Prediction of Compaction Parameters Using Regression and ANN Tools

The compaction parameters are not only important for quality control of earthworks but are required for designing of various works. The engineering properties of soils such as shear strength, CBR, permeability and consolidation parameters are usually determined by testing the samples prepared at optimum moisture content (OMC) and maximum dry density (MDD). On any project, a large number of soil types are generally used. For each soil, besides the basic soil classification parameters, compaction parameters are also determined. Testing of large numbers of soil samples is a cumbersome and time consuming job and requires trained personnel for the purpose. Further, an independent check on the results obtained in the laboratory is also required to make use of test values on site with confidence. Therefore, a need is felt to obtain the required compaction parameters from the basic soil test which are used for the classification of soil viz. Atterberg�s limits, gradation, specific gravity etc. In this endeavour, basic soil parameters were collected from literature and Artificial Neural Network (ANN) techniques have been employed on the data collected, as ANN can better model the relation between compaction parameters and basic soil properties than statistical modelling. Present work demonstrates application of five different ANN algorithms viz. LM, GDM, SCG, BR and CFB to predict standard compaction characteristics of varieties of soils with a large range variation in their basic soil properties. Multiple variable non-linear regression analysis was also carried out, in which establishment of an empirical relationship for prediction of compaction characteristics of Modified compaction (i.e., OMC2 and MDD2) by using Standard compaction values (i.e., OMC1 and MDD1) was attempted. Then validation by an independent outside dataset from literature of Horpibulsuk et al. (2009) was also carried out on ANN models and regression analysis done.