Use of Artificial Neural Network to Predict Compaction Characteristics of Soil from Soil Index Properties (Case of Addis Ababa)
No Thumbnail Available
Date
2021-10
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Addis Ababa University
Abstract
Soil compaction is the most commonly practiced mechanical method used to improve soil behavior and has a significant impact on earthwork structures. However, determining the compaction characteristics of soil in the laboratory requires considerable time and effort. For the purpose of attempting the problem of spending too much time and effort, ANN models that can predict the values of compaction parameters namely Maximum Dry Density (MDD) and Optimum Moisture Content (OMC) from soil index properties are developed using Artificial Neural Network (ANN). A total of 300 secondary data divided as coarse and fine grained soils were used to develop the ANN models. All data were soil laboratory test result records of different soil samples taken from Addis Ababa, Ethiopia. Percent of grain size distribution was used as input parameter for the coarse grained soils while plasticity index along with the percent fine were used as input variables for the fine grained soils. The two variables MDD and OMC were the desired outputs from the models. Supervised learning with a backpropagation algorithm was implemented to train the models using MATLAB R2020a. The models were validated using 15 primary data and the models’ performance was evaluated using statistical values. The fifteen soil samples were collected from different locations in Addis Ababa. Three soil laboratory tests namely grain size analysis, Atterberg limits, and modified compaction tests were done on each sample. The outputs of the developed models were compared with the actual experimental soil laboratory test results and showed a good accuracy with a determination coefficient value of R=0.92 and R=0.81 for both maximum dry density and optimum moisture content respectively. Equations were derived for the ANN models. The models were also compared with existing regression equations developed using Addis Ababa soil type. From the research, it was concluded that the developed ANN models can be applied to predict the value of compaction parameters from soil index properties for both coarse and fine grained soils of Addis Ababa.
Description
Keywords
ANN, Model, Compaction, Density, Optimum, Index