Professional Documents
Culture Documents
2010_Vol48_3-15
2010_Vol48_3-15
2010_Vol48_3-15
To cite this article: Florian Fleissner , Alexandra Lehnart & Peter Eberhard (2010) Dynamic
simulation of sloshing fluid and granular cargo in transport vehicles, Vehicle System
Dynamics: International Journal of Vehicle Mechanics and Mobility, 48:1, 3-15, DOI:
10.1080/00423110903042717
Taylor & Francis makes every effort to ensure the accuracy of all the information (the
“Content”) contained in the publications on our platform. However, Taylor & Francis,
our agents, and our licensors make no representations or warranties whatsoever as to
the accuracy, completeness, or suitability for any purpose of the Content. Any opinions
and views expressed in this publication are the opinions and views of the authors,
and are not the views of or endorsed by Taylor & Francis. The accuracy of the Content
should not be relied upon and should be independently verified with primary sources
of information. Taylor and Francis shall not be liable for any losses, actions, claims,
proceedings, demands, costs, expenses, damages, and other liabilities whatsoever or
howsoever caused arising directly or indirectly in connection with, in relation to or arising
out of the use of the Content.
This article may be used for research, teaching, and private study purposes. Any
substantial or systematic reproduction, redistribution, reselling, loan, sub-licensing,
systematic supply, or distribution in any form to anyone is expressly forbidden. Terms &
Conditions of access and use can be found at http://www.tandfonline.com/page/terms-
and-conditions
Downloaded by [University of Hong Kong Libraries] at 13:22 11 October 2014
Vehicle System Dynamics
Vol. 48, No. 1, January 2010, 3–15
The driving stability of transport vehicles is significantly affected by the type of cargo that is transported
and the design of the tank. Cargo motion can have both beneficial and negative aspects in terms of
driving stability and braking performance. Neglecting the influence of the dynamically moving cargo
in driving simulations leads to significant errors in the simulation results. We propose a new method
for the dynamic simulation of tank trucks carrying fluids and silo vehicles carrying granulates. The
method couples Lagrangian particle methods, such as smoothed particle hydrodynamics for fluids or
the discrete element method for granular media, and multibody systems using co-simulations. The
capability of the new approach is demonstrated by providing simulation results of two benchmark
manoeuvres and the behaviour of fluid and granular cargo is compared.
1. Introduction
The motion of fluids in tank trucks and granular materials in silo vehicles may have a sig-
nificant influence on the driving dynamics. Tank geometries and vehicle suspension systems
have to be designed carefully in order to provide stability during braking and lane change
manoeuvres as well as to improve driving comfort. Dynamic simulations are an attractive
tool to predict the influence of different design parameters during the design process. The
size of separate tank compartments is only one example for an important design parame-
ter, which affects the wavelength of the sloshing motion and thus the eigenfrequencies of
the system. Approaches for the simulation of tank vehicles have been proposed in the past
decades, mostly based on the coupling of multibody systems (MBS) and pendulum mod-
els or Eulerian fluid simulation methods such as the finite volume method [1]. However,
such approaches have several drawbacks. First, sloshing liquids involve free surface flows
which are difficult to handle by Eulerian approaches. Secondly, dry granular materials do
not behave like fluids in terms of their dynamic behaviour. Piling, avalanches and pressure-
dependent shear stiffness are only some of the numerous phenomena and characteristics of
granular materials, which require a different simulation approach. To allow for the simulation
of both, sloshing fluids with free surfaces and granular materials in transport vehicles, a new
co-simulation approach is implemented and tested that couples Pasimodo [2], a Lagrangian
simulation framework for the 3D simulation of granular materials and fluid models, with
Simpack [3], a commercial MBS simulation software. The approach is flexible and robust
and theoretically enables coupled simulations between Pasimodo and any other MBS sim-
ulation software that provides a TCP/IP-co-simulation interface. The coupled simulation
approach is tested on the basis of driving manoeuvres of tank trucks and silo vehicles and
Downloaded by [University of Hong Kong Libraries] at 13:22 11 October 2014
the influence of some characteristic design parameters on the stability of driving dynamics are
studied.
2. Cargo models
Dynamic simulations of sloshing cargo in moving containers require cargo models that are
suitable to incorporate moving boundary conditions, free surfaces and three-dimensional cargo
motion. All three requirements are best met by Lagrangian particle methods that model the
cargo as a set of particles that carry certain model-dependent state variables. Particle motion
is unconstrained, and boundary conditions and particle interactions are modelled via applied
forces such as contact forces or force laws that emerge from constitutive state equations, for
example the dynamic equations of fluid motion. To model a granular cargo, we chose the
discrete element method (DEM) [4] and to model a fluid cargo, we chose smoothed particle
hydrodynamics (SPH) [5]. In Figures 1 and 2, snapshots from simulations with two different
cargo models are depicted.
The DEM models the granulates by a huge number of rigid bodies with a simple geometry, for
example spheres, whose interactions are modelled via a penalty contact approach. Although
fairly simple on the microscopic level, on a macroscopic level with millions of contacts,
such granular systems can exhibit a very complex behaviour featuring phenomena such as
jamming, avalanches and piling. As there are typically from several thousands up to millions
of particles involved in a particle simulation, efficient contact detection is a key feature to
identify adjacent particle pairs whose contact forces have to be calculated. There is a variety
of different time-integration methods [6] that may be used to integrate the dynamics of particle
systems. Among those, step-size controllable energy conserving unconditionally stable one-
step methods such as the Newmark-schemes are especially well suited as they allow for large
time-step sizes even for very stiff materials, they are robust and they do not require to store as
many particle states as it is necessary when using multistep methods. This is especially useful
as, due to the high amount of particles, memory consumption is one of the limiting factors
in particle simulations. In Pasimodo, complicated freeform surface boundaries are defined
as triangular surface meshes. The contacts between particles and the mesh geometry, which
is defined relative to a moving coordinate system, are handled by the same penalty approach
as the inter-particle contacts. Thus, to represent the moving geometry of an MBS body, it is
sufficient to compute the motion of the body’s associated coordinate system. Contacts between
particles and the MBS geometry are detected automatically. The resulting contact forces and
torques are accumulated with respect to the centre of gravity of the MBS bodies.
Vehicle System Dynamics 5
Downloaded by [University of Hong Kong Libraries] at 13:22 11 October 2014
Figure 2. Transport vehicle loaded with a fluid cargo (SPH). Left: SPH particles displayed as kernel-spheres; right:
reconstruction of the free fluid surface.
For the simulations, an artificial granulate material is used. The material consists of spherical
particles that interact via visco-elastic penalty forces. The friction between the particles is
modelled by a Coulomb model. Moreover, switching between sticking and slipping friction
is considered. By changing the contact parameters (Table 1), it is possible to model a large
variety of different granular materials.
Neglecting viscosity, the motion of a fluid can be described by the Euler equations, which are
given by the acceleration equation
dv 1
= − ∇P + f + g, (1)
dt ρ
where ρ denotes the density of the fluid, P the pressure field, v the velocity, g the vector of
gravity and f external forces, for example from contacts with rigid objects or solid boundaries,
and the continuity equation
Downloaded by [University of Hong Kong Libraries] at 13:22 11 October 2014
dρ
= −ρ∇ · v. (2)
dt
Unlike many other methods in computational fluid dynamics such as the finite difference or
finite element method, the SPH method does not need a grid to calculate the required spatial
derivatives. Instead, they are found by analytical differentiation of interpolation formulae,
which allow any function to be expressed in terms of its values at a set of disordered points,
called particles in this method [5]. As the particles move along the velocity field of the fluid,
free surfaces are automatically tracked [7,8]. This marks a great advantage in comparison with
grid-based methods, where free surfaces lead to great difficulties such as only partially filled
cells.
SPH is very suitable as a complement to the DEM method in Pasimodo, as the computational
procedure and the data management are quite the same: contact detection, force evaluation
between pairs of particles and time integration. In fact, SPH particles, which correspond to
moving kernel functions, can be seen as a kind of non-rotational spheres with an interaction
radius defined by the radius of the support of the kernel function. Thus, contacts between
the SPH particles and the triangular surface meshes can be handled by the same approach as
contacts between DEM particles and the mesh geometry presented in the previous section.
However, for the time integration of the SPH equations, a different algorithm has been chosen
since it also has to follow the changes in density. In this work, a predictor–corrector scheme
is used that is based on the leapfrog algorithm and allows time-step control [9].
Designing the interpolation formula, the value of an arbitrary function A at position r can
be understood as the shifted delta distribution δ applied to the function A, i.e.
Therefore, the value A(r) can be approximated by approximating the delta distribution. In
order to do so, a kernel function W (x, h) is chosen, which is smooth enough and compactly
supported. Moreover, it needs to satisfy the conditions
W (r − r ) dr = 1 (4)
and
lim f (r )W (r − r , h) dr = δ(r − r), f (r ) = f (r) (5)
h→0
for any function f . The so-called smoothing length h > 0 determines the size of the support
of the kernel function W and, therefore, the number of neighbour particles that have influence
on a certain particle. For the implementation in Pasimodo, h corresponds to the radius of
the particle, which is used for contact detection. A common choice as a kernel function is
Vehicle System Dynamics 7
defined by
AI (r) = A(r )W (r − r ) dr . (6)
where Ab is the value of function A at the position rb of a particle b with mass mb and density
ρb . In theory, the summation interpolant (7) is a sum over all particles, but in fact it is just a
sum over the neighbour particles that are located within the support of the kernel function.
For approximating the pressure gradient in the acceleration equation (1) and the divergence
of the velocity field in the continuity equation (2) using the summation interpolant (7), it is
now only necessary to calculate the according spatial derivatives of the kernel function W .
The approximations can be chosen to yield equations that are Galilean invariant and conserve
linear and angular momentum. Mass conservation is given automatically as the particles do not
change their masses over time. For further details, see [10]. This leads to ordinary differential
equations in time and makes the pressure gradient a force between pairs of particles. We chose
in the following simulations water as a cargo fluid. Parameters for the SPH fluid model are
summarised in Table 2.
3. Dynamic co-simulation
For the simulation of the vehicle, the classical MBS approach [11,12] is chosen. The com-
mercial MBS modelling and simulation program Simpack is used to create a model of the
truck with 17 degrees of freedom. Simpack provides a co-simulation interface that allows for
data interchange with Matlab/Simulink. Any kind of state or force/torque data can thus be
exchanged. To couple Pasimodo via Matlab/Simulink with Simpack, advantage is taken of
Pasimodo’s plugin interface. This interface enables users to programme custom subroutines in
C++ that can, for example, be used to dynamically modify particle coordinate system states.
For the particle-MBS co-simulations, the plugin interface is used to transfer the state variables
of the MBS tank from Simulink to Pasimodo to position the tank’s surface geometry. In the
same way, the resulting contact forces and torques with respect to the tank’s centre of gravity
are transferred back to Simulink. The data exchange is carried out via a TCP/IP interface both
between Simulink and Pasimodo and between Simulink and Simpack. The Simulink model
for the silo vehicle co-simulation is depicted in Figure 3. To enable a synchronous simula-
tion of both dynamic models, a fixed time interval for data exchange is used. However, as
particle dynamics usually has a much smaller time scale than the dynamics of the MBS truck
8 F. Fleissner et al.
Downloaded by [University of Hong Kong Libraries] at 13:22 11 October 2014
system, a smaller time-step size is used for the time integration of the cargo, i.e. the particle
system. For the simulation of the hard granular medium considered in this work and to model
the incompressibility of water approximately, small time-step sizes t ≈ 10−5 s are required.
Thus, the simulation of a double lane change manoeuvre of 8 s duration on a Pentium IV
requires 16–20 h of computing time.
4. Vehicle model
Different aspects of the vehicle are of interest in the MBS model and in the DEM or SPH
particle model. In the MBS, the particle forces acting on the tank are considered as applied
Figure 5. Different tank geometries. Granulate particles are depicted in their equilibrium positions.
forces. As the particle forces are calculated by the particle simulator, no representation of
the tank geometry is required in the MBS. The particle simulator, on the other hand, only
considers the constrained motion of the tank whose state variables are integrated as part of
the MBS. Thus, in Pasimodo, no representation of the remaining MBS except of the tank is
required. The silo vehicle is modelled as an MBS with 17 degrees of freedom (Figure 4).
The tyre forces are computed by means of the tyre similarity model [13]. To account for a
driver, two additional degrees of freedom are added, which are influenced by a feedback-
controlled driver model. All relative rotations are defined as Cardan angles. The rotations of
the front wheel around the vertical axis are controlled by a PID driver model [14] that uses
the lateral deviation from the ideal trajectory as a controller input. In Table 3, some important
simulation parameters of the vehicle model are listed. The contact detection between particles
and the tank geometry requires a geometric representation of the rigid tank. This geometry
description is supplied as a triangular surface mesh. To observe the influence of a subdivision
of the tank into compartments, we performed simulations with different tank geometries. The
first configuration is a cubical tank with only one compartment, the second configuration is a
cubical tank with two compartments of equal volume and the third configuration is a cylindrical
tank with three compartments (Figure 5).
First, tanks filled with granular media are considered. For the simulation of driving manoeu-
vres, the cargo particles in the tank are considered to be initially at rest. Therefore, initial
settling simulations are performed first. The cargo particles that are initially arranged in a
regular cubic-face-centred lattice settle under gravity. To accelerate the settling process, the
dissipation of kinetic energy was temporarily drastically increased by using larger damping
parameters of the particle contact model. The settling process is considered to be finished
when the rate of change of the total kinetic energy of the particle system is negligible. Using
the results of the settling simulations as initial conditions, the actual driving manoeuvres
were started. To investigate the influence of different tank designs on the driving behaviour
10 F. Fleissner et al.
and to show the influence of the particle cargo model on the driving stability, especially
when compared with a model that regards the cargo as rigid, two classical benchmark driving
manoeuvres are performed.
As the first manoeuvre, a full braking is performed to observe how a dynamically changing
load distribution can lead to tyre locking. During the braking process, the centre of gravity
of the cargo moves to the front of the tank. This results in a discharge on the back wheels
leading to a severe reduction of normal contact force and thus tyre friction force resulting in
Downloaded by [University of Hong Kong Libraries] at 13:22 11 October 2014
brake locking. A subdivision of the tank into compartments reduces the relative longitudinal
particle motion and thus reduces the unloading of the rear wheels. To determine the limit of
braking stability in terms of brake locking, a series of simulations are performed with different
brake torques. The brake torque configurations are listed in Table 4. The sloshing motion of
the particle cargo is depicted in Figures 6 and 7.
The results of the full braking manoeuvres with a single compartment silo vehicle are
depicted in Figure 8. Load cases 1 and 2 show a stable braking motion with constant decel-
eration. For load case 3, the load shift from the rear to the front wheels leads to a significant
reduction of friction force at the rear wheels and thus to a locking of the rear tyres. We inves-
tigated whether a baffle that divides the tank into two compartments can remedy the tyre
Braking torques
Load case Front wheels (Nm) Rear wheels (Nm)
1 10,000 5000
2 13,000 6500
3 16,000 8000
Figure 6. Snapshots of the sloshing motion of a granular cargo during a full-braking manoeuvre with load case 3
and a cubical tank with one compartment.
Figure 7. Snapshots of the sloshing motion of a granular cargo during a full-braking manoeuvre with load case 3
and a cylindrical tank with three compartments.
Vehicle System Dynamics 11
4
x 10 longitudinal force rear tyre longitudinal velocity
2
20
[m/s]
[N]
1
10
0 0
0 1 2 3 0 1 2 3
t[s] t[s]
normalised normal force rear tyre relative angular velocity rear tyre
1 40
Downloaded by [University of Hong Kong Libraries] at 13:22 11 October 2014
[rad/s]
0.8
[−]
20
0.6
0
0 1 2 3 0 1 2 3
t[s] t[s]
Figure 8. Full braking manoeuvre with different braking torque load cases with a single compartment cubical
tank.
15000 20
[m/s]
10000
single compartment 10
5000
double compartment
0 0
0 1 2 3 0 1 2 3
t[s] t[s]
normalised normal force rear right tyre relative angular velocity rear right tyre
1 40
[rad/s]
0.8
[−]
20
0.6
0
0 1 2 3 0 1 2 3
t[s] t[s]
Figure 9. Influence with respect to full braking of a subdivision of the tank into compartments.
locking. Figure 9 shows simulation results of a comparison for a full braking with load case
3. As expected, the baffle allows for a higher deceleration without brake locking.
The second benchmark manoeuvre is a double lane change. The sloshing motion of the
cargo has a significant impact on the driving stability, as the whole particle system behaves
as a huge non-linear spring-dashpot system that severely influences the lateral motion. We
compared the particle cargo with a rigid cargo of equal weight. Its position relative to
12 F. Fleissner et al.
the tank and its inertia properties were chosen in such a way that they reflect the proper-
ties of the particle system at rest. The lateral sloshing of the particle cargo is depicted in
Figure 10.
The influence of the lateral motion of the particle cargo on the driving stability turned out to
be positive in comparison to the rigid cargo for the investigated scenarios, see Figure 11. The
rolling motion of the frame-tank system is significantly reduced. The dynamic reduction of
Downloaded by [University of Hong Kong Libraries] at 13:22 11 October 2014
Figure 10. Snapshots of the sloshing motion of a granular cargo during a double lane change manoeuvre.
4
0.2
[rad]
[m]
2 0
−0.2
0
−0.4
0 2 4 6 8 0 2 4 6 8
t[s] t[s]
normalised vertical force rear right tire 4 lateral force rear right tire
x 10
3 2
2 0
[N]
[−]
−2
1
−4
0
0 2 4 6 8 0 2 4 6 8
t[s] t[s]
Figure 11. Double lane change manoeuvre with two different initial velocities and a comparison to a rigid cargo.
Vehicle System Dynamics 13
the normal contact force that is crucial for tyre friction is generally larger for the simulations
with a rigid cargo.
Next it should be analysed with some simulations whether fluid cargo and granular cargo yield
a different dynamic behaviour. To investigate the dynamics of the truck loaded with a liquid
cargo, we performed again a full braking and a double lane change manoeuvre. In Figure 12,
the fluid motion during a full braking is depicted. The macroscopic behaviour of the fluid is
Downloaded by [University of Hong Kong Libraries] at 13:22 11 October 2014
Figure 12. Sloshing motion of a fluid cargo (depicted as particles) during a full braking manoeuvre
with load case 3 and a cylindrical tank with three compartments.
2 0
−0.2
0
−0.4
0 2 4 6 8 0 2 4 6 8
t[s] t[s]
4
normalised vertical force rear right tyre x 10 lateral force rear right tyre
2 2
0
[N]
[−]
1
−2 granular cargo
−4 liquid cargo
0
0 2 4 6 8 0 2 4 6 8
t[s] t[s]
Figure 13. Comparison of a granular and a liquid cargo in a double lane change manoeuvre with v = 17.5 m/s.
14 F. Fleissner et al.
Downloaded by [University of Hong Kong Libraries] at 13:22 11 October 2014
Figure 14. Snapshots of the sloshing motion during a double lane change manoeuvre with a single compartment
tank and a fluid cargo.
7. Conclusion
It was demonstrated how a co-simulation approach that couples Lagrangian particle methods
and classical multibody dynamics can be used to predict the stability of driving manoeuvres
of transport vehicles. This method can be used to investigate the impact of different tank
designs on the stability of the transport vehicle system. Comparisons between two different
tank designs showed the positive effect of a subdivision of the tank into compartments in
terms of braking stability. Moreover, the simulations show that the lateral motion of a sloshing
cargo can under certain circumstances be beneficial with respect to the rolling stability in lane
change manoeuvres. The comparison of the granular and the fluid cargo during a double lane
change manoeuvre suggests that the type of cargo, i.e. continuous fluid or non-continuous
granular medium, does not have a strong influence on driving dynamics. The mass and density
properties of the transported cargo are of greater influence.
Acknowledgements
The authors would like to thank M.Sc. Vincenzo D’Alessandro from Polytecnico Milano, Italy, for providing the
Simpack vehicle model.
References
[1] R.J. LeVeque, Finite Volume Methods for Hyperbolic Problems, Cambridge University Press, New York, 2002.
[2] Pasimodo. software available at http://www.itm.uni-stuttgart.de/research/pasimodo/pasimodo_en.php.
[3] Simpack. software available at http://www.simpack.com.
[4] P.A. Cundall and O.D.L. Strack, A discrete numerical model for granular assemblies, Geotechnique 29 (1979),
pp. 47–56.
[5] J.J. Monaghan, Smoothed particle hydrodynamics, Rep. Progr. Phys. 68 (2005), pp. 1703–1759.
[6] F. Fleißner, T. Gaugele, and P. Eberhard, Applications of the discrete element method in mechanical engineering,
Multibody Syst. Dyn. 18(1) (2007), pp. 81–94.
Vehicle System Dynamics 15
[7] J.J. Monaghan, Simulating free surface flows with SPH, J. Comput. Phys. 110 (1994), pp. 399–406.
[8] R. Dalrymple and B. Rogers, Numerical modeling of water waves with the SPH method, Coastal Eng. 53 (2006),
pp. 141–147.
[9] J.J. Monaghan, A. Kos, and N. Issa, Fluid motion generated by impact, J. Waterway 129(6) (2003), pp. 250–259.
[10] A. Lehnart, A smoothed particle hydrodynamics approach for treating the Euler-equations (in German), Thesis
DIPL-125, Institute of Engineering and Computational Mechanics, Universtity of Stuttgart, 2007.
[11] W. Schiehlen and P. Eberhard, Technische Dynamik - Modelle für Regelung und Simulation (in German), Teubner,
Wiesbaden, 2004.
[12] A.A. Shabana, Dynamics of Multibody Systems, Cambridge University Press, Cambridge, 1998.
[13] H.B. Pacejka, Tyre and Vehicle Dynamics, Butterworth Heinemann, Oxford, 2002.
[14] M. Plöchl and J. Edelmann, Driver models in automobile dynamics application, Veh. Syst. Dyn. 45 (2007),
pp. 699–741.
Downloaded by [University of Hong Kong Libraries] at 13:22 11 October 2014