You are on page 1of 3

VOLUME 79, NUMBER 23 PHYSICAL REVIEW LETTERS 8 DECEMBER 1997

Granular Convection and Transport due to Horizontal Shaking

Kurt Liffman,1 Guy Metcalfe,1 and Paul Cleary2


1
CSIROyDBCE, P.O. Box 56, Highett, Victoria 3190, Australia
2
CSIROyMIS, Private Bag 10, Rosebank MDC, Victoria 3169, Australia
(Received 7 February 1997)
Vibrated horizontally, granular material has different convection patterns from those observed for the
equivalent vertical vibration. Among the patterns that develop, above the onset of bulk motion, is a four
roll convection pattern. Surface sloshing produces two smaller convective rolls above two lower interior
rolls. Horizontal vibration creates intermittent gaps between the sidewalls and the heap of material.
Avalanches of material into these gaps drives the lower convection rolls. This is a different physical
mechanism from that observed in convection driven by vertical vibration. As is well known, convection
patterns can drive size segregation, and we illustrate some of these effects. [S0031-9007(97)04607-3]

PACS numbers: 46.10. + z, 83.10.Hh, 83.50. – v, 83.70.Fn

Convection patterns arise in beds of granular material cluded for soft particle-wall collisions, though not for soft
subject to vibration [1]. Knight et al. [2] demonstrated this particle-particle collisions. So, our model simulates the
convection and additionally how this convection can seg- full range of frictional interactions between the walls and
regate particles by size, when the vibration is vertical. The the particles. It does static and slip friction for impacts
physical mechanisms driving vertical bulk granular motion between particles, but it has no static, slip, or rolling fric-
are still somewhat uncertain, but frictional interaction of tion for particles that are resting upon each other. This
the particles, both with themselves and with the walls, ap- particle interaction model has been adopted because the
pears to be a prime candidate [1]. Although the behavior standard method for simulating static friction is difficult to
of granules subject to vertical shaking is well documented, implement for particles that are allowed to roll upon one
there have been very few studies that consider the effects another [10]. The lack of such frictional contacts makes
of horizontal shaking. Given that industrial handling of the simulated microstructure somewhat more deformable
granular materials usually involves a combination of hori- than reality, though this effect is probably swamped by
zontal and vertical shaking, additional basic knowledge of the intrinsically more deformable characteristics of round
what occurs with horizontal shaking is of interest. particles, relative to the roughness and angularity of real
In this paper, we document our computational studies of particles. Particle-wall interactions drive the motion, and
a horizontally shaken bed. We find that, for similar levels our inclusion of all the frictional forces should improve
of shaking force, horizontal shaking produces different the simulation of the bulk flow.
convection patterns from what is observed in the vertical To examine the convection generated by horizontal
case. Moreover, the mechanism driving the convection forcing, we consider a box of identical particles. The
rolls is different from that in the vertically shaken case. initial packing of the box has 900 disks, of unit diameter,
We use a numerical code that directly simulates particle on a square 30 3 30 lattice. The box is 30 particle
interactions by a hybrid hard sphere and particle dynam-
ics scheme [3,4], where particles have “hard” or “soft”
collisions. Hard collisions occur between particles with
large relative velocities. Momentum conservation deter-
mines the postcollisional velocities, while energy loss is
set by coefficients of restitution and slip friction [5,6].
Soft collisions occur between particles with small relative
velocities. Here particles are assumed to be connected by
a damped spring stretched by virtual interparticle pene-
tration. Postcollisional velocities and energy losses are
determined by a damped, harmonic-oscillator equation
governing the interaction. The coefficient of restitution
is kept the same for soft and hard collisions by adjusting
the oscillator spring constant and damping factor [7]. The
transition between hard and soft collisions occurs when
the recoil speed from the hard collision can no longer FIG. 1. The average displacement per cell J as a function of
keep two particles apart, and the particles begin to inter- G. Below G ­ 1, there is one convection cell (see inset for
penetrate. Static, slip, and rolling friction ([8,9]) are in- G ­ 0.895 flow), while for G . 1.2 two convection cells start.

4574 0031-9007y97y79(23)y4574(3)$10.00 © 1997 The American Physical Society


VOLUME 79, NUMBER 23 PHYSICAL REVIEW LETTERS 8 DECEMBER 1997

FIG. 2. Displacement vectors sG ­ 2.01d for (a) the beginning of the shaking cycle, (b) halfway through the cycle, and (c) sum
over all phases of the cycle. The arrow indicates the approximate separation point of the heap from the wall. The distance between
the bottom of the bin and the separation point is called the separation distance, which we denote by ys . That displacement vectors
appear to originate from the walls is an artifact of averaging over a complete shaking cycle.

diameters across; the walls are very high, and there is convection rolls [Fig. 2(c)], which flow down the center
effectively no box top. This is a fairly narrow, tubelike of the box and are the result of surface sloshing. The
box, and we would expect some of the details of results lower convection rolls flow up at the center of the box,
to vary with aspect ratio. The box shakes horizontally which is consistent with the limited experimental evidence
according to xstd ­ A sinsvtd, where A is the amplitude available in the literature [13].
and v the angular frequency. The distinguished parame- The mechanism driving the lower convection rolls is
ter is the nondimensional acceleration G ­ Av 2 yg; in avalanching. During each half cycle, particles pile up
what follows A ­ 5 particle diameters. against one wall and a gap appears between the heap
Figure 1 shows the onset of collective motion. Fol- and the other wall [Fig. 2(a)]. The gap allows space for
lowing Taguchi [11], Fig. 1 plots the average particle dis- particles to fall into. There are two ways for particles to
placement J vs G as a convenient way to detect the onset fill this gap. The pile’s surface becomes sufficiently steep
of bulk convection. The initial bifurcation, at G ­ 0.5, to cause an avalanche, or sloshing material can be thrown
is to a 1-roll state with modest surface agitation (inset into the gap. During the next half cycle, the gap closes
to Fig. 1). This is significantly below the G * 1.2 onset and the avalanched particles push other particles into the
for vertically shaken convection [11]. Above G ­ 1.2 the heap interior [Fig. 2(b)]. This happens repeatedly on both
1-roll state changes to a 4-roll state with a large amount of sides to drive the 2 lower convection rolls. To corroborate
surface agitation. Above G ­ 2 there is significant change this convection scenario, consider the relation of gap size
in the shape of the pile, which we will not discuss here. to lower convection roll size. The gap does not open all
In this paper, we will focus on the behavior of the 4-roll the way to the bottom of the container; cf. the arrow in
state for 1.2 , G , 2.2. In a vertically shaken bed, Aoki
et al. [12] have also observed a 4-roll state; however, the
rolls were four across instead of one on top of the other,
and occurred at higher accelerations sG . 4.5d.
In the 4-roll state the pile has a cyclic sloshing motion
with a period Tp equal to the forcing period. The particles
pile up against the walls during different phases of the
cycle. Figures 2(a) and 2(b) show particle displacement
plots made at different phases. These plots are made
by noting the particle positions at some particular phase
of the forcing cycle. The box then goes through one
shaking cycle. Once the box has returned to the same
phase of the cycle, the subsequent position of the particle
is determined and the resulting displacement is computed
and plotted. Figures 1(inset) and 2(c), are obtained in the
same way, except we now divide the forcing cycle into FIG. 3. The distances to the separation points, ys (squares),
and convection points, yc (triangles), versus G. The correlation
50 segments. Each phase segment produces its own set between ys and yc is close to linear (inset). In this case, the
of particle displacements, which can then be averaged separation point is the value of y where the heap is one particle
and plotted. Such phase-averaged plots show the upper diameter away from the side wall.

4575
VOLUME 79, NUMBER 23 PHYSICAL REVIEW LETTERS 8 DECEMBER 1997

FIG. 4. Smoothed values for the height of a large particle s yd


versus time std for horizontal and vertical shaking. Time is in
units of the shaking period sTp d, while the height is in units FIG. 5. Particle displacement vectors for a 10-particle-high
of the original height of the pile (h: in particle diameters). heap sG ­ 2.01d.
The smoothing period is two time units. The solid lines show
the downward movement of the large particle for horizontal
sG ­ 2.0d shaking. The dashed line shows the upward motion Horizontal forcing of granular material in a box pro-
of the large particle for vertical sG ­ 2.0d shaking. This line duces bulk convection by several mechanisms. Surface
is offset by one. sloshing produces smaller surface rolls. Avalanching into
the intermittently formed gaps between the heap and the
Fig. 2(a). Call ys the height measured from the bottom of walls produces larger rolls in the interior of the box. Size
the container at which the heap separates from the wall. segregation in these flows tends to move larger particles
Figure 3 shows that ys decreases with increasing G. Also, from the top of the heap downwards. This effect depends
the lower convection rolls do not extend to the bottom of on the heap height and has implications for granular trans-
the container; cf. Fig. 2(c). At distances lower than ys the port control strategies.
heap touches the wall, so particles cannot tumble below
this point. We can then ask the question: is the distance [1] H. M. Jaeger, S. R. Nagel, and R. P. Behringer, Rev. Mod.
between the bottom of the container and the lowest point Phys. 68, 1259 (1996), and references therein.
of the convection rolls (a distance we call yc ) proportional [2] J. B. Knight, H. M. Jaeger, and S. R. Nagel, Phys. Rev.
to ys ? Figure 3 plots ys and yc vs G, and the inset plots Lett. 70, 3728 (1993).
their correlation. The good linear correlation is consistent [3] C. S. Campbell and C. E. Brennan, J. Fluid Mech. 151,
with our simple model for the physical mechanism of 167 (1985).
horizontal convection. [4] P. A. Cundall and O. D. L. Strack, Geotechnique 29, 47
How do these horizontal convection patterns affect (1979).
segregation? Figure 4 shows the height y of a larger [5] K. Liffman, D. Y. C. Chan, and B. D. Hughes, Powder
particle (twice the others’ diameter) vs time for 15 and Technol. 72, 255 (1992).
[6] M. A. Hopkins, Clarkson University, Potsdam, Report No.
30 particle deep heaps shaken horizontally (solid lines)
87-7, 1987.
and for comparison, a 30 particle deep heap shaken [7] The soft collision coefficient of restitution, obtained
vertically (dashed line). Horizontal shaking drives the from a homogeneous solutionpof the damped harmonic
larger particle (initially placed on top of the heap) down. oscillator, is er ­ exps2pcy 4mk 2 c2 d, where m is
Vertical shaking drives the same particle (initially placed the particle mass, c the damping factor, and k the spring
on the bottom of the heap) up [14]. Note the heap height constant. In this simulation, we set er ­ 0.7 for both
dependence of the horizontal segregation. Starting on top hard and soft collisions.
of the deeper heap, a larger particle swiftly descends to [8] The slip-friction coefficient smslip d is 0.5. The coefficient
the bottom of the upper convection rolls. There it can of rolling friction smroll d is 0.01.
stay for many shaking cycles, but eventually it follows [9] J. Witters and D. Duymelinck, Am. J. Phys. 54, 80 (1985).
the convective flow back to the top and resubmerges. A [10] H. J. Herrmann, in Disorder and Granular Media, edited
by D. Bideau and A. Hansen, Random Materials and
decrease in heap height causes a decrease in the height of
Processes (North-Holland, Amsterdam, 1993).
the lower convection rolls, but the behavior of the surface- [11] Y-h. Taguchi, Phys. Rev. Lett. 69, 1367 (1992).
convection rolls remains relatively unchanged. Figure 5 [12] K. M. Aoki, T. Akiyama, Y. Maki, and T. Watanabe,
shows the convection pattern for a 10 particle deep heap. Phys. Rev. E 54, 874 (1996).
Larger particles sinking in shallower piles can remain near [13] P. Evesque, Contemp. Phys. 33, 245 (1992).
the bottom of the box for very long times before being [14] Movies available at http://fluids.csiro.au/kurtl/Granule/
recycled by the flow to the top. granular.html

4576

You might also like