Gap-graded soil erosion modes predicted by µ-CT informed graded erosion model and multi-criteria assessment Authors: Bin Zhu, Yu-Fei Xie, Dai-Rong Su, Jia-Qi Wang Published: 2026 ABSTRACT This study investigates internal erosion in gap-graded gravelly soils, focusing on the critical fine content transition zone (30–35%). By integrating µ-CT imaging, geometric criteria analysis, and a novel predictive model based on the graded erosion principle, we demonstrate that fine particle content governs the transition between suffusion and piping erosion modes. Experimental results reveal that exceeding the 30–35% threshold shifts the soil fabric from a stable coarse-grained skeleton to an “over-filled” structure, where coarse particles float within a fine-dominated matrix, drastically increasing piping susceptibility. A multi-criteria assessment framework validates the limitations of traditional geometric criteria (e.g., Cu, Kenney & Lau) for transitional soils, while the grading entropy criterion offers enhanced robustness. The proposed graded erosion model successfully predicts particle size distribution evolution toward stable Fuller limits and porosity changes by incorporating particle-size-sensitive erosion rates with physical thresholds. This integrated methodology advances the understanding and prediction of internal erosion in gap-graded soils, supporting the design of resilient geotechnical structures and improving infrastructure risk assessment. Introduction In engineering practice, erosion problems of granular soil typically encompass two main concerns: Seepage Stability Evaluation and Erosion Mode Prediction. The former assesses water seepage and its effects on soil structure. Factors like permeability, pore pressure, and fine particles influence soil stability, helping identify risks like suffusion or suffosion (piping), In this study, suffosion and piping are considered synonymous terms, representing the same type of erosion hazard.which weaken soil and cause instability1. The latter predicts erosion types based on soil conditions, such as suffusion (transportation of fine particles) or suffosion (formation of piping channels and sinkhole). In the Xiacun Town case2, groundwater under intense rainfall or pumping conditions developed high hydraulic gradients, continuously eroded fine particles, and formed seepage channels, ultimately leading to soil erosion and ground collapse.Accurate prediction is vital for designing structures that resist erosional damage and ensure long-term stability. The earliest and most enduring geometric criterion, which is based on the geometric shape of the gravel soil gradation curve, is an empirical evaluation method. Due to its intuition and practicality, these methods continue to serve as the primary approaches for assessing the seepage stability of gravel soils. Gravel soils with gap gradation, which are susceptible to seepage erosion, can experience a transition in erosion modes due to the influence of their fine particle content, specifically from suffusion to piping erosion, accompanied by changes in permeability3,4. Previous studies have indicated that a fine grain fraction of 30% (GAQSIQ-CHN, 2022)5 is a critical state for gap-graded gravel soils. Beyond this threshold, coarse grains cannot form a dominate fabric or skeleton and tend to “float” within the finer grains, which limit is up to 35% fraction defined by Kenney and Lau (1985)6. Vaughan (2006)7 described this as an ‘‘over-filled’’ structure. However, this process requires a higher hydraulic gradient. Here, increased pore pressure results in a loss of shear strength. Zhong et al.8 found through their research that both specimen size and hydraulic loading history affect the critical hydraulic gradient, with the critical hydraulic gradient decreasing as the specimen length (seepage path) increases. Li et al.9 examined how hydraulic gradient, confining pressure, and initial density influence internal erosion and post-erosion behavior of gap-graded soils.In contrast, suffusion erosion occurs with lower hydraulic gradient (smaller than 1.0) and typically happens in what is considered a stable grading soil10. Therefore, the content of fine grains plays a crucial role in determining the erosion mode and internal stability of granular soil11. A fine particle content between 30% and 35% represents a critical and sensitive transition zone where the erosion mode shifts from suffusion to piping5,6. Recent studies on suffusion in gap-graded soils show that the erosion process can be divided into distinct stages and that soils with transitional microstructures are more sensitive to stress conditions12, while triaxial tests indicate that preferential flow paths dominate suffusion kinetics under constant mean effective stress, with a limited role of stress deviator13. In recent years, the development of coupled Discrete Element Method and Computational Fluid Dynamics numerical simulations1,12,13 and micro-observation techniques such as tomography14,15 have provided new approaches for studying the hydrodynamic mechanisms and microstructural evolution of erosion at the particle scale16., However, from the perspective of geometric criteria, traditional methods struggle to precisely delineate the mesoscale processes of pore pressure or fluid flow variations with changing fine particle content. They also face challenges in directly observing the dynamic evolution of how fine particles gradually participate in the internal structural action of soils under near-critical conditions in large-scale engineering, particularly within the critical transition zone of fine particle content (30%–35%). Existing gradation-based criteria exhibit limitations and inconsistencies when evaluating the internal stability of gravelly soils18. While emerging approaches, such as entropy-based criteria19,20 and energy-based erosion theory21, attempt to provide more unified quantitative assessments, a deeper understanding is still required regarding the direct link between the microscopic mechanisms governing erosion mode transition and the corresponding macroscopic hydraulic responses within this critical fine-content transition zone. From a geometric criterion perspective, it is challenging to precisely visualize how pore pressure or fluid flow changes with varying fine particle content. Moreover, directly observing variations in pore pressure on a large engineering scale, especially near critical state—specifically in the range of fine particle content from 30% to 35% where fine particles are believed to gradually participate in the internal structural action of granular soil—is challenging. Currently, an effective approach is to observe these changes directly at the pore scale using tomography techniques such as Scanning Electron Microscopy (SEM), Computerized X-ray Tomography (CT), and other advanced technologies. These methods offer new avenues for studying the internal structure of granular soils22,23. By measuring the geometry structure of micro-pores and particles in µ-CT images, researchers can obtain the macroscopic hydraulic behaviors of sandstone by mechanic simulation, which is also applicable to the seepage erosion evaluation. Nguyen et al. showed through X-ray CT tests that piping induces heterogeneous evolution of the microstructure and physical properties of granular soils24,25. This study employs µ-CT imaging to examine the hydraulic behavior of gap-graded gravel soils under near-critical conditions. Three distinct groups of granular soil specimens were analyzed, with incremental increases in fine-grain content and specific exclusions of particle sizes. First, a criterion integrated evaluation index is proposed mainly based on graded erosion principle. It innovatively links µ-CT microstructural information with macroscopic hydraulic responses and particle gradation evolution, overcoming the limitations of traditional methods that rely on single indicators, (e.g., Uniformity coefficient, Kenney & Lau index, optimal fine particle content) and the incompatibility in criteria for assessing the internal instability of gravelly soils, which has been a persistent challenge.The proposed model can predict the evolution of gradation and porosity during erosion and is cross-validated with multiple geometric criteria, achieving a more systematic assessment of erosion stability and modes such as Kenney and Lau. Second, the Micro-Mechanism of Erosion Mode Transition is researched derived from hydraulic criteria. It elucidates how fine-particle content dominates the flow intensity indirectly controlling the erosion mode by influencing the coarse-particle skeleton structure and quantitatively analyzes the high sensitivity of parameters such as hydraulic conductivity, deepening the understanding of the transition mechanism and the critical conditions. This study aims to elucidate, through CT imaging analysis, the transformation laws of erosion modes and their corresponding imaging characteristics in gravelly soil subjected to progressive particle erosion within critical gradation ranges. The rationality and effectiveness of the methodology will be validated against geometric criteria. Furthermore, an evaluation method will be established to correlate microscopic imaging features with the permeability effects and erosion modes of gravelly soil. Materials Preparation of specimens Three distinct groups of granular soil specimens were prepared and remoulded, each containing varying proportions of fine particles—30%, 32%, and 34%. These specimens were carefully packed into cylindrical Acrylic containers, each measuring 35 mm in height and diameter. They represented typical characteristics of gap-graded gravel soils, with specific particle size ranges deliberately omitted: 1–2 mm, 0.5–1.5 mm, and 0.25–0.5 mm. To capture detailed internal structures, high-resolution 3D X-ray microimaging was performed using the Zeiss Xradia 510 Versa system. This advanced imaging technology generated voxel datasets for each group, producing volumetric data sets of 1004 × 1014 × 1024 voxels. So, the imaging resolution achieved was 35 µm per voxel, ensuring comprehensive visualization and analysis of the soil specimens at the microscale. These datasets are crucial for studying the distribution and interaction of fine particles within the granular matrix, particularly in relation to the hydraulic behavior and permeability of gap-graded gravel soils near critical conditions. The principle of real-time X-ray CT scanning devices is shown in Fig. 1. To thoroughly investigate the internal erosion behavior of gap-graded gravelly soils approaching critical conditions, we employed micro-computed tomography (µ-CT) imaging and processing techniques. These included region of interest (ROI) extraction, watershed threshold segmentation, coarse and fine particle statistics, and annotation methods to process the acquired CT scan images of soil samples (Fig. 2). These techniques enable high-resolution extraction and detailed analysis of the soil’s internal structure. The study utilized three sets of granular soil samples with varying fine particle contents, strictly controlled within the range of 30% to 35%. The samples were categorized based on the absence of certain particle sizes. The particle sizes missing from each sample fell into three specific ranges: 1–2 mm, 0.5–1 mm, and 0.25–0.5 mm. By identifying and documenting these missing sizes, we were able to classify the samples into distinct groups. Fig. 1Schematic illustration of X-ray CT scanning. Schematic illustration of X-ray CT scanning. -specimen lacks finer grains in the 1–2 mm size range;-specimen lacks finer grains in the 0.5–1 mm size range;-specimen lacks finer grains in the 0.25–0.5 mm size range. -specimen lacks finer grains in the 1–2 mm size range; -specimen lacks finer grains in the 1–2 mm size range; -specimen lacks finer grains in the 0.5–1 mm size range; -specimen lacks finer grains in the 0.5–1 mm size range; -specimen lacks finer grains in the 0.25–0.5 mm size range. -specimen lacks finer grains in the 0.25–0.5 mm size range. We further classified the samples within each main group into three sub-groups (L, M, H) based on the accumulated content of sub-2.0 mm fines, with the labels L (Lower), M (Medium), and H (Higher) assigned according to their grain size distribution curves: (1) A specimen labeled as L refers to gap-graded gravel soil with a lower finer-grain accumulated content of 30%; (2) A specimen labeled as M refers to gravel soil with a median finer-grain content of 32%; (3) A specimen labeled as H refers to gravel soil with a higher finer-grain content of 34%. The particle size distribution curves for these groups and sub-groups classified samples are shown in Fig. 3. Fig. 2Gap-graded gravel soil models: (a) Raw 3D images of three sub-groups labeled M from -specimen, -specimen, and -specimen, respectively. In these images, the filling of finer grains (smaller than 2 mm) gradually decreases from left to right. The ‘floating’ coarse grains are visible, illustrating the lack of mutual support among them. (b) Treated 3D image after applying the segmentation, removing occluded porosities. (c) Further segmentation and labeling to the different microstructures, such as small pores (red), large pores (blue) and grains (green). (d) Filling finer particles (smaller than 2 mm). Gap-graded gravel soil models: (a) Raw 3D images of three sub-groups labeled M from -specimen, -specimen, and -specimen, respectively. In these images, the filling of finer grains (smaller than 2 mm) gradually decreases from left to right. The ‘floating’ coarse grains are visible, illustrating the lack of mutual support among them. (b) Treated 3D image after applying the segmentation, removing occluded porosities. (c) Further segmentation and labeling to the different microstructures, such as small pores (red), large pores (blue) and grains (green). (d) Filling finer particles (smaller than 2 mm). Single indicator for gravelly soil erosion analysis The materials of these specimens come from an open-pit site on the banks of the Li River in Guilin, China. Each component of every specimen, including both main and sub-groups, undergoes carefully sieving, weighing, and is finally placed into Acrylic transparent containers for CT scanning. Figure 3 illustrates the particle size distribution (PSD) of the three groups of gap-graded gravel soils (-specimen, -specimen, -specimen) and their respective sub-groups. They are used to analyze the meso-structure and internal erosion issues. The preparation of each group of specimens was made by mixing coarse particles greater with a corresponding amount of fine particles. A set of parameters describing the soil grade characteristics were defined as follows, their values are shown in Table 1.1\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{C}_{u}={d}_{60}/{d}_{10},\:\:{C}_{c}={d}_{30}^{2}/\left({d}_{60}{d}_{10}\right)$$\end{document} where Cu is the uniformity coefficient; Cc is the curvature coefficient; d10 is the particle size at which 10% of all particles by weight pass through the filter (mm), also called the effective particle size of granular soils; and d30, d60 have similar geometrical meanings to d10. However, d60 is the controlled particle size of granular soils. Fig. 3Grading curves of three group gap-grading gravel soils, (a)~(d) DSP curves combined with bar plot. Grading curves of three group gap-grading gravel soils, (a)~(d) DSP curves combined with bar plot. According to the early study by Istomina (1957)26, as cited in Busch et al. (1993)27, it is widely accepted that when the coefficient of uniformity (Cu) exceeds 20, gravelly soils tend to exhibit suffusion characteristics and internal instability, where turbid flow can occur under relatively low hydraulic gradients (0.2–0.5) due to the transport of fine particles11. In contrast, when Cu is less than 10, the interlocking effect between similarly sized particles enhances soil stability, generally resulting in suffosion characteristics and a structurally stable internal state, with erosion typically requiring higher hydraulic gradients (greater than 1.0). Table 1Initial evaluation of erosion mode and basic parameters of all specimens measured by test according to Istomina’s criteria26,44.Soil type/[mm]Specimen No.Porosity\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\boldsymbol{\varphi\:}/\left[\boldsymbol{\%}\right]$$\end{document}Uniformity \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\boldsymbol{C}}_{\boldsymbol{u}}/[-]$$\end{document} Curvature \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\boldsymbol{C}}_{\boldsymbol{c}}/[-]$$\end{document} Characteristics of particle size/[mm]Erosion modeInternal instabilityd 10d 30d 60d 70-spec.L19.419.04.20.212.004.05.0TransitionS/UM1920.50.60.190.753.94.9SuffosionUH18.421.10.40.180.583.84.8SuffosionU-spec.L21.619.04.20.212.004.05.0TransitionS/UM20.620.53.10.191.523.94.9SuffosionUH19.521.11.90.181.153.84.8SuffosionU-spec.L23.66.21.50.652.004.05.0SuffusionSM22.56.51.50.591.833.94.9SuffusionSH21.26.91.40.551.703.84.8SuffusionSNote: Transition (10 < Cu < 20), Suffusion (Cu ≤ 10) and Suffosion or piping (Cu ≥ 20). S for geometric Stable; U for geometric Unstable; S/U geometric Stable or Unstable (transition zone). L for lower finer-grain accumulated content of 30%; M for median finer-grain content of 32%; H for higher finer-grain content of 34%. Table 1Initial evaluation of erosion mode and basic parameters of all specimens measured by test according to Istomina’s criteria26,44.Soil type/[mm]Specimen No.Porosity\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\boldsymbol{\varphi\:}/\left[\boldsymbol{\%}\right]$$\end{document}Uniformity \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\boldsymbol{C}}_{\boldsymbol{u}}/[-]$$\end{document} Curvature \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\boldsymbol{C}}_{\boldsymbol{c}}/[-]$$\end{document} Characteristics of particle size/[mm]Erosion modeInternal instabilityd 10d 30d 60d 70-spec.L19.419.04.20.212.004.05.0TransitionS/UM1920.50.60.190.753.94.9SuffosionUH18.421.10.40.180.583.84.8SuffosionU-spec.L21.619.04.20.212.004.05.0TransitionS/UM20.620.53.10.191.523.94.9SuffosionUH19.521.11.90.181.153.84.8SuffosionU-spec.L23.66.21.50.652.004.05.0SuffusionSM22.56.51.50.591.833.94.9SuffusionSH21.26.91.40.551.703.84.8SuffusionSNote: Transition (10 < Cu < 20), Suffusion (Cu ≤ 10) and Suffosion or piping (Cu ≥ 20). S for geometric Stable; U for geometric Unstable; S/U geometric Stable or Unstable (transition zone). L for lower finer-grain accumulated content of 30%; M for median finer-grain content of 32%; H for higher finer-grain content of 34%. Initial evaluation of erosion mode and basic parameters of all specimens measured by test according to Istomina’s criteria26,44. Note: Transition (10 < Cu < 20), Suffusion (Cu ≤ 10) and Suffosion or piping (Cu ≥ 20). S for geometric Stable; U for geometric Unstable; S/U geometric Stable or Unstable (transition zone). L for lower finer-grain accumulated content of 30%; M for median finer-grain content of 32%; H for higher finer-grain content of 34%. However, while the Cu-based criterion offers certain reference value for macroscale erosion mode assessment, it has notable limitations in evaluating internal stability. These limitations are mainly reflected in the following aspects: First, as a global gradation parameter, Cu fails to accurately capture the mechanical interaction between fine particle fillings and the coarse-grained skeleton in gap-graded soils. Second, the method does not adequately account for the influence of fine particle content—even with the same Cu value, soils with different fine contents may show significant differences in internal stability and erosion mechanisms. Finally, microstructural characteristics (e.g., particle arrangement, pore connectivity) play a decisive role in internal stability, yet Cu cannot reflect such structural information. For instance, in the case shown in Fig. 2, where coarse particles are “floating” within a fine-grained matrix, the Cu value may exceed 20, but due to the lack of a stable coarse skeleton, the erosion mode is more likely to be internal instability-induced suffosion (structural collapse due to intensified seepage erosion) rather than stable suffusion (mere transport of fines). Therefore, in practical engineering applications, it is essential to incorporate multidimensional indicators such as fine particle content and pore structure parameters to establish a more systematic evaluation framework for the internal stability of gravelly soils. Advanced watershed threshold segmentation technology was applied to conduct a more detailed segmentation of three types of gap-graded gravel soil samples (Fig. 4), identifying large pores, small pores, matrix particles, and sporadically distributed mineral particles. This segmentation method allows for a clearer distinction of the microstructural characteristics of gravel soils, particularly the porosity and particle distribution, which are crucial for understanding the material’s erosional behavior under different environmental conditions28. From the Fig. 4, it can be observed that there is a higher content of fine particles corresponding to the Fig. 3: (1) -specimen, from top to bottom, these fine particles are mainly concentrated at the bottom of the sample, with some also present at the top. (2) -specimen contains the least amount of fine particles, all of which are concentrated at the bottom of the sample, forming a structure characteristic similar to a filter layer. (3) -specimen’s structural characteristics are intermediate between the above two samples, leaning more towards -specimen, as can also be seen from their particle gradation curves. Fig. 4Demonstration of three groups of specimens and their sub-groups. Micro-structural and macro-structural characteristics of three types of gravel soils are identified and labelled. (a) Soil specimen with a gap in the 1.0–2.0 mm fraction; (b) Soil specimen with a gap in the 0.5–1.0 mm fraction; (c) Soil specimen with a gap in the 0.25–0.5 mm fraction. Grain annotations (in green) represent geometric feature classifications, while mineral annotations (in yellow) represent compositional classifications. Due to the relatively homogeneous mineral composition of the sample and the resolution limits of CT imaging, only typical mineral examples are annotated. Demonstration of three groups of specimens and their sub-groups. Micro-structural and macro-structural characteristics of three types of gravel soils are identified and labelled. (a) Soil specimen with a gap in the 1.0–2.0 mm fraction; (b) Soil specimen with a gap in the 0.5–1.0 mm fraction; (c) Soil specimen with a gap in the 0.25–0.5 mm fraction. Grain annotations (in green) represent geometric feature classifications, while mineral annotations (in yellow) represent compositional classifications. Due to the relatively homogeneous mineral composition of the sample and the resolution limits of CT imaging, only typical mineral examples are annotated. In practical engineering, sieve analysis can obtain the gradation curves of gravelly soils, thereby providing rapid assessment indicators such as the coefficient of uniformity (Cu)4,29,30. However, through detailed image analysis of the soil microstructure, engineers can more accurately predict seepage-induced erosion modes. From a visual analysis perspective: Both Type - and Type - specimens exhibit distinctly unstable gradation characteristics, with significantly high fine content and localized occurrences of coarse particles being “floating” within the fine matrix. These specimens demonstrate clear suffosion susceptibility, though relatively high hydraulic gradients are required to trigger such erosion. In contrast, while Type - specimens maintain relatively stable overall gradation curves, they still possess potential suffusion risks even under low hydraulic gradients. Theorietical background Graded erosion principle for gravel soils For some erodible gravel soils with wider or gap-graded particle sizes, there exists a principle known as graded internal erosion. This process involves the selective erosion of finer grain sizes, which occurs at specific critical hydraulic gradients. Essentially, the finer the particles, the lower the hydraulic gradient required to mobilize and transport them within the range of granular, non-cohesive soils (e.g., sands and silts). This trend reverses for cohesive clay particles, where electrochemical binding forces significantly increase the resistance to erosion, often requiring higher hydraulic gradients or prior structural breakdown for transport to occur.This principle highlights how different soil particle sizes behave differently under flowing water, influencing erosion patterns accordingly4,17,45. Assuming that the critical hydraulic gradient for a certain grain size is linearly related to its weight proportion in the PSD curve, the principle can be shown as follows:2\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\left({i}_{cr}\right)}_{j}=\frac{0.85{d}_{j}}{{P}_{j}{d}_{85}}\left(1-n\right)\frac{{\gamma\:}^{{\prime\:}}}{{\gamma\:}_{w}}$$\end{document} where dj is the diameter of the jth grain-size grade that has been eroded out of the soil (mm), and Pj is the content corresponding to the jth grain size grade (%). Figure 5 shows the erosion process of gravel soil across states from i−1 to i + 1. Fig. 5Illustration of state transition caused by internal erosion in which it can be seen that finer particles are lost gradually. Illustration of state transition caused by internal erosion in which it can be seen that finer particles are lost gradually. According to Indraratna & Vafai (1997)31, a series of geometric-based formulas to predict various soil characteristics like particle size distribution (PSD), porosity, and permeability are developed. These formulas are derived from the mass balance of mixed liquid flowing out of eroded soil. The calculating volumes such as the mixed liquid flushed out (VSout), soil eroded from the unit soil, and water flowing out of the unit soil (VWout) calculated from the in situ measured flow rate (Q) are specified. This method helps understand how gravel soil properties change as water erodes it over gravel soil erosion process. So, for the unit soil, water and components of mixed liquid meet the overall mass balance as follows:3\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\left\{\begin{array}{c}{V}_{out}=Q\varDelta\:t={V}_{{s}_{out}}+{V}_{{w}_{out}}\\\:{\rho\:}_{m}{V}_{out}={\rho\:}_{s}{V}_{{s}_{out}}+{\rho\:}_{w}{V}_{{w}_{out}}\end{array}\right.$$\end{document} where ρm, ρs, ρw is the density of the mixed liquid, soil grains, and water, respectively. The value ρm can be measured in situ. Indraratna & Vafai’s (1997)31 demonstrates how the content of gravel particles changes under different erosion states at a small scale (test sample scale). From the point of view of solving the smaller scale problem, if we use numerical methods to make a prediction of the PSD curves at a specimen scale, the above equations seem to be sufficient, just as done by the literature31. However, for some inverse problems, it is not practical, such as assessing the water-inrush volume or flow rate (Q). Estimating the soil erosion rate would be a practical and alternative method to simplify the prediction of PSD. Zhu et al. (2023) [4] derived formulas for predicting the evolution of PSD curve, porosity as shown in the following:4\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{P}_{j}^{\left(i+1\right)}=\left\{\begin{array}{cc}\frac{{P}_{j}^{\left(i\right)}-{\alpha\:}^{\left(i\le\:{\boldsymbol{I}}_{\boldsymbol{G}}\right)}}{1-{\alpha\:}^{\left(i\le\:{\boldsymbol{I}}_{\boldsymbol{G}}\right)}};&\:\:\:\:\:\:\:\:\:\:\:{\alpha\:}^{\left(i\le\:{\boldsymbol{I}}_{\boldsymbol{G}}\right)}={\left({P}_{j}^{i\le\:{\boldsymbol{I}}_{\boldsymbol{G}}}\right)}_{{P}_{j}^{\left(i\right)}\le\:{S}_{{I}_{G}}\le\:{P}_{x}}\cdot\:\frac{\stackrel{\sim}{m}}{{S}_{{I}_{G}}}\\\:\frac{{P}_{j}^{\left(i\right)}-{\alpha\:}^{\left({\boldsymbol{I}}_{\boldsymbol{G}}\right)}}{1-{\alpha\:}^{\left({\boldsymbol{I}}_{\boldsymbol{G}}\right)}};&\:{\alpha\:}^{\left({\boldsymbol{I}}_{\boldsymbol{G}}\right)}={\left({P}_{j}^{{\boldsymbol{I}}_{\boldsymbol{G}}}\right)}_{{P}_{j}^{\left(i\right)}>{P}_{x}}\cdot\:\frac{\stackrel{\sim}{m}}{{S}_{{I}_{G}}}\end{array}\right.$$\end{document}(PSDprediction)5\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\varphi\:}^{(i+1)}={\varphi\:}^{\left(i\right)}+\left(1-{\varphi\:}^{\left(i\right)}\right)\cdot\:{\varDelta\:\varphi\:}^{\left(i\right)}$$\end{document}(Porosityestimation)6\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\varDelta\:\varphi\:}^{\left(i\right)}=\frac{{V}_{{S}_{out}}}{{V}_{T}};\left({S}_{i}\le\:\stackrel{\sim}{m}\le\:{S}_{i+1}\right),\:{V}_{{S}_{out}}=\raisebox{1ex}{$\stackrel{\sim}{m}$}\!\left/\:\!\raisebox{-1ex}{${\rho\:}_{m}$}\right.$$\end{document} where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\stackrel{\sim}{m}$$\end{document} is the erosion rate of soils corresponding to a grain size range, which is similar to the concept of the erosion rate of soils per unit pore area suggested by Reddi et al. (2000)32; The “IG” represents a specific grading state in which the fine grains are completely eroded from the i state to the i + 1 state under the corresponding hydraulic gradient; SIG denotes the total amount of fine particles that can be fully eroded in the IG state. This threshold denotes a specific grading state (IG) at which the fine particles are considered to be completely eroded from the current grain-size distribution under the prevailing hydraulic conditions.; ϕ is the porosity; VT is the total volume of gravel soil; \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\alpha\:}^{\left(i\right)}$$\end{document} represents the cumulative mass fraction (in percentage) of fine particles that have been eroded up to the ith erosion state. It is an indicator that quantifies the amount of fine particles lost within a specific grain-size range, provided that the erosion limit (e.g., below the Fuller curve) has not been exceeded. Differently with Indraratna & Vafai’s (1997)31 method, introducing a parameter \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\stackrel{\sim}{m}$$\end{document} for each state, where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\stackrel{\sim}{m}$$\end{document} = SIG indicates complete erosion of particles with diameter j, and\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\stackrel{\sim}{m}$$\end{document} < SIG denotes partial remaining of those particles. Given an estimation of the erosion rate of soil \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\stackrel{\sim}{m}$$\end{document} (or, calculable by some empirical model), the quasi-piecewise function (Eq. 4), α(i), depends on Pj(i) when the condition of Pj(i) ≤ SIG ≤ Px is met; otherwise, α(IG) is a constant, this conveniently determines the PSD of granular soils by volume percentage. The first formula in Eq. 4 may yield a negative value, which can be set to zero if mass conservation is not considered. Of course, erosion only occurs in the fine particle fraction of granular soils, specifically where Pj ≤ Px and d0 > d10. Actually, during the erosion process of gravel soil, the value of d0 continually changes, altering the erosion limit SIG (%) for each state. Thus, an implicit procedure constantly monitors these changes. Once if fine grains are not completely eroded and the erosion conditions persist, the erosion process continues to guide the PSD curve to a stable state. To facilitate this analysis, a computer subroutine (Granular_soil_PSD) has been developed in MATLAB to plot the PSD curve for any given state. Grading State Index IG The changes in soil particle gradation can occur through two main mechanisms: first, particle breakage leads to an increase in fine particles, and second, the slow transport of fine particles leads to a decrease in fine particles. These changes can affect the fundamental properties of the soil, particularly the characteristics of its critical state33. Obviously, the issue of graded erosion in gap-graded gravel soil discussed in this paper pertains to the latter mechanism. In the process of internal erosion, the Fuller curve generally served as a limiting grading to determine the ultimate eroded state of the granular soil with a particle size distribution. Muir and Maeda (2008)33propose a grading state index to conveniently describe the changing grading of the soil, along with the rules governing how this index changes with compression or erosion, as well as its impact on the soil’s mechanical properties. A definition of the grading state index, IG, in a semi-logarithmic gradation plot is derived by comparing the areas under the current particle size distribution and under a limiting distribution, i.e. Fuller limit (Fig. 6a). The IG34 ranges from 0 to 1, reflecting the transition of soil grading from a uniform particle size to a limiting grading, which may potentially exhibit fractal characteristics35. Fig. 6(a) Definition of IG as ratio of areas ABC and ABD under current and limiting gradings; (b) gradings prediction (solid curves) based on graded erosion principle, grading reached by removal of particles from initial grading with RD = 10 (dotted curve) (cited from Muir et al., 201034. (a) Definition of IG as ratio of areas ABC and ABD under current and limiting gradings; (b) gradings prediction (solid curves) based on graded erosion principle, grading reached by removal of particles from initial grading with RD = 10 (dotted curve) (cited from Muir et al., 201034. The definition of IG is exact for the linear grading AC (dot line) in Fig. 6a. However, symmetrical grading of granular soils (solid curve) in Fig. 6a is commonly observed in real-world scenarios. To simply and approximate, the particle size ratio, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{R}_{D}={d}_{max}/{d}_{min}$$\end{document}, is introduced as a characteristic ratio of the maximum to minimum particle size. The area under this curve, which forms a triangle on the particle size distribution chart, is given by (ln RD)/2. This value is closely approximated by the area under the corresponding RD = 10 grading curve, where the underestimate of area for D > D50 is balanced by the overestimate of area for D < D50.7\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{I}_{G}\approx\:\frac{ln{R}_{D}}{2B}$$\end{document} where B is the power of maximum particle size, treated as a constitutive parameter. The two measures of grading between IG and RD are essentially interchangeable. Erosion mass rate \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\stackrel{\sim}{\boldsymbol{m}}$$\end{document} Sterpi (2003)36 conducted a series of laboratory tests to investigate the effects of erosion and the transport of fine particles induced by seepage flow, referencing the experimental setup established by Kenney and Lau (1985)6. An empirical model for the erosion and transport of fine particles in granular soils, induced by seepage flow, has been presented. This model is based on an erosion-transport law developed and calibrated using data from laboratory seepage tests on reconstituted silty sand samples. It considers local volume reductions and changes in mechanical properties as potential direct consequences of fine particle erosion. The experimental data applied for calibration and simulation are cited in Fig. 7. Fig. 7Variations of the percentage by weight of eroded fine particles with time and with hydraulic gradient. (a) modification from the data of Sterpi, 200336; (b) simulation based on the specimens in this study (m0 = 32%). Variations of the percentage by weight of eroded fine particles with time and with hydraulic gradient. (a) modification from the data of Sterpi, 200336; (b) simulation based on the specimens in this study (m0 = 32%). For fitting the laboratory test data, the erosion law about \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\stackrel{\sim}{m}$$\end{document} was proposed between the percentage by weight of eroded fine particles µe, the hydraulic gradient i, and time t, expressed in hours8\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\stackrel{\sim}{m}={m}_{0}\left[1-exp\left(-{\left(\frac{t}{{t}_{0}}\right)}^{b}\cdot\:\frac{{i}^{c}}{a}\right)\right]$$\end{document} where m0 is initial fines content by weight, t0 is one hour; a, b and c are nondimensional fitting parameters according to type of granular soils adopted. It is important to emphasize that Eq. 8 leads to two primary implications. The first implication is that erosion occurs without threshold of hydraulic gradient below which the process does not occur. In soils with a well-graded particle distribution, particularly those with fine components within the silt range, this assumption is reasonable, as even a minimal hydraulic gradient can induce erosion of the finest particles. However, a different behavior may be anticipated in more uniform coarse soils, where a specific threshold for the hydraulic gradient, beyond which erosion occurs, can be identified38. The second implication is that, if sufficient time is provided, complete erosion will occur, meaning the process will entirely remove the fine particles (i.e., \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\stackrel{\sim}{m}$$\end{document} → m0 as t → ∞). However, this second limitation is unlikely to be realistic in engineering practice. In fact, when the gradient values are within those typically encountered in practical applications, the theoretical time required for complete erosion becomes exceedingly long (hundreds or thousands of years), thus losing its engineering significance. The emergence of these implications is due to the fact that the erosion model itself fails to incorporate the hydraulic gradient based on particle size threshold segmentation. As outlined in the propose in Eq. 6, it explicitly accounts for the critical hydraulic gradient necessary to drive particles of different sizes. Thus, an improved erosion model integrating the principles of graded erosion is as follows:9\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\left\{\begin{array}{c}{\stackrel{\sim}{m}}_{j}={m}_{j0}\left[1-exp\left(-{\left(\frac{t}{{t}_{0}}\right)}^{b}\cdot\:\frac{{{\left({i}_{cr}\right)}_{j}}^{c}}{a}\right)\right]\\\:\stackrel{\sim}{m}=\sum\:_{k}{\stackrel{\sim}{m}}_{k}\end{array}\right.$$\end{document} where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\left({i}_{cr}\right)}_{j}$$\end{document} is the critical hydraulic gradient needed to drive the jth grain-size particles. \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\stackrel{\sim}{m}}_{j}$$\end{document} is the erosion mass rate of jth grain-size particles. Therefore, this improved erosion model is a time-dependent exponential function with an upper limit determined by the grain size, reflected by the critical hydraulic gradient. When applied to predict the changes in particle size distribution (PSD) due to the erosion process, it is necessary to assess whether the erosion process has reached the lower limit, i.e., whether the PSD curve is below the Fuller curve. If not, erosion continues until the stopping condition is met. Methods and validation Temporal effects on erosion rate variation This section, based on the grading erosion principle we have proposed, conducts graded erosion PSD curve and porosity predictions for the three soil types under the effects of particle removal. The results are then compared with the predictions obtained from the entropy-based criteria and the optimal fine particle content criteria, proposed by Chinese code GAQSIQ-CHN (2022)5, for gravel soil erosion, providing a comprehensive assessment of the potential erosion modes for the three soil types. Based on the data from the three experimental specimens in this study and in accordance with the graded erosion theoretical analysis model presented in the previous section, the graded erosion process of and specimens are analyzed. The complete erosion processes of each specimen and their corresponding subsamples are described separately as follows: All three subgroups of -specimen will experience graded erosion at rates of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\stackrel{\sim}{m}$$\end{document}= 5% to 10%.The L-subgroup of 2-specimen will also be subjected to erosion at \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\stackrel{\sim}{m}$$\end{document} = 5% to 10%.The M-subgroup of 2-specimen will undergo graded erosion at \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\stackrel{\sim}{m}$$\end{document} = 5%, 10%, and 13%.The H-subgroup of 2-specimen will undergo graded erosion at \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\stackrel{\sim}{m}$$\end{document} = 5%, 10%, and 15%.-specimen remains naturally stable and is not affected by erosion. All three subgroups of -specimen will experience graded erosion at rates of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\stackrel{\sim}{m}$$\end{document}= 5% to 10%. All three subgroups of -specimen will experience graded erosion at rates of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\stackrel{\sim}{m}$$\end{document}= 5% to 10%. The L-subgroup of 2-specimen will also be subjected to erosion at \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\stackrel{\sim}{m}$$\end{document} = 5% to 10%. The L-subgroup of 2-specimen will also be subjected to erosion at \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\stackrel{\sim}{m}$$\end{document} = 5% to 10%. The M-subgroup of 2-specimen will undergo graded erosion at \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\stackrel{\sim}{m}$$\end{document} = 5%, 10%, and 13%. The M-subgroup of 2-specimen will undergo graded erosion at \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\stackrel{\sim}{m}$$\end{document} = 5%, 10%, and 13%. The H-subgroup of 2-specimen will undergo graded erosion at \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\stackrel{\sim}{m}$$\end{document} = 5%, 10%, and 15%. The H-subgroup of 2-specimen will undergo graded erosion at \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\stackrel{\sim}{m}$$\end{document} = 5%, 10%, and 15%. -specimen remains naturally stable and is not affected by erosion. -specimen remains naturally stable and is not affected by erosion. According to the parameters of the seepage erosion experimental setup reported by Sterpi (2003)36 as cited in Uzuoka et al. (2012)37, the seepage length in the prototype is 14 cm. In the present experiment, the seepage length of the specimen is 3.5 cm, resulting in a geometric scale ratio of 1:4 between the model and the prototype. Based on the time-scaling law for diffusion-type (or constant-pressure-drop) Darcy flow:10\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\frac{{t}_{p}}{{t}_{m}}={\left(\frac{{L}_{p}}{{L}_{m}}\right)}^{2}={\left(\frac{1}{4}\right)}^{2}=\frac{1}{16}$$\end{document} where tp and Lp denote the characteristic time and length of the prototype (from reference36), and tm and Lm represent the corresponding values for the present specimens. Thus, the characteristic seepage time of the specimen model is 1/16 of that of the prototype. Given that the particle size distribution of the present experimental specimens is similar to that of the samples tested by Sterpi (2003) (cited in Uzuoka et al. 201237), the fitting parameters provided in Ref36. (a = 4.02, b = 0.5, c = 4.02) are adopted directly. Using Eq. (8), the erosion rate \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\stackrel{\sim}{m}$$\end{document} of the present experimental specimens is simulated, m0 = 32% for example, under the different hydraulic gradient, and the results are shown in Fig. 7b. Effects of particle removal According to the principle of grading erosion, gravel soil with coarse grains as the primary structural framework initiates internal erosion from the fine-grain fraction. Within a specific hydraulic gradient range, erosion progressively affects larger grain particles until it reaches its limit. Throughout this internal erosion process, the particle size distribution (PSD) curve, porosity, and permeability of the gravel soil continually change (Fig. 8). Based on the Cu values in Table 1, a preliminary assessment indicates that the - and -specimens fall under the category of unstable gravel soils (Cu > 20). CFD simulations further support this finding [14,36 ~ 38], showing uneven distributions of pore pressure and flow field in both gravel soil specimens, making them susceptible to suffusion erosion. PSD curves prediction The PSD curves for the to -specimens, categorized by fine particle content, are shown in Fig. 8a and g. These curves illustrate the evolution of PSD in process of successive particle removal, and its alignment with the Fuller limit criteria. The transition from unstable to stable grain size distributions, driven by graded internal erosion and the loss of smaller particles, results in a more stable seepage control structure. Figure 8a and c show that the -specimen stabilizes after two erosion phases, regardless of fine grain content (High, Medium, or Low). The -specimen with Low fine particle content stabilizes after two phases (Fig. 8d), while the High and Medium specimens require three phases (Fig. 8e and f). The -specimen, a stable gravel soil, shows PSD curves entirely below Fuller limit. The definition of Fuller limit is as follow1111\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:P\left(d\right)=100\cdot\:{\left(\frac{d}{{d}_{max}}\right)}^{n}$$\end{document} where dmax is the maximum value of particle size (mm); d is an arbitrary particle size between d0 and dmax; and P(d) is the mass percent of particles smaller than d (mm); n is an exponent value normally ranged from 0.3 to 0.741,42. In Fig. 8, the range is set from 0.5 to 0.7, with the upper limit defined as 0.7, the lower limit as 0.5 and the median as 0.6. Each sub-figure includes a background of Fuller’s curve, commonly used in granular material formulations like gravel, sand, and concrete. Fuller’s curve describes the ideal particle size distribution for optimal packing and performance. The shaded region, between powers of 0.5 and 0.7 in Eq. 11, indicates the stable range for gravel soils against seepage erosion under certain conditions. Based on Lubochkov’s (1965) lab tests, cited by Kovacs (1981)41, the power value n was determined to be 0.6. Figs. 8e8f illustrate the three phases of internal erosion in the -specimen with High (H) and Medium (M) fine particle contents. For the H-labeled specimen, fine particles are fully eroded after three erosion phases, reaching stability at \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\stackrel{\sim}{m}$$\end{document} = 15%. In contrast, for the M-labeled specimen, stability is reached with only partial erosion (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\stackrel{\sim}{m}$$\end{document} = 13%). This demonstrates that the derived formula (Eqs. 4 and 5) can predict the geometric shape of the gravel soil PSD curve based on the parameters (IG and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\stackrel{\sim}{m}$$\end{document}), without needing to delve into complex mechanical or hydraulic details.The PSD calculation results based on the graded internal erosion method are shown for three granular soil specimens in figures (a) through (g). (a)~(c): The -specimen undergoes two erosion phases with three states. (d)~(f): The -specimen experiences three erosion phases with four states, excluding the L sub-group. (g): The -specimen is not eroded with a stable grading. (h): The porosity is classified by specimen and state along the X-axis. The Median (M) trends are shown as connected lines, and statistical plots (violin with box) illustrate the range of values, with the Highest (H) and Lowest (L) values being key points. The violin and box plots together help visualize the change in porosity of each specimen across different erosion states. 8f illustrate the three phases of internal erosion in the -specimen with High (H) and Medium (M) fine particle contents. For the H-labeled specimen, fine particles are fully eroded after three erosion phases, reaching stability at \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\stackrel{\sim}{m}$$\end{document} = 15%. In contrast, for the M-labeled specimen, stability is reached with only partial erosion (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\stackrel{\sim}{m}$$\end{document} = 13%). This demonstrates that the derived formula (Eqs. 4 and 5) can predict the geometric shape of the gravel soil PSD curve based on the parameters (IG and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\stackrel{\sim}{m}$$\end{document}), without needing to delve into complex mechanical or hydraulic details.The PSD calculation results based on the graded internal erosion method are shown for three granular soil specimens in figures (a) through (g). (a)~(c): The -specimen undergoes two erosion phases with three states. (d)~(f): The -specimen experiences three erosion phases with four states, excluding the L sub-group. (g): The -specimen is not eroded with a stable grading. (h): The porosity is classified by specimen and state along the X-axis. The Median (M) trends are shown as connected lines, and statistical plots (violin with box) illustrate the range of values, with the Highest (H) and Lowest (L) values being key points. The violin and box plots together help visualize the change in porosity of each specimen across different erosion states. Figure 8 The PSD calculation results based on the graded internal erosion method are shown for three granular soil specimens in figures (a) through (g). (a)~(c): The -specimen undergoes two erosion phases with three states. (d)~(f): The -specimen experiences three erosion phases with four states, excluding the L sub-group. (g): The -specimen is not eroded with a stable grading. (h): The porosity is classified by specimen and state along the X-axis. The Median (M) trends are shown as connected lines, and statistical plots (violin with box) illustrate the range of values, with the Highest (H) and Lowest (L) values being key points. The violin and box plots together help visualize the change in porosity of each specimen across different erosion states. Porosity changes Figure 8h illustrates the evolution of porosity in different specimens during the internal erosion process, emphasizing the changes in porosity for each erosional state. The violin plot employs connected median lines to depict the porosity evolution for each erosional state. Additionally, the plot includes statistical data, such as the higher (H) and lower (L) values, represented using violin plots with box plots to indicate the distribution. The “Higher” specimen corresponds to the maximum porosity value, while the “Lower” specimen represents the minimum. As the fine particle content decreases across the specimens, the range of porosity values broadens, suggesting a correlation between grain composition and porosity variation. When porosity reaches approximately 46%, it marks the loosest state for gravel soils, indicating significant erosion and structural changes in the soil. This progression aligns with the physical process of internal erosion, where increased porosity reflects the breakdown of the material’s internal structure due to erosion. In summary, these figures provide valuable insights into the relationship between erosion phases (represented by the PSD curve movement) and the changes in porosity for each specimen. They help in understanding the effects of granular soils under internal erosion conditions. The Particle Size Distribution (PSD) results, based on the graded internal erosion method, are presented for three granular soil specimens in figures (a) through (g). Below is a summary of the data: Figures (a) to (c): The -specimen undergoes two erosion phases, progressing through three states. Each phase marks a distinct change in the grading and erosion behavior.Figures (d) to (f): The -specimen experiences three erosion phases and four states (excluding the L sub-group), which allows for a detailed analysis of how the soil’s grading changes with each phase.Figure (g): The -specimen remains unaffected by erosion and maintains a stable grading throughout, serving as a control or baseline specimen to compare against the other two.Figure (h): The porosity of each specimen is categorized by specimen and state along the x-axis. The median (M) trends are represented by connected median lines for each state. Violin plots combined with box plots visualize the distribution of porosity values, with the highest (H) and lowest (L) values indicating key points in the data. This allows for a clearer understanding of how porosity changes with erosion, showing the spread and central tendency of the data across the different erosion states. Figures (a) to (c): The -specimen undergoes two erosion phases, progressing through three states. Each phase marks a distinct change in the grading and erosion behavior. Figures (a) to (c): The -specimen undergoes two erosion phases, progressing through three states. Each phase marks a distinct change in the grading and erosion behavior. Figures (d) to (f): The -specimen experiences three erosion phases and four states (excluding the L sub-group), which allows for a detailed analysis of how the soil’s grading changes with each phase. Figures (d) to (f): The -specimen experiences three erosion phases and four states (excluding the L sub-group), which allows for a detailed analysis of how the soil’s grading changes with each phase. Figure (g): The -specimen remains unaffected by erosion and maintains a stable grading throughout, serving as a control or baseline specimen to compare against the other two. Figure (g): The -specimen remains unaffected by erosion and maintains a stable grading throughout, serving as a control or baseline specimen to compare against the other two. Figure (h): The porosity of each specimen is categorized by specimen and state along the x-axis. The median (M) trends are represented by connected median lines for each state. Violin plots combined with box plots visualize the distribution of porosity values, with the highest (H) and lowest (L) values indicating key points in the data. This allows for a clearer understanding of how porosity changes with erosion, showing the spread and central tendency of the data across the different erosion states. Figure (h): The porosity of each specimen is categorized by specimen and state along the x-axis. The median (M) trends are represented by connected median lines for each state. Violin plots combined with box plots visualize the distribution of porosity values, with the highest (H) and lowest (L) values indicating key points in the data. This allows for a clearer understanding of how porosity changes with erosion, showing the spread and central tendency of the data across the different erosion states. Erosion mode identification: A grading entropy framework The grading entropy-based diagram has emerged as a significant tool for predicting the internal erosion stability of soils, becoming a focal point of research in recent years. By calculating the base entropy and entropy increment of the soil particle distribution, this approach facilitates the prediction of the system’s thermodynamic state and the degree of disorder. These metrics provide insights into the internal erosional stability of the soil, as well as potential erosion mode and their evolutionary direction20,44. The fundamental principles underlying this methodology have been proposed and applied in various representative cases, with a detailed exposition provided in the Appendix A. Figure 9a shows that all of -specimen (L, M, or H) are unstable during the first two stages of erosion, indicating a vulnerability to erosion or failure in the initial phase. However, by State 2, these groups stabilize, meaning they reach a condition where they no longer experience significant destabilization or erosion during this phase. The erosion mode in this phase is piping, where water flows through internal voids or channels, further degrading the soil. This piping process continues until it reaches Zone II, where the erosion behavior stabilizes, and the specimens show resistance to further internal erosion. In the Zone II, the fine particles are fulfilling the voids between coarse particles and participate in the interaction among the structural skeleton. It is worth to note that the A-value for relative base entropy in the -specimen (L) lies just to the right of the 2/3 line, signaling critical stability. Comparing this to Fig. 8a, the grading curve aligns with the upper limit of the Fuller curve (n = 0.5). Meanwhile, the final stable grading curve corresponds to the Fuller curve (n = 0.6), suggesting that two erosion phases are required, as outlined by the graded internal erosion analysis. The results from the graded erosion calculation closely match those derived from the entropy-based diagram, with only slight discrepancies. Figure 9b presents more conclusive results, showing that each of the -specimen reaches stability (2/3 line) after one erosion phase (two states), except for the -specimen (H), which undergoes two erosion phases (three states). These results, compared with the corresponding DSP diagram (Fig. 7d and f), align with the upper limit of the Fuller curve (n = 0.5). This suggests that the Fuller curve’s criterion (n = 0.6) for gravel soil erosion stability may be somewhat conservative, as it overestimates the number of erosion phases needed for stability in some cases. Fig. 9Normalized entropy data for three groups of the gravel soil specimens. Normalized entropy data for three groups of the gravel soil specimens. In Fig. 9c, all of the -specimen remain stable for internal erosion, positioned to the right of the 2/3 line in Zone II of the entropy diagram. This indicates that the skeleton or fiber of soil structure of the -specimen is composed of coarse particles completely with the fine particles loose stacking in the voids. This arrangement enhances soil stability under the skeleton mainly supporting by coarse particles, as the free-floating fine particles may be carried away by seepage erosion, but they are unlikely to affect the overall structural stability. Overall, in Sample Groups and, specimens with high (H) and medium (M) fines content will undergo suffusion mode. The fine particles will be flushed out until the sample reaches a state of seepage stability. In contrast, specimens with low (L) fines content in Groups and, along with all specimens in Group, will remain in a state of seepage stability. However, they may still experience suffusion mode, which causes the movement of fine particles without collapsing the overall soil skeleton structure. Identification of erosion modes with respect to optimal fine-grain content As an additional evaluation, it is valuable to consider the optimal fine-grain content (Equation B1) to assess the internal erosion risk for each specimen group. For the scenario of the initial state, the grain size distribution curve is split to find the optimal point. Figure 10 illustrates this splitting approach, using every specimen (L) as an example. Fig. 10Illustration of split-up approach for every specimen (L). (a) -specimen; (b) -specimen; (c) -specimen. The dashed red vertical and horizontal lines represent the particle size boundary and the content boundary between coarse and fine grains, respectively. Illustration of split-up approach for every specimen (L). (a) -specimen; (b) -specimen; (c) -specimen. The dashed red vertical and horizontal lines represent the particle size boundary and the content boundary between coarse and fine grains, respectively. The key parameter of this criterion is the selection of the uniformity coefficient Cu′​ for coarse grains. Specifically, data within the 2 ~ 10 mm range (representing coarse grains) are used directly to plot their grading curves. From these curves, parameters like D60​ and D10​ for the part of coarse grains are measured, which are then used to compute Cu′​ as outlined in Equation B2. The formula of separating size between coarse and fine grains to determine the split-up position is shown as follow:12\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{d}_{q}=\sqrt{{d}_{70}{d}_{10}}$$\end{document} where dq is the geometrical average particle size representing the separating size between coarse and fine grains (mm); and d70 is the particle size at which 70% of all particles by weight pass through the filter (mm). Additionally, another parameter is porosity (ϕ), which is detailed in Table 1. The porosity values ϕ0​ range from 26% to 48%, with a general assumption of 35%, presumed to represent the porosity for homogeneous gravel soils with a uniformity coefficient Cu​=1. This approach provides a comprehensive way to evaluate internal erosion risk by considering both grain size distribution and porosity characteristics. Table 2 shows the evaluated results based on the Equation B1, in which the related parameters listed in Equation B2 are considered. Table 2Evaluation of Internal Erosion Influenced by Coarse Grain State and Optimal Fines Content Based on Equation B1.Soil typeSub-groupC u’ϕ 0/[%]ϕ c/[%]d q/[mm]P X/[%]Percentage of optimal finer-grain/[%]Erosion modeInternal instibilityP op0.9Pop1.1Pop-spec.L2.13531.90.9730.029.526.632.5TransitionS/UM0.9632.029.326.432.2TransitionS/UH0.9234.029.026.131.9SuffosionU-spec.L2.13531.90.9725.031.027.934.1SuffusionSM0.9627.030.327.233.3SuffusionSH0.9229.029.626.632.5TransitionS/U-spec.L2.53531.21.827.731.828.735.0TransitionS/UM1.728.230.827.833.9TransitionS/UH1.6228.829.826.832.8TransitionS/UNote: The light green highlighted section indicates that the optimal fine-grain content, based on Equation B2, takes into account more factors when determining the porosity (ϕc) of coarse grains. Table 2Evaluation of Internal Erosion Influenced by Coarse Grain State and Optimal Fines Content Based on Equation B1.Soil typeSub-groupC u’ϕ 0/[%]ϕ c/[%]d q/[mm]P X/[%]Percentage of optimal finer-grain/[%]Erosion modeInternal instibilityP op0.9Pop1.1Pop-spec.L2.13531.90.9730.029.526.632.5TransitionS/UM0.9632.029.326.432.2TransitionS/UH0.9234.029.026.131.9SuffosionU-spec.L2.13531.90.9725.031.027.934.1SuffusionSM0.9627.030.327.233.3SuffusionSH0.9229.029.626.632.5TransitionS/U-spec.L2.53531.21.827.731.828.735.0TransitionS/UM1.728.230.827.833.9TransitionS/UH1.6228.829.826.832.8TransitionS/UNote: The light green highlighted section indicates that the optimal fine-grain content, based on Equation B2, takes into account more factors when determining the porosity (ϕc) of coarse grains. Evaluation of Internal Erosion Influenced by Coarse Grain State and Optimal Fines Content Based on Equation B1. Note: The light green highlighted section indicates that the optimal fine-grain content, based on Equation B2, takes into account more factors when determining the porosity (ϕc) of coarse grains. Table 3 summarizes the evaluation outcomes for the three specimens based on Equation B3, following the Chinese code GAQSIQ-CHN (2022)5, in which the porosity ϕc = 30% of coarse grains is determined directly, rather than being computed by Equation B2. Ideally, the erosion mode for all three specimens should align with predictions based on the entropy-based criterion. Therefore, we believe that the evaluation based on the optimal fine-grain content criterion based on the Chinese code (GAQSIQ-CHN 2022)5 and the entropy-based criteria mutually support each other effectively when ϕc​ is constant, specifically set at 30%. This is confirmed by comparing the results in Table 3 with those in Fig. 9. However, the erosion mode assessment results are not identical, particularly for the -specimen (L) and (M), when comparing the results in Tables 2 and 3. Table 3Evaluation of the Impact of Optimal Fines Content Based on Equation B3 (GAQSIQ-CHN, 2022)5.Soil typeSub-groupϕ C/[%]d q/[mm]P X/[%]Percentage of optimal finer-grain/[%]Erosion modeInternal instibility P op 0.9Pop1.1Pop-spec.L30.01.0230.027.324.429.9SuffosionUM0.9632.027.024.329.6SuffosionUH0.9234.026.724.029.3SuffosionU-spec.L30.01.0225.028.625.731.4SuffusionSM0.9627.027.925.130.7TransitionS/UH0.9229.027.224.529.9TransitionS/U-spec.L30.01.8027.730.227.233.3TransitionS/UM1.7028.229.326.332.2TransitionS/UH1.6228.828.325.531.1TransitionS/U Table 3Evaluation of the Impact of Optimal Fines Content Based on Equation B3 (GAQSIQ-CHN, 2022)5.Soil typeSub-groupϕ C/[%]d q/[mm]P X/[%]Percentage of optimal finer-grain/[%]Erosion modeInternal instibility P op 0.9Pop1.1Pop-spec.L30.01.0230.027.324.429.9SuffosionUM0.9632.027.024.329.6SuffosionUH0.9234.026.724.029.3SuffosionU-spec.L30.01.0225.028.625.731.4SuffusionSM0.9627.027.925.130.7TransitionS/UH0.9229.027.224.529.9TransitionS/U-spec.L30.01.8027.730.227.233.3TransitionS/UM1.7028.229.326.332.2TransitionS/UH1.6228.828.325.531.1TransitionS/U Evaluation of the Impact of Optimal Fines Content Based on Equation B3 (GAQSIQ-CHN, 2022)5. Upon calculations to the other scenarios of erosion state with fine grain removal, it has been found that the parameter ϕ0 — representing the porosity of gravel soil, assumed to consist of uniform-sized particles when considering the effects of Equation B2 — shows high sensitivity. Higher values of ϕ0​, meaning loose skeleton of gravel soil, tend to push all specimens towards the suffusion mode, whereas lower values tend towards the piping mode. Another sensitive parameter is the original porosity ϕ, it has the more significant effects for determining the erosion mode as a constant ϕ0. When the gravel soil is gap-graded and in a state of medium density (ϕ0​ ranging from 0.33 to 0.37), the two criteria yield similar evaluation results. The pore characteristics directly reflect the compaction behavior of gravel soil, manifested as a progressive decrease in the area of Zone II and Zone III, as shown in Fig. 11. The discrepancy arises because these geometric criteria, whether based on Cu, Fuller-curve or grading-entropy, are unable to consider the compaction factors of gravel soils, unlike the criterion based on optimal fine-grain content. This distinction highlights the importance of accounting for compaction effects when assessing the internal erosion behavior of gravel soils, particularly in terms of porosity and its influence on the erosion mode. Thus, understanding and evaluating the compaction degree of gravel soil is essential in predicting its behavior under different conditions, especially in relation to erosion mechanisms. Fig. 11Radar charts based on two versions of the evaluation criterion for optimal fine-grain content. The first column displays radar charts generated by the original version, characterized by a larger Pop area representing the optimal fine-grain content, along with corresponding larger Zone II and Zone III. The second column shows radar charts produced by the improved version (GAQSIQ-CHN, 2022)5, which features a reduced Pop range and, to some extent, reflects the compaction characteristics of gravelly soils. Radar charts based on two versions of the evaluation criterion for optimal fine-grain content. The first column displays radar charts generated by the original version, characterized by a larger Pop area representing the optimal fine-grain content, along with corresponding larger Zone II and Zone III. The second column shows radar charts produced by the improved version (GAQSIQ-CHN, 2022)5, which features a reduced Pop range and, to some extent, reflects the compaction characteristics of gravelly soils. Erosion mode evaluation: a Kenney & Lau framework Further assessment of the erosion stability has been carried out by integrating other methodologies, such as the one proposed by Kenney and Lau (1985)6, which is based on the geometric characteristics of the fine-grain part of the DSP curve. In Fig. 12a, the curves of F versus H’ (mass fraction between a specified grain size D and 4D) are plotted, along with the unstable boundary defined by Kenney and Lau’s criterion (shadowed area). According to this criterion, soil instability is indicated if the F versus H’ curve lies partially or entirely within the shadowed region. However, relying solely on the F versus H’ curve for accurate judgment may not always be sufficient. A more reliable interpretation involves considering the value of (H’/F)min, which is more informative. Figure 12b, a dual diagram corresponding to Fig. 12a, offers a more precise evaluation by referencing the position of the (H’/F)min value. From Fig. 12a, it can be seen that for and -specimens, the F values in the range of 10% to 30% lie within the shadowed area. In Fig. 12b, it can also be observed that they have almost the same (H′/F)min, both of which are located within the shadowed area. Based on this result, it can be concluded that internal erosion may cause instability in these specimens. The behavior of -specimen is consistent with previous evaluations. However, -specimen differs from the optimal fine particle method in terms of erosion mode, as the latter follows a transformation mode. The reason for this is believed to be related to the density of the fine particles, which has had a certain impact and requires further assessment. Nonetheless, it can be reasonably inferred that -specimen tends to transition toward a piping erosion mode that could destabilize the structure. The -specimen, although within the shadowed area in the F value range of 5% to 28%, can still be judged as critically stable. This conclusion can be further confirmed through the (H′/F)min value on the F ~ H′/F curve in Fig. 12b, where the values of this sample are located near the stability boundary. Based on its (H′/F)min value, the sample can be classified as critically stable. Based on this, we can also make a reasonable inference, which is that the subgroups of -specimen, classified as the transition mode, tend to exhibit a suffusion erosion mode that contributes to the structural stability. Fig. 12Judgment of instability of internal erosion for three granular soil specimens by Kenney & Lau’s criterion. NG, narrowly graded; WG, well graded: (a) F ~ H’ diagram; (b) F ~ H’/F diagram. Judgment of instability of internal erosion for three granular soil specimens by Kenney & Lau’s criterion. NG, narrowly graded; WG, well graded: (a) F ~ H’ diagram; (b) F ~ H’/F diagram. Comparative assessment of erosion mode criteria Table 4 summarizes the erosion stability assessment results for the reference soil specimens. The results demonstrate general consistency in the evaluations of specimens and across the four criteria: grading entropy (A), the Kenney and Lau method (H’/F)max, optimal fine content (Pₓ), and the uniformity coefficient (Cu). However, the assessment of erosion stability for the gravelly soil specimen (rows highlighted in light green in Table 4) shows significant variability. This is particularly evident when the optimal fine content (Pₓ) criterion is applied, where considerable divergence in the results is observed. For example, when the optimal fine content (Pₓ) criterion is applied to specimens (M) and (H) from Sample Group, the assessment results in a “transitional state,” meaning the erosion mode could either be stable suffusion or unstable suffosion. In contrast, the other three criteria predominantly classify these specimens as unstable. This discrepancy can be traced back to the original formulation of the optimal fine content (Equations B1-B2). In this version, the optimal filling space for fine particles, ϕₒ, is directly related to the grain size distribution characteristics of the coarse fraction. Specifically, ϕₒ is a function of the uniformity coefficient of the coarse grains, Cu′, and the porosity under the idealized assumption of uniformly coarse grains, n₀. Here, Cu′ defines the theoretical lower limit of ϕₒ (when Cu′ = 1, representing perfectly uniform grains), while ϕ₀ governs its upper limit (with ϕ₀ = 0.26 corresponding to dense packing and ϕ₀ = 0.48 to loose packing). Table 4An Evaluation Matrix of Internal Erosion Characteristics for Three Specimens Under Four Distinct Criteria.UnstableStableSoil typeA(H’/F)min P x CuSoil typeA(H’/F)min P x CuBy four criteriaBy two criteria-spec.L0.6140.20@25Px > 1.1Pop10 < Cu < 20-spec.L10 < Cu < 20M0.6030.19@27Px > 1.1PopCu > 20H0.5910.17@29Px > 1.1PopCu > 20-spec.L0.6450.20@2510 < Cu < 20-spec.L Px < 0.9P op 10 < Cu < 20M0.6330.19@270.9Pop < Px < 1.1PopCu > 20M0.9Pop < Px < 1.1PopH0.6200.17@290.9Pop < Px < 1.1PopCu > 20H0.9Pop < Px < 1.1Pop By one criterion By four criteria -spec.0.9Pop < Px < 1.1Pop-spec.L0.6971.14@130.9Pop < Px < 1.1PopCu < 100.9Pop < Px < 1.1PopM0.6871.00@140.9Pop < Px < 1.1PopCu < 10H0.89@150.9Pop < Px < 1.1PopH0.6770.9Pop < Px < 1.1PopCu < 10Note: The rows highlighted in light green represent the erodibility results of gravel soil, determined by different criteria, with certain discrepancies. Table 4An Evaluation Matrix of Internal Erosion Characteristics for Three Specimens Under Four Distinct Criteria.UnstableStableSoil typeA(H’/F)min P x CuSoil typeA(H’/F)min P x CuBy four criteriaBy two criteria-spec.L0.6140.20@25Px > 1.1Pop10 < Cu < 20-spec.L10 < Cu < 20M0.6030.19@27Px > 1.1PopCu > 20H0.5910.17@29Px > 1.1PopCu > 20-spec.L0.6450.20@2510 < Cu < 20-spec.L Px < 0.9P op 10 < Cu < 20M0.6330.19@270.9Pop < Px < 1.1PopCu > 20M0.9Pop < Px < 1.1PopH0.6200.17@290.9Pop < Px < 1.1PopCu > 20H0.9Pop < Px < 1.1Pop By one criterion By four criteria -spec.0.9Pop < Px < 1.1Pop-spec.L0.6971.14@130.9Pop < Px < 1.1PopCu < 100.9Pop < Px < 1.1PopM0.6871.00@140.9Pop < Px < 1.1PopCu < 10H0.89@150.9Pop < Px < 1.1PopH0.6770.9Pop < Px < 1.1PopCu < 10Note: The rows highlighted in light green represent the erodibility results of gravel soil, determined by different criteria, with certain discrepancies. An Evaluation Matrix of Internal Erosion Characteristics for Three Specimens Under Four Distinct Criteria. Note: The rows highlighted in light green represent the erodibility results of gravel soil, determined by different criteria, with certain discrepancies. However, the assumption of uniform coarse grains is inherently ambiguous and highly sensitive, making it difficult to establish a definitive interpretation. This assumption is not only closely tied to the compaction characteristics of coarse and fine particles in gravelly soils but may also be influenced by the material properties of the grains themselves. For instance, a study by Slangen & Fannin(2017)44 demonstrated significant differences in the sensitivity to various erosion modes between glass beads and natural soil particles. The improved version of Equation B3 (GAQSIQ-CHN, 2022)5 simplifies the calculation of ϕₒ by setting it to a fixed empirical value of 0.30. This change essentially transforms the criterion into an empirical assessment. It is these theoretical and empirical uncertainties that lead to the ambiguous classification of the critically positioned Sample Group in this study. Discussion Based on the comprehensive experimental and analytical results presented, the following key points warrant in-depth discussion regarding the internal erosion behavior of gap-graded gravelly soils, particularly within the critical fine content transition zone (30–35%). Interplay between fine particle content and erosion mode transition The experimental data robustly confirm that fine particle content is the predominant factor governing the transition between suffusion and suffosion (piping) erosion modes. For specimens in Groups and (with higher initial uniformity coefficients, Cu > 20), the erosion process initiated as suffusion. However, as the fine content approached and exceeded approximately 30–35%, a shift towards suffosion was observed. This transition is mechanistically linked to the structural role of fines. Below the critical threshold, fines primarily fill voids within a stable coarse-grained skeleton. As the fine content increases, this skeleton becomes progressively “over-filled“7, leading to a state where coarse particles “float” within the fine matrix. This transition fundamentally alters the soil’s fabric from a coarse-dominated skeleton to a fine-dominated matrix, drastically reducing its shear strength. The -specimen group, with its inherently stable gradation (Cu < 10), consistently exhibited suffusion behavior without transitioning to suffosion, underscoring the stability provided by a well-interlocked coarse skeleton. Validation and limitations of geometric criteria This study provides a critical evaluation of various geometric criteria for erosion assessment. The results generally align with the Kenney & Lau6 and grading entropy criteria, which successfully identified the instability of - and -specimens and the stability of -specimens. The grading entropy framework, in particular, offered a thermodynamic perspective, effectively mapping the evolution of specimens towards stability as fines were eroded. However, significant discrepancies emerged with the optimal fine-grain content criterion, especially for the critically positioned -specimen group. The high sensitivity of this criterion to the assumed porosity of the coarse skeleton (ϕc or ϕ0) highlights a major limitation. The theoretical calculation is highly dependent on an idealized characterization of the coarse skeleton’s porosity, which is influenced by grain shape, packing, and compaction—factors not fully captured by gradation curves alone. This underscores that while geometric criteria are valuable for initial screening, their accuracy can be compromised for soils in transitional states. The role of µ-CT imaging in elucidating micro-mechanisms The application of µ-CT imaging was pivotal in bridging the gap between macroscopic hydraulic behavior and microscopic soil fabric. The visualizations directly revealed the “floating” coarse grains in - and -specimens with high fine content, providing tangible evidence for the “over-filled” fabric concept. This microstructural insight explains why these specimens required higher hydraulic gradients to initiate erosion but were then prone to more catastrophic suffosion failure. For the -specimen, the images confirmed a stable coarse skeleton with fines acting as void fillers. The ability to segment and quantify pores and particles paves the way for future quantitative correlations between microstructural parameters and macroscopic erosion susceptibility. Implications of the graded erosion principle and predictive model The proposed graded erosion principle and its mathematical formulation offer a dynamic framework for predicting the evolution of soil properties during erosion. The model’s success in forecasting the progression of PSD curves towards a stable Fuller limit and the concomitant increase in porosity is a significant step forward. By incorporating a time- and gradient-dependent erosion rate that is sensitive to particle size, the improved model addresses a key limitation of previous empirical models that lacked a physical erosion threshold. This allows for a more realistic simulation of the erosion process. The prediction that - and -specimens required multiple erosion phases to reach stability, while -specimens remained stable, aligns well with physical expectations. Limitations and future research Regarding the surface morphological characteristics of particles, Didier Marot et al43. investigated the influence of coarse particle angularity on internal seepage erosion in cohesive sandy soils. They found that increased angularity could enhance erosion resistance by up to five times and reduce hydraulic conductivity. As suggested by Slangen and Fannin (2017)44, future research should incorporate the effects of particle shape and mineralogy. Additionally, the sensitivity of the parameter ϕ₀ in the optimal fine-content criterion warrants experimental validation across various soil types. Finally, integrating real-time monitoring with µ-CT imaging holds the potential to bridge the gap between laboratory models and field-scale erosion dynamics. It is important to note that the graded erosion model proposed in this study is a phenomenological model. Its core contribution lies in utilizing microstructural information to predict the evolution of macroscopic soil properties (such as gradation and porosity) with varying fine particle content, and in achieving cross-validation with multiple traditional geometric criteria. It is not designed to simulate the fluid-solid interaction mechanisms at the particle scale (as revealed, for example, by CFD-DEM approaches). This clarification defines the positioning of our model within the current research framework. Its strength lies in establishing a bridge between microscopic observations and macroscopic engineering assessments, while its simplification of fundamental mechanical processes points the way for future development of multi-scale models that integrate microscopic mechanisms with macroscopic responses. It should be noted that the validation of the multi-criteria integrated evaluation framework established in this study is primarily based on a series of carefully designed laboratory tests and numerical model cases. Although these cases are sufficient to clearly reveal the controlling influence of fine particle content on the transition of erosion modes within the critical transition zone and to verify the logical consistency of the proposed method, the number of analyzed specimens remains relatively limited. Future research is necessary to further validate and refine the universality and robustness of this framework across a broader range of soil types and under more complex stress paths and hydraulic conditions. In conclusion, the transition of erosion modes in gap-graded gravelly soils is a complex process predominantly controlled by fine particle content. Integrating µ-CT imaging with the proposed graded erosion principle provides a powerful, microstructurally-informed framework for understanding and predicting internal erosion behavior, ultimately contributing to the construction of more resilient geotechnical structures. Conclusions This study systematically investigated internal erosion in gap-graded gravelly soils, focusing on the critical fine content transition zone (30–35%). Through integrated µ-CT imaging, geometric criteria analysis, and a newly developed predictive model, key conclusions emerge: (1) When the fine particle content exceeds 30–35%, the transition of erosion modes is dominated by the controlled morphology of the soil structure. In this study, the proposed hierarchical erosion model, within a multi-scale integrated framework, successfully predicts the evolution of gradation toward the Fuller stability limit and the associated porosity changes, thereby providing a robust and innovative approach to elucidating and forecasting the critical relationship between erosion modes and structural evolution. (2) The study validates a multi-criteria assessment framework. While traditional geometric criteria (Cu, Kenney & Lau) provide useful screening, they show limitations for transitional soils. The grading entropy criterion offers more robust stability evaluation, demonstrating that no single method suffices for comprehensive assessment. (3) The proposed graded erosion model successfully predicts PSD evolution toward stable Fuller limits and porosity changes. By incorporating particle-size-sensitive, time-dependent erosion rates with physical thresholds, it overcomes previous limitations and provides a realistic framework for forecasting long-term soil evolution. This research establishes that erosion modes directly reflect fine-content-controlled soil fabric. The integration of advanced imaging, multi-faceted analysis, and physically-based modeling significantly enhances erosion prediction capabilities, contributing to more resilient geotechnical structures. Future work should focus on model refinement and quantitative µ-CT analysis to strengthen micro-macro correlations.