Volume 24, Issue 4 (2024)
MCEJ 2024, 24(4): 69-81 |
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:
Hashemi Salanghouch S, Fakhimi A A. Prediction of soil friction angle using a deep learning model: A case study of Firuzkuh sand. MCEJ 2024; 24 (4) :69-81
URL: http://mcej.modares.ac.ir/article-16-72688-en.html
URL: http://mcej.modares.ac.ir/article-16-72688-en.html
1- M.Sc. Student, School of Civil and Environmental Engineering, Tarbiat Modares University, Tehran, Iran.
2- Professor, School of Civil and Environmental Engineering, Tarbiat Modares University, Tehran, Iran. ,A.fakhimi@modares.ac.ir
2- Professor, School of Civil and Environmental Engineering, Tarbiat Modares University, Tehran, Iran. ,
Abstract: (904 Views)
Friction angle of soil is a critical parameter in the geotechnical engineering and has a direct impact on the design of various structures, such as retaining walls, slopes, and piles. This parameter plays a crucial role in determining the overall safety and performance of these structures, making it a key player in the geotechnical analysis and design. In recent years, there have been some impressive advancements in the field of artificial neural networks and deep learning models. These advancements have transformed these models into the highly effective tools for predicting the properties and behavior of soil. By using a powerful deep learning model, it is now possible to save a considerable amount of time and money when it comes to estimating and predicting soil properties. In this particular study, a convolutional neural network was developed to predict the peak friction angle of Firuzkuh sand using some soil images and the dry density as the input parameters. The network itself consisted of five consecutive convolutional layers, as well as a pyramid pooling module that utilized four different pooling sizes arranged in parallel. In addition, two fully connected layers were incorporated into the network's design, which enabled it to satisfactorily process the input parameters of the images and the dry densities with respect to the speed and precision. This network converts the soil image into a scalar (number) by using these 5 convolutional layers, the pyramid pooling module and a fully connected layer. Then, this scalar is concatenated with the dry density of the soil, is passed through a fully connected layer, and the peak friction angle of the soil is obtained as an output. For data generation, a total of ten samples of Firuzkuh sand were prepared. These samples had different gradation curves, which are referred to as S1 to S10 specimens. Each soil specimen was compacted at three different dry densities. The peak friction angle associated with the 30 different densities for the 10 different particle size distributions (S1 to S10 specimens) was determined using the direct shear test apparatus. The direct shear test box was 100 ×100 × 25 mm in size. For network training and testing, the soil specimens were spread on a flat surface and 50 photos in different light environments with varying distances of the camera from the soil surface, were taken from each specimen. Since in the network training process, three dry densities were considered for each sample, a total of 1500 images were prepared for the network database. Of these, 1125 photos were used for training and 375 photos were saved for testing the network. The network was trained for 1000 epochs on the training data, and the mean square error after 1000 epochs was reduced to 1.84. The outcome of the assessment conducted on the designed convolutional neural network in this study, using 375 test data, revealed that the network can predict the peak friction angle of Firuzkuh sand by incorporating the image and dry density of the soil as input variables. The total normalized relative error was 3.0%, while the maximum normalized relative error was 10%. This indicates that the network has the ability to quickly predict the peak friction angle of the Firuzkuh sand with a good accuracy.
Keywords: Deep learning, Soil friction angle, Firuzkuh sand, Convolutional neural Network, Digital image processing
Article Type: Original Research |
Subject:
Geotechnic
Received: 2023/11/30 | Accepted: 2024/02/28 | Published: 2024/10/1
Received: 2023/11/30 | Accepted: 2024/02/28 | Published: 2024/10/1
Send email to the article author
| Rights and permissions | |
 |
This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License. |