Characterization of Geotechnical Properties Using Bayesian Theory and Multi-Source Data Fusion Methods

基於貝葉斯理論和多源數據融合方法的岩土參數表徵研究

Student thesis: Doctoral Thesis

View graph of relations

Author(s)

Detail(s)

Awarding Institution
Supervisors/Advisors
  • Lulu Zhang (External person) (External Supervisor)
  • Yu WANG (Supervisor)
Award date3 Nov 2022

Abstract

Growing urbanisation worldwide has driven the construction of numerous new urban infrastructure projects, requiring large amounts of geomaterials (soil and rock) as essential building materials. Geomaterials are natural materials whose geotechnical properties show apparent uncertainty across sites and even between different soil layers at the same site. This uncertainty can be modelled by a random variable with a probability distribution function (PDF) or a multivariate probability distribution model for multiple geotechnical properties. Furthermore, geotechnical properties also exhibit spatial variability within the same soil layer at a site. These characteristics of underground geomaterials play a vital role in engineering design and analysis.

Characterization of geotechnical properties can be divided into two scenarios: 1) when the spatial variability is not considered explicitly (e.g., using a PDF model), and 2) when the spatial variability is explicitly considered (e.g., using a random field with explicit consideration of spatial coordinates). In these scenarios, geotechnical properties usually exhibit non-Gaussian and non-stationary characteristics. In addition, site-specific measurement data are often sparse and limited in engineering practices, sometimes even comprising just a few data points. Because of these characteristics, characterization of geotechnical properties is challenging.

This study is intended to develop innovative methods to address these challenges. When the spatial variability is not explicitly considered, a non-parametric method is proposed to construct a site-specific multivariate probability distribution model by integrating site-specific limited and sparse geotechnical data with regional database information. The proposed method is non-parametric and does not pre-specify the PDF types. When the spatial variability is explicitly considered, a Bayesian inference method based on Markov chain Monte Carlo (MCMC) simulation is proposed to estimate the spatial variability of non-stationary geotechnical data. Furthermore, the Bayesian inference method is combined with the cokriging method to estimate the spatial variability by integrating two correlated geotechnical properties. To overcome the challenges from extremely sparse geotechnical data, a novel 1D data fusion method, called collaborative Bayesian compressive sampling (Co-BCS), is proposed to integrate the extremely sparse geotechnical data with another correlated and abundant geotechnical data. Finally, to estimate the two-dimensional (2D) subsurface spatial variability based on extremely sparse geotechnical data, a new 2D data fusion method, called multi-source Bayesian compressive sampling (MS-BCS), is proposed to integrate the extremely sparse geotechnical data with abundant geophysical data. The proposed Co-BCS and MS-BCS can improve the performance of characterization of spatial variability by multi-source data fusion.

The methods developed in this study relax the assumptions of stationarity and normality, which conventional methods often depend on, and can effectively handle sparse measurement data of geotechnical properties. The developed new methods perform well in numerical and real-life cases, and promote the characterization of geotechnical properties in engineering design and analysis.