Deep mining of big data and model tests on dilatancy characteristics of dilatant soils

The dilatancy of soils is an important basis for constitutive models, and the current dilatancy models do not fully reveal their common laws, which is also an important reason why the existing constitutive models cannot well reflect the deformation mechanism of soils. Based on the Hadoop and Spark computing platform, a distributed Levenberg Marquardt regression (DLMR) algorithm for deep mining of big data with strong global optimization, fast convergence and computational stability is proposed. Based on a large number of experimental data of dilatancy characteristics of dilatant soils, according to the DLMR algorithm and the basic mechanical properties of soils, the big data characteristics of dilatancy of dilatant soils are obtained. It is found that there are obvious nonlinear characteristics between dilatancy ratio and stress, strain and stress increment, and the correlation functions between them are established respectively. On this basis, a dilatancy model which can reflect the common law of dilatancy characteristics of dilatant soils is constructed. Through model comparison, it is shown that the proposed model is superior to the dilatancy model of modified Cambridge model and Rowe model. By simulating the triaxial compression experimental data of dilatant soils under different stress paths, it is shown that the new model can well reflect the dilatancy under different stress paths.