Deformation Characteristics And Fragility Analysis Of Geosynthetic-encased Coal Slag Column Through Machine Learning

Weak soil foundation can’t be used directly as the foundation of a structure,for its high compressibility,low bearing capacity,low shear strength and poor seismic performance.Using geosynthetic-encased coal slag column(GECSC)as a foundation treatment technology can significantly improve the bearing capacity and seismic performance.However,the structure of the GECSC is complex and few studies concentrate on the deformation of GECSC installed in the loose sand.Moreover,no fragility analysis has been conducted on the GECSC,limiting the wide application of this technology.This research responds to the initiative of‘carbon neutrality and carbon peaking’,exploring the environmental protection of using coal slag as the aggregate of GECSC,the microscopic tests were conducted to detect the content of the heavy metal.A large scale physical model test and numerical simulation of GECSC in loose sand have been carried out,focusing on its bearing capacity and deformation characteristics;Qualitative analysis of parameters was then performed;machine learning methods were adopted to reduce the feature parameters and analyze the deformation of GECSC;fragility analysis was achieved through Logistic regression.The main conclusions were reached as follow:1.Verify the environmental protection of coal slag.Coal slag and fly as are homologous.The uppermost elements are silicon and aluminum,which account for more than 80%.Quartz and mullite are the main crystalline minerals.The concentrations of heavy metal of coal slag far below the threshold required by code and its superiority in the economy and environmental protection.2.1g physical modeling and numerical analysis on GECSC.(1)The bearing capacity of the composite foundation is significantly improved compared with the untreated foundation.The predominate bulging of the column occurs between 0～3D below the top,and the maximum radial strain is observed at 1.7D.(2)The geosynthetic encasement constraint the radial strain and increase the stiffness of GECSC,which results in greater bearing capacity and smaller radial strain;(3)The effectiveness of GECSC in loose sand is remarkable.The loose sand foundation’s bearing capacity is increased by 74%,compared with 8%of very dense sand foundation.(4)The depth of 1.7D is a significant expansion position for all samples,and another prominent bulge position occurred below the encased area.The maximum radial strain is significantly influenced by the load.3.Establishing a series of regularization models and optimized neural network models to analyze the settlement and deformation characteristics of GECSC.(1)According to the regularization method,the stiffness of encasement has the most significant influence on settlement and deformation of GECSC,followed by diameter;(2)The settlement and radial deformation were predicted by LM-BP,GA-LM-BP,and PSO-LM-BP neural networks.The determination coefficient R~2 on the test set of three methods exceed 95%and the robustness of GA-LM-BP is the best.(3)The sequence model of LSTM can accurately predict the load-settlement deformation of GECSC in future.4.The fragility analysis is conducted based on Logistic regression.(1)Logistic regression requires only a small amount of data to generate vulnerability curves with high accuracy;(2)The existence of geosynthetic encasement can improve the seismic performance of GECSC composite foundation significantly;(3)Conservative parameters of GECSC should be selected for soil with large variability;(4)Logistic regression can quantitatively analyze the failure probability of composite foundation considering multiple parameters.