Statistical Analysis Of Compression And Shear Characteristics Of Cohesive Soil

Cohesive soil is an unavoidable foundation soil for many projects,the deformation and strength of which have been widely concerned by scholars.As one of the most commonly used deformation indexes,compression index is often used to evaluate soil deformation and calculate foundation compression deformation.Stress-strain curve of soil is an important basis for expressing its deformation and strength characteristics,and it is also the data basis for studying its mechanical constitutive relationship and calculating model parameters.However,the undisturbed soil is not only difficult to sample but also to carry out the deformation tests.Therefore,the statistical work of deformation and strength is of great significance.At present,there are some problems in the prediction of deformation and strength indexes of cohesive soil,such as regional limitations,difficulty in carrying out undisturbed soil deformation test,lack of engineering soil information and so on.This study is carried out in combination with the Fundamental Research Funds for the Central Universities（DUT21TD106）.So,based on a large number of existing cohesive soil deformation and strength test data,the empirical relationships between different physical indexes and compression index,stress-strain curve are established in this paper.The compression index and stress-strain curve can be obtained quickly through corresponding physical indexes,which lays a foundation for the digitization of geotechnical engineering.The main research contents of this paper are as follows:（1）706 groups of test data of cohesive soil compression index and physical indexes in12 different regions in the existing literature are collected and sorted out.Based on this,the correlation between different physical indexes（w_L 、e₀ 、I_P 、w）n 、γ_d)and compression index Cc is analyzed.The results of correlation analysis show that the correlation coefficients of w_L,I_P and Cc are about 0.75,which is a significant correlation;The correlation coefficients of e₀,w_n,γ_d and Cc are greater than 0.90,belong to high correlation;Among of which,the correlation between e₀ ands Cc is the highest,the correlation coefficient is 0.944.Then,based on the existing formulas,the single parameter and multi parameter empirical regression formulas between cohesive soil compression index and various physical indexes with regional generality are proposed,and the results are tested.After testing,it is recommended to take the single parameter（Cc ^e₀）and multi parameter（Cc ^w_L 、e₀）fitting formulas with the highest prediction accuracy as the regression prediction formula of compression index,and the prediction standard deviations of them are0.074 and 0.051 respectively.（2）A variety of machine learning algorithms are used to predict the compression index under the condition of multiple parameters.The results show that the prediction accuracy of each machine learning algorithm is different under varieties parameter sets.Among them,the prediction accuracy of {e₀,w_L,w_n } parameter set based on extreme learning machine（ELM）algorithm is the highest,and the prediction standard deviation of that is 0.023.Comparing the prediction accuracy of machine learning algorithm and multiple regression model,it can be seen that when the number of input items n < 3,there is little difference between the two prediction accuracy,and machine learning algorithm has no obvious advantage;However,when the number of input items n ≥ 3,the prediction accuracy of machine learning algorithm is significantly better than that of regression method,reflecting its advantages in complex multidimensional data processing.Therefore,when the amount of data is large and there are many input and output parameters,it is recommended to use machine learning algorithm as a geotechnical parameter prediction tool.Finally,the network structure and training parameters of different machine learning algorithms are given,and the compression index Cc can be predicted by using the data set in this paper.（3）A total of 173 sets of stress-strain curves of cohesive soil in 28 areas among 43 literatures under the condition of consolidated undrained triaxial test are collected and sorted out,and the exponential function expressions of plasticity indexPI and strength（σ₁-σ₃）_f of normally consolidated saturated undisturbed cohesive soil and remolded cohesive soil are established respectively.Based on the collected stress-strain curves of normally consolidated saturated cohesive soil under consolidated undrained triaxial shear test,the normalization formulas of stress-strain curves of undisturbed cohesive soil and remolded cohesive soil are obtained by selecting strength（σ₁-σ₃）_f as the normalization factor.Finally,the strength calculation formula and stress-strain curve normalization formula of normally consolidated saturated undisturbed cohesive soil and remolded cohesive soil are combined,and the stress-strain curve fitting formula of cohesive soil under the condition of consolidated undrained triaxial test is obtained.Through verification,the prediction error of stress-strain curve is controlled within 10%.