Search for: All records

Backward erosion piping (BEP) has been recognised as a major cause of failures in water-retaining structures. However, the fundamental mechanisms controlling the phenomenon are not well understood. This research applies the theory of rate processes to develop a constitutive relationship between energy density of the seepage flow and the erosion rate of soils during the evolution of BEP. The resulting equation is used to analyse four datasets of previously reported experimental observations. The mechanical parameters estimated through the proposed model fall into the ranges of values that were reported in the literature. To validate the proposed approach, the constitutive model was incorporated into a multiphase numerical framework to simulate evolution of BEP in embankment soil and compared with reported experimental observations. The numerical framework with the proposed constitutive model is shown to be capable of reproducing both the observed evolution of local hydraulic gradients and pipe progression in the structure. 

Large Language Models (LLMs) have demonstrated unprecedented capabilities in code generation. However, there remains a limited understanding of code generation errors that LLMs can produce. To bridge the gap, we conducted an in-depth analysis of code generation errors across six representative LLMs on the HumanEval dataset. Specifically, we first employed open coding and thematic analysis to distill a comprehensive taxonomy of code generation errors. We analyzed two dimensions of error characteristics -- semantic characteristics and syntactic characteristics. Our analysis revealed that LLMs often made non-trivial, multi-line code generation errors in various locations and with various root causes. We further analyzed the correlation between these errors and task complexity as well as test pass rate. Our findings highlighted several challenges in locating and fixing code generation errors made by LLMs. In the end, we discussed several future directions to address these challenges. 

Abstract Cardiomyocytes are viscoelastic and key determinants of right ventricle (RV) mechanics. Intracellularly, microtubules are found to impact the viscoelasticity of isolated cardiomyocytes or trabeculae; whether they contribute to the tissue-level viscoelasticity is unknown. Our goal was to reveal the role of the microtubule network in the passive anisotropic viscoelasticity of the healthy RV. Equibiaxial stress relaxation tests were conducted in healthy RV free wall (RVFW) under early (6%) and end (15%) diastolic strain levels, and at sub- and physiological stretch rates. The viscoelasticity was assessed at baseline and after the removal of microtubule network. Furthermore, a quasi-linear viscoelastic (QLV) model was applied to delineate the contribution of microtubules to the relaxation behavior of RVFW. After removing the microtubule network, RVFW elasticity and viscosity were reduced at the early diastolic strain level and in both directions. The reduction in elasticity was stronger in the longitudinal direction, whereas the degree of changes in viscosity were equivalent between directions. There was insignificant change in RVFW viscoelasticity at late diastolic strain level. Finally, the modeling showed that the tissue's relaxation strength was reduced by the removal of the microtubule network, but the change was present only at a later time scale. These new findings suggest a critical role of cytoskeleton filaments in RVFW passive mechanics in physiological conditions. 

Large Language Models (LLMs) have recently been widely used for code generation. Due to the complexity and opacity of LLMs, little is known about how these models generate code. We made the first attempt to bridge this knowledge gap by investigating whether LLMs attend to the same parts of a task description as human programmers during code generation. An analysis of six LLMs, including GPT-4, on two popular code generation benchmarks revealed a consistent misalignment between LLMs' and programmers' attention. We manually analyzed 211 incorrect code snippets and found five attention patterns that can be used to explain many code generation errors. Finally, a user study showed that model attention computed by a perturbation-based method is often favored by human programmers. Our findings highlight the need for human-aligned LLMs for better interpretability and programmer trust. 

ABSTRACT Backward erosion piping (BEP) is a significant contributor to failures in global flood protection infrastructure, yet it remains among the least understood geotechnical phenomena, particularly concerning the fundamental mechanisms driving its initiation. This study focuses on the development of a novel stochastic framework for the prediction of critical hydraulic gradients causing BEP initiation. The novelty of the study lies in the following: (1) the development of a grain‐scale probabilistic model based on fundamental mechanisms by means of the theory of rate processes, (2) quantification of the influence of soil variability on BEP initiation probability by introducing an initiation probability function, and (3) an analytical framework reconciling grain kinetics of BEP initiation with the Weibull distribution. A particle‐scale BEP initiation probabilistic model is first established based on fundamental grain kinetics under seepage flow by using the theory of rate processes. To investigate how soil variability influences initiation, a stochastic dual random lattice modeling framework is exercised, complemented by direct x‐ray computed tomography measurements of soil variability conducted on sand samples. The analytical probabilistic model for BEP initiation closely aligns with the Weibull distribution, also demonstrating that soil variability influences both the scale and shape parameters of the distribution. This work establishes the linkage between probability of BEP initiation as described by the theory of rate processes and phenomenological Weibull statistics. Findings presented herein bring the potential to develop a multiscale probabilistic framework by means of Weibull statistics for evaluating the probability of BEP initiation at multiple scales. 

ABSTRACT Backward erosion piping (BEP) is a leading internal erosion mechanism for flood protection system failures. A model capable of predicting critical hydraulic conditions for BEP initiation at multiple scales while also incorporating soil variability is a pressing need. This study formulates and validates a novel multiscale probabilistic BEP initiation framework with incorporation of soil variability. The framework is based on a grain‐scale probabilistic model and the weakest link theory, and the theory of rate processes. The multiscale framework proposed herein is validated through a wide range of available experimental data from independent sources, encompassing tests performed at multiple scales. Following calibration with small‐scale experimental data, the model demonstrates accurate prediction of critical hydraulic gradients at larger scales (3–6 orders of magnitude difference), including the ability to capture the grain size dependence of BEP initiation and providing uncertainty estimates. A systematic analysis is performed to uncover the effects of different soil properties on multiscale critical hydraulic conditions.