A framework for estimating the matric suction of unsaturated soils using two artificial intelligence techniques

ABSTRACT: A major obstacle in implementing the state-of-the-art understanding of the mechanics of unsaturated soils into geotechnical and geo-environmental engineering practice is associated with a lack of quick, reliable, and economical techniques for matric suction measurement. Matric suction is a key stress state variable that significantly influences the hydro-mechanical behaviour of unsaturated soils. For this reason, in this paper, two artificial intelligence (AI) models were developed for estimating matric suction based on the support vector regression (SVR) and multivariate adaptive regression spline (MARS) algorithms. The results suggest that both models can reasonably estimate matric suction. Compared to the MARS model, the SVR model is capable of achieving a higher accuracy. Nonetheless, the MARS model facilitates sensitivity analysis and the selection of essential input parameters. An empirical equation is proposed based on the MARS model for the matric suction estimation of low plastic soils with a plasticity index equal to or less than seven. Finally, a framework is proposed for the estimation of matric suction, which combines the strengths of both SVR and MARS models. The study is promising for engineers to implement the mechanics of unsaturated soils into practice because the input parameters used in this study can be determined quickly from conventional soil tests in the laboratory.