Abstract

Cement stabilisation has been widely used for improving the engineering properties of soft soils. The unconfined compressive strength (UCS) is the most common strength parameter used for the current design practice of cement-stabilised soil due to its simplicity and cost-effectiveness. However the UCS test does not take into account the effect of confining stress on material strength, and thus it is considered to be conservative, variable and lacking in reliability. The undrained triaxial compression test, although less straightforward to conduct and more costly, is more representative in terms of simulating actual field conditions. The first part of this work presents data collection and collation from six research-based projects involving both UCS tests and undrained triaxial tests, performed on the same laboratory-prepared cement-treated soil samples. Results from the UCS tests were compared and correlated to those from the undrained triaxial tests by normalising the data and developing contour plots to illustrate the relationships between the two strengths. A Bayesian neural network model was developed to provide better estimation of the strength correlations as a function of aforementioned input parameters. In order to further explore the applicability of artificial neural network modelling for cement-stabilised soils, the second part of the study extends its application to predicting the UCS and stiffness of stabilised soils. The correlations between stiffness and strength from both the UCS test and undrained triaxial compression test were also studied as part of the work. Relationships between the UCS and stiffness for laboratory-stabilised soils and field deep mixing were found to be consistent with the findings from existing literature. The overall research highlighted the potential of using artificial intelligence for providing preliminary design parameters of cement-stabilised soils.