J.-C.
Tsai
*a,
M.-R.
Chou
ab,
P.-C.
Huang
ab,
H.-T.
Fei
a and
J.-R.
Huang
*a
aInstitute of Physics, Academia Sinica, Taipei, Taiwan. E-mail: jctsai@phys.sinica.edu.tw; ajrhuang@gate.sinica.edu.tw
bDepartment of Physics, National Taiwan University, Taiwan
First published on 21st July 2020
We study the responses of fluid-immersed soft hydrogel spheres that are sheared under controlled volume fractions. Slippery, deformable particles along with the density-matched interstitial fluid are sandwiched between two opposing rough cones, allowing studies for a wide range of volume fraction ϕ both above and below the jamming of granular suspension. We utilize sudden cessations of shearing, accompanied by refraction-matched internal imaging, to supplement the conventional flow-curve measurements. At sufficiently high volume fractions, the settling of particles after the cessations exhibits a continuous yet distinct transition over the change of the shear rate. Such changes back out the qualitative difference in the state of flowing prior to the cessations: the quasi-static yielding of a tightly packed network, as opposed to the rapid sliding of particles mediated by the interstitial fluid whose dynamics depends on the driving rate. In addition, we determine the solid–fluid transition using two independent methods: the extrapolation of stress residues and the estimated yield stress from high values of ϕ, and the settling of particles upon shear cessations as ϕ goes across the transition. We also verify the power law on values of characteristic stress with respect to the distance from jamming ϕ − ϕc, with an exponent close to 2. These results demonstrate a multitude of relaxation timescales behind the dynamics of soft particles, and raise questions on how we extend the existing paradigms of the flow of a densely packed system when the softness is actively involved.
Granular materials at high ϕ present interesting examples showing the solid–fluid duality. In the past decades, studies using shear flows have established dimensionless quantities such as the inertia number and viscous number that successfully capture behaviors of both dry grains and particle–fluid mixture for a wide range of volume fractions ϕ up to 0.58.8–11 As reviewed by Guazzelli and Pouliquen,9 the current paradigm in theories has assumed Newtonian behaviors, i.e., the stress varying linearly with the shear rate (but with a ϕ-dependent viscosity) at constant volumes, while most supporting experiments and numerical studies are stress-controlled. On the other hand, it is not uncommon that at values of ϕ near the solid–fluid transitions, particle suspensions can exhibit shear-thinning behaviors1 such that the shear stress is proportional to n with an index n significantly less than 1 and in many cases close to 0.5.12–14 It is also worth noting that all hard-sphere theories would inevitably lose their predictive power on the flow behaviors at volume fractions beyond the “jamming point” – the current consensus for frictionless particles is the random-close-packing (RCP) ϕRCP ≈ 0.635, despite the existence of some debates on the exact meaning of RCP.15 Understanding the yielding and flow of densely packed soft particles remains a profound challenge. Therefore, further experimental information on the rheological behaviors as well as the shear-induced structural changes across the solid–fluid transition is of vital importance.
In this paper, we start with an Overview of our experiments on shear flows of centimeter-sized hydrogel particles immersed in a density-matched interstitial fluid. This section provides a survey over a wide range of volume fractions with the conventional flow-curve measurements, providing hints on the solid–fluid transition and an intriguing rate-dependent dynamics at densities well above the jamming point. These behaviors are systematically analyzed in two subsequent sections, using sudden cessations to the steady shearing in combination with our internal imaging of particle movements. Other factors such as a transition of fluid dynamics that coincides with the solid–fluid transition, the weakening of the hydrogel under stress over a long time and its possible effects are reviewed in our Discussion. In the conclusion, we summarize our findings and open questions that call for further understanding.
In Fig. 1c, the flow curves are displayed in logarithmic scales in a wider range of volume fractions. We also show the particle Reynolds number , which quantifies the relative importance of the inertial force in comparison with the viscous stress in a conventional suspension. In our experiments, the particles and fluids are density-matched at ρ = 1.2 g cc−1, while both the particle diameter d and the interstitial fluid viscosity η are fixed. Therefore Rep can be regarded simply as a dimensionless shear rate. Meanwhile, as the volume fraction decreases, particles would eventually lose direct contact and the system becomes a true suspension. Although all measurements are performed over both the low-Rep (<1) and the high-Rep (>1) regimes, data for low-volume fraction experiments (ϕ < 0.51) are all below our instrumentational limit for Rep < 3 and are thus not shown. In these cases wtih ϕ < 0.51, the presented data are dominated by the fluid dynamics and exhibit a simple, constant logarithmic slope with σS ∼ 1.7. However, this should not be interpreted as shear thickening. Based on our previous work in a similar geometry,18 we have demonstrated that in this regime of Rep secondary vortices are induced. Inertia is important in this regime, and that also explains why the shear stress appears insensitive to the viscosity at a limit of ϕ = 0 in this experiment – see the good overlap of data points for pure water (η = 1 mPa s) and those for our standard PVP-360 solution (η = 8 mPa s).
The good collapse of data points for ϕ = 0.73, 0.64, and 0.55 seems to extend the previous finding that the stress ratio can indeed be well characterized by a single dimensionless number J for multiple volume fractions in granular suspensions.9 Nevertheless, our measurements are made for much denser packing of particles than in previous studies. Reduction of volume fraction to ϕ = 0.51 creates a clear separation of the curve from those at higher volume fractions, in consistence with the picture that the system has become a suspension. However, a more accurate determination of the solid–fluid transition might require other indicators, which we would describe in the subsequent section with alternative measurements based on residual stress and particle movements.
It is worth noting that such transient rearrangements have been predicted in numerical studies targeting the dynamics of microgels.3 Nevertheless, our experiments present direct visual demonstrations of such effects but in the context of a centimeter-sized granular packing. As a rough indicator for these transient movements, we set the origin t = 0 upon the cessation of shearing, subtract the time series of image I(t) by a reference frame set at t = 5 s, compute the mean pixel intensity of each frame and average over 40 repeated cessations. Such “phase average” 〈|ΔI(t)|〉 is computed for experiments in a wide range of driving rates and at different volume fractions. Imaging conditions are kept identical (using the same camera with 876 × 876 pixels at a grayscale of 256, at the same exposure time of 8 ms with a fixed illumination). Fig. 3(a and b) illustrates the decay of 〈|ΔI(t)|〉 in a typical case above jamming. In particular, with the noise level ε removed, the semi-logarithmic plot Fig. 3b reveals interesting clues on the qualitatively different dynamics behind the “slow” versus “rapid” flows, which results in distinct outcomes of the settling in response to abrupt stopping of the boundary:
(1) At shear rates of 0.1 s−1 and below, Fig. 3b shows that the decay of particle movements is close to a simple exponential function. The curves are nearly a smooth continuation from t < 0 in which the packing is driven to flow, presumably, with successive local yielding of the elastic network. The time evolution after the cessation makes little distinction on whether the boundary has stopped. However, in this “quasi-static” regime, the cumulative magnitude of relaxation (f0), as well as the decay rate (β), still has a subtle dependence on the driving rate (ΩOn/2π). This subtle rate dependence and its relationship to the force balance behind the flows post delicate questions for theories.
(2) At shear rates above 1 s−1, Fig. 3b presents the settling of particles as a two-stage process. Upon the abrupt stopping of the boundary, |ΔI(t)| shows a steep descent within the first half second. Images reveal that particles exhibit rapid movements in finding neighbors for building up a network, based on which further adjustments take a much slower pace. The follow-up adjustments, cumulatively, are in general much smaller than 1d for each individual particle in reaching their final position – see Fig. 2b for two snapshots of the image ΔI. Interestingly, the good overlap of 〈|ΔI(t)|〉 in the second stage indicates that the cumulative relaxations are, in average, insensitive to the initial shear rates. This suggests that, with sufficiently high values of ΩOn, the combination of the rapid boundary shearing and the fluid dynamics at intermediate-to-high Reynolds numbers creates a set of fluid-like, unsupported initial configurations. It is then understandable that the outcomes of the relaxation, in average, appear insensitive to the history prior to the cessations (t < 0).
Furthermore, for the three high-volume-fraction cases that exhibit similar behaviors, we fit the long tails as shown in Fig. 3b by an exponential decay, f0exp(−β(t − tstop)), in which 1/β features a characteristic time of decay. The constant f0 can be regarded as the magnitude of total relaxation for the second stage. The results are shown in Fig. 4(a and b). In addition, the ratio (〈|ΔI(t)|〉 − ε)/f0 in Fig. 4c indicates the difference of the full curve from a simple exponential decay: a large value indicates the occurrence of the aforementioned two-stage process, while unity corresponds to the case with a straight exponential. In short, Fig. 4 summarizes a gradual transition of dynamical regimes over the driving rate: the incremental yielding of elastic “solids” in the slow regime, in contrast to the more “fluidic” sliding of particles mediated by interstitial fluid flows in the fast regime. Note that such a transition over driving rate coincides with the aforementioned observation of the steady-state flow curves (Fig. 1b), although a rigorous theoretical explanation demands further work. And such rate dependence appears insensitive to the exact value of ϕ, as long as it is high enough to develop solidity.
Fig. 4 Summary on the driving-rate-dependent settling of particle movements, for experiments at three volume fractions above 0.6 as indicated by the legend. (a and b) Values of 1/β and f0 for the best fits as shown in Fig. 3b. (c) 〈|ΔI(t)|〉 − ε/f0, featuring the degree of deviation from a simple exponential decay. All are shown as functions of the corresponding shear rate prior to the sudden cessation. The vertical dot-dash line marks the same 0.3 s−1 as in Fig. 1(b and c) to guide the eyes. |
It is also worth noting that, once the volume fraction goes below the jamming point and the system becomes a true suspension, the evolution of images is dramatically different. This is demonstrated by Fig. 3c, with further information on the solid–fluid transition to be discussed in the next section.
Fig. 5 Stress measurements with cyclic shear cessations – (a) protocol of CSC as demonstrated by the time-dependent shear rate (t); (b) the time-dependent shear and normal stress in response to the CSC, as well as illustrations of the plateau values, |σ(P)S| and σ(P)N, and stress residues, |σ(R)S| and σ(R)N. Sample signals are shown for the case of ϕ = 0.73, with smoothing by 0.1 s to suppress the random noise. (c) Plateau values as functions of ϕ, in experiments with two different values of (P), ΔOff = 120 s, and accumulated strain ΔOn = 5.5 for each half-cycle. Statistical variations among different cycles are all within the size of the symbols. Corresponding steady-state stress (adapted from Fig. 1c) is also displayed for comparison. |
Fig. 5c presents our measurements of plateau values as functions of ϕ at two rotation rates, for both |σ(P)S| and σ(P)N. In addition, we adapt data of steady-state stress σS from Fig. 1 to show that, with an accumulated strain ΔOn = 5.5 for each half-cycle, the values of |σ(P)S| in our CSC are in fairly good agreement with the true steady-state stress.
In the attempts of approaching the solid–fluid transition, we perform a series of CSC experiments with incremental changes of ϕ and measure the stress residues with care using four different rotation rates spanning over three decades for each case. Results are shown in Fig. 6, which describes how “yield stress” would vanish as ϕ decreases. The rationales are as follows: (A) for our system with hydrogels, we have verified that both σS and |σ(P)S| decrease monotonically as the shear rate is reduced, within the range of our observations. (B) It is shown in previous studies of microgels that the stress residue increases over the decrease of the shear rate,3 and we have also verified that this is also true with our fluid-immersed macroscopic hydrogel particles – see data of |σ(R)S| in Fig. 6a. (C) It is only natural that the dynamical yield stress σY must be bounded by the relationship |σ(R)S| < σY < |σ(P)S|, and that σY can be estimated simply by progressively reducing the driving rate (not necessarily to zero). Therefore, by measuring how |σ(R)S| or |σ(P)S| changes as functions of ϕ, one shall be able to get a reasonable estimate of how σY approaches zero. In fact, as shown by Fig. 6, we find that both |σ(R)S| and |σ(P)S| present a good fit to be proportional to (ϕ − ϕc)2. However, the uncertainty on ϕc is significant, because instrumental noise has limited our stress measurements to be reliable only at values of ϕ that are at a substantial distance above the critical value.
Fig. 7 Settling of particles, in response to the cessation of shearing at t = 0, for four different volume fractions ϕ as indicated. Definitions of 〈|ΔI(t)|〉 − ε, and time t follow those from Fig. 3. Data include runs with ΩO(t < 0)/2π = 1 rps (squares), 0.1 rps (circles), and 0.01 rps (triangles), respectively. |
Intriguingly, the solid–fluid transition in our system shows a critical volume fraction that coincides with the random-loose-packing (RLP) value reported in prior studies.23 However, as RLP is usually discussed in the context of frictional packing, whether there is a real connection between the solid–fluid transition of our system which is well lubricated, and the value of RLP, demands further studies for clarification.
We choose a low viscosity for our interstitial fluid, which has made the effect of yield stress explicit. However, due to instrumentational limitations described above, an accurate determination of the dynamical yield stress from flow curves, Fig. 1c, still would count on extrapolations, and complications arise around the transition ϕc (estimated to be between 0.549 and 0.554, Fig. 7) and below: The convexity of these curves comes mainly from that the driving rate goes across Rep = 1 such that the logarithmic slope can be anticipated to change even for a pure fluid. The existence of a yield stress contributes relatively little to the convexity, which is crucial in “determining” the yield stress. To model the flow curves correctly, one might have to use a less restrictive HB model, in which the index n also depends on the Reynolds number (or ) – and that is beyond the scope of our discussion. In this work, we have explored two alternative routes to detect ϕc, by observing the change of (a) the residual stress in CSC experiments and (b) the particle settling after the sudden cessation of driving. It turns out that the latter provides a more sensitive detection on the solid–fluid transition ϕc than the former.
The shear-to-normal stress ratio, σS/σN, shows a rapid initial decay in the first 102 s, but is stabilized beyond 103 s. The rapid decay implies a reduction of structural anisotropy due to particle rearrangements, which we will demonstrate with internal imaging subsequently. The substantial decay of total stress between 102 s and 105 s but with a fixed stress ratio, on the other hand, suggests the weakening of the hydrogel particles themselves with an isotropic decay of mean stress over time. In what follows, we test our conjecture on the weakening of hydrogels with an experiment at the single-particle level.
To probe whether the strength of individual grains gradually weakens, we monitor the response of a single hydrogel particle to a fixed compressional strain over long time. As shown in Fig. 8b, the particle is compressed to d − ε0 and stays fixed. The initial condition F(0+) = 0.6 N is about ten times of d2σN to mimic the situation of those load-bearing particles inside a high-volume-fraction packing. Multiple long-time experiments confirm that (1) the hydrogel maintains its strength up to about 100 s, and that (2) hydrogel particles do weaken significantly at larger timescales. The fact (2) confirms our conjecture that material weakening is involved in the long-time decay of stress. In addition, fact (1) in combination of the stabilized stress ratio at 102 s justifies our choice of ΔOff = 120 s (30 s) for evaluating the residual stress in Fig. 5 (Fig. 6).
We anticipate that such mechanical weakening can also play a role in steady-state flows at extremely slow shear rates. This suggests that the conventional strategy of obtaining the dynamical yield stress by extending the window of observation in toward the “slow limits” might involve undesired complications for materials like hydrogels, which by themselves evolve slowly.
One intriguing question remains open: what determines the transitional zone of shear rates (between 0.1 and 1 s−1) for the rate-dependent behaviors that consistently occur in both the conventional flow-curve measurements (Fig. 1) and the response to sudden cessations (Fig. 3 and 4)? Inverting the transitional shear rates gives the timescale between 10 and 1 s. However, with the material parameters in our experiment, time constants constructed from dimensional analyses such as η/σN or η/E are well below 10−5 s. These values are obviously too small to account for the relaxation we have observed, even though they are commonly used as the time unit in theories and simulations.24 We believe that a suitable explanation on what we find must include the dynamics of lubrication and, most likely, in conjunction with draining between particle surfaces and/or the permeability of the hydrogel. Full answers demand further works. However, there are a few clues: Our internal imaging of steady-state flows at different driving rates does reveal non-smoothness of particle trajectories that depends on the shear rate. In addition, we have also noted that such rate dependence can become dramatic when slippery particles are replaced with frictional grains.25 Another interesting puzzle is on the reason behind the subtle rate dependence of the total relaxation (f0, Fig. 4b) in describing the settling of particles at the slow limit. We leave this as another open question.
We hope that our survey of the shear flow and relaxation of frictionless soft particles raises questions worthy of attention and sheds light on extending the existing paradigms, for our understanding on the macroscopic rheology of packed grains with the finite rigidity and fluid-mediated interactions between particles taken into account.
Footnote |
† By precision optical monitoring, we have verified that our rotating boundary stops abruptly within 10 ms in all cases. |
This journal is © The Royal Society of Chemistry 2020 |