Analysis of Earth Embankment Structures using Performance-based Probabilistic Approach including the Development of Artificial Neural Network Too

Earth embankment structures such as dams and levees are essential in our nation's infrastructure; being used for flood protection, water storage and hydropower generation. Frequent storms showed the deteriorating condition of dams and levees and urgent the need for a better condition assessment and stability evaluation. Past dams and levees failures have proved that the use of factor of safety approach is not an effective indicator of these structures' performance given the uncertainties in design input parameters (soil properties, loading history, etc.). An evaluation of the existing FEMA risk prioritization tool was performed along with sensitivity analysis. Numerical analysis of the Howard A. Hanson dam using PLAXIS 2D is preformed and its results within the context of limit states and FEMA risk tool is presented and discussed. A contour approach is proposed for risk estimation using various combinations of probabilities of exceeding limit states for seepage, earthquake and LSIII. A new risk tool is developed with MATLAB graphical user interface (GUI) to implement the finite element program PLAXIS 2D results. A total of 363 cases of numerical analysis using the finite element program PLAXIS 2D are performed varying key parameters including: i) geometry and properties of embankment ii) cycles of rising and falling water levels simulating the consecutive storms. MATLAB was used to develop a neural network which showed a training and prediction regression (R2) of 98%. A parametric study on the effect of geometry, soil parameters and cycles of loading is performed using a base embankment model. As side slopes becomes steeper, shear strain increased from 0.1 to 0.5 to 3.8% going from 4:1 to 3:1 to 2:1 slope at time of 30 days. As embankment friction angle increased from 25° to 36°, there was a decrease in shear strain at toe from 2.4 to 0.5%. Cohesion of embankment, foundation and alluvial had a limited effect on the shear strain at toe. The embankment and alluvial permeability had a slight effect on the shear strain values. Increasing the foundation permeability from 6.8*10-5 to 1*10-3 cm/s caused an increase in toe shear strain to exceed LSIII. Increasing cycles of loading from 1 to 6 cycles, increased the shear strain by a factor of 10. Relative importance of the input parameters, indicated that side slopes and foundation permeability have the most effect on the shear strain at toe. The results of an integrated remotesensing program and finite element modeling for a Sherman Island levee section is presented. Sherman island levee is constructed over peat deposit which experience large land subsidence and decomposition over time. Remote sensing data were used for the calibration of a numerical model using the finite element program PLAXIS 2D with mesh updating. Amorphous peat showed stiffer response and lower compressibility than fibrous peat. The assumption of amorphous peat led to computed displacements that ranged from 10 to 30% less than those with the fibrous peat properties, depending on the location within the domain. The analyses indicated a relatively small mechanistic deformation induced by an "extreme" water level under transient conditions. A study on the effect of peat different states of decomposition varies from H1 to H10 on the Von Post scale is performed. Deformed shape and probability of exceedance for both shear strain and gradient is shown. H1-H3 case reached probability of exceeding LSI of 1 after 270 days while H4-H7 peat took approximately 10,000 days to reach PE of 1. H8-H10 peat reached a PE of 1 at around 300,000 days.