Professional Documents
Culture Documents
Non-linear Belt
Transient Analysis
A Hybrid Model for Numerical
Belt Conveyor Simulation
A. Harrison, USA
Transient conveyor belt behaviour if
often the course of belt failure. Many
belt tension calculation models used to
avoid such failure are likely to result in
an overestimate of belt running tension
and, thus, in expensive over dimension-
ing. The hybrid model presented in this
paper provides a solution to this and
several other belt conveyor design
problems.
F
rictional and rolling losses along a running conveyor are dis-
cussed due to their important influence on wave propaga-
tion during starting and stopping. Hybrid friction models
allow belt rubber losses and material flexing to be included in
the initial tension calculations prior to any dynamic analysis. forces along the belt that may have deleterious affects on the
Once running tensions are defined, a numerical integration belt and its structure.
method using non-linear stiffness gradients is used to generate
transient forces during starting and stopping. A modified Euler As a first step in the analysis of dynamic forces, the steady state
integration technique is used to simulate the entire starting and forces need to be quantified. Many methods exist for comput-
stopping cycle in less than 0.1 seconds. The procedure enables a ing the tension distribution along a running belt, however not
faster scrutiny of unforeseen conveyor design issues such as low all models will behave equivalently because the specification of
belt tension zones and high forces at drives. rolling and indentation losses are often difficult to quantify in
practice.
1. Introduction Basic methods for calculation belt tensions exist in the DIN
Standards [1] and CEMA [2]. The CEMA method has advantages
Dynamic analysis of long conveyor belts with distributed mass in analysis since idler rolling losses can be used as an input (the
and elasticity depends on initial conditions and the tension dis- Kx factor) however the flexural and indentation losses at the idler
tribution around the running belt. Conveyor belt tension is a interface (the Ky factor) is not always easy to predict. For a dy-
complex function of friction and gravitational loads acting on namic analysis to properly describe the inertial fluctuations along
every element of the belt. Transient behaviour of a belt occurs a belt during the transient phase, the static tensions are a critical
during the staring and stopping phase, leading to high and low precursor to solving the problem.
Assuming that a static analysis model can 2. Hybrid Belt Friction Model
provide a sufficient number of input variables
that define the losses, the initial conditions for a 2.1 Friction Equations
dynamic analysis can be set. Inertial fluctuations
on starting or stopping are compounded by the up- As suggested in the Introduction, running tensions form the ini-
hill or downhill parts of the profile, as well as the addi- tial conditions for a dynamic simulation. To accommodate re-
tional non-linear wave dispersion affects caused by high belt cent developments in low indentation-loss rubber materials, a
sag at low tension areas. To make dynamic models tractable, the hybrid friction model [9] is developed to combine the basic me-
simplest solution is to ensure that belt sag is controlled to within chanics of motion with new methods utilizing rubber properties.
a few percent. Such a condition can be ensured in the steady For a hybrid friction model, length segment L has a tension
state but not always in the dynamic state. change Fj defined by Eq. (1):
A number of methods and techniques have been published that Fj = L · g [R + (BI + BF) + MF,z + V] + Q · v + P + O (1)
describe various aspects of dynamic analysis. The realization that
starting dynamic effects could be minimized using an “S-curve” with:
velocity history for starting was published as early as 1983 [3].
The solution applied a well known method in the Calculus of H
V = m• · __
L
Variations to prove that a cycloid-shaped velocity curve will pro-
duce smooth belt starting, free of sudden belt element displace- PQ = vb − vl
ments that lead to wave propagation. In this way, the dynamic
starting problem becomes tractable and a wave-equation model where:
[4] can be applied along with the convolution integral to predict
the dynamic response of the starting motion. L length of conveyor [m]
Stopping and braking are more complex to analyze since R Rotating resistance of all idlers [kg/m]
this condition cannot be controlled in most cases. Pow-
ered stopping using a reverse “S-curve” can be ap- g gravitational acceleration, g = 9.81 [m/s2]
plied to smoothly bring a belt to a stop, however
this solution does not cater to power outages. BI Belt indentation [kg/m]
Some aspects of the profile such as valleys
cause potential dynamic problems that can BF belt flexure [kg/m]
add to the impulse problem due to non-lin-
ear effects similar to jamming. A conveyor MF,z Material flexure resistance [kg/m]
having a significant material lift compo- (z = c or z = r carry and return runs)
nent or high rolling resistance, or both,
will stop rapidly. The resulting impulse m mass to be lifted [kg/m]
can be severe.
H lifting height, H can be negative [m]
The impulse can be calculated as if the en-
tire belt were inelastic without a distributed PQ force to accelerate material [N]
mass, as defined in the 1980s [5]. In this refer-
ence, criteria were established to inform the Q mass flow [kg/s]
belt designer when the conveyor would have a
high stress condition on stopping. In that work, any vb belt speed [m/s]
stopping time quicker than 5 wave cycles will present a
high-stress problem. vl loading speed [m/s]
Given that a static tension analysis can be achieved with some P force to rotate all pulleys [N]
reliability, the addition of dynamic forces to the static conditions
is still an ongoing part of international research. Many papers O Forces for all other accessories, scrapers and [N]
discuss alternatives to wave solutions, such as mass-spring mod- special losses (eg. turnovers)
els [6] and more recently non-linear mass-spring models [7,8]. In
the past, a closed-form wave solution was a convenient way of With respect to each element j, it is:
obtaining the modal oscillations in a belt, but the problem has
always been that positional variations in damping and strain are
very difficult to treat with mathematically-based solutions.
[ e
R(j) = Kta · Ks(j) + Az(j) · (_ 2v ) b · ____
−c · T
S (j)
z
]
where a, b, c are fractions.
With modern computers, simulation of the dynamic behaviour
d · tanδ · v · g · M + Mm,z(j) ] · Sz(j) ] 1.33
e
of multi-element distributed mass systems can be rapid. Assum- BI(j) = ________
g · f(d ) [ [ b
ing a mass-spring model can be formulated to describe a damped i,z
belt element, there will be practical advantages over wave mod- where d, e are fractions, and
els particularly in the way boundary conditions are mathemati-
cally treated in areas such as take-ups and multiple drives. tanδ rubber loss (T) = f(v,Sz(j),t) + f(T) as functions.
t viscoelastic time constant of rubber cover at a [Hz] It is not necessary to show belt tensions around the conveyor
frequency; t = v/Sz profile, since they are somewhat obvious from Fig. 1. However, a
display of power as a function of temperature for various model
f(di,z) a function of idler diameter di,z, types is instructive. Fig. 2 shows the relationship between power
f(di,z) = k1. di,z + k2 (effective tension) and temperature for the hybrid model in rela-
(z = c or z = r carry and return runs) tion to the other models used for comparison. On examination
of the analysis for the particular example, one notes that:
S idler spacing [m]
a) C
EMA and ViscoAH methods over estimate the power by a
A idler drag [kg] significant amount. A design based on CEMA would result in
large expensive drives and the belt cost would be substantial.
v speed [m/s] The consequence would be a larger than required starting
impulse, which would substantially increase starting ten-
T temperature [°C] sions, further impacting pulley loads, splice loads and the
S (j)n
development of more severe take-up oscillations.
BF(j) = m* · Kta · (Mb + Mm,z) · ____
z
L(j)
b) Th
e Non-Lin solution is similar to a DIN solution when the
where m*, n are fractions. DIN input frictions are set to Fc = 0.01 and Fr = 0.008. In fact,
site measurements of this conveyor yields a DIN ff = 0.0075.
For material on the carrying side it is:
[
c) W
hen empty, a Non-Lin model requires less power than DIN
]
s
S (j)
MF(j) = [p + q · v] · 1.66 · ___
c
S (j)
would predict. This finding may explain some unpredictable,
r
almost chaotic low tensions and dynamic slack at the tail un-
with: der certain operating conditions. Low tail-zone tension will
cause non-linear dynamic effects that stem from wave dis-
(0.003 < m < 0.008), (n =3/2), (0.2 < p < 3), persion in this zone of the conveyor.
(0.02 < q < 0.6), (1 < s < 3) is a real value
Fig. 2: omparison of power consumption of a 5 % slope conveyor,
C
using various models, including CEMA, DIN and non-lin.
2.2 Application
In the above equations, Mb and Mm represent belt and material
mass in kilogram per meter, respectively. To determine the driv-
ing power Pe and maximum belt tension I1 in relation to slack
side tension, a summation of each element of force Fj around the
belt results in:
3. Dynamic Simulation
One way of conducting a transient analysis is to superimpose
transient wave model solutions on the static tension analysis
[10], treating a belt as a continuum. Applying the superposition
principle to add modal vibrations to static conditions during Fig. 3: Model of a simulation element j.
transient phases can be mathematically problematical when
there is a mismatch of wave velocities near zero of the impulse
function. Wave energy that remains active after the belt mass Each element of a belt loop is analyzed by summing the inertial,
comes to rest during starting or stopping has to be properly body, restoring, damping and externally imposed forces. The so-
treated otherwise the predicted belt dynamic force histories may lution contains a diagonal mass matrix [M], non-linear stiffness
be incorrectly modeled. and damping matrices [K] and [C] respectively, as well as a col-
umn matrix array for external and body forces [F + W + R]. Col-
On the other hand, finite element models used to simulate the umn arrays for all elements of the problem include the accelera-
dynamic motion of a belt require the addition of boundary tion [a], velocity [V] and displacement [U]. Note that the accel-
conditions that have to be tracked and maintained throughout eration for each element is aj = Üj.
the calculations. Cumulative errors in velocities or displace-
ments at element boundaries can result in erroneous dynamic Two approaches are available to simulate belt running with in-
solutions. cluded dynamic oscillations of all elements. In a relative motion
solution, a driving force Fj(t) is applied to one of the masses in
Continuous simulation during the entire process of starting and the F matrix (the drive):
stopping uses numerical integration with a small time step. The
modeling presented here uses time steps of between 0.001 s to [M] [a] + [C] [V] + [K] [U] = [F + W + R] (2)
0.01 s. In the past, slower computers meant that longer time
steps were required so that modeling time was manageable. In Alternatively, the belt loop motion can be simulated using a
that situation, well known 5th–order Runge-Kutta approxima- moving support where applied motion is independent of the ele-
tions were devised to reduce “look-forward” error in the simu- ment motion (wave action). Support displacement (of selected
lated displacements and velocities. masses) is defined by a new array [𝛙]. Derivatives of [𝛙] form the
support velocity array [𝛟]. Absolute element motion is the re-
Modern computers are now significantly faster and higher preci- sult:
sion processors reduce computational truncation. In effect, these
combinations of technological development enable much small- [M] [a] + [C] [V] + [K] [U] = [C] [𝛟] + [K] [𝛙] + [W + R] (3)
er time steps in numerical integration, reducing the need for er-
ror correction. Simpler and faster Euler-type numerical integra-
tors now work as well, if not better, than approximating methods 3.2 Simulating the Dynamic Motion
of the past.
To simulate the motion of each element j along a belt by the
relative motion method, Eq. (2) is used. Each element’s displace-
3.1 Elemental Forces ment Uj and velocity Vj is computed in a program loop by incre-
menting variable s:
Euler-X is the term used to define an integration engine devel-
oped for the purpose of high-speed simulation of the entire f(Uj(s)) = 2 Uj(s) – Uj-1(s) – Uj+1(s) (4a)
starting and stopping process of a conveyor belt. To begin, Fig. 3
shows a typical element of the simulation, in which body and f(Vj(s)) = 2 Vj(s) – Vj-1(s) – Vj+1(s) (4b)
external forces are accommodated. Wj + Rj – Kj f(Uj(s)) – Cj f(Vj(s)) + Fj(t)
aj(s) = _________________________
Mj
(4c)
Long belts can be divided into “j” distributed mass-elastic seg-
ments, where j > 6. Each segment “j” is defined in terms of its th ) = Uj(s) + Vj(s) ts
Vj(s + h/ts) = Vj(s) + aj(s) ts : Uj(s + __ (4d)
s
mechanical characteristics such as mass Mj, positional depend-
ent non-linear stiffness Kj connecting elements relative to posi- Initial conditions on U(s) and V(s) are required before aj is calcu-
tion Uj, body forces Wj, external time-dependent forces Fj(t) ap- lated. Each calculation is separated by a unit time step ts with a
plied to mass Mj, velocity damping Cj between masses and rolling perturbation h (a variable). When Eq. (3) is used, support dis-
force Rj for each mass, with: placement ψ(t) in the [𝛙] array is required. For a linear velocity
ramp or constant acceleration on starting and stopping, the
Rj = f(R(j), BI(j), BF(j), MF(j)) function takes the form ψ(t) = f(a·t, b·t2), a and b are constants.
Each element may have different mass, stiffness and damping