Statistical and Predictive Analyses of the Strength Development of a Cement-Treated Clayey Soil

Abstract

The mechanical efficiency of soil stabilization with cement is mainly controlled by various parameters, namely, the amount of binder, the compaction soil state and the curing conditions. The strength of the treated soil is the result of a complex combination of several factors that condition the physicochemical processes involved in cement hydration, which are difficult to monitor. The objective of this study is to identify the relevant parameters governing the bonding in cement-treated soil and suggest a predictive model for strength evolution using these parameters as input. To this purpose, an extensive testing program is presented to assess the impact of the initial water content (11–18%) and dry density (1.6–1.87 Mg/m3) as well as cement dosage (3 and 6%) in sealed curing conditions for 0, 7, 28 and 90 days. The water content variation, the total suction and the compressive strength were measured after different curing durations. The experimental results are first discussed in the parameters’ space, and then through a principal components analysis to overcome the complexity due to the parameters’ interdependency. The PCA revealed that the cement dosage explained 40% of the dataset variance, the remaining 60% being related to a combination of the initial state and curing time. Finally, a predictive model based on an artificial neural network was developed and tested. The predicted results were excellent, with an R2 of +0.99 with the training data and +0.93 with the testing data. These results should be improved by extending the dataset to include different soils and additional compaction conditions.

Keywords

• soil cement stabilization
• statistical analysis
• artificial neural networks
• unconfined compressive strength