Prediction of California Bearing Ratio of Soils Using Artificial Neural Network

Abstract

California Bearing Ratio (CBR) value is an important parameter in indexing the resistance offered by soils in the sub grade layers or in the foundation of a structure. The laboratory and field tests are extensively used for its determination to assess the strength of soils. An attempt has been made in the present study to estimate the soaked CBR from the test result values of sieve analysis, Atterberg limits, optimum moisture content (OMC) and maximum dry density (MDD). Simple, multiple linear regression (SRA, MLR) and artificial neural networks techniques were applied for analysis of data to develop the most suitable model. The data were collected from road construction works of Madhya Pradesh region of India. Simple linear regression analysis gave satisfactory correlation (R 2 =0.62) with MDD variable only, however artificial neural networks and multiple linear regression analysis resulted in strong correlations (R 2 =0.88-0.98) of CBR with sieve analysis, Atterberg limits, optimum moisture content and maximum dry density. The obtained correlation equation as a result of regression analysis is in satisfactory agreement with the test results. Gravel, sand content, OMC and MDD indicated to be the governing factors and developed model can be used for determination of CBR for a preliminary design of a project having financial and time limitations.