Z. Guan
Please Note
15 records found
1
Characterisation of failure modes and their associated consequences is crucial for assessing the performance and reliability of embankments. Among the many factors influencing embankment stability, surcharge loads and soil spatial variability play significant roles in determining both the probability and the consequence of failure. In this work, a parametric study is conducted to investigate the effects of surcharge load length (e.g. as in rail traffic loading), distance from the load to the slope crest, and horizontal scale of fluctuation of soil strength variability on the failure behaviour of a 3D embankment, using the Random Finite Element Method. Five hundred realisations are generated for each stochastic scenario, for which the ultimate bearing capacity of the embankment is calculated under undrained conditions and the failure consequences are quantified in terms of failure volume and length. The results indicate that the surcharge load distribution and soil spatial variability jointly influence the system response. While the load-to-crest distance typically governs the dominating failure mode, the ratio between the horizontal scale of fluctuation and the load length governs the likelihood of localised failure. It is found that the most adverse scenario corresponding to the lowest mean bearing capacity is associated with relatively smaller failures.
In geotechnical site investigation, Robertson’s soil behavior type (SBT) chart is widely used for soil classification based on two quantities measured during a cone penetration test (CPT), the normalized cone resistance Qt and the normalized friction ratio FR. Qt and FR are negatively correlated and provide complementary information for soil classification. However, this cross-correlation between Qt and FR has not been explicitly modelled in previous studies of subsurface soil classification and stratification using an often-limited number of CPT soundings from a specific site. This study aims to leverage such cross-correlation for improving CPT-based stratification and zonation by a joint sparse representation of Qt and FR in a vertical cross-section, as well as quantifying their uncertainty under a Bayesian framework. In addition, direct application of the SBT chart to a vertical cross-section often leads to noisy results (e.g., SBTs fluctuate rapidly and unrealistically within short distances). The noises are subsequently removed mainly by subjective engineering judgment in current practices. In this study, a randomization of input measurements is proposed to filter out the noise and improve computational efficiency simultaneously. Both simulated and real data examples are used to illustrate the proposed method. The results indicate that the proposed method significantly improves accuracy of the soil classification and stratification and automatically removes the noise.
Several global or regional databases for various types of soils have been developed due to their importance in engineering design and analysis. However, a database is not yet available for collapsible loess in which severe geohazards often occur. In this study, a comprehensive loess database with twelve soil parameters is compiled by collecting results of field and laboratory tests on collapsible loess from the city of Xi’an, China. Basic statistics, marginal probability distribution functions (PDFs), and a correlation matrix for loess parameters are estimated from the database. To the best of the authors’ knowledge, this is the first collapsible loess database at a municipal level. In addition, existing databases often lack sufficiently complete multivariate measurement data for a proper estimation of statistical correlations among multiple soil properties. In this study, this incomplete multivariate measurement data problem is tackled by Bayesian methods (i.e., Bayesian Gaussian mixture model and Bayesian compressive sampling (BCS) with Karhunen–Loève (KL) expansion, BCS-KL), which are illustrated and validated using the incomplete and complete subsets of the loess database, respectively. Both the Bayesian Gaussian mixture model and BCS-KL are non-parametric, and they offer a flexible way of modeling marginal PDFs and a correlation matrix from incomplete measurements in a realistic manner.