1. Introduction
Granular flows driven by gravity are common in both geophysical (Schneider et al. Reference Schneider, Bartelt, Caplan-Auerbach, Christen, And and Mcardell2010; Wang et al. Reference Wang, Dai, Nakahara and Sonoyama2018; Hürlimann et al. Reference Hürlimann, Coviello, Bel, Guo and Yin2019; Dong & Su Reference Dong and Su2024) and industrial processes (Ottino & Khakhar Reference Ottino and Khakhar2001; Chung et al. Reference Chung, Hunt, Huang and Liao2024). However, understanding these flow behaviours remains a significant challenge, particularly when inertial and viscous effects occur upon particles submerged in a fluid. Unlike dry granular materials, wet granular flow depends not only on particle characteristics (Roland et al. Reference Roland, Marisa, Palucis, Yohannes and Hill2016; Li et al. Reference Li, Wang, Wu and Niu2021; Shi et al. Reference Shi, Yang, Kwok and Jiang2024), but also on the properties of interstitial fluids such as density (Sun et al. Reference Sun, Zhang, An, Liu and Wang2021), viscosity (Kasper et al. Reference Kasper, Magnanimo, de, Sjoerd, Beek and Jarray2021) and others. In wet granular flows, fluid–particle interaction emerges as a more complex process involving dilatancy (Pouliquen & Renaut Reference Pouliquen and Renaut1996; Wang et al. Reference Wang, Wang, Peng and Meng2017), vortex motion (Tang, Lin & Wang Reference Tang, Lin and Wang2024) or non-local effects (Perrin et al. Reference Perrin, Wyart, Metzger and Forterre2021). As seen in recent geophysical observations, such an interaction can dramatically alter the packing stability and flow dynamic behaviours. For example, submarine landslides can happen on extremely low slide angles (
$\lt 5^\circ$
) and travel with a long run-out distance (up to 1000 km) (Haflidason et al. Reference Haflidason, Sejrup, Nygård, Mienert, Bryn, Lien, Forsberg, Berg and Masson2004; Gee et al. Reference Gee, Uy, Warren, Morley and Lambiase2007; Zheng et al. Reference Zheng, Dai, Mao and Huang2023). Consequently, a scientific investigation into these flows is essential for understanding their physical mechanisms and minimising the potential damage.
A typical and practical configuration for studying the underlying dynamics of granular materials is the rotating drum. Previous studies have primarily focused on dry granular flow, examining factors such as rotation speed, configuration size and particle properties. Building on observations from drum experiments, Henein, Brimacombe & Watkinson (Reference Henein, Brimacombe and Watkinson1983) first characterised different flow regimes using a normalised parameter known as the Froude number (
$F_{r}$
), which is defined as follows:
\begin{equation} F_{r} =\sqrt {\frac {\omega ^{2}\boldsymbol{\cdot }D}{2g}}, \end{equation}
where
$\omega$
is the rotation speed,
$\textit {D}$
is the drum diameter and
$\textit {g}$
is the gravitational acceleration. The Froude number represents the ratio of centripetal to gravitational forces in a rotating drum system. As
$F_{r}$
increases, the granular flow regime transition occurs, successively exhibiting six distinct regimes including slipping, slumping, rolling, cascading, cataracting and centrifuging. An important quantitative parameter that illustrates these regimes is the surface angle of a granular pile, denoted as
$\theta$
. At low Froude number (
$F_{r}\lt$
$10^{-3}$
), hysteresis and avalanches (Tegzes, Vicsek & Schiffer Reference Tegzes, Vicsek and Schiffer2003; Yang et al. Reference Yang, Zhang, Li, Zheng and Zivkovic2017; Li et al. Reference Li, Yang, Zheng and Sun2018; Perrin et al. Reference Perrin, Clavaud, Wyart, Metzger and Forterre2019) are observed. The granular pile flow starts at the maximum surface angle,
$\theta_{\textit{start}}$
, and stops at a smaller angle,
$\theta _{\textit{stop}}\lt \theta _{\textit{start}}$
(Daerr & Douady Reference Daerr and Douady1999; Pouliquen & Forterre Reference Pouliquen and Forterre2001; Balmforth & McElwaine Reference Balmforth and McElwaine2018). When
$F_{r}$
reaches a moderate range (
$10^{-3}\sim 10^{-1}$
), granular materials exhibit the rolling or cascading motion, characterised by a continuous and stable surface morphology (Mellmann Reference Mellmann2001). At very high Froude number (
$F_{r}\gt$
$10^{-1}$
), the material slides along the drum wall, entering the centrifuging regime. This movement mode is particularly suitable for industrial applications, such as drying (Chaudhuri, Muzzio & Tomassone Reference Chaudhuri, Muzzio and Tomassone2006; Kwapinska, Saage & Tsotsas Reference Kwapinska, Saage and Tsotsas2008). For wet granular materials in a drum, recent studies have indicated that similar regimes and dynamic behaviours are observed when interstitial fluid is present (Chou & Hsiau Reference Chou and Hsiau2012; Chen et al. Reference Chen, Suo, Dong, Zhong, Wei and Gan2024). However, the role of fluid in the transition of wet granular flow remains unclear.
Wet granular flow in a rolling drum is considerably complex due to the interplay of inertial, viscous and gravity-driven effects acting on the particles. Sylvain et al. (Reference Sylvain, Gondret, Perrin and Rabaud2003) conducted pioneering experiments to investigate the influence of interstitial fluid on the granular avalanche dynamics. They introduced two dimensionless numbers: the Stokes number
$S_{t}$
and the density ratio
$r=\sqrt {\rho _{s}/\rho _{f}}$
(where
$\rho _{s}$
and
$\rho _{f}$
are the particle and fluid densities, respectively) to evaluate the relative importance of the inertial, viscous and free fall regimes. The Stokes number is defined as
$S_{t} =\sqrt {\rho _{s}\Delta \rho gd^{3}} /18\eta$
, where
$\Delta \rho =\rho _{s} -\rho _{f}$
represents the solid–fluid density difference,
$\textit {d}$
is the particle diameter and
$\eta$
is the fluid viscosity. Subsequent experimental and numerical studies further investigated the effects of fluid viscosity (Brewster, Grest & Levine Reference Brewster, Grest and Levine2009; Preud’Homme et al. Reference Preud’Homme, Lumay, Vandewalle and Opsomer2021; Dong et al. Reference Dong, Wang, Marks, Chen and Gan2023), water content (Yuan et al. Reference Yuan, Moscariello, Cuomo and Chareyre2019), particle size (Jarray, Magnanimo & Luding Reference Jarray, Magnanimo and Luding2019; Zhou et al. Reference Zhou, Cui, Jing, Zhao, Song and Huang2020) and filling level (Liao, Hsiau & Nien Reference Liao, Hsiau and Nien2015; Vu et al. Reference Vu, Amarsid, Delenne, Richefeu and Radjai2024). These existing results have established the qualitative relationship between the flow dynamics within individual regimes and the fluid properties, as well as the particle characteristics. However, the major feature controlling flow regimes transitions has been seldom studied. The classic phenomenological criterion for the rolling–cascading transition was introduced by Mellmann (Reference Mellmann2001),
$F_{r} ({D}/{d} ) \approx 2$
, the validity of which, however, is not proven for wet granular flow. Furthermore, when considering more realistic scenarios of wet granular flows in nature, such as the higher sea-water density encountered by submarine landslides, fluid density is rarely isolated for analysis. This distinction is significant in submerged granular systems, as demonstrated in recent experiments conducted by Perrin et al. (Reference Perrin, Clavaud, Wyart, Metzger and Forterre2019). The frictional and frictionless conditions in their studies essentially reflect the relative density between the particles and ambient fluid, which certainly influences the interparticle friction and suspension rheology.
In this article, we investigate the rolling and cascading dynamics of wet granular materials in a rotating drum. An extensive series of experiments are conducted with varying fluid densities and rotation speeds. Using image recording and image processing technology, we first obtain steady granular observations and significant features including the free surface angle, velocity field and flow regime. It can be found that the dynamic repose angle and equivalent friction coefficients increase monotonically with the Froude number. Additionally, we propose a morphological parameter called the angle divergence to quantitatively distinguish between two investigated flow regimes. Finally, we constructed the flow phase diagrams against the Froude number and density ratio to depict the flow transition in submerged granular systems. The resulting nonlinear phase boundary highlights the significant effects of fluid density.
(a) Schematic diagram of the rotating drum test. (b) Spherical POM particles used in the experiments and (c) snapshot of the experimental process during rotation. Here, x and y respectively correspond to the flowing direction and the direction perpendicular to the flowing surface.

2. Experimental method
2.1. Experimental set-up and procedure
Figure 1 illustrates the experimental apparatus and materials used to investigate the flow characteristics of wet granular materials. The experiments are conducted in a rotating drum with an inner diameter of 150 mm and a length of 200 mm. To facilitate direct observation and accurate measurements, both the cylindrical drum and two matching plates are made of Plexiglass. For each experiment, particles are first placed into the drum and secured with four screws. Fluid is then slowly injected through a small pipe at the bottom of the drum to prevent air bubble formation, until the entire system is filled. The particles used in this study are polyoxymethylene (POM) with a density of 1.439 g
$\textrm {cm}^{-3}$
and a diameter of 2 mm, as shown in figure 1(b). To investigate the influence of fluid density, four types of interstitial fluids are employed: water, 15 wt. % NaCl, 20 wt. % NaCl and 24 wt. % NaCl. And the corresponding fluid density varies from 0.975 to 1.182 g
$\textrm {cm}^{-3}$
. All experiments are confirmed to operate firmly within the inertial regime (
$S_{t}\gt 1$
and
$S_{t}/r \gt 2.5$
), so that the viscous effects of fluid can be neglected. The properties of particles and fluids are summarised in table 1. For all experiments, the same mass of POM particles is used to maintain a constant filling level of approximately 50.3 %. The total mass of POM spheres for all cases is approximately 2542.92 g. In addition, the drum is driven by a controllable step motor (OUBONG 5IK90RGU-CF) via a conveyor belt. This allows for a wide range of rotation speeds from 0.66 to 5.00 rpm (revolutions per minute), facilitating the investigation of rolling and cascading regimes in submerged granular materials. Note that 0.66 rpm represents the minimum reliably achievable rotation speed of our current apparatus.
Properties of different fluid types and corresponding measurements.

After sample preparation, the cylindrical drum is initially pre-rotated at a speed of 3.44 rpm for five minutes to achieve a relatively uniform initial packing state (see figure 12(a) in Appendix B). Following this, the wet granular pack transitions from the horizontal position to the rolling and cascading regimes. We record the continuous granular flow using a high-speed camera (Basler acA2000–165uc) with a resolution of 2040
$\times$
1086 pixels. Figure 1(c) provides a snapshot of the rotation process. The experimental images have a scaling factor of 0.24 mm per pixel, and the frame rate is set at 50 frames per second, allowing for clear capture of particle motion and the interface between grains and the liquid during flow. For each experiment, we acquire 3000 frames (equivalent to 1 minute) for the initiation stage and 12 000 frames (equivalent to 4 minutes) during the stabilisation stage. A full evolution of free surface angles including the initiation and stabilisation stages can be seen in Appendix B. Each experiment is conducted three times to ensure the repeatability of results.
The main steps of particle image processing. (a) Experimental original image. (b) Grey scale image after Gaussian filtering. (c) Binarised image by global thresholding. (d) Edge detection and surface angle measurement. Here,
$\ast$
represents the reference point. The blue and red dashed lines correspond to the best linear fits of the upper (
$\theta _{\textit{top}}$
) and lower (
$\theta _{\textit{lo}w\textit{er}}$
) segments in the flowing surface.

2.2. Image processing and data acquisition
During the granular flow experiments in the drum, distinct flow regimes are observed depending on the drum speed (Henein et al. Reference Henein, Brimacombe and Watkinson1983; Govender & Indresan Reference Govender2016) and the nature of the particles (Chen et al. Reference Chen, Li, Xiu, Zheng and Yang2020; Wang et al. Reference Wang, Barés, Barés and Azéma2024) and fluid (Sun et al. Reference Sun, Zhang, An, Liu and Wang2021). One significant feature to distinguish these regimes is the surface angle. In our study, each experimental image is digitally processed to obtain the morphology of granular packing, as shown in figure 2. The experimental original image (figure 2 a) is first converted into a grey scale image (figure 2 b) after Gaussian filtering and denoising. Then the image is binarised (figure 2 c) through global thresholding to segment the white POM particles from the background. The edges and surfaces of the flowing particles are then extracted using the Bwboundaries function in Matlab. After recognising the boundary, we fit the coordinate points of the surface and calculate the surface slope for each frame. Figure 2(d) shows the criteria for defining the surface angles. This process yields two distinct parameters: the starting angle, defined as the maximum angle during the initiation stage, and the dynamic repose angle, defined as the angle when the flow reaches steady state. Further details are provided in figure 12 (Appendix B).
The velocity evolution and flow field of the particles provide insights into different flow regimes. To measure the velocity field, we use PIVlab (Thielicke & Sonntag Reference Thielicke and Sonntag2021), which has been demonstrated to be effective in similar works (Roman et al. Reference Roman, Soulaine, Alsaud Abu, Moataz and Tchelepi2016; Sarno et al. Reference Sarno, Carravetta, Tai, Martino, Papa and Kuo2018; Niu et al. Reference Niu, Zheng, Yuan, Mao and Huang2024), to analyse experimental images across all cases. Cross-correlation optimisation is performed in the frequency domain using a fast Fourier transform algorithm. Through particle image velocimetry, we have extracted the velocity components along the flow direction and the vertical surface for all frames. For each case, the velocity fields during the stabilisation stage are smoothed and averaged to obtain a stable flow field, as shown in figure 3.
Velocity vector (a) and velocity field (b) of the cascading regime at
$\rho$
= 1.182 g cm−3,
$\omega$
= 4.00 rpm. (c) Evolution of free surface angle and the probability density functions (PDFs) of
$\theta _{\textit{top}}$
and
$\theta _{\textit{lo}w\textit{er}}$
. The red and blue dotted lines respectively correspond to the average values of
$\theta _{\textit{top}}$
and
$\theta _{\textit{lo}w\textit{er}}$
.

3. Experimental results
3.1. Observation of two typical flow regimes
Through a series of drum tests, two flow regimes of wet granular flow including the rolling and cascading regimes are investigated. When the high rotation speed is employed, the cascading regime emerges due to the enhanced inertia relative to gravity (Wang et al. Reference Wang, Barés, Barés and Azéma2024). Figure 3 illustrates the dynamic process of a typical cascading regime. In this regime, the velocity field exhibits three types of local flow characteristics. At the bottom of granular packing, particles move in a clockwise direction, displaying a magnitude and orientation similar to the roller. This particle movement is referred to as the passive region in figure 3(b). The top left portion of the packing also moves clockwise along the rolling direction but shifts away from the wall, resembling a parabolic motion. Then the rapid particle movement gradually slows down under the action of the ambient fluid, finally forming a counterclockwise movement area in the top right part. These two local movements are classified as the active region. Due to continuous rotation of drum, the closed flow stream is formed and an obvious S-shaped surface appears, which is the main identification of the cascading regime. Figure 3(c) presents the time evolution and PDF statistics of
$\theta _{\textit{top}}$
and
$\theta _{\textit{lo}w\textit{er}}$
, corresponding to the angles of the top and lower surfaces, respectively. It is evident that the PDFs of
$\theta _{\textit{top}}$
and
$\theta _{\textit{lo}w\textit{er}}$
are distinctively bimodal and both exhibit a Gaussian-like distribution. This indicates that the granular system has reached a stable state. And there is a significant divergence between the average of
$\theta _{\textit{top}}$
and
$\theta _{\textit{lo}w\textit{er}}$
, with
$\theta _{\textit{top}}$
being
$8.61^\circ$
larger than
$\theta _{\textit{lo}w\textit{er}}$
.
Velocity vector (a) and velocity field (b) of the rolling regime at
$\rho$
= 1.182 g cm−3,
$\omega$
= 1.00 rpm. (c) Evolution of free surface angle and the PDFs of
$\theta _{\textit{top}}$
and
$\theta _{\textit{lo}w\textit{er}}$
.The red and blue dotted lines respectively correspond to the average values of
$\theta _{\textit{top}}$
and
$\theta _{\textit{lo}w\textit{er}}$
.

By contrast, the surface of granular packing is nearly flat for the rolling regime. The particle movement in the active region exhibits a uniform flow along the free surface, as shown in figure 4. In this regime, the surface angle of granular flow remains constant, with
$\theta _{\textit{top}}$
and
$\theta _{\textit{lo}w\textit{er}}$
differing by only approximately
$0.42^\circ$
. This behaviour contrasts sharply with the distribution characteristics observed in the cascading regime. Considering both the velocity field and surface morphology, we qualitatively classify the flow regimes of wet granular material across various
$\omega$
and
$\rho _{f}$
. Preliminary comparisons indicate that the system transitions from the rolling regime to the cascading regime as
$\omega$
increases, consistent with previous studies on dry granular materials (Henein et al. Reference Henein, Brimacombe and Watkinson1983; Govender & Indresan Reference Govender2016; Wang et al. Reference Wang, Barés, Barés and Azéma2024). However, it is interesting to notice that the fluid density significantly affects the rolling–cascading flow transition.
Starting angles (a) and dynamic repose angles (b) vs rotation speed at different fluid densities.

3.2. Angle of repose and equivalent friction
One of the key elements in studying the dynamic behaviour of wet granular flow is the inclination surface angle. This allows direct comparison with different flow regimes. Figure 5 displays the starting angles (
$\theta _{m}$
) and dynamic repose angles (
$\theta _{d}$
) as a function of rotation speed at different fluid densities. Here,
$\theta _{m}$
is characterised as the maximum surface angle during the initial state, i.e. the inclination angle on beginning to flow under gravity. From figure 5(a), it is apparent that
$\theta _{m}$
increases with
$\omega$
. This is primarily attributed to the enhanced inertia of the increasing rotation speed. In addition, fluid density significantly affects the evolution of
$\theta _{m}$
, particularly at the higher rotation speed condition. For higher fluid density,
$\theta _{m}$
increased more dramatically. At the lowest
$\omega$
of 0.66 rpm,
$\theta _{m}$
rises by approximately
$3^\circ$
, which is an order of magnitude less than the increase (approximately
$13^\circ$
) at the largest
$\omega$
of 5.00 rpm. These findings align with the experimental results of Chen et al. (Reference Chen, Suo, Dong, Zhong, Wei and Gan2024), who took into account the relative density by varying the density of the particles. For submerged granular packing, the buoyancy of the fluid counteracts a portion of the gravitational forces acting on the particles. Thus, as the fluid density increases, the enhanced buoyancy effect results in a more significant inertial behaviour under the same
$\omega$
. Moreover, at the largest
$\omega$
= 5.00 rpm and
$\rho _{f}$
= 1.182 g
$\textrm {cm}^{-3}$
, the initiation of granular flow occurs at an inclination angle of
$54^\circ$
, which is rare in dry spherical granular flow (Sylvain et al. Reference Sylvain, Gondret, Perrin and Rabaud2003; Jarray et al. Reference Jarray, Magnanimo and Luding2019).
The dynamic repose angles,
$\theta _{d}$
, are obtained when the wet granular flow reaches a steady state. Here, for the rolling regime,
$\theta _{d}$
is defined as the average value of the inclination angle of the free grain surface during rotation. For the cascading regime, the average value of
$\theta _{\textit{top}}$
serves as
$\theta _{d}$
because it more accurately reflects the spontaneous movement of particles compared with
$\theta _{\textit{lo}w\textit{er}}$
and
$\theta_{top}$
and
$\theta_{lower}$
. This is derived from comprehensive analysis of the velocity fields and statistical characterisation of the surface angles (figure 3). Figure 5(b) presents the relationship between
$\theta _{d}$
and
$\omega$
for variational fluid densities. As
$\omega$
increases,
$\theta _{d}$
exhibits a consistent monotonic increase alongside
$\theta _{m}$
. The rise in rotation speed supplies additional external energy to the granular system, resulting in greater inertia and higher centrifugal acceleration. Consequently, the inertial effect causes particles to adhere to the wall, leading to an increase in the slope of the granular pile. Additionally, the relationship between
$\theta _{m}$
and
$\omega$
in figure 5(a) is well represented by a linear function with a constant slope. In contrast, a transitional phenomenon is observed in
$\theta _{d}$
. Specifically, the slope of
$\theta _{d}$
as a function of
$\omega$
changes dramatically at certain points for different fluid densities. These critical points are situated near the flow transition from table 2. This previously unreported phenomenon underscores the complexity of flow transitions and the dynamics of wet granular materials.
Equivalent dynamic friction coefficient,
$\mu _{d}$
(a) and reduced friction,
$\delta \mu$
(b) versus Froude number,
$F_{r}$
, for various fluid densities.

To further explore the inter-particle properties of wet granular flow, we extracted two equivalent friction coefficients,
$\mu _{d}$
and
$\mu _{m}$
, corresponding to the dynamic repose angle and the starting angle, respectively. The reduced friction is then defined as the difference between these coefficients,
$\Delta \mu =\mu _{m} -\mu _{d}$
. Herein, the equivalent friction coefficient is determined according to
$\mu =\tan \theta$
. While these parameters maybe not be directly applicable to the evaluation of the granular rheology, the relative variation of the equivalent friction coefficients provides insight into interparticle contact and interaction in the submerged condition, which we believe is significant. Figure 6 gives the relationship between
$\mu _{d}$
and
$\Delta \mu$
with
$F_{r}$
. A notable trend is observed:
$\mu _{d}$
increases monotonically with rising
$F_{r}$
. In our configuration, the particulate mobility is a result of the competition between the drag and gravity effects. The drag force arises primarily from the inertia of drum rotation and fluid resistance. As a result, when
$F_{r}$
increases, the drag force becomes more dominant, leading to a higher equivalent friction coefficient. This also explains the correlation between a steeper free surface and greater rotation speed. The relationship between reduced friction and Froude number shows a consistent collapse in figure 6(b), irrespective of the fluid properties. The striking result is that, at high
$F_{r}\gt$
0.03,
$\Delta \mu$
seems to yield an asymptotic value. This is because the cascading regime has been close to transforming to cataracting regime at such high
$\omega$
, forming a strong, inertial-dominant and centrifugal effect (Henein et al. Reference Henein, Brimacombe and Watkinson1983; Arseni et al. Reference Arseni, Monaco De, Greco and Maffettone2020).
Top and lower surface angles vs rotate speed at different fluid densities: (a) 0.975,(b) 1.058, (c) 1.149 and (d) 1.182 g cm−3.

3.3. Free surface angle and angle divergence
The flow transition is frequently observed in both industrial applications and various geophysical phenomena, such as snow avalanches, debris flow and submarine landslides. However, the shift from continuous flow to cascading flow remains poorly understood and lacks a quantitative description. To address this, we have introduced
$\theta _{\textit{top}}$
and
$\theta _{\textit{lo}w\textit{er}}$
from the perspective of the surface morphology, to quantitatively characterise the dynamic behaviours of wet granular flow. The above statistical analysis of
$\theta _{\textit{top}}$
and
$\theta _{\textit{lo}w\textit{er}}$
is indeed found to follow two separate distribution patterns with distinct flow regimes. Building on these observations, we plot the evolution of these surface angles as fluid density varies in figure 7. For all
$\rho _{f}$
cases investigated,
$\theta _{\textit{top}}$
exhibits a noticeable steepening compared with
$\theta _{\textit{lo}w\textit{er}}$
. At high fluid density and sufficiently large rotation speed, the flow field exhibits strong fluid–particle interactions and significant inertia, leading to higher granular packing and a steeper
$\theta _{\textit{top}}$
. Conversely, we notice that
$\theta _{\textit{lo}w\textit{er}}$
remains relatively unchanged, regardless of fluid properties or rotation speed. The classical fluctuation of
$\theta _{\textit{lo}w\textit{er}}$
is around
$30^\circ$
, larger than the usual value (
$\simeq$
$25^\circ$
) for frictional particles (Clavaud et al. Reference Clavaud, Berut, Metzger and Forterre2017a
). Note that, in the case of higher
$\rho _{f}$
and larger
$\omega$
, the divergence between the
$\theta _{\textit{top}}$
and
$\theta _{\textit{lo}w\textit{er}}$
becomes more pronounced. Such variation indicates that fluid density significantly influences the formation of surface topography in wet granular flow, in conjunction with rotation speed. The divergence is very small and close to zero in the case of low
$\rho _{f}$
and small
$\omega$
. This corresponds to the rolling regime. One expected behaviour is the occurrence of negative angle divergence, where
$\theta _{\textit{top}}$
is smaller than
$\theta _{\textit{lo}w\textit{er}}$
and the flow surface is slightly convex. This phenomenon is consistently observed in our preliminary dry granular flow experiments and previous studies (Brewster et al. Reference Brewster, Grest and Levine2009; Chen et al. Reference Chen, Li, Xiu, Zheng and Yang2020). For very low rotation rates there are negligible inertial effects, and the surface shape is thus flat or slightly convex due to interparticle cohesion.
Flow phase diagram against the Froude number and density ratio for wet granular matter.

4. Analysis and discussion
4.1. Rolling–cascading flow regime transition
For wet granular matter, experimental and numerical studies have explored the dynamics and morphology of granular flow when submerged in water or with varying particle properties (Brewster et al. Reference Brewster, Grest and Levine2009; Liao et al. Reference Liao, Hsiau and Nien2015; Yuan et al. Reference Yuan, Moscariello, Cuomo and Chareyre2019). One can observe that the various regimes in a rotating drum primarily depend on the Froude number (Sylvain et al. Reference Sylvain, Gondret, Perrin and Rabaud2003). This normalised parameter is directly related to the rotation speed for a given drum size. Therefore, there is a critical existing rotation speed where flow can transition from the rolling to the cascading regime. However, when considering more complex scenarios, such as the higher sea-water density associated with submarine landslides, significant difference is observed from our experimental findings. To show this, we define by
$\Delta \theta$
the variations in surface angle divergence, where
$\Delta \theta = \theta _{\textit{top}} -\theta _{\textit{lo}w\textit{er}}$
. This provides a quantitative measure of the dynamic separation of granular flow induced by particle inertia. Figure 8 presents the flow phase diagrams against the Froude number and solid–liquid density ratio, integrating all quantitative observables. According to the experimental phenomenon, the cascading regime occurs in high
$F_{r}$
and low
$r$
cases, located on the lower right side of the phase diagram. Fluid density significantly influences the rolling–cascading flow regime transition. By smoothing the experimental results, an iso-line can roughly be drawn for a typical value of
$\Delta \theta \simeq$
$2^\circ$
in the centre of the scale. This nonlinear character reflects the complexity of the rolling–cascading regime transition, although it does not delineate the precise transition boundary. To our knowledge, this phenomenon has never been reported in previous studies. When
$\Delta \theta$
exceeds the threshold, the system enters the cascading regime. It shows that higher fluid densities facilitate the transition of particles into the cascading regime. Consequently, we conclude that the flow transition of wet granular materials exhibits a nonlinear phase transition boundary, also depending on the fluid density. This indicates that the critical rotation speed required for flow transition does not apply to all scenarios.
Normalised angle divergence
$\Delta \theta ^{\ast }$
versus normalised Froude number
$F_{r}^{\ast }$
for wet granular matter.

Although we have identified the quantitative characteristics of the rolling–cascading regime transition from a morphological perspective, further explanation is still required to fully elucidate the underlying mechanisms. Our experimental results show that the fluid density manifests as an obvious buoyancy effect within the inertial regime from the (
$S_{t}$
,
$r$
) domain. To combine the effects of fluid density and rotation speed, we thus introduce a buoyancy parameter,
$B=\Delta \rho /\rho _{s}=1-1/r^{2}$
. Then we normalise the Froude number by a buoyancy parameter as
$F_{r}^{\ast }=F_{r} \boldsymbol{\cdot }B^{-1} =F_{r}/ (1-1/r^{2} )$
. This dimensionless parameter essentially further incorporates the equivalent gravity caused by the solid–fluid density differences. Figure 9 presents the scaling of normalised angle divergence
$\Delta \theta ^{\ast }=\Delta \theta /\theta _{d}$
with
$F_{r}^{\ast }$
. It is obvious that the scaling follows two distinct behaviours for different flow regimes. The critical Froude number observed here is approximately 0.1, which is bigger than the usual value for dry granular flow (Mellmann Reference Mellmann2001). For low
$F_{r}^{\ast }$
, below 0.1,
$\Delta \theta ^{\ast }$
is very small and close to zero. Conversely,
$\Delta \theta ^{\ast }$
increases significantly with
$F_{r}^{\ast }$
when
$F_{r}^{\ast }\gt$
0.1, showing a monotonically increasing curve of the cascading regime in wet granular flow. When
$F_{r}^{\ast }\gt$
0.1, the particle inertia is dominant, resulting in the dynamic separation between the active zone and the passive zone of wet granular flow. Strong particle inertia leads to the S-shaped surface of cascading regime. This is consistent with the transition boundary shown in figure 8. We also validate that the increase of
$\Delta \theta ^{\ast }$
at high
$F_{r}^{\ast }$
is primarily due to the increase in
$\theta _{\textit{top}}$
, while
$\theta _{\textit{lo}w\textit{er}}$
remains nearly unchanged (see figure 7). For the granular flow in fluids, the grain kinetic energy is totally dissipated by the fluid in the collision process at low
$F_{r}^{\ast }$
, resulting in a smoother collision of whole systems. The obtained slope surface is thus certainly flat. However, once a large enough
$F_{r}^{\ast }$
is reached, the fluid–solid mobility in the active region performs different patterns. The top and lower particles of the surface are dominated by rolling inertia and dissipation by the fluid, respectively. An obvious surface angle divergence of the slope is thus observed, explaining the increase of
$\Delta \theta ^{\ast }$
.
In addition, the motion of grains submerged in the fluid is primarily attributed to buoyancy and viscous effects. However, in our rotating system, we believe the rolling–cascading transition is unlikely to be attributable to adhesive forces between particles and the fluid. First, we have employed the (
$S_{t}$
,
$r$
) domain framework to confirm that all experiments operate firmly within the inertial regime. The fluid density and viscosity are chosen so that
$S_{t}\gt$
1 and
$S_{t}/r\gt$
2.5, indicating that they are governed by the inertial regime (Sylvain et al. Reference Sylvain, Gondret, Perrin and Rabaud2003; Andreotti, Forterre & Pouliquen Reference Andreotti, Forterre and Pouliquen2013; Clavaud et al. 2017b). In this regime, the viscous effects of the fluid are relatively weak and negligible compared with the gravitational forces acting on the particles. In addition, the fluid viscosity change is sufficiently small. Therefore, the primary influence of the fluid on the particles can be roughly considered to be an effective buoyancy, which acts like an equivalent gravitational force. In fact, the modified
$F_{r}^{\ast }$
considers the equivalent gravitational effect brought about by the fluid buoyancy, which is also applicable to dry granular material immersed in air. As a result, when the density of the ambient fluid increases, the equivalent gravity acting on the particles decreases. This means that a small external energy input can lead to significant changes in their dynamic behaviour. This finding is consistent with the observation of the remarkably long-runout landslides on other planetary bodies, which occur in low-gravity environments (Singer et al. Reference Singer, Mckinnon, Schenk and Moore2012; Lucas, Mangeney & Ampuero Reference Lucas, Mangeney and Ampuero2014; Johnson & Campbell Reference Johnson and Campbell2017; Johnson & Sori Reference Johnson and Sori2020).
(a) Typical temporal evolution of dynamic repose angle fluctuations for the rolling regime (
$\rho _{f}$
= 0.975 g cm−3,
$\omega$
= 1.00 rpm) and the cascading regime (
$\rho _{f}$
= 1.182 g cm−3,
$\omega$
= 4.00 rpm). (b) Time series of normalised dynamic repose angle fluctuations extracted from the local rectangular area in (a).

4.2. Flow fluctuation of two flow regimes
Flow fluctuation of particles quantifies the stability of a wet granular system and the interaction among particles or between particles and fluid (Jarray et al. Reference Jarray, Magnanimo and Luding2019). The measurement of dynamic repose angle fluctuations,
$\delta \theta _{t}$
, is important to elucidate the variable dynamics and morphology of wet granular flows. In figure 10, we examine the temporal evolution of
$\delta \theta _{t}$
and normalised
$\delta \theta _{t}/\theta _{d}$
for the rolling regime (with
$\rho _{f}$
= 0.975 g
$\textrm {cm}^{-3}$
and
$\omega$
= 1.00 rpm) and cascading regime (with
$\rho _{f}$
= 1.182 g
$\textrm {cm}^{-3}$
and
$\omega$
= 4.00 rpm). It is evident that the cascading regime experiences significantly more intense fluctuations compared with the rolling regime. In the cascading regime, the peak fluctuation
$\delta \theta _{p}$
during steady state exceeds
$3^\circ$
, while in the rolling regime, it is only
$1.5^\circ$
. This suggests that particle motion is more vigorous and unstable in the cascading regime, particularly in the active region. This instability is further highlighted in the local time series of
$\delta \theta _{t}/\theta _{d}$
, as illustrated in figure 10(b).
Standard deviation of angle fluctuations and peak fluctuations in wet granular flow for different fluid densities: (a) 0.975, (b) 1.058, (c) 1.149 and (d) 1.182 g cm−3.

The standard deviations of
$\delta \theta _{t}/\theta _{d}$
,
$\Delta \theta _{t}$
and
$\delta \theta _{p}$
demonstrate a systematic transition from the rolling to cascading regime. As shown in figure 11,
$\Delta \theta _{t}$
remains consistently below
$0.015^\circ$
and
$\delta \theta _{p}$
hovers around
$2^\circ$
in the rolling regime. However, upon transitioning to cascading regime,
$\Delta \theta _{t}$
increases above
$0.02^\circ$
and
$\delta \theta _{p}$
increases by approximately three times. Through the analysis of fluid fluctuations, we identified the dynamic signature of the rolling–cascading transition, where both
$\Delta \theta _{t}$
and
$\delta \theta _{p}$
perform a significant jump. Such a jump underscores the distinct flow dynamics and closed streamline between the two regimes, as also indicated in figures 3 and 4. In general, flow fluctuations in granular systems primarily result from the breaking and reconstruction of internal force chains, as well as collisions between particles during flow (Niu et al. Reference Niu, Zheng, Yuan, Mao and Huang2024). As
$\rho _{f}$
and
$\omega$
increase, the enhanced energy input and fluid disturbance will accelerate this process and strengthen these fluctuations, leading to an increase of macroscopic
$\mu _{d}$
in wet granular systems (see figure 6).
Details and corresponding flow regimes of all the cases.

5. Conclusions
We have experimentally investigated the role of fluid density in the dynamics and flow regime transitions of wet granular materials. The results demonstrate that a solid–fluid density difference has a significant effect on both the starting angle and dynamic repose angle of a granular pile. The equivalent friction coefficient increases monotonically with the Froude number, which is mainly associated with the inertia of the drum rotation and fluid resistance. Importantly, we have also quantitatively characterised the rolling–cascading regime transition based on the evolution of the granular surface morphology. The PDFs of these morphological parameters show two distinct patterns for the rolling and cascading regimes. The flow phase diagram and flow curve presented here clearly delineate the transition rules governing these regimes. This implies that, for a wet granular material, higher fluid densities significantly enhance the likelihood of the system transitioning into the cascading regime. A notable jump observed in the angle fluctuation is believed to be an important dynamic signature, further supporting the rolling–cascading regime transition rule we proposed. This jump indicates the enhanced energy input and fluid disturbance strengthen the flow fluctuations, leading to significant angle divergence within the granular packing.
Our study highlights the significant influence of fluid density on submerged granular media, which has not received much attention before. However, further exploration with a wider range of particle sizes is necessary to verify the universality of these findings in complex application scenarios. Additionally, the fluid used in this study had a relative low viscosity coefficient, which simplifies the analysis and allows for a focused examination of the impact of the fluid density. However, cohesive forces induced by the fluid viscosity may still play a significant role at high enough rotation speed. Therefore, more refined experiments and a numerical study are needed to comprehensively investigate the viscosity effects in future research.
Acknowledgements
We thank M.Y. Zhu and K. Li for many helpful discussions, and X.H. Lin for careful instruction in drum operation.
Funding
This research was supported by the National Natural Science Foundation of China (Grant No. 42120104008) and the Fundamental Research Funds for the Central Universities.
Data availability statement
The data supporting this study’s findings are openly available from the corresponding author upon reasonable request.
Declaration of interests
The authors report no conflict of interest.
Appendix A.
Considering both the velocity field and surface morphology, we have classified the flow regimes of wet granular materials across all cases involving varying rotation speeds and fluid densities. Detailed descriptions about the rolling and cascading regimes are provided in the main text. Below, we summarise the statistical results in table 2.
Appendix B.
(a) Schematic diagram of the experimental procedure, (b) Evolution of free surface angles during the initiation and stabilisation stages.

The drum experiments we carried out were carefully designed and verified. Figure 12(a) presents the detail of our experimental procedure. Before the experiment began, we initially pre-rotated at a speed of 3.44 rpm for five minutes to achieve a relatively uniform initial packing state. Then, we acquired 3000 frames (1 min) for the initiation stage and 12 000 frames (4 min) during the stabilisation stage. By image processing, we can obtain the starting angle (
$\theta _{m}$
) from the initiation stage and dynamic repose angle (
$\theta _{d}$
) during the stabilisation stage, as shown in figure 12(b).































