Вы находитесь на странице: 1из 14

Engineering Structures 24 (2002) 15611574

www.elsevier.com/locate/engstruct

Simulation of axisymmetric discharging in metallic silos. Analysis


of the induced pressure distribution and comparison with different
standards
M.A. Martnez, I. Alfaro, M. Doblare
Department of Mechanical Engineering, Centro Politecnico Superior, University of Zaragoza, Mara de Luna, 3, 50018, Zaragoza, Spain
Received 2 April 2001; received in revised form 26 June 2002; accepted 27 June 2002

Abstract
The main objective of this paper is to present a new contribution to the problem of dynamic continuum simulation of discharge
of cylindrical silos by the Finite Element Method where many attempts have been made in the past by other researchers. We start
with a study of the bulk solid constitutive behaviour, the analysis of the stored-solid to silo-wall contact interaction and with a
discussion of the remeshing and rezoning algorithms needed to appropriately take into account the large displacements and the
associated mesh distortions that appear in this type of problems. Some restrictions of the simulation are due to the axisymmetry
of the model and the constitutive assumptions. First of all, a static analysis is accomplished using the usual hypotheses included
in different standards to check the ability of the method to reproduce standard available solutions. After this calibration stage, a
dynamic analysis is carried out to take into account the effects induced by the silo quaking phenomena, computing the overpressure
factor and comparing the obtained results with the pressure estimations established by different standards like the European standard
Eurocode ENV 1991-4, the French AFNOR P22 630, the German DIN 1055 or the American ACI 313-97 and R313-97. 2002
Elsevier Science Ltd. All rights reserved.
Keywords: Metallic silos; Dynamic pressure distribution; Discharging processes in silos; Bulk solid materials; Finite element simulation; Rezoning
and remeshing

1. Introduction
In spite of considerable experience in the construction
of metallic and concrete silos, their design still lacks of
a global theory generally accepted by researchers. This
affirmative sentence is confirmed by the important differences between the proposals included in the standards
of several countries that, in some cases, lead to very different designs.
One of the most important problems for a silo
designer is the accurate prediction of the external load
distribution acting on the shell, with special care on the
wall pressures induced by the stored material. This
pressure distribution depends on the bulk solid constitutive behaviour, the interaction between the stored solid

Corresponding author. Tel.: +34-976-761912; fax: +34-976762578.


E-mail address: mdoblare@posta.unizar.es (M. Doblare).

and the silo walls and the flow properties during the filling and discharging processes [1].
Bulk solids are composed by individual solid particles
inside a continuous phase, usually gaseous. The interaction among these particles and the continuous phase is
complex, being very difficult to formulate a complete
and accurate theoretical description of this problem.
This behaviour is a kind of combination between
liquids and solids. A liquid under static conditions cannot transmit shear forces so its pressure increases linearly with depth, independent of the direction. Bulk solids, on the contrary, can form surfaces up to a certain
slope, corresponding to the natural angle of frictional
stability. They are able to transmit static shear forces and
the pressures on the silo wall do not increase linearly
with depth, but they quickly reach a maximum, due to
the wall friction forces (Fig. 1) [2,3,47]. In addition,
these pressures depend on the direction and vary according to whether the solids are being filled, stored or discharged. Bulk solids cannot be considered as solid either,

0141-0296/02/$ - see front matter 2002 Elsevier Science Ltd. All rights reserved.
PII: S 0 1 4 1 - 0 2 9 6 ( 0 2 ) 0 0 1 0 0 - 1

1562

M.A. Martnez et al. / Engineering Structures 24 (2002) 15611574

Fig. 1. Different pressure distributions of a liquid and a bulk solid.

since they are not able to carry significant vertical loads


without lateral support, more like fluids.
The bulk solid within a silo is subjected to two main
pressure states, that, using usual terms in soil mechanics,
may be named as active and passive stress states [8,9].
The active state develops during filling and remains until
material is removed from the silo. It is identified by the
fact that the maximum principal stress s1 at the axis of
symmetry is the vertical compression stress component.
As soon as the bulk solid starts to flow towards the
outlet, it deforms plastically compressing horizontally
and expanding vertically. This is the typical situation of
a passive stress state that mainly appears in the area of
convergent flow, in which the major principal stress s1
at the axis of symmetry is now directed horizontally.
Pressures acting on the silo walls during discharge
depend very much on the way the convergent flow channel is formed. If it approximately coincides with the hopper geometry, the whole contents of the silo enters in
motion, exhibiting a flow pattern known as mass flow,
in contrast to the so-called funnel flow, where the area
of convergent flow, although well extended into the vertical part of the silo, is bounded by stationary material
inside the so-called dead zone. A fully-developed passive stress state therefore applies only to hoppers or hopper-like areas of convergent flow during discharge.
The determination of these pressures is a well-known
problem that has long been studied. The first method
proposed by Janssen [2], known as the slice-element
method, basically consists of finding the solution of the
differential equation corresponding to the vertical equilibrium in a horizontal grain slice of the silo. This problem is not well-posed, needing additional hypotheses in
order to allow for a solution to the problem. The most
usual of these hypotheses are:
The independence of the vertical pressures with
respect to the horizontal co-ordinates, that is, the
pressure distribution depends only on the vertical
coordinate.
A constant ratio between vertical and horizontal
pressures.
With these assumptions, it is possible to solve the
problem very easily, getting an analytical expression of
the pressure distribution, valid for any silo. Many other

authors [10,47,3,1115] have contributed in different


ways to the improvement of this method that finally has
become the basement of most national standards. However, they usually consider significant additional safety
coefficients (see ACI [16], Eurocode [17], DIN [18] or
AFNOR [19]) due to the inability of the method to take
into account the important dynamic effects that appear
during discharge.
Due to its intrinsic hypotheses, the slice-element
method accurately estimates the wall pressure, but the
hypothesis of constant vertical stress is incorrect. In
order to get some additional insight, Jenike and Johanson
[47] included a yield criterion for the steady state flow.
The two resulting partial differential equations were converted into four ordinary differential equations, which
were solved by the method of characteristics [20]. This
approach can be applied to filling and discharge processes, although its main disadvantage is the difficulty
of extending it to complex geometries, constitutive
behaviours or interactions between walls and grain.
More recently, the possibility of applying numerical
methods to obtain approximate solutions of very complex problems has pushed the scientists to look at this
old challenge from a new point of view. For instance,
the use of finite elements [21] allows the solution of the
dynamic problem in discharging processes for arbitrary
geometries and elaborated constitutive models, simulating the plastic and viscous behaviour of the grain solid
and including the contact interaction between grain and
silo walls.
Finite element dynamic studies of the discharge process in silos have been performed by Kolymbas [22], or
Rombach [23]. Ruckenbrod and Eibl [24] successfully
used the constitutive models of Lade [25] and the more
complex mathematical model of Kolymbas [22] to
describe silo discharge and flow patterns by using a viscoelastoplastic constitutive relationship.
Ooi et al. [26,27] also obtained predictions of the
pressure laws in the hopper and studied the influence of
wall imperfections considering elastic behaviour of the
bulk material. Rong, Ooi and Rotter [28], Holst, Ooi,
Rotter and Rong [29,30] have extensively worked in
filling and discharging processes in silos. In filling states
they employed elastic constitutive behaviour, while in
discharging cases a MohrCoulomb criterion was used.
Meng, Jofriet and Negi [31,32] proposed a special
class of secant constitutive relationship for isotropic and
cohesionless granular material. They employed an elasto-plastic relationship with a DruckerPrager yield criterion and a PrandtlReuss flow rule. A simple finite
element model served as benchmark for the correlation
with experimental results, and a parametric study was
carried out.
Wie ckowski [33] employs a particular Arbitrary LagrangianEulerian formulation [34,35] to overcome the
problem of mesh distortion. He also uses a Drucker

M.A. Martnez et al. / Engineering Structures 24 (2002) 15611574

Prager yield condition and a non-associative flow rule


to describe the mechanical behaviour of the bulk
material and an explicit algorithm to integrate the resultant equations with respect to time.
Despite all of these works, there still remain important
difficulties in getting accurate finite element solutions,
caused by the extreme distortion of the finite element
meshes near the silo outlet or the correct calibration of
the constitutive model of bulk solid.
In order to overcome the intrinsic difficulties of
methods like the Finite Element Method (FEM) based
on continuum assumptions, the discrete element method
(DEM) (Cundall [36] among many others) has recently
been applied to this problem. This method treats the
grain mechanical behaviour at the individual particle
level, and although most researchers use spherical particles, some work on non-spherical particles has also
been undertaken [37]. In the DEM, Newtons equation of
motion for each single particle replaces the differential
equilibrium equations of continuum mechanics, and the
model describing the particle contacts replaces the
constitutive continuum equations.
Developed during the last ten years, its actual potential of practical application is still unknown, since it
depends strongly on the available computer power,
being, for the moment, constrained to the analysis of
small models. Some researchers who have employed this
method to estimate pressure laws in silos have been
Yoshida [38] and Rong, Ooi and Rotter [28].
Some interesting papers have been published by Holst,
Ooi, Rotter and Rong [29,30] and Sanad, Ooi, Holst and
Rotter [39] consisting of a summary of several independent calculations made by researchers from different
countries. A standard problem of a silo filling and discharging was defined and two methods employed: the
finite element and discrete element method. Both commercial finite element codes and specifically written programs were used and very different constitutive relationships were applied (purely elastic, DruckerPrager with
and without dilatant flow, hyperelastic relations, etc.).
Pressure distributions were compared with classical
theories. A recent state of the art can be found in a special issue of the Journal of Engineering Mechanics concerning The statics and flow of dense granular systems [40].
In this paper, we focus on three of the main aspects
of the application of finite elements to the discharging
process in silos: the constitutive behaviour (both elastic
and perfect elastoplastic models have been compared),
the contact between the stored solid and silo walls
(friction contact elements have been employed) and the
solution of the mesh distortion problem (a remeshing
rezoning algorithm has been adopted for the first stages
of discharge). A reference example is solved using axisymmetric finite elements, considering both static and
dynamic situations, discussing the influence of the

1563

constitutive law and the flexibility of the wall on the


pressure distribution and overpressure dynamic factors.
Finally, these results are compared with the pressure
laws proposed by different standards.

2. Finite element modelling of the simulation of the


discharging process
As was pointed out in the previous section, the most
important limitations in the application of the FEM to
the dynamic simulation of the discharging process in
metallic silos are: 1. The use of an appropriate constitutive law for the stored grain; 2. The excessive mesh distortion near the hopper; 3. The contact interaction
between the bulk solid and the silo walls. In this section
we focus on these three problems, since the fundamentals of the finite element method on non-linear problems
are considered well known.
2.1. Bulk solid constitutive behaviour
The first problem concerning the bulk solid constitutive relation is essential in the discharging process simulation and the determination of the pressure distribution.
Two different bulk solid behaviours have been considered here: a purely linear elastic constitutive law and
a more complex elastoplastic constitutive model including a yield or fracture DruckerPrager criterion usually
used in soil mechanics and granular materials [9]. The
yield surface expression for the linear Drucker-Prager
criterion is:
F tptgbd 0

(1)

Fig. 2 depicts the representation of the linear Drucker


Prager criterion with isotropic hardening for a plane
strain state, with

Fig. 2.
terion.

Graphic representation of the plastic DruckerPrager cri-

M.A. Martnez et al. / Engineering Structures 24 (2002) 15611574

1564

p s1 s3

t s1s3

(2)

The relationships between the cohesion c and the


internal friction angle f of the classic MohrCoulomb
criterion with the parameters d and b of the Drucker
Prager criterion (for the extreme cases of associate flow
and non-associate flow with null dilatant angle y) are
written as:
Associate flow
tanb

(3)

3sinf

1 1 / 3sin f
2

3cosf

1 1 / 3sin f

Nonassociate flow
tanb 3sinf

,
(4)

d
3cosf.
c

For granular materials, non-associate flow with y


b is usually used. In this work, we have considered nondilatant flow (y 0), that is, incompressible inelastic
strain. Although grain solids are essentially cohesionless,
a small value of c must be used to avoid numerical
singularities. Relation between cohesion in Drucker
Prager criterion d and compression yield stress in the
hardening law is defined as:

1
d 1 tanb sc
3

(5)

2.2. Mesh distortion and rezoning algorithm


A total Lagrangian formulation has been used to perform the simulations. In this approach, due to the large
strains appearing, the elements can become so distorted
that the results are useless. This is exactly what happens
in the simulation of discharging processes in silos, due
to the high velocity of the grain exiting the cone, in comparison to the velocity of the inside grain, leading to
important local distortions of the mesh near the outlet.
A possibility to solve this problem is to remesh
locally, projecting the results from the original to the
new mesh. This projection approach is known in the
specialised literature as rezoning (ABAQUS, [41]) The
analysis process, therefore, consists of several steps, i.e.
1. Running the analysis with a specific mesh until it
becomes distorted, then 2. The results are mapped onto
a new non-distorted mesh, and 3. The calculation continues with this new mesh. This rezoning procedure may
be done as many times as needed during the analysis.
The rezoning is a projection technique composed of
three steps, as detailed in the following:
The values of the element variables are extrapolated
from the integration points to the nodes of the
deformed old mesh, by performing some kind of
patch recovery technique.

Each integration point and node of the new mesh is


detected to belong to an element of the deformed old
mesh. Therefore, it is necessary that the integration
points and nodes of the new mesh are located within
the boundary defined by the old mesh.
The values of the variables at the integration points
(element variables) and nodes (nodal variables) of the
new mesh are obtained by direct interpolation of the
nodal values of the deformed old element to which
they belong.
The accuracy of the rezoning technique highly
depends on the distortion of the initial mesh and the size
of the elements. If the initial mesh is not distorted
excessively and if the meshes are fine enough, this
rezoning approach works well. Otherwise, a discontinuity on the solution is clearly detected.
The generation of the new mesh may be done manually, by using the mesh generation capabilities of the
code used or, alternatively, an external automatic algorithm may be used. In this case, an in house code that
automatically generates the new mesh and the input file
for ABAQUS has been implemented.
This code reads the geometry of the deformed old
mesh, the nodal and element variables at the time step
decided by the user, and other data such as groups of
elements and nodes, applied loads, boundary conditions,
computation method, etc. It also reads the data of the
new mesh that, in our case, is a subset of a fixed mesh
filling all the space in which there will be grain during
discharging. This fixed mesh will be denoted as pattern
mesh (Fig. 3). Once the new mesh has been defined
inside the current boundary, it is written on the
ABAQUS input file, together with all the data needed
for the rezoning, continuing the simulation automatically.
As has been explained already, the new mesh is the
particular region of the pattern mesh placed inside the
boundary of the deformed mesh. Therefore, the location
of the nodes of the new mesh are exactly the same as
the ones of the pattern except those close to the boundary
that have to be adapted in order to get coincidence
between the edges of the new elements with the current boundary.
The process therefore may be summarised as follows:
The data of the pattern mesh are obtained from an
ABAQUS output file.
The data of the deformed mesh, at the time increment
decided by the user, and other necessary data are read
from different ABAQUS output files.
A polyline is generated using the nodes of the boundary as vertexes. The black mesh in Fig. 3a represents
the pattern mesh and the grey mesh the deformed
mesh, with the thicker line defining the current
deformed boundary polyline.

M.A. Martnez et al. / Engineering Structures 24 (2002) 15611574

1565

All the elements involved in the rezoning process have


been considered to have the same geometric type and
the same number of nodes, for instance, bilinear fournoded elements. As is clear in Fig. 3f, the new mesh is
composed of elements with three and four nodes. However, triangular elements are defined as degenerated
four-noded elements.
The properties of the nodes and elements are mapped
to the new nodes and elements of the deformed mesh
by suitable interpolation. For instance, if a node of the
new mesh is at the same location as a node of the
deformed mesh, belonging to a certain group, this new
node will belong to the same group. In the same way,
if the node is placed between two nodes of the deformed
mesh belonging to the same group, this intermediate
node is also assigned to the group. Once all the elements,
nodes, materials and geometry groups of the new mesh
have been defined, it is possible to generate automatically the ABAQUS input file.
The final new mesh obtained has the same quality as
the pattern mesh (except near the current boundary,
which coincides exactly with the one of the deformed
old meshes), then the analysis may continue after rezoning (see Fig. 4 for an example of this remeshing
algorithm).
Fig. 3.

Automatic creation of the new mesh.

The nodes closest to each vertex of the polyline are


translated to the location of the vertex, as shown in
Fig. 3b. If a node is closest to two different vertexes,
the node is translated to the boundary point such that
the adjacent elements become less distorted.
All the nodes of the pattern mesh are classified
according to their location in the initial mesh as
inside, outside or on top of the polyline.
Then the elements of the pattern mesh are classified
according to their nodes. A three digit number is
assigned to each element: the first digit corresponds
to the number of nodes that are inside the polyline,
the second to the number of nodes that are located on
the line and the third one to the number of nodes outside the boundary. Several examples may be observed
in Fig. 3.
All the elements that have nodes inside and outside
the boundary are modified. In Fig. 3c, for instance,
the element on the lower left corner, whose type is
211, is transformed into an element type 220 by translating one of its nodes to the polyline (see Fig. 3d).
In Fig. 3d, all the elements with type 121 are transformed into triangular elements type 120 (see Fig. 3e).
At this moment there is no element with nodes inside
and outside, so the external elements are removed and
the internal elements are associated to the new mesh
(Fig. 3f).

2.3. Contact interaction with the silo wall


The interaction between the stored solid and the silo
walls has been modelled with contact surfaces, avoiding,
therefore, the penetration of the grain nodes into the wall
surface. All the simulations have been performed considering finite strains and displacements, that is, a fully
geometrically non-linear approach.
2.4. Finite element model
A silo with usual geometric dimensions shown in Fig.
5a has been chosen as reference. Wheat has been considered as the bulk material.
We have considered an axisymmetric model composed by 840 bilinear quadrilateral axi-symmetrical solid
elements with complete integration, CAX4 in the

Fig. 4. Different meshes of the bulk solid at the silo outlet before
and after applying remeshing algorithm.

M.A. Martnez et al. / Engineering Structures 24 (2002) 15611574

1566

Table 1
Elastic material parameters for bulk material (AFNOR)
Internal angle of friction (fv)
Friction coefficient (mv)
Material density
Young modulus
Poisson coefficient

Fig. 5.

Geometry of the silo.

ABAQUS nomenclature, for the stored grain and rigid


(RAX1) or flexible axisymetric shell elements (SAX1)
with two nodes for the silo wall depending on the
example. The finite element mesh has been progressively
refined near the outlet zone, as is clearly shown in
Fig. 5b.
Four different meshes have been employed in the
dynamic analysis following the ideas described in the
previous section for remeshing. They are shown in Fig.
6, being easy to notice how the deformed bulk solid that
passes through the outlet is kept inside the mesh, but as
will be proved later it has a negligible influence on the
rest of the stored solid. The pattern mesh is always maintained as was stated before, allowing us to get accurate
results during the evolution of the discharging process.
For the elastic behaviour of the bulk solid, average
values from different standards were computed for the
wheat constitutive parameters. These values were very

23.80
0.308
835 kg/m3
2.0E+05 N/m2
0.4

close to the ones proposed by the AFNOR standard [19],


so, finally, we decided to use the values proposed by this
standard that are included in Table 1.
From these parameters (the internal friction angle in
particular) the material parameters for the Drucker
Prager criterion were computed using Eq. (4) giving the
values established in Table 2.
Different simulations have been performed to analyse
the influence of different constitutive models (elastic vs
elasticplastic), the flexibility of the silo walls or the differences between the results associated to static and
dynamic situations. The different examples that we will
consider are:
Static analyses with linear elastic and elasticplastic
behaviour of the bulk solid, both with rigid and flexible silo walls and both under filling and discharging.
Dynamic simulation of the discharging process considering elasticplastic behaviour or failure behaviour
of the bulk solid.

3. Results of the static filling analyses


3.1. Elastic behaviour and rigid silo wall
In this first case, a static geometrically non-linear
analysis has been performed, considering initially purely
elastic properties for the grain, using the values included
in Table 1. The silo wall was initially considered as rigid
and the outlet closed. Contact surfaces have been
included along the cylindrical body, the hopper and the
outlet. The friction coefficients are also included in
Table 1.
Vertical displacements and the horizontal stress distribution are shown in Fig. 7a,b. The maximum displacement appears near the axis of symmetry and the minimum near the silo wall, due to friction. Horizontal
stresses are almost constant at each height as predicted
by the Janssen theory. Only in the region near the joint
Table 2
Plastic behaviour parameters for bulk material

Fig. 6. Different meshes employed for the dynamic simulation for


t 0, 0.14, 0.27, 0.40 s.

Internal angle of the Drucker


Prager criterion
Yield stress

34.95
1.0E+01 N/m2

M.A. Martnez et al. / Engineering Structures 24 (2002) 15611574

1567

Fig. 7. Vertical displacements and horizontal stresses in a static analysis with elastic behaviour of the bulk solid (rigid silo wall and closed outlet).

between the cylindrical body and the hopper, is this violated, due to normal discontinuity that strongly changes
the local contact stress distribution.
The pressure distributions along the cylindrical body
and the hopper and the comparison with the ones proposed by the four standards mentioned in the introduction are shown in Fig. 8. From them we can conclude
that the finite element prediction in the cylindrical body
is in close agreement with the one proposed by the
AFNOR standard. This is not surprising since this is the
only code that follows the Janssen theory strictly, while

the rest of them (DIN, ACI and ENV) include overpressure factors. On the other hand, in the hopper, the finite
element approximation is not very accurate. For
instance, the expected limit to a zero value of the pressure when approaching the outlet is not predicted appropriately due to the wrong assumption of closed outlet
that induces a non-negligible pressure value at this point.
The tangential stress along the walls follows the same
tendency. Perfect sliding is obtained along almost the
whole wall, so a constant ratio between the friction stress
and the horizontal pressure is obtained. Only the nodes

Fig. 8. Horizontal pressure and tangential stress distributions over the silo wall for a static analysis with elastic behaviour of the bulk solid and
rigid walls, and comparison with the different standards.

1568

M.A. Martnez et al. / Engineering Structures 24 (2002) 15611574

close to the body-hopper joint remain tied to the silo


wall. Therefore friction stresses are smaller than the ones
predicted by the Janssen theory (AFNOR code).
3.2. Elastic behaviour and flexible silo wall
If we now consider flexible walls, in order to reproduce a more realistic situation, (standard properties for
the steel of E 200 GPa and n 0.3 have been used)
the vertical displacements and the horizontal stresses are
strongly modified as expected.
The elastic behaviour of the grain and the flexibility
of the walls produce a high reduction of the pressure
distribution of 70% when compared to the ones previously obtained under the rigid wall hypothesis and
therefore much smaller than the ones proposed by the
different standards (Fig. 9a,b). In this case, the contact
is lost near the body-hopper joint, leading, therefore, to
null pressure in that region. This fact is due to the
inability of the grain to follow the wall displacements,
since the grain horizontal displacements are essentially
caused by the values of the high vertical stresses and the
Poisson effect. The consequence is the increase of the
vertical pressures inside the bulk solid and the decrease
of the horizontal pressure and tangential stresses along
the cylindrical body. Complementary, this also induces
a strong increase of the wall pressure and friction stress
in the hopper, in order to balance vertical forces.
The reason for these unrealistic results is, no doubt,
the elastic behaviour of the bulk solid, since without a
horizontal restriction (due to the wall flexibility) the
grain should tend to form a cone with the natural frictional slope angle, while the simulation with elastic
behaviour tends to maintain the initial rectangular form.
Therefore it seems necessary to include a more realistic
material behaviour in order to avoid this problem.

Fig. 9.

3.3. Plastic behaviour


We have considered a perfectly plastic behaviour with
a DruckerPrager criterion as has been previously
explained in section 2. The corresponding results are
shown in Fig. 9a,b, being clear now that the differences
between the situations corresponding to the rigid and
flexible walls are much smaller than in the elastic simulation. Only some small waves in the normal pressure
distribution near the body-hopper joint appear for the
flexible case, due to the interaction with shell bending
effects. In this region the need of additional reinforcements is well known in order to absorb the local bending
stresses, keeping the membrane state predominant in the
rest of the structure.
A small pressure reduction of about 510% is produced in the flexible hypothesis, due to the use of a small
value for the cohesion in order to avoid numerical problems.
Rigid or flexible wall hypotheses show similar results
except near the body-hopper joint where the bending
waves, already mentioned, appear (Fig. 9a,b). Under the
hypothesis of flexible walls, the normal pressures along
the hopper increase slightly, since they may balance the
reduction of a 5% of the vertical stresses and the friction
forces along the cylindrical body. Finally, and since the
results obtained using plastic constitutive behaviour are
very similar for rigid and flexible walls, we will consider
only the first situation, due to its greater simplicity.
Some partial conclusions for the simulation of filling
processes employing static analyses are:
The flexible wall hypothesis must be always used
together with a plastic behaviour of the bulk material,
in order that the stored grain can follow the wall displacements. In this case the obtained pressure distri-

Horizontal and tangential stresses for different simulations.

M.A. Martnez et al. / Engineering Structures 24 (2002) 15611574

bution is in agreement with the standards and wellknown experimental data.


Shell local bending causes pressure waves near the
body-hopper joint.
A decrease of the pressure values of about 5% in the
cylindrical body is obtained when using flexible
walls, when compared to the pressures computed
under the rigid wall hypothesis.
A perfect plastic criterion is a good choice for the
bulk material constitutive behaviour. In fact almost
the whole grain region becomes plastic, with the
maximum plastic strains located at the symmetry axis,
near the silo wall and near the body-hopper joint.
Plastic equivalent stress distribution is very similar
both for the static and the dynamic analyses. (Fig.
12b).

4. Results of the dynamic discharge analyses


Once the geometric model, the contact interaction and
the bulk material behaviour have been checked in the
static case, it is possible to accomplish a detailed
dynamic analysis of the discharging process. Rigid walls
have been considered in this case, since, as has been
stated, when plastic failure behaviour is considered for
the stored material, the influence of the wall flexibility
is not so important, and the computational cost is considerably reduced.
An implicit time integration algorithm with a certain
amount of numerical damping to dissipate the undesired
effects of spurious higher time frequencies, the Hilber
HughesTaylor method (HHT) [42], has been employed
to integrate the evolution equations. This numerical

Fig. 12. Vertical displacements inside the silo for a time t 0.08
sand equivalent plastic strains for t 0.14 s.

1569

damping is useful in these types of problems to improve


the algorithmic stability. It is important in this case due
to the intermittent contact of the bulk material with the
silo walls. A value of the damping parameter, a
0.20, for the HHT algorithm has been chosen in order
to damp out the low frequency response.
Two steps are applied: first a static analysis is performed considering the previously mentioned boundary
conditions corresponding to a closed outlet. The second
step corresponds to the dynamic problem, which appears
after removing the outlet boundary conditions, simulating its instantaneous opening. The first moments are the
most important ones reaching a stationary flow very rapidly. Small time increments have been chosen, according
to this high velocity problem, so a time step of t
3.0 106 has been used.
The analysis goes on until the mesh distortion is too
high to get useful results, then the remeshing procedure,
previously explained, is used and the associated projected results on the new mesh are computed to continue
with the analysis. Four different remeshing intervals
have been analysed defined by three remeshing processes that, in our case, have been needed at times t
0.14, 0.27, 0.40 s. The four meshes have been shown
in Fig. 6. The distribution of vertical displacements,
equivalent plastic strains and vertical stresses are
presented in Fig. 10a,b, and 10c respectively for t
0.15 s, before the first remeshing process.
Some interesting aspects are easily observed in Fig. 10
which support the validity of this simulation: the vertical
displacement contour plot (Fig. 10a) reproduces qualitatively well the experimental results obtained by Kvapil
[10], known as Kvapil ellipses, shown in Fig. 11. Kvapil
considered that the moving particles follow two different
movements: the primary and the secondary. Particles
displace downwards only, due to gravity forces in the
primary movement, while particles inside the secondary
movement region have vertical and horizontal displacements. These movement regions have elliptical forms.
This secondary movement is extremely important near
the outlet appearing in every type of silo, while the primary movement arises only in very tall silos.
When the grain begins to flow out of the silo through
the outlet, both ellipses start to grow and when the primary ellipse reaches the free surface a repose zone is
formed. The secondary ellipse grows until arriving at the
deepest point of the repose zone (Fig. 11b), then it starts
to decrease. Finally, a discharging funnel is formed that
constitutes the interface between the slow and the fast
moving areas.
Fig. 10a shows how the FE simulation reproduces the
formation of the first Kvapil ellipse during the initial
discharging process. At this time, the maximum vertical
displacement through the outlet is about 0.13 m, which
represents an average velocity at this region of about
1 m/s.

1570

M.A. Martnez et al. / Engineering Structures 24 (2002) 15611574

Fig. 10. Distribution of vertical displacements, equivalent plastic strains and vertical stresses in the outlet region at t 0.15 s before remeshing.

Fig. 11. Ellipses of movement for the bulk solid proposed by Kvapil.

The equivalent plastic strain distribution (Fig. 10b)


shows that the maximum plastic strains appear in the
region near the outlet, representing the local failure of
the stored grain, due to its inability to support tension
stresses. The maximum plastic strain component appears
along the vertical axis just above the outlet. This
inability to support tension stresses is also observed in
the vertical stress distribution (Fig. 10c); the tension
stresses in almost all this area are equal to the limit of
10 N/m2 that has been established as the tension failure
stress for this material (Table 2).
The immediate consequence is that the bulk solid that
has passed through the outlet does not affect the rest
of the grain inside the silo. This is the reason why the
corresponding finite elements are not removed from the
current mesh. On the contrary, they are retained in order
to make the boundaries of the two meshes coincident
(constituting the current deformed boundary polyline),
which is a condition needed in ABAQUS to perform the
rezoning process.
A portion of the grain remains tied to the hopper, for-

ming a slope angle of about 60 (Fig. 12a). This region


corresponds to the funnel flow in the discharging process
as it is described by the AFNOR standard for a silo of
these dimensions. Most elements have arrived at the
state of failure at t 0.14 s (Fig. 12b). The maximum
values of the plastic strains appear in the hopper region
due to the tension stresses as has been explained before.
There also appear high plastic strains in the cylindrical
body near the wall due to the high friction forces.
In Fig. 13a normal wall pressure distribution is shown
for t 0.02 s, just after outlet opening. The best-fit
curve to Janssens theory is also shown in Fig. 13a with
the values of the associated coefficients. A least mean
square method has been used for the cylinder zone.
Results are compared with the static FEM simulation.
The dynamic overpressure is shown in Fig. 13b.
The dynamic nature of the simulation leads to stress
waves due to the contact-impact between the grain and
the rigid walls. The high-frequency contribution of the
waves is damped by the numerical damping supplied by
the integration algorithm and by the smoothing produced
by the rezoning algorithm.
The distribution is similar both for the dynamic and
static simulations over the upper part of the silo, while
at levels below y 6 m, the increase becomes
important, from 15% to 50% in the hopper-cylinder transition. The dynamic overpressure on the hopper wall
reaches values close to 60%, showing a tendency to zero
pressure near the outlet. Obviously, in this zone the
dynamic overpressure coefficient is negative because
static FEM does not show this zero tendency.
Similar results for a further instant, t 0.50 s, are
shown in Fig. 14a,b. The wall pressure distribution is
compared with the best-fit Janssens curve and with the
static simulation. Clearly, the most critical zone is the

M.A. Martnez et al. / Engineering Structures 24 (2002) 15611574

1571

Fig. 13. Horizontal pressure and dynamic overpressure coefficient for a time t 0.02 s compared with the static pressure distribution.

Fig. 14. Horizontal pressure and dynamic overpressure coefficient for a time t 0.50 s compared with the static pressure distribution.

cylinder body-hopper transition, where an increase of


85% is reached, both in cylinder and hopper zone. A
small increase of the pressure occurs from the initial
time, t 0.02 s to t 0.5 s, but these values remain
practically constant until a significant reduction of the
grain level is produced. The normal pressure distribution
at levels above y 6 m does not present this increase.
These results are in agreement with previous references in the literature Rotter et al. [43] and Sanad et al.
[39], using the discrete element and finite element
methods. Specifically in the FEM simulations a pressure
increase appears in the bottom of silo cylinder body.
Unfortunately, a deeper comparison is not possible due
to the different silo dimensions and the presence of a
hopper in the example of this paper.
The represented wall friction coefficient was computed as:
m

tangencial stress
,
normal stress

(6)

and is shown in Fig. 15a,b for the static and the dynamic
analyses for t 0.02 s and t 0.50 s respectively. In
the static case the wall developed friction coefficient
coincides with the input wall friction, m 0.308, (Table

2) for the cylinder zone, which means that perfect sliding


occurred. However this ratio is reduced to m 0.18 at
the hopper, so bulk material remains tied to the hopper
wall (Fig. 7a) and tangential contact stresses are lower
than the product of normal stresses and input friction
data.
This tendency is completely different for the dynamic
analysis. For t 0.02 s, the developed wall friction
coefficient is higher on the top of the cylinder and in the
outlet, because bulk material begins to go out, while m
is lower just in the cylinder-hopper transition, where
grain remains fixed. This zone is called dead zone by
AFNOR in the case of mixed discharging. For t
0.50 s, m coincides with input data on the hopper due
to the discharging movement of the material and, while
this friction ratio is practically constant in the cylinder
m 0.10, where the stored material remains attached to
the wall. Logically there is a point where the slip-stick
condition changes, and this point rises when the discharging process advances [47].
These distributions show a behaviour very similar to
the one observed by Jenike and Johanson [47], in
whose results peak pressure appears when material
changes from an active to a passive state. This peak of

1572

Fig. 15.

M.A. Martnez et al. / Engineering Structures 24 (2002) 15611574

Developed wall friction coefficient (tangential-normal stress ratio) for static and dynamic (t 0.02 s and t 0.50 s) F.E. predictions.

normal pressure appears at h 1 m for t 0.2 s and


at h 2 m for t 0.50 m (Figs. 13a and 14a) approximately. This change from an active to a passive state is
produced a little above the beginning of slipping in the
bulk material (Fig. 15a,b), and its mission is to compensate this activepassive stress change. This peak rises
along the silo wall when grain begins to slip.
A comparison between the numerical normal pressure
distributions and the ones predicted by national standards is included in Fig. 16. Firstly in the cylinder body
the static simulation does not provide accurate results,
while the dynamic simulation shows a behaviour equal
to AFNOR for t 0.02 and between AFNOR and DINENV for t 0.50 s. The pressure distribution predicted

by ACI standard is too high. The safety coefficients


used by ACI applied to the static prediction of the
Janssen theory are 1.5 and 1.7 for dynamic and live
loads respectively, with a total multiplicative coefficient
of 2.55. ACI is, therefore, the most conservative of the
four standards studied, even when ACI considers a lower
ratio between the horizontal and vertical stresses, which,
in theory, should lead to smaller pressures along the silo
body (i.e. ACI uses a ratio of 0.42 and AFNOR 0.60).
The behaviour of the FEM simulations on the hopper
(Fig. 16) is again near the DIN standard or even to ENV
standard, while AFNOR and ACI do not always estimate
pressure on the safety side. However the dynamic simulation shows a tendency to zero pressure just above the
outlet, while DIN and ENV are more conservative and
predict non zero pressure on the outlet. It is remarkable
that both standards always predict wall normal pressure
higher than the static or dynamic simulation, though the
procedure of calculation is different. The DIN standard
considers overpressure coefficients during the discharging process, while the ENV incorporates local pressure
concentration factors in every region.

5. Conclusions

Fig. 16. Horizontal pressure distributions over the silo wall for a
static and dynamic analysis (t 0.02 s and t 0.50 s), and comparison with the different standards.

The computation of the pressures acting on metallic


silos, due to the effect of the grain during the discharging
process, is one of the most important aspects to be taken
into account in the design of silo structures and the main
objective of this paper. Numerical simulations have been
made using the finite element method with some specific
features, regarding the bulk solid constitutive behaviour,
the treatment of the contact between the stored solid and
the silo walls or the remeshing and rezoning processes
needed to solve the problem of excessive mesh distortion
during discharge.
The first problem addressed has been the simulation
of the filling process under the Janssen hypotheses that
are the basis of most European and American codes.

M.A. Martnez et al. / Engineering Structures 24 (2002) 15611574

They consist of a static calculation, considering a rigid


silo wall assuming a Coulomb friction model between
the grain and the wall. The results obtained with our FE
analysis are in close agreement with the AFNOR standard estimations, while the ones of the DIN, Eurocode
and ACI are always on the safety side, showing that
these last three standards impose too high safety coefficients.
Studying the influence of the flexibility of the wall, it
is concluded that the initial model, which assumes a perfect elastic behaviour of the grain, is not valid. The
pressure along some areas of the wall is reduced to
approximately zero, implying that the grain is not able
to follow the horizontal displacements of the wall. To
solve this problem a DruckerPrager perfectly plastic
constitutive model, that is, a failure criterion with null
cohesion, has been employed and compared to the elastic
one. The pressure distribution coincides again with the
one given by the standards, but, due to the flexibility of
the wall, its value is reduced by about 5%. In addition,
the pressure distributions exhibit some waves, especially
near the joint between the cylindrical body and the hopper.
A dynamic analysis has been performed using the
rigid wall and plastic behaviour for the grain as model
properties in order to simulate the discharging process.
As different authors have commented before, this work
confirms that the most unfavourable time appears during
the first steps of the discharge. The obtained pressures
are substantially higher than the static ones in some
zones of the silo. The dynamic pressure increase is about
1030% for the cylinder zone and reaches 85% in the
cylinderhopper transition, while just above the outlet it
can be negative, due to the non-zero tendency of the
static pressure distribution. These values are close to the
ones established by many authors by experimental
methods [47] or by numerical techniques [39,43]. The
developed wall friction coefficient for the dynamic
analysis clearly shows the zones where grain remains
fixed to the wall or the zones of relative movement. A
peak of normal pressure accompanies this change of slipstick condition, as it is explained in several papers [4
7] whose mission is to compensate activepassive stress
change. This peak rises along the silo wall when grain
begins to slip.
A comparison between different standards shows that
AFNOR leads to the best results for the static filling process, while the rest of the standards are on the safety
side. The DIN and ENV standards provide the best accuracy in the dynamic simulation of the discharging process, while AFNOR predicts lower values and ACI
higher ones.
We can finally conclude that complex phenomena like
the discharging of grain in silos, can be analysed by
numerical simulations, although there are two essential
aspects that have to be taken into account: a correct

1573

constitutive model for the grain, including failure and an


adequate treatment of the mesh distortion by means of
appropriate algorithms.
Nevertheless, this simulation presents some restrictions. Firstly, a full 3-D model must substitute the axisymmetric model to incorporate eccentric effects in the
discharging process. The constitutive assumptions are
another model limitation, i.e. effects like variation of the
grain density during the analysis are not considered, and
should be verified by experimental results. The contact
algorithm must be improved to get a more realistic ratio
between friction and normal stresses.
The remeshing and rezoning techniques here used are
able to reproduce the discharging process although with
a high computational cost. Perhaps a meshless method
can avoid this inconvenience. However the main problem of this simulation is the difficulty of verification of
the assumptions of the numerical model.
References
[1] Gaylord EH, Gaylord CN. Design of steel bins for storage of
bulk solids. Englewood Cliffs, NJ: Prentice Hall, 1984.
[2] Janssen HA. Versuch u ber getreidedruck in silozallen (research
on pressure in silos). Zeitschrift des vereines Deutscher Ingenieure 1895;39:10459.
[3] Walker K. An approximate theory for pressures and arching in
hoppers. Chemical Engineering Science 1966;21:97597.
[4] Jenike AW, Johanson JR. Bin loads. Journal of Structures
Division ASCE 1968;94(ST4):101141.
[5] Jenike AW, Johanson JR. On the theory of bin loads. Journal of
Engineering for Industry, ASME, Series B 1969;94(2):33944.
[6] Jenike AW, Johanson JR, Carson JR. Effect of initial pressures
on flowability of bins. Journal of Engineering for Industry,
ASME, Series B 1969;94(2):3959.
[7] Jenike AW, Johanson JR. Bins loadspart 2: Concepts; part 3:
mass flow bins. Journal of Engineering for Industry, ASME, Series B 1973;95(1):112.
[8] Terzaghi K. Theoretical soil mechanics. New York: John Wiley
and Sons, 1943.
[9] Atkinson JH, Bransby PL. The mechanics of soils. An introduction to critical state soil mechanics. McGraw Hill, 1978.
[10] Kvapil R. Theorie der Schuttgutbewegung. Berlin: VEB-Verlag
Technik, 1959.
[11] Walters JW. A theoretical analysis of stresses in axially-symmetric hoppers and bunkers. Chemical Engineering Science
1973;28:77989.
[12] Motzkus U. Belastung von Silobo den und Auslauftrichtern durch
Ko rnige Schu ttgu ter, PhD Dissertation, TU, Braunschweig, 1974.
[13] Enstad GG. On the theory of arching in mass flow hoppers.
Chemical Engineering Science 1975;30:127383.
[14] MacLean AG. Initial stress fields in converging channels. Bulk
Solids Handling 1985;5(2):4954.
[15] Schulze D. Untersuchungen zur gegenseitigen Beeinflussung von
Silo und Austragorgan, PhD Dissertation, TU Braunschweig,
1991.
[16] ACI 313-97 & ACI 313R-97, Standard practice for design and
construction of concrete silos and stacking tubes for storing
granular materials (ACI 313-97) and CommentaryACI 313R97, ACI International, 1997.
[17] Eurocode 1: Basis of design and actions on structures. Part 4:
Actions in silos and tanks. CEN, 1995.

1574

M.A. Martnez et al. / Engineering Structures 24 (2002) 15611574

[18] Din 1055, Teil 6, Lastannahmen fu r bautenLsten in silozellen, 1987.


[19] AFNOR P-22 630, Construction metallique: Silos, BNCM-3,
1992.
[20] Wilms H. Spannungsberechnung in Silos mit der Charakteristikenmethode, PhD Dissertation, TU Braunschweig, 1984.
[21] Zienkiewicz OC, Taylor RL. The finite element method, 5th ed.
John Wiley & Sons, 2000.
[22] Kolymbas D. An outline of hypoplasticity. Arch. Appl. Mech.
1991;61:14351.
[23] Rombach GA. Schu ttguteinwirkungen auf Silozellen, Exzentrische Entleerung. Vero ffentlichungen Heft 14, PhD Dissertation,
University of Karlsruhe, 1991.
[24] Ruckenbrod C, Eibl J. Proceedings RELPOWFLO II, Oslo. 1993;
503516.
[25] Lade PV. Elasto-plastic stress strain theory for cohesionless soil
with curved yield surface. Int. J. Solids Structure
1977;13:101935.
[26] Ooi JY, Rotter JM. Elastic predictions of pressures in conical silo
hoppers. Engineering Structures 1991;13(1):212.
[27] Ooi JY, She KM. Finite-element analysis of wall-pressure in
imperfect silos. International Journal of Solids and Structures
1997;34(16):206172.
[28] Rong G, Ooi JY, Rotter JM. Discrete element modeling of particulate solids in silos. Mechanics of Deformation and Flow of
Particulate Materials ASME-ASCE-SES. Joint Summer Meeting
1997. ASCE.: 321334.
[29] Holst JMFG, Ooi JY, Rotter JM, Rong GH. Numerical modeling
of silo filling. I: continuum analyses. Journal of Engineering
Mechanics, ASCE 1999;125(1):94103.
[30] Holst JMFG, Ooi JY, Rotter JM, Rong GH. Numerical modeling
of silo filling. II: discrete element analyses. Journal of Engineering Mechanics, ASCE 1999;125(1):10411.
[31] Meng QG, Jofriet JC, Negi SC. Finite-element analysis of bulk
solids flow. Part 1. Development of a model-based on a secant

[32]

[33]

[34]

[35]

[36]
[37]
[38]

[39]

[40]

[41]

[42]

[43]

constitutive relationship. Journal of Agricultural engineering


Research 1997;67(2):14150.
Meng QG, Jofriet JC, Negi SC. Finite-element analysis of bulk
solids. Part 2. Application to a parametric study. Journal of Agricultural Engineering Research 1997;67(2):1519.
Wie ckowski Z, Youn SK, Yeon JH. A particle-in-cell solution to
the silo discharging problem. International Journal for Numerical
Methods in Engineering 1999;45(9):120325.
Hirt CW, Amsden AA, Cook JL. An arbitrary LagrangianEulerian computing method for all flow speeds. Journal of Computational Physics 1974;14:22753.
Liu WK, Ghang H, Chen J, Belytschko T. Arbitrary Lagrangian
Eulerian PetrovGalerkin finite elements for nonlinear continua.
Comput. Methods Apl. Mech. Engrg. 1988;68:259310.
Cundall PA, Strack ODL. Geotechnique 1979;29:47.
Rothenburg L, Bathurst RJ, Geotechnique 42 (192) 3, 7995.
Yoshida J. Analytical study on the pulsation phenomenon of
granular materials in silos during discharge. Advanced Powder
Technology 1994;5(1):8598.
Sanad AM, Ooi JY, Holst JMFG, Rotter JM. Computations of
granular flow and pressures in a flat-bottomed silo. J. Eng. Mech.
2001;127(10):103343.
Journal of Engineering Mechanics. Special issue on The statics
and flow of dense granular systems, ed. Ooi JY, Hopkins M,
Sture St, J. Eng Mech., 2001; 127 (10).
Abaqus/Standard Theory Manual, Abaqus/Standard Users Manual & Abaqus/Explicit Users Manual. Version 5.8. Hibbit, Karlsson & Sorensen, Inc, 1998.
Hilbert HM, Hughes TRJ, Taylor RL. Improved numerical dissipation for time integration algorithms in structural dynamics.
Earthquake Engineering and Structural Dynamics 1977;5:28392.
Rotter JM, Holst JMFG, Ooi JY, Sanad AM. Silo pressure predictions using discrete element and finite element analyses. Philosophical Trans. Royal Soc., London 1747;356:2685712.

Вам также может понравиться