Downloads: 128 | Views: 208
M.Tech / M.E / PhD Thesis | Civil Engineering | India | Volume 7 Issue 10, October 2018
Study of ANN Program for Prediction of California Bearing Ratio of Fine-Grained Soils
Vasu. Nagamalli | Suresh Praveen Kumar.P
Abstract: Expansive soils, popularly known as black cotton soils in India, undergo swelling by absorbing water and shrinking by loss of moisture. Expansive soils are a boon to former but problematic to civil Engineers as they are prone to high volume changes even due to natural processes of climatic and environmental changes. The shear strength of an expansive soil is very high in dry state and it reduces considerably upon wetting. So statistical models (i. e. , Regression Analysis, ANN, Fuzzy Logic and Genetic Algorithm) are developed to assess or estimate the engineering properties from basic soil properties. In this paper study both Regression Analysis and Artificial Neural Networks are used for estimating CBR of fine-grained soils can be determined by carrying out laboratory CBR test on undisturbed samples, however, the test is quite time-consuming and laborious. Therefore, many empirical formulas based on regression analysis have been presented for estimating the CBR by using soil Index properties. In the present study a statistical regression model is developed for estimating CBR of soils. Artificial Neural Network (ANN) model is suggested for prediction of CBR, considering basic soil properties like WL, PL and MDD, OMC of the soils as the input parameters. An NN Code is developed for prediction of output parameter by using Back-Propagation Algorithm. Also a model is developed for predicting Compression Index of soils neural fitting tool which is part of MATLAB Software. For development of this model, Levenberg-Marquardt (LM) Back-propagation Algorithm (trainlm) is considered. Comparative studies were done between the observed and predicted values obtained using MR Model, developed Program, and Neural Fitting tool (NF tool), using same input parameters and to predict same output parameter. The predicted values of CBR obtained from the developed code model is found closer to actual/observed values of CBR when compared to that of from Neural Fitting tool (NF tool). So the developed NN Program is successfully executed for the prediction of CBR of fine-grained soils.
Keywords: Expansive Soils, CBR, ANN, ANF, MATLAB Software
Edition: Volume 7 Issue 10, October 2018,
Pages: 880 - 884