A novel approach to predicting young’s modulus of jet grouting laboratory formulations over time using data mining techniques

Abstract

Many geotechnical projects today have demonstrated a need for improved soil foundation properties, namely physical and mechanical properties. There are currently several soil improvement methods available for this task, including jet grouting (JG) technology. In this method, a slurry grout is injected into the subsoil at high pressure and velocity to destroy the soil structure. The injected slurry (normally cement) and the fragmented soil together create an improved soil mass with better strength, deformability and permeability characteristics. However, due to the inherent geological complexity and high number of parameters involved in this improvement process, the design of its physical and mechanical properties is a very complex task, especially in the initial project stages and in small-scale geotechnical projects, when information is scarce. Consequently, the economics and quality of the improvement can be adversely affected, and it would be beneficial to develop an accurate model to simulate the effects of the different parameters involved in the JG process. In many geotechnical structures, advanced design incorporates the ultimate limit state and the serviceability limit state design criteria, for which the uniaxial compressive strength and deformability properties of the improved soils are needed. A previous study by the author proposed some regression models based on data mining (DM) techniques to predict the uniaxial compressive strength of JG laboratory formulations (JGLF) over time. In the present study, similar tools such as multiple regression, artificial neural networks, support vector machines and functional networks are trained to predict the deformability modulus of JGLF over time. Additionally, the mathematical expressions proposed by the Eurocode 2 andModel Code 1990 that are currently used to estimate concrete stiffness over time are adapted to the JG material. The results showthat the novel soft computing model ismore accurate and capable of learning the complex relationships between JGLF deformability and its contributing factors. A novel visualisation approach is also applied to the work based on a sensitivity analysis method. Such an approach enables the identification of the most important input parameters and their average influence on deformability predictions for JGLF. Moreover, through the application of DM techniques, a novel approach capable to predict JGLF stiffness based on its unconfined compressive strength and three other variables related to soil and mix properties is proposed.