Chemical Engineering Science 129 (2015) 4957

Contents lists available at ScienceDirect

Chemical Engineering Science

journal homepage: www.elsevier.com/locate/ces

Finite element investigation of the ow and stress patterns in conical

hopper during discharge
Q.J. Zheng a,b, A.B. Yu a,n
a
Department of Chemical Engineering, Monash University, Clayton, Vic. 3800, Australia
b
Laboratory for Simulation and Modelling of Particulate Systems, School of Materials Science and Engineering, The University of New South Wales, Sydney,
NSW 2052, Australia

H I G H L I G H T S G R A P H I C A L A B S T R A C T

An Eulerian-formulation nite ele-

ment method is used to describe
granular ow in hopper.

This FEM outperforms the ordinary
Lagrangian computation in resolving
the mesh distortion problem.

Satisfactory results are obtained in
mass and funnel ow modes, mass
owrate and wall pressure.

art ic l e i nf o a b s t r a c t

Article history: Precise evaluation of the dynamics of granular material during hopper discharge, particularly the
Received 24 December 2014 velocity eld coupled with the stress eld, has been an important area of research for many years. In this
Received in revised form paper, a nite element method (FEM) based on the Eulerian formulation is described and validated to
10 February 2015
meet this need. It is demonstrated that this method outperforms the ordinary Lagrangian-formulation
Accepted 19 February 2015
Available online 27 February 2015
method in resolving the problem of mesh distortion, thereby is capable of simulating the complete
emptying process of a hopper. On this basis, various discharge behaviours of a conical hopper are
Keywords: studied, including the mass and funnel ow modes, the mass ow rate and the wall pressure. The results
Granular materials are in general agreement with those from experiments and recognized correlations in all the examined
Hopper
aspects, which validates the applicability of the Eulerian FEM. This continuum method ought to be of
Elastoplastic theory
practical signicance because it is computationally tractable, and viable in dealing with complex silo
Eulerian nite element method
Flow pattern geometries and variable ow patterns of granular materials.
Wall stress & 2015 Elsevier Ltd. All rights reserved.

1. Introduction
Flow of granular materials in hoppers is a common occurrence
in many industrial applications. This conned ow has some
unique ow and stress characteristics, which should be fully
understood in order to design a device satisfying operational and
n
Corresponding author. Tel.: 61 3 99050582; fax: 61 3 99055686.
E-mail address: Aibing.Yu@monash.edu (A.B. Yu).
safety requirements. It has been a subject of research over the past
century as seen from reviews (Nedderman et al., 1982; Tuzun et al.,
1982; Drescher, 1991; Nedderman, 1992; Roberts, 1998; Rotter,
2001).
A few salient representative studies had been done by Janssen
(1895), Beverloo et al. (1961) and Jenike (1961, 1964), which have
shed light on the general ow behaviours. However, how to
precisely evaluate the velocity and stress elds within a hopper
continues to present a challenge in practice. One reason for it is
that the classic theories have considered solely concentric

http://dx.doi.org/10.1016/j.ces.2015.02.022
0009-2509/& 2015 Elsevier Ltd. All rights reserved.
50 Q.J. Zheng, A.B. Yu / Chemical Engineering Science 129 (2015) 4957
geometries such as conical and wedge-shaped hoppers, where the
centerlines of the outlet and hopper coincide. But owing to some
particular needs, complex or eccentric hoppers with an off-
centered outlet are also being widely used in industrial applica-
tions (Ketterhagen and Hancock, 2010), which may have more
complicated ow and load behaviours (Ramrez et al., 2010).
Furthermore, the previous hopper theories were all derived under
certain assumptions, which may or may not hold depending on
the situation. For example, Ding et al. (2013) pointed out that the
pressure theories (Janssen, 1895; Walker, 1966; Walters, 1973) are
mainly applicable to steep hoppers because the effect of wall
friction in such hoppers is fully developed during the discharge as
assumed in the theory, whereas for shallow hoppers with only
partially mobilized wall friction, considerable emendations to the
theories are required.
In general, two types of numerical methods are used in the
modelling and simulating granular ow: discrete and continuum.
In continuum-typed methods, granular material is treated as a
continuum medium assuming that the system scale is always
much greater than component particles or grains. Its dynamics is
governed by conservation equations of mass, momentum, and
energy, with a constitutive model describing its intrinsic charac-
teristics as well as the initial and boundary conditions. The
granular ow is thus modelled at a macroscopic or global scale.
Discrete methods, in particular the so-called discrete element
method (DEM), can on the other hand achieve a particle-scale
simulation of particulate system. DEM simulates the motion of
each individual particle in a system based on the second New-
tonian law and some practical contact or non-contact force models
(Zhu et al., 2007, 2008). Unlike common theoretical treatments,
DEM does not enforce assumptions on constitutive relationships
or hopper geometry and can provide important microdynamic
information such as the trajectories of and transient forces acting
on individual particle. Consequently, it has been increasingly
employed in recent decades to explore hopper behaviours, e.g.
the mass discharge rate (Zhu and Yu, 2004; Anand et al., 2008),
particle ow patterns (Ketterhagen et al., 2009) and internal bulk
stresses (Langston et al., 1995; Rotter et al., 1998; Zhu and Yu,
2002). Nevertheless, it is known that to a large degree, DEM
simulation can only deal with small-scale systems because it is
computationally very demanding. How to overcome this problem
remains a challenge, although some measures such as the multi-
domain continuum and discrete methods (Parisi et al., 2004;
Rousseau et al., 2009; Nitka et al., 2011) are proposed to mitigate
this computational tension.
As a continuum approach, Finite Element Method (FEM) also
prevailed in previous hopper investigations. FEM associated with
elastoplastic constitutive theories can give satisfactory predictions
of the internal stress and wall pressure in hoppers (Ooi and Rotter,
1990; Rotter et al., 1998; Goodey et al., 2003, 2006; Goodey and
Brown, 2004; Vidal et al., 2006, 2008; Ding et al., 2013). However,
its application to reproduce the ow conditions of granular
material is usually not easy (Rotter et al., 1998; Sanad et al.,
2001; Tejchman and Klisiski, 2001) and requires some particular
schemes or treatments such as the re-meshing and re-zoning
scheme (Sanad et al., 2001), adaptive meshing technique (Yang
et al., 2011), viscoplastic granular uid models (Haussler and Eibl,
1984; Karlsson et al., 1998; Elaskar et al., 2000; Bhrnsen et al.,
2004), smoothed particle hydrodynamics (SPH) method (Sugino
and Yuu, 2002), and material point method (MPM) (Wieckowski
et al., 1999).
In our recent study (Zheng and Yu, 2014), an Eulerian-
formulation FEM was implemented in the analysis of a represen-
tative problem, i.e. stress dip beneath sandpile. This approach can
describe the dual solid- and uid-like behaviours of granular
material and satisfactorily reproduce the critical stress dip
phenomenon under various conditions. It not only explains the
mechanism of dip but may also open up a new direction to
describe granular materials in nature and many industrial pro-
cesses. The goal of this paper is to extend its application to a
common industrial process, i.e. hopper discharge; and to compare
the FEM results of granular dynamics with the state-of-art hopper
theories and experimental measurements. The paper is organized
as follows. First, an elaborate description of the computational
methods as well as the constitutive relationship and boundary
conditions used in simulations are given in Section 2. The simula-
tion results of mass ow rate, ow pattern and wall stress are
presented and discussed in Section 3. Finally, the main conclusions
of this work are given in Section 4.
2. Set up for numerical simulation
2.1. MohrCoulomb elastoplastic relationship
It is known that the robustness of a continuum approach
depends largely on the choice of the constitutive model used for
describing material behaviours. To date, a number of such con-
stitutive models have been proposed for granular materials in the
literature, ranging from the classic plastic theories (Coulomb, 1776;
Drucker and Prager, 1952; Lade, 1977) to more recent polar
elastoplastic theory (Tejchman and Wu, 1993), hypoplasticity
theory (Wu et al., 1996), double shearing theory (Spencer and
Hill, 2001) and others. But at the bulk scale where micromechanics
such as shear banding is not concerned, the classic elastoplastic
models are believed to be adequate for capturing stress character-
istics of granular materials (Goodey et al., 2003, 2006; Zheng and
Yu, 2014). Hence, this paper utilizes the traditional MohrCoulomb
elastoplastic model to characterize the mechanical behaviours of
particulate materials stored in a hopper. The model mainly
consists of three parts: a linear isotropic elastic law to dene the
material behaviour at small loads, a yield criterion to determine
the transition of material into yielding, and a non-associated
plastic ow potential to determine the ow directions after
material yielding. Exact expression of this MohrCoulomb elasto-
plastic model can be found in the literature (Abaqus 6.10, 2010)
and is briey described below.
This elastic behaviour is completely determined by Young's
modulus E and Poissons ratio in isotropic elastic law:
ij Del el
ijkl kl 1
el
where ij is the total stress; el
kl is the elastic strain; and Dijkl is the
fourth-order tensor of elasticity.
The yield condition is given by
Rmc q  p tan  c 0 2
where
1   1  
Rmc p sin cos tan 3
3 cos 3 3 3
 
p  13 trace ij is the rst q
invariant of stress representing the
equivalent pressure; q 32 Sij Sij is the Mises equivalent stress
and Sij is deviatoric stress; and and c are known as the angle of
internal friction and the cohesion of granular material. is the
 3
deviatoric polar angle and is dened by cos 3 9 r=q where
1=3 r
is an invariant measure of deviatoric stress r 2 Sij Sjk Ski .
The ow potential G is chosen to be a hyperbolic function in the
meridional stress plane and a smooth elliptic function in the
deviatoric stress plane, dened mathematically by the following
formula:
q
G cj0 tan 2 Rmw q2 p tan 4
 Q.J. Zheng, A.B. Yu / Chemical Engineering Science 129 (2015) 4957 51

where discharging behaviours. Given the axial symmetry of the silo

  geometry, only a quarter of the silo is modelled in the simulations
4 1  e2 cos 2 2e 12 3  sin
Rmw   q
  for computational efciency. In doing so, each of the side surfaces
2 1  e cos 2e 1 4 1  e cos 5e  4e 6 cos
2 2 2 2 should be assigned a plane symmetry condition as shown in Fig. 1.
5 In the following sections, however, a half silo may be presented to
give a more illustrative view, which is created by a mirror image
where is the dilatancy angle of material; cj0 is the initial of the existing quarter model. Note that in practice, the ow in a
cohesion yield stress; and is a parameter characterizing the discharging hopper may be asymmetrical due to some transient
eccentricity of ow potential, with 0.1 in all the present effects, or the asymmetric arrangement or operation. In this case,
simulations. Note that e depends on internal friction angle, given the above symmetric treatment is not valid. However, for the
by e 3  sin =3 sin . common cylindrical silo/conical hopper such as the one consid-
With the use of a non-associated ow potential, the volume ered here, the symmetrical model is applicable and can effectively
dilatancy of granular material in the process of shear ow can be reduce the computational cost.
controlled through the parameter of dilatancy angle . In this The simulation of the discharge process is executed with the
study, is set to 51 in all the simulations, because dilatancy is Eulerian analysis module available in the commercial package
usually small for common granular materials. Only cohesionless Abaqus (Version 6.10). In the simulation, silo wall is modelled as a
materials are considered in this work, i.e. cohesion c 0. The other rigid shell (element type R3D4) by assuming that its thickness and
material properties are given in Table 1. deection under the lateral pressure of bulk material are all
negligible. Its translational and rotational freedoms of motion are
2.2. Discretization mesh and boundary conditions fully constrained throughout the simulation. The bulk material
contained in the silo is modelled using three-dimensional Eulerian
The conguration of the silo considered is shown in Fig. 1. It elements (EC3D8R) integrated with the default hourglass control.
consists of a cylindrical bunker of diameter D at the top, and a Due to the Eulerian characteristic, the procedures and bound-
conical hopper of angle at the bottom, with an orice of diameter ary conditions used for modelling the hopper discharge are
D0 and a total height H. The values for these geometrical para- different from those used in the ordinary Lagrangian FEM simula-
meters are listed in Table 1. Besides a basic case, a range of tions. The Eulerian FEM approach does not impose specic
parametric values are considered to explore their inuences on the requirements on the shape or dimensions of the computational
space (discretized as mesh), as far as it can cover the regions of
Table 1 interest. To facilitate data collection on the material stress and
Parameters used in the current FEM simulations. velocity near the silo wall, the computational mesh is here given
the same shape as the silos and a slight increase in the mesh
Parameters Basic values Ranges of variation
width to surround the silo wall. Fig. 1 displays the mesh used in
Silo height, H (m) 2.48 simulations of basic silo case. For other silo geometries (or cases),
Silo diameter, D (m) 1.0 the FEM mesh needs to be varied accordingly. Under default
Orice diameter, D0 (m) 0.1 0.050.5 circumstances, no materials exist in Eulerian elements. In other
Half angle of hopper, (deg) 30 1590
words, the elements are occupied by voids (note that although
Bulk density, (kg/m3) 856
Youngs modulus, E (MPa) 10 voids have no physical properties, they are treated as material in
Poissons ratio, 0.3 Eulerian FEM scheme). Therefore, prior to the start of a simulation,
Internal friction angle, (deg) 30 3050 it is essential to assign granular materials to these Eulerian
Dilatancy angle, (deg) 5 elements as an initial condition, which is analogous to the lling
Cohesion, c (Pa) 0
Wall friction, 0.3 00.84
process of a practical silo. Moreover, as illustrated in Fig. 1, two
Eulerian-type boundaries should be created at the top and at the
outlet of the hopper mesh to allow granular material to ow freely
in and out of the computational space. Voids entering from the top
surface, slowly take the place of granular material during the silo
discharge. The materials moving out of the outlet will not be
considered in subsequent calculation steps because they are out-
side the computational domain.
To summarize, the procedure to simulate silo discharge consists of
four major steps: (i) creating the silo wall and Eulerian mesh using the
parameters desired; (ii) feeding granular material into the mesh up to
an expected level; (iii) applying boundary conditions and gravity as
described above; and (iv) nally, executing the simulation.

2.3. Numerical method and a primary verication

Most FEM approaches are formulated in the Lagrangian frame of

Fig. 1. Conguration of the silo under consideration (left) and the discretization
mesh used in the FEM simulation (right). An enlarged view of the hopper section
can be found in Fig. 2.
reference in which the computational mesh is essentially combined
with the solid material and undergoes deformation as the material
moves. Such Lagrangian FEMs are good at computing the deformation
gradient of solids but may encounter severe distortions and entangle-
ments of the mesh when remarkable material deformation (or
granular ow) occurs. The commercial package Abaqus (Version
6.10) provides an Eulerian method which helps resolve such a mesh
distortion problem. This method is essentially a Lagrangian plus
remap scheme, calculating the dynamics of a material in two sub-
52 Q.J. Zheng, A.B. Yu / Chemical Engineering Science 129 (2015) 4957
steps: a Lagrangian-like step and a convection step. In the former, the
governing equations of granular ow are solved in a similar way to the
ordinary Lagrangian FEM, which allows the FEM mesh deform
temporarily to a new conguration. Thereafter a convection algorithm
with the second-order accuracy is used to remap all state variables of
material from the new mesh conguration back to the original mesh.
In this way, the computational mesh can always maintain the original
conguration and granular material to ow through the mesh
effectively like a uid.
Therefore, unlike that in the traditional Lagrangian FEM, the
computational mesh in the Eulerian FEM does not stand for granular
material. To determine the volume and positions of materials in the
mesh, an additional variable named as Eulerian Volume Fraction (EVF)
is dened, which equals unity in elements occupied by granular
material and zero in void elements. Based on the values of EVF, the
interface between granular materials and voids can be identied and
visualized in the post-processor. The contact between granular mate-
rial and wall is dealt with by an enhanced immersed boundary
method (Abaqus 6.10, 2010), where setting up the contact requires
the silo wall placed inside the Eulerian mesh; otherwise, the contact
between them cannot be sensed by the FEM solver. The friction at the
silo wall is assumed to obey the AmontonsCoulomb law as com-
monly done in the literature. The adopted solver is of explicit type and
takes into consideration the inertial effect of material. In default
circumstances, all Eulerian elements are void (EVF 0) and need to
be lled with granular material (EVF1). It is done by creating an
initial eld of material assignment in the model with the aid of a
volume fraction tool in the pre-processor of Abaqus.
Fig. 2. Snapshots of the discharge process simulated with the current Eulerian FEM
(left) and ordinary Lagrangian FEM (right). Colours indicate the magnitude of
vertical particle velocity. Only parts of mesh close to the outlet are shown for
clarity.
Fig. 3. Temporal responses of particle velocity in vertical direction Vy at the centre
of orice.
As a primary test, the Eulerian FEM approach is used to simulate
the basic silo discharge case described in Table 1. To make a
comparison, the same exercise is also carried out with the ordinary
Lagrangian FEM (Abaqus 6.10/Explicit), details of which can be found
elsewhere (Tejchman and Klisiski, 2001; Goodey et al., 2003, 2006;
Yang et al., 2011). The difference between the two methods in
simulation of discharge can be clearly seen from Fig. 2. As observed
by Tejchman and Klisiski (2001) and other researchers, Lagrangian
FEM leads to increasing distortions of the computational mesh with
the continuous outow of material in the vicinity of outlet, which is a
distinct contrast to the present Eulerian FEM.
The distortions of the mesh can cause serious problems in the
simulation of a discharge process. Fig. 3 shows the temporal response
of particle velocity Vy at the centre of outlet. It can be observed that, at
the very beginning of discharge, there is an almost perfect coincidence
of velocity curves obtained by the Eulerian FEM and Lagrangian FEM,
implying that the distortion of the Lagrangian meshes is not signi-
cant, and both methods are valid. However, with the continuation of
the material ow, the two velocity curves begin to depart from each
other. The particle velocity predicted by the Eulerian FEM gradually
approaches a plateau indicating a steady discharge rate; while the one
by the Lagrangian FEM experiences a wavy and lasting decrease until
the end of simulation (further simulation is rather difcult because the
mesh distortion signicantly reduces the stable size of the time-step).
The latter response by the Lagrangian FEM is obviously unphysical
because for the considered granular material (c 0) and hopper
opening (D0 0.1 m), in no way can granular ow be choked by
cohesive arching or jamming effect. It should therefore be attributed to
the loss of numerical reliability due to the over-distorted mesh. The
Eulerian FEM, on the other hand, results in a more reasonable velocity
response and is more desirable for studying the discharge behaviour.
3. Results and discussion
3.1. Velocity prole
The velocity eld in a hopper is a fundamental indicator of
discharge dynamics and also represents a basic benchmark test for
evaluating the validity of a continuum model. Fig. 4 shows the
contour and vector plots of a typical velocity eld in a discharging
hopper, which is obtained at 10 s of discharge when the ow of
particulate material becomes steady. It is obvious that in most
parts of the conical hopper, the granular ow approximately
follows a radial ow pattern, where all velocity vectors point to
the imaginary vertex of the hopper along radial branches r. Despite
 Q.J. Zheng, A.B. Yu / Chemical Engineering Science 129 (2015) 4957 53

Fig. 4. Particle ow velocity in the conical hopper: contour plot (left) and vector
plot (right). A polar coordinate system (r, ) is dened to aid the description of
velocity prole with r being measured from the imaginary apex of hopper.

some degree of departure close to the hopper outlet, such a

pattern of hopper ow is consistent with the experimental
observations (Tuzun and Nedderman, 1985a, 1985b; Nedderman,
1988; Cleaver and Nedderman, 1993a; Xie and Shinohara, 1999;
Chen et al., 2005; Choi et al., 2005; Steingart and Evans, 2005).
A quantitative verication of the resulting velocity eld is made
against the results of a previous experiment (Cleaver and
Nedderman, 1993a). Note that the particle velocities in the
experiment have not been measured directly; rather, they have
been obtained as the average speeds of some marker particles
traveling from the top to the outlet of hopper. A similar averaging
procedure is performed on the FEM outcome in order to compare
the experimental results. Imagine a marker particle is initially
placed at the top of the hopper (radial coordinate r1). Following
the ow eld V r; in the hopper, it will reach the outlet (radial
coordinate r0) in a period of t (the residence time). On average, the
velocity V~ of the marker particle over the residence time can be
expressed as
Z r1 Fig. 5. Comparison of the velocity proles between the present FEM and the
dr
V~ r 1 r 0
previous experiment (Cleaver and Nedderman, 1993a) in a hopper with: (a) wall
6
r 0 V r;
friction tan(15.61), and (b) tan(28.51). The abscissa is the angular coordinate
as dened in Fig. 4. The ordinate denotes the normalized velocity V~ ()/V~ (0) as
Note that the average velocity V~ is a function of angular described in the text.
coordinate only. It is further normalized by the average velocity
at the centerline 0, i.e. V~ =V~ 0, as done in the experiment. analysed the discharging velocity in a conical hopper by treating the
Because of the above averaging and normalizing treatments, bulk materials as a MohrCoulomb-type plastic solid. A good agree-
although the present FEM simulation considers hoppers of differ- ment of their results with the experimental measurements has also
ent sizes from the experiment, the nal velocity proles of the been observed. Yet in that work, the elastic compressibility of the bulk
hoppers in the dimensionless form remain to be comparable. material was neglected and an assumption of radial stress eld in the
Fig. 5 compares the predicted velocity prole V~ =V~ 0 across the hopper needed to be used in their analysis. In contrast, the current
hopper with the solid velocities measured experimentally (Cleaver and FEM simulation does not involve such an assumption and conrms
Nedderman, 1993a). In accordance with the experiment, two specic more convincingly that the traditional elastoplastic models are effec-
cases of wall friction are under consideration in the numerical tive to describe the behaviours of granular ow in hoppers, provided
modelling, i.e. tan(15.61) and tan(28.51), which are laid out in that an appropriate simulation method is used.
Fig. 5a and b, respectively. The internal friction angle of the tested
bulk material was not precisely determined in the experiment. 3.2. Flow mode
Approximately, it should fall in the range between 30.41 and 37.11
(Cleaver and Nedderman, 1993a). As much, here, three values of are The ow pattern of particles within a hopper is one of the
attempted in simulations including the upper and lower limits of major concerns in industries because it is not only directly related
estimation as well as an intermediate value 351. As is evident from to the service performance and repeatability but also considerably
Fig. 5, the velocity proles created by 30.41 and 37.11 well affects the integrity of hopper structure. Normally, two particle
encompasses the experimentally measured data in both considered ow patterns are likely to occur in a hopper: (i) mass ow, where
cases. An even closer agreement is achieved when internal friction all particles move simultaneously upon opening the hopper
angle 351. This comparison suggests that, within the condence orice, approximately conforming with a rst in rst out princi-
level of the experiment (Cleaver and Nedderman, 1993a), the FEM ple (Anand et al., 2008); and (ii) funnel ow, in which only a
simulation can be deemed as valid in predicting the ow eld of a portion of particles can be mobilized during discharge while
discharging hopper. Previously, Cleaver and Nedderman (1993b) also others form stagnant zones. In practical applications, it is crucial
54 Q.J. Zheng, A.B. Yu / Chemical Engineering Science 129 (2015) 4957
Fig. 6. A typical chart of particle ow modes in conical hoppers. Dashed line
represents the criterion proposed by Jenike (1961, 1964). The inset snapshots,
which are taken from FEM simulations at 20 s after hopper outlet opened, show the
movements of different material layers during discharging process (colours have no
special meaning other than for distinguishing particles at different layers). Para-
meters used for these insets are respectively: top, 301, tan() 0.7; middle, ow modes of bulk materials in various conditions can be inter-
tan() 0.3 and from left to right, 151, 301, 451; bottom, 301, tan() preted from the coloured solid layers shown in Fig. 6. Clearly, just
0.1. (For interpretation of the references to colour in this gure legend, the reader is
referred to the web version of this article.)
Fig. 7. Contours of MFI over the domain of hopper angles and wall friction angles
for granular material with internal friction angle 301. The red line of MFI 0.3
represents the mass/funnel transition criterion obtained by FEM. The black line is
the criterion proposed by Jenike (1961). (For interpretation of the references to
colour in this gure legend, the reader is referred to the web version of this article.)
to forecast under what condition each ow pattern will occur, so
that engineers can modulate the ow conditions by adjusting
hopper parameters according to the specic needs of an operation.
Previously the particle ow patterns in conical hoppers have been
theoretically explored by Jenike (1961) and Drescher (1992). More
recently, this topic was also probed by Ketterhagen et al. (2009)
based on DEM simulations of a rectangular hopper. In this section,
the mass and funnel ow behaviours of granular material in a
conical hopper is studied based on continuum mechanics and FEM
simulation, and compared with the widely-accepted design
charts (Jenike, 1961, 1964).
Fig. 6 shows a prototypical criterion proposed by Jenike (1961,
1964) for predicting the transition between mass ow and funnel
ow in a 3D conical hopper. It is suggested from the graph that the
appearance of mass ow or funnel ow largely depends on the tilt
angle of hopper and the friction of particle against hopper wall
(tan()). Smaller (steep hoppers) and wall friction tend to
create a mass ow pattern in hopper, whereas greater hopper
angle (shallow hoppers) and wall friction , usually lead to a
Fig. 8. Contours of MFI for granular materials with internal friction angle 401.
Indications of lines are the same as those in Fig. 7.
funnel ow pattern in the hopper. To demonstrate the movements
of solids in the FEM simulations, the initial silo bed is partitioned
into a number of layers using different colours. In this way, the
as suggested by the so-called Jenike chart, the hopper angle and
wall friction play a dominant role in determining the ow mode
increasing either the wall friction at a xed hopper angle, or the
hopper angle at a constant wall friction can both aggravate the
unevenness of ow conditions in the hopper and effect the
transformation from mass to funnel ow mode.
To make a quantitative comparison with the Jenike criterion, a
Mass Flow Index (MFI) (Johanson and Jenike, 1962; Ketterhagen
et al., 2009) is used to quantify the ow patterns obtained by FEM.
MFI is dened as a ratio of the mean velocity at the wall ( ) to
the mean velocity at hopper centerline (0):
V
MFI 7
V 0
where V denotes the statistically mean value of the resultant
velocity along the wall or centreline of hopper (Ketterhagen et al.,
2009). Normally, 0.0 oMFI o1.0 because the particles near the
wall always move slower than those at the centreline. Previously,
it has been recognized that MFI 0.3 is practically a critical point
to judge the ow modes in a hopper, with MFI 4 0.3 being a mass
ow and MFIo 0.3 being a funnel ow (Johanson and Jenike, 1962;
Ketterhagen et al., 2009). Therefore, by evaluating the MFI, various
ow modes observed in the FEM simulations can be identied and
a mass/funnel ow transition criterion can be established based on
the FEM results. Figs. 79 show contour plots of MFI values with
regard to hopper angle and wall friction for granular materials
with different internal friction angles. It can be seen that the
predicted line of MFI 0.3 can match the Jenike mass/funnel
transition criterion in the case of internal friction angle 301
a parametric value corresponding to the most common granular
materials. In cases of higher (Figs. 8 and 9), however, the FEM
results show considerable discrepancies from the Jenike chart. We
note that such discrepancies are in fact reasonable because the
Jenike chart was known to be slightly conservative in estimating
the mass ow regime. A similar trend has been observed by
Ketterhagen et al. (2009) in their DEM simulations.
3.3. Mass ow rate
The mass of particles that are released from the hopper per unit
time, W, depends on numerous parameters such as orice diameter
D0, hopper angle , wall friction , and internal friction angle
(Nedderman et al., 1982). This study does not intend to be a full
 Q.J. Zheng, A.B. Yu / Chemical Engineering Science 129 (2015) 4957
Fig. 9. Contours of MFI for granular materials with internal friction angle 501.
Indications of lines are the same as those in Fig. 7.
investigation of the mass ow rate regarding these variables but
rather be a basic validation of the proposed computational method.
Thus, only the mass discharge rate W from a at-bottomed cylindrical
silo ( 901) is focused in this section, whose correlation is estab-
lished by Beverloo et al. (1961):
p considered the shape factor of conical hopper and provided a
W C g D0  z5=2 8
where C is an empirical constant within 0.550.64 depending on wall
friction, and it typically takes a value of 0.58. z is a microscopic
quantity correlated to the particle size and shape and commonly
interpreted with the concept of empty annulus (Brown and Richards,
1965; Nedderman, 1992). The acceleration due to gravity g 9.8 m/s2,
and bulk density is given in Table 1.
The FEM results of the mass ow rate W are plotted in Fig. 10 as
a function of the outlet diameter D0. Here, W is obtained through a
product of bulk density and the area integration of vertical
 
particle velocity Vy across the hopper outlet. The Beverloo equa-
tion given by Eq. (8) is also added to the gure, whose coefcient C
is treated as an adjustable parameter in order to t the FEM
results. z is xed at zero because FEM is based on the continuum
assumption of granular material and thus should not contain any
microscopic lengths. Moreover, for large-scale hoppers with outlet
D0z, neglecting z in the Beverloo equation will not cause much
error in the discharge rate. Evidently from the comparison, FEM
predictions comply with the known Beverloo correlation, that is,
the mass ow rate W is proportional to the outlet diameter D0
raised to the power of 52 . Moreover, we see through the tting that
the FEM results correspond to roughly a value of C 0.59, which is
quite close to the commonly agreed value of 0.58 for a cylindrical
silo. This agreement imparts additional condence to the quanti-
tative credibility of FEM simulations for evaluating the mass
discharge rate. Note that if checked in more details, the accuracy
of FEM predictions could also be inuenced by the choices of
constitutive models and material parameters such as dilatancy
angle, which may be part of future work.
3.4. Wall pressure
The lateral pressure acting on silo wall has an intensively
studied topic since the time of Janssen (1895) who derived the
rst algebraic equation for describing wall pressure by analysing
the equilibrium of a differential slice element in a cylindrical silo. A
known characteristic of Janssens equation is that the wall pressure
may become saturated with the depth in a tall slender silo due to
the upward resistance of friction at the wall. Walker (1966)
improved the original differential slice method of Janssen and
extended its application to conical hoppers. Walters (1973) further
55
Fig. 10. Mass ow rate W as a function of outlet diameter D0 in a at-bottomed
hopper ( 901): points, present FEM results; line, the least-square t of FEM
results using the Beverloo relation (Eq. (8)) when C0.59 and z 0. The micro-
scopic length z is xed at zero in data tting because FEM results are obtained
based on the continuum assumption. In addition, zD0 in large-scale hoppers and
has little inuence on the discharge rate.
solution to account for the sharp switch stress occurring at the
transition section in silos. Nedderman (1992) reviewed these
hopper stress solutions and summarized the following procedures
to calculate the pressure distribution in a silo:
(i) Under assumptions that the frictional traction at the wall is
fully developed during discharge stage and granular material
is at an active Rankine state, the wall pressure in the bunker
section of the silo can be expressed as (Nedderman, 1992)
gD 4K A
w 1  exp  y 9
4 D
where y is the depth measured downward from the top
surface of stored material, is the coefcient of wall friction,
and KA is the so-called lateral stress ratio of active Rankine
state related to wall friction, internal friction angle of granular
material and some distribution factor.
(ii) In the lower hopper, where a passive Rankine state is assumed
for the stored granular material, the wall pressure has the
following form:
(  "  m  1 #)
m
h gh h
w Cp Q 0 1 10
h0 m1 h0
where h is the material height measured upward from the
imaginary vertex of the conical hopper, h0 is the height of
hopper, Q0 is the surcharge pressure from solid material in
upper bunker, and Cp and m are parameters connected to
hopper angle , wall friction and internal friction angle
(Nedderman, 1992).
A recent experimental measurement of silo pressure carried
out by Couto et al. (2013) is chosen to compare with the above
theoretical correlations and our FEM simulation. The model silo
tested was similar to the conguration shown in Fig. 1 but had a
different hopper angle 34.31 and a larger outlet D0 0.35 m.
The material under test was common wheat, with internal friction
angle 30.171, Youngs modulus E 10,674.33 kPa and mean bulk
density 856 kg/m3. The coefcient of friction between wheat
and silo wall was estimated as 0.2. Strictly speaking, there is a
certain amount of cohesive interaction between contacting grains
due to the existence of humidity. But a precise evaluation of this
56 Q.J. Zheng, A.B. Yu / Chemical Engineering Science 129 (2015) 4957
Fig. 11. Comparison of silo pressure between experiment (Couto et al., 2013),
current FEM simulation at discharge time of 0.5 s, and theoretical correlations, Eqs.
(9) and (10). The graph on the left shows contour of the horizontal component of
normal stress xx in discharging silo.
effect was not provided in the experiment, thus c is assumed to be
zero in the FEM model. Other parameters involved use the values
given in Table 1.
Fig. 11 (left) shows the simulated pattern of internal stress
(horizontal component xx) in the discharging silo. It can be
observed that the stress distribution is divided into two regimes.
In the bin, as the frictional traction from lateral wall cannot
balance the gravity, the internal stress of bulk material accumu-
lates in terms of depth. In the hopper, however, because the
inclined wall can provide sufcient support to the material above,
the internal stress is gradually released towards hopper outlet. A
transition zone exits between the two regimes in the form of an
arch, and has the maximum stress values. In Walkers analysis,
the transition zone was simply assumed to be a plane surface
separating the active (bin) and passive (hopper) regimes, which is
obviously not as realistic as the arch in the FEM modelling and
causes some inaccuracy of Eqs. (9) and (10) on silo pressure at the
height of 0.50.75 m above the silo transition, as shown in Fig. 11
(right).
The silo pressure w can be simply considered to be the
horizontal component of bulk stress xx (variable SVAVG11 in
Abaqus output) in the bin section where silo wall is vertically
placed. But in the hopper, due to the tilt of hopper wall, w should
be determined through a coordinates transformation of the stress
state xx, yy (SVAVG22) and xy (SVAVG12). Fig. 11 (right) plots the
silo pressure prole obtained from the FEM simulation at time of
discharge 0.5 s when all particles across the silo start owing, i.e.
ow eld forms, but the total lling height does not change yet
that is the maximum height. The agreement between FEM and
experimental results is generally satisfactory, particularly consid-
ering the uncertainty associated with the parameter selection in
FEM simulations. Note that FEM appears to give an accurate
estimation of the pressure peak at the silo transition, which is an
important quantity affecting silo failures.
The hopper pressure theories, i.e. Eqs. (9) and (10), are also
compared in Fig. 11. The coefcients involved in the equations are
determined based on the above-mentioned silo and material
parameters according to Nedderman (1992): lateral stress ratio
KA 0.338, hopper height h0 0.738 m, surcharge Q0 13.0 kPa and
Cp 2.23. Regarding the calculation of coefcient m, different
formula are available (Nedderman, 1992) and can lead to different
values. The largest value is chosen here, m 3.73. As seen from
Fig. 11, the prediction of Eqs. (9) and (10) also match the
experimental measurement approximately but its overall accuracy
is not as good as the FEM prediction.
Another notable discrepancy between the theoretical correla-
tion and FEM simulation is that the Nedderman approach assumes
a fully-developed frictional traction throughout silo wall so that
the resulting pressure curve is always smooth; in contrast, FEM
considers the non-uniform development of frictional traction at
different positions of wall and thus can represent the oscillations
of wall pressure. Moreover, the FEM simulations can provide
detailed internal velocity and stress results, which are not avail-
able from the Nedderman approach.
It should be noted that wall pressure of silos is signicantly
inuenced by the particle ow patterns inside (Savage, 1965;
Tuzun and Nedderman, 1985b; Langston et al., 1995). Previous
pressure theories have been established based on the assumption
of mass ow. However, in practice, it is the condition of funnel
ow which causes problems and deserves most research atten-
tions (Ooi et al., 1998). While such situations are difcult to handle
with theories, they can be readily accounted for by the current
FEM simulation as discussed in Section 3.2.
4. Conclusions
This work evaluates the applicability of our recent continuum
approach (Zheng and Yu, 2014) for modelling the granular
dynamics in discharging hoppers. This method is based on the
Eulerian-formulation nite element method and treats granular
material as an elastoplastic solid described by the traditional
MohrCoulomb model. Various discharge conditions have been
examined, leading to the following conclusions:
1) The proposed Eulerian FEM overcomes the previous problem
associated with distorted mesh and is therefore applicable to
modelling the complete process of hopper discharge. In contrast
to the conventional Lagrangian FEM which often leads to an
unphysical clogging of material, the Eulerian FEM gives rise to a
continuous and steady outow of granular material through the
small orice, as expected in a practical course of hopper discharge.
2) The MohrCoulomb model, as a traditional constitutive
description of dense granular material, quantitatively captures
the ow and stress behaviours of granular material in a hopper.
Both mass ow and funnel ow regimes can be reproduced by
the FEM and found to be satisfactorily comparable to the Jenike
design chart. Consistent with the known Beverloo correlation,
the mass discharge rate obtained in the FEM simulation also
varies with the orice diameter raised to the power of 52 . The
constant C corresponding to the FEM results is found to be
roughly 0.59, which is very close to the empirical value of 0.58
in a at-bottomed cylindrical silo. The FEM prediction of the
wall pressure is also satisfactory in comparison with an
experimental measurement on a laboratory-scale silo.
3) It is argued that the previous difculties in continuum research
of hopper discharge are more likely induced by the simulation
method rather than by constitutive models of granular materi-
als. Although the existing constitutive models may not be
perfect when examined in detail, they already work for most
industrial applications where the micromechanics of material
is less important. The proposed FEM approach can play a
signicant part in practice because it is computationally ef-
cient and not constrained by the silo geometry or discharge
method. It can serve as a real-time simulator to aid the design
and assessment of silos or hoppers in industrial applications.
Acknowledgement
The authors are grateful to the Australian Research Council for
the nancial support of this work.
 Q.J. Zheng, A.B. Yu / Chemical Engineering Science 129 (2015) 4957
References
Abaqus 6.10, 2010. Analysis User Manual. Dassault Systmes Simulia Corp.,
Providence, RI.
Anand, A., Curtis, J.S., Wassgren, C.R., Hancock, B.C., Ketterhagen, W.R., 2008.
Predicting discharge dynamics from a rectangular hopper using the discrete
element method (DEM). Chem. Eng. Sci. 63, 58215830.
Beverloo, W.A., Leniger, H.A., van de Velde, J., 1961. The ow of granular solids
through orices. Chem. Eng. Sci. 15, 260269.
Bhrnsen, J.U., Antes, H., Ostendorf, M., Schwedes, J., 2004. Silo discharge:
measurement and simulation of dynamic behavior in bulk solids. Chem. Eng.
Technol. 27, 7176.
Brown, R.L., Richards, J.C., 1965. Kinematics of the ow of dry powders and bulk
solids. Rheol. Acta 4, 153165.
Chen, J.F., Rotter, J.M., Ooi, J.Y., Zhong, Z., 2005. Flow pattern measurement in a full
scale silo containing iron ore. Chem. Eng. Sci. 60, 30293041.
Choi, J., Kudrolli, A., Bazant, M.Z., 2005. Velocity prole of granular ows inside silos
and hoppers. J. Phys.: Condens. Matter 17, S2533.
Cleaver, J.A.S., Nedderman, R.M., 1993a. Measurement of velocity proles in conical
hoppers. Chem. Eng. Sci. 48, 37033712.
Cleaver, J.A.S., Nedderman, R.M., 1993b. Theoretical prediction of stress and velocity
proles in conical hoppers. Chem. Eng. Sci. 48, 36933702.
Coulomb, C.A., 1776. Essai sur une application des rgles de maximis & minimis
quelques problmes de statique, relatifs larchitecture. Mem. Math. Phys.,
pp. 343381.
Couto, A., Ruiz, A., Aguado, P.J., 2013. Experimental study of the pressures exerted
by wheat stored in slender cylindrical silos, varying the ow rate of material
during discharge. Comparison with Eurocode 1 part 4. Powder Technol. 237,
450467.
Ding, S., Rotter, J.M., Ooi, J.Y., Enstad, G., Xu, D., 2013. Normal pressures and
frictional tractions on shallow conical hopper walls after concentric lling:
predictions and experiments. Chem. Eng. Sci. 89, 264272.
Drescher, A., 1991. Analytical Methods in Bin-load Analysis. Elsevier, Amsterdam.
Drescher, A., 1992. On the criteria for mass ow in hoppers. Powder Technol. 73,
251260.
Drucker, D.C., Prager, W., 1952. Soil mechanics and plastic analysis or limit design.
Q. Appl. Math. 10.
Elaskar, S.A., Godoy, L.A., Gray, D.D., Stiles, J.M., 2000. A viscoplastic approach to
model the ow of granular solids. Int. J. Solids Struct. 37, 21852214.
Goodey, R.J., Brown, C.J., 2004. The inuence of the base boundary condition in
modelling lling of a metal silo. Comput. Struct. 82, 567579.
Goodey, R.J., Brown, C.J., Rotter, J.M., 2003. Verication of a 3-dimensional model
for lling pressures in square thin-walled silos. Eng. Struct. 25, 17731783.
Goodey, R.J., Brown, C.J., Rotter, J.M., 2006. Predicted patterns of lling pressures in
thin-walled square silos. Eng. Struct. 28, 109119.
Haussler, U., Eibl, J., 1984. Numerical investigations on discharging silos. J. Eng.
Mech. ASCE 110, 957971.
Janssen, H.A., 1895. Versuche ber getreidedruck in silozellen. Z. Ver. Dtsch. Ing. 39,
10451049.
Jenike, A.W., 1961. Gravity Flow of Bulk Solids. University of Utah.
Jenike, A.W., 1964. Storage and Flow of Solids. University of Utah.
Johanson, J.R., Jenike, A.W., 1962. Bulletin 116: Stress and Velocity Fields in Gravity
Flow of Bulk Solids, Salt Lake City, UT.
Karlsson, T., Klisinski, M., Runesson, K., 1998. Finite element simulation of granular
material ow in plane silos with complicated geometry. Powder Technol. 99,
2939.
Ketterhagen, W.R., Curtis, J.S., Wassgren, C.R., Hancock, B.C., 2009. Predicting the
ow mode from hoppers using the discrete element method. Powder Technol.
195, 110.
Ketterhagen, W.R., Hancock, B.C., 2010. Optimizing the design of eccentric feed
hoppers for tablet presses using DEM. Comput. Chem. Eng. 34, 10721081.
Lade, P.V., 1977. Elasto-plastic stressstrain theory for cohesionless soil with curved
yield surfaces. Int. J. Solids Struct. 13, 10191035.
Langston, P.A., Tzn, U., Heyes, D.M., 1995. Discrete element simulation of internal
stress and ow elds in funnel ow hoppers. Powder Technol. 85, 153169.
Nedderman, R.M., 1988. The measurement of the velocity prole in a granular
material discharging from a conical hopper. Chem. Eng. Sci. 43, 15071516.
Nedderman, R.M., 1992. Statics and Kinematics of Granular Materials. Cambridge
University Press, Cambridge.
Nedderman, R.M., Tzn, U., Savage, S.B., Houlsby, G.T., 1982. The ow of granular
materialsI: Discharge rates from hoppers. Chem. Eng. Sci. 37, 15971609.
Nitka, M., Combe, G., Dascalu, C., Desrues, J., 2011. Two-scale modeling of granular
materials: a DEMFEM approach. Granul. Matter 13, 277281.
57
Ooi, J.Y., Chen, J.F., Rotter, J.M., 1998. Measurement of solids ow patterns in a
gypsum silo. Powder Technol. 99, 272284.
Ooi, J.Y., Rotter, J.M., 1990. Wall pressures in squat steel silos from simple nite
element analysis. Comput. Struct. 37, 361374.
Parisi, D.R., Masson, S., Martinez, J., 2004. Partitioned distinct element method
simulation of granular ow within industrial silos. J. Eng. Mech. 130, 771779.
Ramrez, A., Nielsen, J., Ayuga, F., 2010. Pressure measurements in steel silos with
eccentric hoppers. Powder Technol. 201, 720.
Roberts, A.W., 1998. Particle technologyreections and horizons: an engineering
perspective. Chem. Eng. Res. Des. 76, 775796.
Rotter, J.M., 2001. Guide for the Economic Design of Circular Metal Silos. Spon,
London.
Rotter, J.M., Holst, J.M.F.G., Ooi, J.Y., Sanad, A.M., 1998. Silo pressure predictions
using discrete-element and nite-element analyses. Philos. Trans.: Math., Phys.
Eng. Sci. 356, 26852712.
Rousseau, J., Frangin, E., Marin, P., Daudeville, L., 2009. Multidomain nite and
discrete elements method for impact analysis of a concrete structure. Eng.
Struct. 31, 27352743.
Sanad, A., Ooi, J., Holst, J., Rotter, J., 2001. Computations of granular ow and
pressures in a at-bottomed silo. J. Eng. Mech. 127, 10331043.
Savage, S.B., 1965. The mass ow of granular materials derived from coupled
velocity-stress elds. Br. J. Appl. Phys. 16, 1885.
Spencer, A.J.M., Hill, J., 2001. Non-dilatant double-shearing theory applied to
granular funnel-ow in hoppers. J. Eng. Math. 41, 5573.
Steingart, D.A., Evans, J.W., 2005. Measurements of granular ows in two-
dimensional hoppers by particle image velocimetry. Part I: experimental
method and results. Chem. Eng. Sci. 60, 10431051.
Sugino, T., Yuu, S., 2002. Numerical analysis of ne powder ow using smoothed
particle method and experimental verication. Chem. Eng. Sci. 57, 227237.
Tejchman, J., Klisiski, M., 2001. FE-studies on rapid ow of bulk solids in silos.
Granul. Matter 3, 215230.
Tejchman, J., Wu, W., 1993. Numerical study on patterning of shear bands in a
Cosserat continuum. Acta Mech. 99, 6174.
Tuzun, U., Houlsby, G.T., Nedderman, R.M., Savage, S.B., 1982. The ow of granular
materialsII Velocity distributions in slow ow. Chem. Eng. Sci. 37, 16911709.
Tuzun, U., Nedderman, R.M., 1985a. Gravity ow of granular materials round
obstacles I: investigation of the effects of inserts on ow patterns inside a silo.
Chem. Eng. Sci. 40, 325336.
Tuzun, U., Nedderman, R.M., 1985b. Gravity ow of granular materials round
obstacles II: investigation of the stress proles at the wall of a silo with inserts.
Chem. Eng. Sci. 40, 337351.
Vidal, P., Cout, A., Ayuga, F., Guaita, M., 2006. Inuence of hopper eccentricity on
discharge of cylindrical mass ow silos with rigid walls. J. Eng. Mech. 132,
10261033.
Vidal, P., Gallego, E., Guaita, M., Ayuga, F., 2008. Finite element analysis under
different boundary conditions of the lling of cylindrical steel silos having an
eccentric hopper. J. Constr. Steel Res. 64, 480492.
Walker, D.M., 1966. An approximate theory for pressures and arching in hoppers.
Chem. Eng. Sci. 21, 975997.
Walters, J.K., 1973. A theoretical analysis of stresses in axially-symmetric hoppers
and bunkers. Chem. Eng. Sci. 28, 779789.
Wieckowski, Z., Youn, S.K., Yeon, J.H., 1999. A particle-in-cell solution to the silo
discharging problem. Int. J. Numer. Methods Eng. 45, 12031225.
Wu, W., Bauer, E., Kolymbas, D., 1996. Hypoplastic constitutive model with critical
state for granular materials. Mech. Mater. 23, 4569.
Xie, H.Y., Shinohara, K., 1999. Measurement of solids velocity in a conical hopper by
mass tracer particles. Chem. Eng. Sci. 54, 455459.
Yang, Y., Ooi, J., Rotter, M., Wang, Y., 2011. Numerical analysis of silo behavior using
non-coaxial models. Chem. Eng. Sci. 66, 17151727.
Zheng, Q., Yu, A., 2014. Why have continuum theories previously failed to describe
sandpile formation? Phys. Rev. Lett. 113, 068001.
Zhu, H.P., Yu, A.B., 2002. Averaging method of granular materials. Phys. Rev. E 66,
021302.
Zhu, H.P., Yu, A.B., 2004. Steady-state granular ow in a three-dimensional
cylindrical hopper with at bottom: microscopic analysis. J. Phys. D: Appl.
Phys. 37, 1497.
Zhu, H.P., Zhou, Z.Y., Yang, R.Y., Yu, A.B., 2007. Discrete particle simulation of
particulate systems: theoretical developments. Chem. Eng. Sci. 62, 33783396.
Zhu, H.P., Zhou, Z.Y., Yang, R.Y., Yu, A.B., 2008. Discrete particle simulation of
particulate systems: a review of major applications and ndings. Chem. Eng.
Sci. 63, 57285770.