Application of a sensitivity analysis procedure to interpret uniaxial compressive strength prediction of jet grouting laboratory formulations performed by SVM model
Jet Grouting (JG) technology is one of the most widely used soft-soil improvements methods around the world. When compared with other methods, JG versatility is highlighted, since it can be applied to several soil types, creates elements with different geometric shapes and, normally, is less expensive. However, due to inherent geological complexity of soil and high number of variables involved in JG process, JG design is a hard task. Nowadays, Us design is essentially based on empirical rules that are often too conservatives, compromising the economy and the quality of the treatment. In the present study, data mining (DM) techniques, particularly the high learning capabilities of support vector machines (SVM) algorithm were used to predict uniaxial compressive strength (UCS) of JG laboratory formulations over time. Furthermore, and by performing a detailed sensitivity analysis, some important information was extracted based on the learned model. The high performance achieved by SVM algorithm in UCS prediction is summarized showing the high predictive accuracy reached (R~2=0.93). In addition, after apply a one- and two-dimensional sensitivity analysis, an important explanation of the model is given in terms of what are the key variables and its effect on UCS estimation, as well as the interaction level between input variables. Hence, it is shown that age of the mixtures, cement content and the relation between mixture porosity and volumetric content of cement have a high influence in strength behavior of JG laboratory formulations. Furthermore, was also possible to observe that water/cement ratio is the variable with higher interaction with age of the mixture and cement content.
展开▼
