-16-8121-en.html
1- T.M.U
2- Tarbiat Modares University
2- Tarbiat Modares University
Abstract: (5161 Views)
In the recent years, new techniques such as artificial neural networks were used for developing of the predictive models to estimate the needed parameters in Geotechnical Engineering such as swelling potential. If over 50% of the particles in a sample are able to pass through a number 200 screen or sieve then the sample is classified as either silt or clay or some combination of both. Regardless of the percentage of “fines” in a particular sample, a significant presence of clay minerals in a sample can indicate a possible expansive soil problem. When they absorb water they increase in volume. The more water they absorb the more their volume increases. Expansions of ten percent or more are not uncommon. This change in volume can exert enough force on a building or other structure to cause damage.
Cracked foundations, floors and basement walls are typical types of damage done by swelling soils. Damage to the upper floors of the building can occur when motion in the structure is significant.
Expansive soils will also shrink when they dry out. This shrinkage can remove support from buildings or other structures and result in damaging subsidence. Fissures in the soil can also develop. These fissures can facilitate the deep penetration of water when moist conditions or runoff occurs. This produces a cycle of shrinkage and swelling that places repetitive stress on structures. Determination of swell potential of soil is difficult, expensive and time consuming and also involves destructive tests. Multi-layer Perceptron model is one of the most sufficient methods of the Artificial Neural Networks in most of the research applications in engineering etc. In this research, Multi-layer Perceptron model and Radial Basis Function model of ANN (artificial neural networks) were used in order to predict expansive behavior of clayey soils (i.e., swell percent). All data have been modeled by using many types of architectural Multi-layer Perceptron network. Then, the output result of these networks are compared with each other according to the assessment indexes which has been leaded to the best architectural network selection in viewpoint of accusation and usage. It is noticeable that the parameters such as Natural Water Content, Plastic Index, Dry Density and Fine Soil Percent are considered as input parameters and swell percent (S%) is considered as output parameter.
The Soils which are selected for this research is clayey soils from different areas of Iran. Consequently this ANN has the ability to predict expansive behavior of diverse types of clayey soils. To train this network, results of previous researches, geotechnical consultant engineering data and the available thesis about Expansive soils are used. It was found that the Multi-layer Perceptron (MLPst and MLPdy) models exhibited a higher performance than Radial Basis Function (RBF) model for predicting expansive behavior of clayey soils. Also, the comparison of the MLPst and MLPdyn network models indicates that their accuracies are almost the same. However, the time taken by MLPst is less than that of MLPst in this study.
Since the population of the analyzed data is relatively limited in this study, the practical outcome of the proposed models could be used with acceptable accuracy at the preliminary stage of design.
Received: 2015/11/8 | Accepted: 2015/08/23 | Published: 2015/11/8
| Rights and permissions | |
 |
This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License. |