Neural modelling of compactibility characteristics of cohesionless soil

Abstract

Compaction is the method of in-situ soil modification to improve its engineering properties. Two key compactibility parameters are: the maximum dry density ρ_{d max} and the corresponding optimum water content w_opt. They are basic parameters for designing, constructing and controlling the compaction quality of earth structures (e.g. earth dams, highway embankments). Soil compactibility can be determined from the laboratory compactibility curve basing on Proctor's test. However, this test is destructive, time-consuming and expensive. To facilitate the determination of the cohesionless soil compactibility parameters, correlations between ρ_{d max} and w_opt and the basic parameters characterizing soil grain-size distribution (C_U, D₁₀, D₂₀, D₃₀, D₄₀, D₅₀, D₆₀, D₇₀, D₈₀, and D₉₀) were developed. Artificial neural networks are applied to determine models with good prediction quality. The neural models have higher accuracy than the classic statistical models.