29

Powder Technology, 9 (1974) 29-37

@ EIsevier Sequoia S.A., Lausanne - Printed in The Netherlands

Experimental Evidence for the Stochastic Theory of Particle Flow Under Gravity*

W.W. MULLINS

Carnegie-Mellon University. Schenley Park, Pittsburgh, Pa. 15213 (U.S.A.)

(Received April 30,1973; accepted August 6, 1973)

Summary appropriate existing data in the literature_ The

elements and relevant results of the theory are
Experimental evidence is presented for a presented in Section 2. In Section 3, certain
recent stochastic theory of flow which repre- flow data available in the literature are plot-
sents the convergent flow of cohesionlesspar- ted, showing fair agreement. with the theoreti-
titles under gravity toward an open orifice as cal predictions. A discussion and interpreta-
equivalent to a counterflow of uoids from the tion of the plots is offered and a qualitative
orifice upward through the bed by biased ran- extension of the theory based on a transient
dom flight; the theory is summarized in a new effect is suggested to explain some of the dis-
phenomenological form. Data, taken from the crepancy. Finally, Appendix A offers addi-
literature, were obtained from cells initially tional theoretical results pertaining to the
loaded with alternate layers of differently finite orifice case.
colored granular material. The theory predicts
that after a fixed flow, each layer develops a 2. STOCHASTIC THEORY OF PARTICLE FLOW
depression such that if zO is the original UNDER GRAVITY
height of a given layer above the orifice and
z, is the corresponding height of the depres- The theory is essentially a kinematic one
sion minimum, then for three-dimensional since mechanics (i.e. forces, stresses) does not
ftow a plot of zz vs. z$ for different layers enter the picture. Furthermore, the orifice is
will yield a straight line of slope one; the in- treated in a purely formal way as a sink for
tercept gives statistical information concern- particles; the mechanics of the flow in this
ing the equivalent void jumps. For two-dimen- region, which must ultimately become a free-
sional flow, the corresponding theoretical plot fall zone for an open orifice, is not consid-
is &I,‘” vs. z~/~. ered.
Data plotted from several sources conform The basic concept is the representation of
closely to the above predictions, provided z,, convergent particle flow toward an open
is not too large for a given flow. The latter orifice, under gravity, as equivalent to a coun-
discrepancy is qualitatively explained by a terflow of voids entering the orifice and
transient effect requiring the density to fall to migrating upwards through the bed by a
a certain level before steady-state flow can oc- biased random flight as they are repeatedly
cur_ filled by particles moving down from above;
the vertical (z) component of void motion is
1. INTRODUCTION always positive (up) and the horizontal com-
ponent is random. In the literal version of the
The purpose of this paper is to compare the theory, the particles are visualized as arranged
predictions of a recent stochastic theory1 of on a random lattice so that voids represent
cohesionless particle flow under gravity with unoccupied lattice sites. Assuming that the
voids move independently with a certain
* This paper was presented at the Second Powtech probability density (p.d.f.) describing the
Conference held in Harrogate, England, in March magnitude and direction of-each jump, the
1973, and organised by the Powder Advisory Centre. steady-state flux of voids (s) and hence of
30

particles (3 = -3) may be deduced from the the steady-state condition that the density is
theory of random flight. The model is closely constant at any given point. Equation (4),
analogous to biased solid-state diffusion’ by which follows from (2) and (3), is parabolic
the vacancy mechanism as well as to the and is identical to the two-dimensional time-
Galton device illustrating the laws of probabil- dependent diffusion equation under the sub-
ity in which balls dropping through an orifice stitution (time) -+ z, (diff. coeff.) 4 CY;with
encounter a lattice of pins and, as a result of appropriate boundary conditions the equa-
chance deflections, fan out into a Gaussian tion determines J,, , and hence 2 from eqn.
distribution; the downward-biased random (2), throughout the entire region of consider-
flight of balls in the Galton device corre- ation. Although the equations used in this
sponds to the upward-biased random flight of paper to fit the experimental data are based
voids in our case. on the solution of eqn. (4) for a point orifice
In the original theory, the basic flow equa- in the floor of a semi-infinite bed, the equa-
tions were stated in terms of the number c of tion has also been solved for other boundary
voids per unit volume. The treatment also conditions including a finite orifice in the
showed, however, that the z component of floor of a semi-infinite bed and a point orifice
particle flux is given by at the base of a vertical pipe.
An alternative model of the particle flow to
J2 = -v Cur) c (1) that of the sand lattice, which leads to the
where v is the effective number of void jumps same phenomenonological equations (2)-( 5),
per unit time and (w) is the expected value is illustrated in Fig. 1. This shows, in two
of the vertical component w of a jump. It is dimensions, an idealized flow chain describing
therefore possible to use eqn. (1) to substitute a component of the total flow induced by the
for c in eqns. (11) and (12) of ref. 1, thereby downward movement of particles at the bot-
obtaining the basic steady-state flow equa- tom_ The total flow is regarded as the super-
tions in terms of the observable quantity Jz : position of a large number of such chains
the result is giving a convergent tree-like pattern of move-
ment. Since, for a short time At, the displace-
ment in each ci- ain may be formally described
by moving an equivalent void along the chain
path from bottom to top, the model leads
exactly to eqns. (2)-( 5) with u and w defined

where

(z12> w> _03 (5)

cx= 2(W)
- = 2(w)
- = 4(w)
and where u and u are the x and y compo-
nents of the equivalent void jump and r = (x 2
‘Y 2 ) % the cylindrical radial component. (Note
v does not enter the steady-state equations.)
These equations constitute a purely phenom-
enological basis for the description of the
steady-state flow.
Equation (2) (assuming a frame at rest with
respect to the orifice) states that particles
move laterally toward regions of greater ORtilCE
downward flux, in proportion to the hori- Fig. 1. Flow chain representing a component of the
zontal gradient of-J, . Equation (3) expresses tote1 flow.
 31

as the components of the independently de-

scribed chain link as shown in Fig. 1. Evi-
dently, eqns. (2)-(5) per se do not require
the physical presence of porosity in the par-
ticle bed. We turn now to the presentation of
the equations used in this paper to analyze
experimental data.
As already stated, we assume steady-state
flow and the point orifice approximation,
justified later in this section. Furthermore, to
convert fluxes to velocities, we assume a con-
stant density ?z (no. particles per unit volume)
throughout the region of steady-state flow.
Then supposing particle motion is adequately
described by the average velocity (ie. ignoring
fluctuations), it can be shown that for a semi-
r,=
infinite bed with orifice at r = z = 0, a particle Fig. 2. Cross-section of calculated consecutive marker
whose cylindrical coordinates are (ro,zo ) at t surfaces for three-dimensional flow out of a point
= 0 will have coordinates (r, z) at time t given orifice (z = r = 0). Curve labels give elapsed time.

by

[1 t 1
1 I2
z=zo 1-p @a) where u is the equivalent volume drained out.
TV0 zo) Equation (7a) predicts that if measure-
ments on a series of initially horizontal layers
and of various heights z. are [Link] after a fixed
l/2 elapsed time of flow, a plot of 2% vs. zz
z
r=rO
[l
7
.O
(6b) should be a straight line of slope one; from
the intercept i, on the 20” axis, CYmay evident-
ly be calculated as
where
2no!n ri a=* (7b)
T(r,z,) = Q exp G Z;
[I I
0
if u is known.
is the exit time for the particle; in this expres- For the two-dimensional case (all quantities
sion Q is the efflux defined as the number of independent of y), the equations corre-
particles draining out of the orifice per unit sponding to eqns. (6) are
time.
If a thin initially horizontal layer of distin- rl - t 213

T(sozo) @aI
z = z.
guishable (e.g. colored) particles is assumed to L 1
be at z = zo, then as r. ranges over all values,
eqns. (6) describe the profile z(r) of the layer
after an elapsed time t of flow. Figure 2 x = x0 z II2 (=I
shows a plot of several consecutive profiles [ 20 1
calculated from eqns. (B), labeled in units of
the time required for a particle on the z axis,
originally at zo, to reach the orifice (z = r = T(x,, zo) = 4fntrrkf
R exp [4s]zz/2
0
(8~)
0).
calling 2,: the minimum value of z for the where Q ’ is the number of particles draining
profile, occurring on the z axis, we have, upon out of the orifice (z = x = 0) per unit time per
squaring eqn. (6a) and using eqn. (6~) with r. unit length normal to the x zplane (i.e. along
= 0, the result the y direction).
Qf-2 u Raising eqn. (8a) to the 3/2 power and us-
*2=z2___z
270X
Ua) ing eqn. (SC) with x0 = 0, we have
m 0 2iTcun O
32

$W _ 3Q’t = z3/2 3a
-3’2=
;rn 0 0 -
112
4(m)““,z 4(5ra)
(W

where a is the equivalent area on the x z plane

drained out.
Equation (9a) shows that for two-dimen-
sional flow, a plot of .zL’” us. i~t12 should
yield a skaight line of unit slope. Again from
the intercept i2 on the 2: 12. axis, (Y may evi-
dently be calculated from

(9b)

if is known.
~7

The point orifice approximation represents

the orifice as a point sink of particles and
therefore ignores the dimensions and details Fig. 3. Reproduction of Guinta’s Fig. 7 illustrating
of flow around the real orifice; we proceed to measuring procedure.
discuss the justification for this approxima-
tion_
Consider first a finite circular orifice be much less than z* so that the free-fall zone
(radius R, area A ) of an idealize3 type such can safely he neglected in the analysis which
that the effius y = Q/A per unit area is con- follows.
stant and of sufficiently small magnitude so Although some of the data analyzed in Sec-
[Link] the stochastic flow model ran be sup- tion III do not fulfill eqn. (10) (for two di-
posed to hold within the immediate vicinity mensions R + X), the agreement between
of the orifice*. Then a rigorous evaluation of theory and experiment shown by Figs. 3-6
the exit time of a particle on the z axis may seems to indicate that the point orifice ap-
be carried out (Appendix A). The results proximation is reasonable and suggests the
(eqns. A.7 and A-12) show that the error in possibility that the “effective” dimensions of
using the point orifice approximation to com- the physical orifices may be somewhat less
pute exit times is less than 10% if than their geometrical dimensions; this would
s > s ri: = 2.5 R”Ja result, for example, from a non-uniform ef-
(10)
Furthermore, if the condition is well satisfied,
the velocity field in general is essentially given
by the point orifice approximation.
For a free orifice, entirely open to the
space below, an order of magnitude calcula-
tion indicates, and experimental data indirect-
ly support, the existence of something resem-
bling a free-fall zone surmounting the orifice.
An estimate of the height h of the free-fall
zone, however, which can be made from the
empirical formula of Rausch,3 shows that for IO0 -

the systems discussed in this paper, h should

*
250 300 350 500 450 500
The condition might be approximated experimen-
2: (inch=1
tally by regulating the particle efflux velocity to be
the order of or less than the free-fall velocity Fig. 4. 22 plot of Guinta’s data on iron ore concen-
through a particle diameter. trate from Table 2.
 33

T----
I
PARTICLE SIZE RANGE = 0.0312” - 0 0625”

Fig. 6. r3/2 plot of Brown and Hnwksle>-‘s data

sand from Table 1.

amined and are found to have the sagging or

plunging shapes shown in Fig. 3; the quanti-
ties z. and z, used in the analysis are illus-
trated in the figure for layer 4. (h?arked layers
Fig. 5. 2312 plot of Gardner’s data on Durite frcm studied in this and in the other data have been
Table 3. numbered consecutively from the top down.)
The z2 plot of Guinta’s data (three-dimen-
flux velocity with higher values toward the sional system) is given in Fig. 4 and the ~a!’
orifice center leading to an Rezf < R, that plots of Gardner’s and of Brown and Hawks-
would better satisfy eqn. (10). ley’s data (two-dimensional systems), subject
The remaining assumption of steady-state to a limitation described below, are given in
flow appears from the data to be violated for Figs. 5 and 6 respectively_ The agreement
short times and large ze. A qualitative explana- with the theoretically predicted 45” line ap-
tion of this observation is offered in Section pears reasonably good. Note also that the plot
3. of Brown and Hawksley’s data is taken from
three different figures representing consecu-
3. EXPERIMENTAL DATA tive stages of flow in the same cell; this pro-
gression corresponds to the three 45” lines of
Experimental data in the literature suitable increasing intercept as described by eqn. (9).
for testing the predictions of the stochastic To calculate CYfrom the data of Guinta, we
theory are very meager. In this section, we note from Fig. 4 that i, = 138 in2; v was
present z2 and z312 plots for data taken from estimated as the average of the approximately
Guinta”, Gardner’, and Brown and Hawks- conical volumes included between the initial
ley6. The systems studied by these investi- and final configurations of layers 5 and 6,
gators are characterized in Table 1 and the which gives 95 in3 2 20 in3 _ Substituting
data used for analysis are given in detail in these values into eqn. (7b), we obtain
Tables 2-4. Figure 3, which is a reproduction (Y = 0.11 in. + 0.02 in., which appears in Table
of Guinta’s Fig. 7, illustrates the typical situa- 1. From this value of Q, a value z*: may be
tion in which a cell is initially loaded with calculated* from eqn. (10): the result t*’ = 5.7
alternate layers of differently colored granular in. is also shcwn in Table 1. Values of o’ and
material. After a certain amount of flow out z4 for the data of Gardner are calculated from
of the orifice has occurred, the layers are ex- eqns. (9b) and (lo), using i, = 5.75 ins/*
from Fig. 5, and (I = 1.4 in2 2 0.2 in’, esti-
mated graphically from several layers of
* The only data given by Guinta” for which eqn. (ia)
can be tested are those on iron ore concentrate.
This material is also reported to have behaved in the
most cohesionless manner of those studied in the * For Guinta’s system R = 0.5 in. and for Gardner’s
investigation. system X = 0.125 in.
TABLE 1
Characteristics of experimental systems analyzed and resulting estimates of CYand t*
____I_
Aulhor Moterial PCC.?i~!t?size Dinaension- Estimate value Estimated z *
(ic) ality 0 f flow of (Y (in.) (in.)
--- -- -
Guinta Iron ore 0.05 (approx. median), 3 0.11 f 0.02 5.7
concentrate range from below
0.016 to 0.125
Gardner Durite: angular 0.0312 to 0.0625 2 0.01 -+ 0.002 3.5
calcined flint
&own and Sand less than 0.007 2 scale factor unknown
Hawksley

Gardnef’s Fig_ 10. The results are Q = 0.01 in. function) for the differential solid angle lying
* 0.002 in. and 2:s = 3.5 in., as recorded in at the angle 0 with respect to thez axis given by
Table 1. Values of CPand zzz could not be
calcuIated from the photographs of Brown P(B ) = * c0Pe, for e c ~12
and Hawksley, as the relevant dimensions
were not given. = 0, for 0 > 7i/2
If the orifices were of the ideal types dis-
cussed in Section 1, with a uniform 4, then representing an increasing upward bias as m
for z > z*, the point orifice approsimation takes on increasing positive values. Then it
should hold reasonably well (exit time error < can be shown from the definition (eqns. 2)
10%). Although most values of z, do exceed that
the corresponding z+, some values do not, as (r’ ) m + 2
shown by Tables 2 and 3. Especially for these cy = (12)
cases, and for the others as well, the closeness
of fit to the theoretical line seems to show According to eqn. (12), for a fixed upward
that the point orifice approximation is con- bias (fixed m), Q is proportional to po. where-
siderably better than z* would indicate, per- as for a fixed po, (Y decreases toward zero as
haps for the reason suggested in Section 1. the upwar:= bias (m) increases. Qualitatively,
Additional support of the theory is pro- one would expect: (1) pa to be the order of a
vided by the approximate agreement between few times the particle diameter d, taking
the two determined values of LYand the order larger values for more uniform particles and
of magnitude of the corresponding particle (2) (Y to therefore be the order of magnitude
sizes. To explore this point, consider a simple of d but to decrease for increasingly irregular
model in which all equivalent void jumps particles both because of an increasing m
(links of the flow chain of Fig. 1) are of (analogous to a steeper angle of repose) and a
length pot but hzve ap.d.f. (probability density
TABLE 3
Data from Gardner’s Fig. 10. VaIues of LO and r,
TABLE 2 were obtained by multiplying direct measurements on
photograph by inferred scale factor 2.4
Data from Guinta’s Fig. 7. Values of zO and z, were
obtained by multiplying direct measurements on Layer 312 3/Z
photograph by inferred scale factor 5.45
20 =m =0 “m
No. (in.) (in.)

Layer =o 2, z; +a 3 8.32 7.68 23.9 21.3

NO. (in.) (in.) 4 7.75 6.48 21.6 16.5
5 6.80 5.40 17.7 12.5
1 21.6 18.7 466 350 6 6.12 4.51 15.1 9.58
2 19.6 15.8 384 250 7 5.55 3.65 13.1 6.97
3 17.6 13.2 310 174 8 4.87 2.74 10.7 4.54
4 15.5 10.2 240 104 9 4.32 1.97 8.97 2.76
5 13.5 6.0 182 36 10 3.60 1.08 6.82 1.12
6 11.6 0 135 0 11 3.00 0 5.19 0
-
 35

TABLE 4
Data from Brown and Hawksley’s Figs. 21-23. Values of q, and z, were obtained by direct measurement of
photographs. Scale factor unknoxvn
_I_______---
Layer Fig. 21 Fig. 22 Fig. 23
No. ---____
=3/z 3*‘2 3“2 3 I 2 312
=0 =rn 203/2 m 20 =rn =0 = II, =0 ‘m =o Irn
(in.) (in.) (in.) (in.) (in.) (in.)
__-_______.-- -___
1 2.15 1.86 3.15 2.5-1 2.15 1.60 3.15 2.03
2 1.96 1.G2 2.74 2.06 1.96 I.-13 2.73 1.71
3 1.79 1.42 2.39 1.69 1.79 1.20 2.39 1.32
4 1.59 1.20 2.00 1.32 1.59 0.93 2.00 o.s9
5 1.40 0.98 1.65 0.97 1.40 0.62 1.65 0.49
6 1.20 0.68 1.32 0.56 1.20 ? 1.32
7 1.02 0.92 1.03 0.86 1.02 ?
8 0.82 0.70 0.74 0.58
9 0.62 0.44 0.49 0.29
10 0.43 0 0.28 0
-- -~- ---

decreasing ~0. These expectations are qualita- totjl number of particles has drained from the
tively borne out since the irregular Durite par- bed. A simplified qualitative description of
ticles of Gardner show CC- l/4 d whereas the this transient effect is obtained by supposing
rounder particles of Guinta show 01- 2d; eqn. that steady-state flow occurs in an expanding
(12) would, for example, approximately ac- vertically elongated region surmounting the
count for the former case with m = 2 and p o orifice defined by an advancing front or boun-
= 2d, and for the latter case with m = 1, p. = dary beyond which flow has not yet occurred_
10d. Of course, these values are merely illus- The boundary might be supposed to advance
trative and there are complications, especially as the normal component of flux at any point
in the latter case, due to the distribution of reduces the paticle densi* from the original
particle sizes. value to the threshold value assumed requisite
Although all layers (with a positive zm ) of for steady-state flow; stopping the efflux
Guinta’s Fig. ‘7 are plotted in Fig. 4, the first would stop the advance of the boundary. This
(top) two layers of Gardner’s Fig. 10 are picture would explain qualitatively the lag in
omitted from Table 3 and Fig. 5, and the first deformation of the upper layers of material as
six layers of Brown and Hawksley’s Fig. 21 well as the rather sharp break (boundary posi-
are omitted from Table 4 and Fig. 6. The rea- tion) between the deformed and undeformed
son for this is that a simple area check indi- parts of a given layer as seen in Fig. 3.
cates that the omitted layers had not de-
formed to nearly the same extent as the layers CONCLUSION
below; therefore the flow had not reached a
steady-state condition. In fact, for sufficiently Evidence presented for the stochastic or
small amounts of flow, a sufficiently distant kinematic theory of particle flow under grav-
layer (large zO) will not deform at all. Even ity consists of the encouraging fit of experi-
the points corresponding to Guinta’s top layer mental points for both two- and three-dimen-
and Gardner’s third layer are seen to be con- sional flow, ranging over an order of magni-
siderably above the theoretical line, indicating tude of particle size, to the straight lines of a
a larger than predicted +, , and hence less de- theoretically prescribed unit slope on the S2
formation than predicted. and 2’ plots respectively. Additional evidence
We conclude from the preceding observa- is the agreement of the values of CX,calculated
tions that for any given region of the bed from the plot intercepts, with the order of
there is a transient stage during which the par- magnitude of the particle size. Finally, the
ticle density falls to a certain level before qualitative shapes of the marker layers resem-
steady-state flow takes place. Furthermore, ble the predicted ideal contours shown in
observation suggests that the onset of flow Fig. 2, more-or-less depending on the system.
occurs for a given region only after a certain The failure of sufficiently distant layers to
36

deform to the extent predicted by the steady- was evaluated using Simpson’s rule and is
state theory for a fixed amount of flow has tabulated in Table 5 for 0 < u. < 2.0; for u.
been noted and tentatively described in terms > 2, the asymptotic approximation
of a transient effect requiring a reduction of
the particle density to a threshold value be- I,=2u; (1+-&J (A7)
fore steady-state flow can occur; this effect
might account for the discontinuities some- holds to better than 1%; for the point orifice
times seen in the deformation of a given layer. approximation, Ia = 2L1: _
The limits of validity of the point orifice For the two-dimensional case of a slit ori-
[Link] were discussed in a prelimi- fice (normal to the plane of flow) of width
nary way, particularly with reference to an 2X, the expression corresponding to eqn.
idealized finite orifice (Appendix A). (A2) for J, at a distance z above the slit cen-
The stochastic theory offers a number of ter is
additional predictions available for experi-
mental testing, particularly on steady-state (A81
flow patterns for various geometries.
Substituting this expression into eqn. (Al)
APPENDIXA and introducing the reduced quantities, $ =
X/a, C = z/a, and the variable u = 5/g2, we
The time T required for a particle marker obtain
at ze on the z axis to exit (reach the orifice) is
T(w,) =
given from the stochastic theory by
K212 (zd (A91

where
(AlI K = --an$2 _ 2nX3
(AlO)
aQ
2
9

where U, is the particle velocity in the z direc- where Q is the number of particles discharged
tion. For a circular orifice of radius R cen- per unit time per unit length of the slit ori-
tered on the z axis, with a discharge rate of q fice. The integral
particles per set per unit area, J, ( = -Vc) is
given by I*(uo) = Jo d”
o erf [ (4u)- ‘I21
J=(z) = -q[ 1 - exp(--R”/4olz)] (A2)
evaluated numerically, is also tabulated in
Substituting eqn. (A2) into (Al) we obtain
Table 5 for 0 < u. < 2.0; for u. > 2, the
dz asymptotic expression
T&z,) = n/q j” (A3)
o 1 - exp(R 2/4cyz)
Introducing the reduced quantities (referred T_4BLE 5
to the length (Y) p = R/a and c = z/e and the Exit-time integrals for two (T2) and three (13) dimen-
variable u = </p2, eqn. (A3) may be written as sions for idealized finite orifices

(A4) uo I2 (uo) 13 fuoj uo I2 (uol 13 (Uo)

T(uo) = K313 (~0)

0 0 0 1.1 1.645 3.051

(subscript denotes dimensionality) where 0.1 0.100 0.102 1.2 1.849 3.563
0.2 0.208 0.226 1.3 2.060 4.115
anp’_ ?ra3np4 _ rnR4
K3=_-_-_
(-45) 0.3 0.327 0.384 1.4 2.279 4.706
4 Q Qa 0.4 0.457 0.580 1.5 2.505 5.337
in which Q is the total discharge of particles 0.5 0.598 0.815 1.6 2.738 6.009
0.6 0.750 1.088 1.7 2.977 6.720
per unit time from the orifice. The integral 0.7 0.911 1.402 1.8 3.223 7.471
0.8 1.082 1.754 1.9 3.475 8.262
du
(A6) 0.9 1.261 2.147 2.0 3.733 9.093
o 1 --exp[-_(1/4u)l 1.0 1.449 2.579
 3’7

REFERENCES
(-412)
_
I W.W. Mullins, Stochastic theory of particle flow
again holds to better than 1%; for the point under gravity, J. Appl. Phys., 43 (2) (1972) 665.
orifice approximation, 2 P.G. Shewmon, Diffusion in Solids, McGraw-Hill,
New York, 1963, p_ 44.
3 3.M. Rausch, Ph.D. Thesis, Princeton Univ., 1949
1 = 2a112 312
3 3uo - (unpublished)_
4 J.S. Guinta, Flow of granular material in flat-
bottom bins, Trans. ASME, Ser. B, 91 (1969) 406.
ACKNOWLEDGEMENTS 5 G.C. Gardner, The region of flow when discharging
granular materials from bin-hopper systems, Chem.
Eng. Sci., 21 (1966) 261.
It is a pleasure to express appreciation to R.L. Brown and P.G.W. Hawksley, The internal
Ms. Mary Holmes Dague for assistance in flow of granular masses, Fuel Sci. Practice, 26 (6)
preparing the manuscript and to Mr. Robert (1947) 159.
Miller and Mr_ Tony Fraticelli for their assis-
tance in preparing the figures.