COMPACTION PARAMETERS PREDICTION AND ANALYSIS FROM ATTERBERG LIMIT BY ARTIFICIAL NEURAL NET OF SOILS ALONG MOROCHO-APOSTO ROAD

Abstract

Urbanization together with large scale industrialization and modernization has created shortage of land for construction in most countries. To overcome this problem and fulfil the need for infrastructure, land is developed using sound and cost effective engineering techniques such mechanical compaction. This method identifies important compaction parameters mainly optimum moisture content and maximum dry density and are acquired from simple laboratory testing. But when there is a limited number of laboratory equipment, budget and man power in case of larger projects which requires reasonable sizable effort, labor, time and budget, implementing other methods such developing linear relationship for determination compaction parameters of soil is a must and economical. In these study attempts to predict optimum moisture content and maximum dry density from Atterberg limit parameters by using artificial neural net analysis of soils in Morocho Aposto road has been made. When there is a need for rehabilitating and upgrading of the road, conducting this study provides future information and used to validate findings. To conduct the study 75 data bank that includes 20 samples as primary and 55 data as secondary source were used to analyze and develop the predictive model. Samples were collected randomly from 20 pits and subjected to both Atterberg limit and Modified proctor test in accordance to AASHTO standard. The data were analyzed and computed using Neural Net. Commands where written for prediction and linear regression modeling. The data bank was trained and optimized using Levenberg-Marquardt, Bayesian regularization and Scaled Conjugate Gradient algorithm based on each criterion. Levenberg-Marquardt training algorithm is selected as the best algorithm by with the two algorithms. The study found that, LM algorithm has mean estimate error of 1.527, training time < 1 second and R value of 0.978 that predicted the models with better accuracy than SCG and BR algorithms. Therefore, the study concludes that OMC and MDD can be predicted from Atterberg limit by artificial neural net with error margin of ± 2 for soils in the study area