Beyond the deterministic approach - on the feasibility of data assimilation methods in geotechnics

Doctoral thesis, 2023

The surge in economic and social development has resulted in significant challenges, especially for linear physical infrastructure. A substantial part of the ageing physical linear infrastructure has been built on problematic soils and often have poorly documented foundation solutions. A typical example is the case of embankments for infrastructure on soft ground conditions. Soft soils possess various important characteristics that contribute to their complex emerging soil response when subjected to hydro-mechanical loading. In recent times, numerous advanced constitutive models grounded in various theories and hypotheses have emerged to capture the behaviour of soft soils. These models differ from models commonly used in geotechnical engineering, as they encompass complex soft soil features, \eg anisotropy, rate-dependency and degradation of bonding that enable reasonably accurate predictions for test data obtained under controlled laboratory conditions. However, their applicability for making informed decisions on large-scale field projects may be limited, as deterministic calculations alone may not adequately consider the variability in the behaviour of geomaterials encountered in real-world scenarios. Furthermore, not all model parameters have direct geotechnical significance as they are derived solely from mathematical expressions, posing challenges in their identification. With the growing utilisation of advanced constitutive models in engineering analysis, the input parameters for these models take on crucial roles as design parameters. This thesis provides a probabilistic methodology that enables the identification of parameters of constitutive models for geotechnics, through inverse analysis using Data Assimilation (DA). The primary objective of this thesis is to evaluate the applicability of existing Data Assimilation concepts in the field of geotechnical engineering. To achieve this, a modular framework that allows the implementation and use of multiple DA methods in conjunction with geotechnical numerical codes is created. A comprehensive and systematic comparison of contemporary state-of-the-art DA schemes specific to geotechnical engineering is performed along with examining the factors influencing their performance. Additionally, hybridisation of meta-heuristic algorithms with classical Data Assimilation methods has also been proposed to improve some of the observed drawbacks. In this thesis, the limitations of the deterministic approach has been demonstrated and the need for a robust probabilistic tool is shown to be paramount. It has also shown that it is time to start embracing the value of monitoring data which can be put to efficient use when a robust probabilistic framework like Data Assimilation is considered.