Application of inverse analysis to geotechnical problems, from soil behaviour to large deformation modelling

Abstract

Large deformation analysis has become recently centre of attraction in geotechnical design. It is used to predict geotechnical boundary value problems such as, excessive movement of soil masses like landslides or soil-structure interaction like pile installations. Wrong understanding and simulation of each mentioned problem could lead to significant costs and damages, therefore, robust approaches of modelling are needed. Throughout the past decades many numerical methods aiming to simulate large deformations have been introduced as for example, Discrete Element Method (DEM), Smooth Particle Hydrodynamics (SPH), Updated Lagrangian Finite Element Method (UL-FEM) and Material Point Method (MPM). They are varying in basic theories, capabilities and accuracy. But, the complexity is the feature which is quite common in all them and it is attributed to the unclear response of soil body under excessive deformations. As a result these methods are involving many uncertainties in input parameters. Determination of these parameters is always difficult, because reproducing larg deformations in the laboratory is difficult and needs advanced and expensive facilities. As a result the introduction of a methodology for estimation of the model parameters adopted for large deformation analysis is extremely needed. Inverse analysis approaches have proved to be able to overcome complex engineering problem in different fields. In geotechnical engineering, inverse analysis is typically employed to back-calculate the input parameter set of a model to best reproduce monitored observations. Accordingly, its application attempts to clarify the effective soil conditions and allows for an update of the design based on the insitu measurements. Numerous researches have been fulfilled to evaluate the performance of this approach in geotechnical problem, however, rarely the application of this methodology to the problems involving large deformations have been addressed. This thesis is addressing these issues by combining inverse analysis methods with advanced numerical methods and soil constitutive models. The proposed methodology is applied to two popular large deformation engineering problem i.e. landslides and soil-structure interaction, particularly cone penetration tests modelling. Different case studies are addressed; two methods of Smoothed Particle Hydrodynamic and Material Point Method are adopted as numerical models, depending on the case study. Similarly, various constitutive models ranging from the simple Mohr-Coulomb to the advanced ones such as Hardening soil and Hypoplastic model are employed. The employed inverse analysis algorithm also varies by the type of the numerical models and required computation time of the forward model. Particularly, two algorithm are selected, a gradient-based method (modified Gauss-Newton method) and an evaluation based one (Species- based Quantum Particle Swarm Optimization). In each case the strength and shortcoming of the adopted methods as well as the role played by the adopted benchmarks and the type of observation in model calibration is assessed. A concept of in-situ recalibration of the model is defined and its importance is highlighted. This method is used to determine advanced constitutive model parameters using in-situ tests and geometrical observations. As a conclusion, the research shows how using an inverse analysis algorithm may improve the modelling of geotechnical problems involving large deformations and, particularly, facilitate model calibration and discovering the shortcoming and strength of the numerical models. [edited by Author]