This example illustrates how Abaqus/Explicit can be used to model the rolling process, even though rolling is normally applied at
relatively low speeds.
Hot rolling is a basic manufacturing technique used to transform preformed shapes into a
form suitable for further processing. Rolling processes can be divided into different
categories, depending on the complexity of metal flow and on the geometry of the rolled
product. Finite element computations are used increasingly to analyze the elongation and
spread of the material during rolling (Kobayashi, 1989). Although the forming process is
often carried out at low roll speed, this example shows that a considerable amount of
engineering information can be obtained by using the explicit dynamics procedure in Abaqus/Explicit to model the process.
The rolling process is first investigated using plane strain computations. These results
are used to choose the modeling parameters associated with the more computationally
expensive three-dimensional analysis.
Since rolling is normally performed at relatively low speeds, it is natural to assume
that static analysis is the proper modeling approach. Typical rolling speeds (surface
speed of the roller) are on the order of 1 m/sec. At these speeds inertia effects are
not significant, so the response—except for rate effects in the material behavior—is
quasi-static. Representative rolling geometries generally require three-dimensional
modeling, resulting in very large models, and include nonlinear material behavior and
discontinuous effects—contact and friction. Because the problem size is large and the
discontinuous effects dominate the solution, the explicit dynamics approach is often
less expensive computationally and more reliable than an implicit quasi-static solution
technique.
The computer time involved in running a simulation using explicit time integration with a
given mesh is directly proportional to the time period of the event. This is because
numerical stability considerations restrict the time increment to
where the minimum is taken over all elements in the mesh, is a characteristic length associated with an element, is the density of the material in the element, and and are the effective Lamé's constants for the material in the element.
Since this condition effectively means that the time increment can be no larger than the
time required to propagate a stress wave across an element, the computer time involved
in running a quasi-static analysis can be very large. The cost of the simulation is
directly proportional to the number of time increments required, if remains constant, where T is the time period of
the event being simulated. ( will not remain constant in general, since element distortion will
change and nonlinear material response will change the effective Lamé
constants and density; but the assumption is acceptable for the purposes of this
discussion.) Thus,
To reduce n, we can speed up the simulation compared to the
time of the actual process; that is, we can artificially reduce the time period of the
event, T. This will introduce two possible errors. If the
simulation speed is increased too much, the inertia forces will be larger and will
change the predicted response (in an extreme case the problem will exhibit wave
propagation response). The only way to avoid this error is to find a speedup that is not
too large. The other error is that some aspects of the problem other than inertia
forces—for example, material behavior—might also be rate dependent. This implies that we
cannot change the actual time period of the event being modeled. But we can see a simple
equivalent—artificially increasing the material density, , by a factor reduces n to , just as decreasing T to does. This concept, which is called “mass scaling,” reduces the ratio
of the event time to the time for wave propagation across an element while leaving the
event time fixed, thus allowing treatment of rate-dependent material and other behaviors
while having exactly the same effect on inertia forces as speeding up the time of
simulation. Mass scaling is attractive because it allows us to treat rate-dependent
quasi-static problems efficiently. But we cannot take it too far or we allow the inertia
forces to dominate and, thus, change the solution. This example illustrates the use of
mass scaling and shows how far we can take it for a practical case.
The trial and error method works well for most generic quasi-static problems; however,
for rolling processes Abaqus/Explicit can set the mass scaling factor automatically based on the rolling geometry and mesh
properties. An acceptable value for the stable time increment is calculated, and the
appropriate mass scaling factor is applied on an element-by-element basis. The value of
the stable time increment is based on the average element length in the rolling
direction, the average velocity of the product through the rollers, and the number of
nodes in the cross-section of the mesh.
Problem description
A steel plate of an original square cross-section of 40 mm by 40 mm and a length of
92 mm is reduced to a 30 mm height by rolling through one roll stand. The radius of
the rollers is 170 mm. The single roller in the model (taking advantage of symmetry)
is assumed to be rigid and is modeled as an analytical rigid surface. The isotropic
hardening yield curve of the steel is taken from Kopp and Dohmen (1990). Isotropic
elasticity is assumed, with Young's modulus of 150 GPa and Poisson's ratio of 0.3.
The strain hardening is described using 11 points on the yield stress versus plastic
strain curve, with an initial yield stress of 168.2 MPa and a maximum yield stress
of 448.45 MPa. No rate dependence or temperature dependence is taken into
account.
Coulomb friction is assumed between the roller and the plate, with a friction
coefficient of 0.3. Friction plays an important role in this process, as it is the
only mechanism by which the plate is pulled through the roll stand. If the friction
coefficient is too low, the plate cannot be drawn through the roll stand. Initially,
when a point on the surface of the plate has just made contact with the roller, the
roller surface is moving faster than the point on the surface of the plate and there
is a relative slip between the two surfaces. As the point on the plate is drawn into
the process zone under the roller, it moves faster and, after a certain distance,
sticks to the roller. As the point on the surface of the plate is pushed out of the
process zone, it picks up speed and begins to move faster than the roller. This
causes slip in the opposite direction before the point on the surface of the sheet
finally loses contact with the roller.
For plane strain computations a half-symmetry model with
CPE4R elements is used. For the
three-dimensional computations a one-quarter symmetry model with
C3D8R elements is used. The roller is
modeled with analytical rigid surfaces for both the two-dimensional and
three-dimensional cases. For quasi-static rolling problems perfectly round
analytical surfaces can provide a more accurate representation of the revolved
roller geometry, improve computational efficiency, and reduce noise when compared to
element-based rigid surfaces.
The roller is rotated through 32° at a constant angular velocity of 1 revolution per
second (6.28 rad/sec), which corresponds to a roller surface speed of 1.07 m/sec.
The plate is given an initial velocity in the global
x-direction. The initial velocity is chosen to match the
x-component of velocity of the roller at the point of
first contact. This choice of initial velocity results in a net acceleration of zero
in the x-direction at the point of contact and minimizes
the initial impact between the plate and the roller. This minimizes the initial
transient disturbance.
In all but one of the analyses performed in this example, the masses of all elements
in the model are scaled by factors of either 110, 2758, or 68962. These scaling
factors translate into effective roller surface speeds of 11.2 m/sec, 56.1 m/sec,
and 280.5 m/sec. An alternative, but equivalent, means of mass scaling could be
achieved by scaling the material mass density by the aforementioned factors. In one
analysis, automatic mass scaling is used.
The element formulation for the two-dimensional (using
CPE4R elements) and three-dimensional
(using C3D8R elements) analyses uses
the pure stiffness form of hourglass control. The element formulation is selected
using section controls. In addition, the three-dimensional model (using
C3D8R elements) uses the centroidal
kinematic formulation. These options are economical yet provide the necessary level
of accuracy for this class of problems. Two- and three-dimensional analyses using
the default hourglass control formulation, the combined hourglass control
formulation, and the enhanced hourglass control formulation are included for
comparison. For the three-dimensional case, both the orthogonal kinematic
formulation and the centroidal kinematic formulation are considered.
For the sole purpose of testing the performances of the modified triangular and
tetrahedral elements, the problem is also analyzed in two dimensions using
CPE6M elements and in three dimensions
using C3D10M elements.
Results and discussion
Table 1 shows the effective rolling speeds and the relative
CPU cost of the cases using the element
formulations recommended for this problem. The relative costs are normalized with
respect to the CPU time for the two-dimensional
model (using CPE4R elements) with the
intermediate mass scaling value. In addition, Table 2 compares the relative CPU cost and accuracy
between the different element formulations of the solid elements using the
intermediate mass scaling value.
Plane strain rolling (CPE4R
elements)
A plane strain calculation allows the user to resolve a number of modeling
questions in two dimensions before attempting a more expensive three-dimensional
calculation. In particular, an acceptable effective mass scaling factor for
running the transient dynamics procedure can be determined.
Figure 1 through Figure 3 show contours of equivalent plastic strain for the three mass scaling factors
using the stiffness hourglass control. Figure 4 through Figure 6 show contours of shear stress for the same cases. These results show that
there is very little difference between the lowest and the intermediate mass
scaling cases. All the results are in good agreement with the quasi-static
analysis results obtained with Abaqus/Standard. The results of the largest mass scaling case show pronounced dynamic
effects. Table 1 shows the relative run time of the quasi-static calculation, and Table 2 compares the different element formulations at the same level of mass
scaling. The intermediate mass scaling case gives essentially the same results
as the quasi-static calculation, using about one-thirteenth of the
CPU time. In addition to the savings
provided by mass scaling, more computational savings are achieved using the
stiffness hourglass control element formulation; the results for this
formulation compare well to the results for the computationally more expensive
element formulations.
Three-dimensional rolling (C3D8R
elements)
We have ascertained with the two-dimensional calculations that mass scaling by a
factor of 2758 gives results that are essentially the same as a quasi-static
solution. Figure 7 shows the distribution of the equivalent plastic strain of the deformed sheet
for the three-dimensional case using the centroidal kinematic formulation and
stiffness hourglass control. Figure 8 shows the distribution of the equivalent plastic strain of the deformed sheet
for the three-dimensional case using the default section controls (average
strain kinematic and relax stiffness hourglass). Table 1 compares this three-dimensional case with the plane strain, quasi-static, and
three-dimensional automatic mass scaling cases; and Table 2 compares the five different three-dimensional element formulations included
here with the two-dimensional cases at the same level of mass scaling. The
accuracy for all five element formulations tested is very similar for this
problem, but significant savings are realized in the three-dimensional analyses
when using more economical element formulations.
Analyses using CPE6M and
C3D10M elements
The total number of nodes in the
CPE6M model is identical to the
number in the CPE4R model. The
number of nodes in the C3D10M
model is 3440 (compared to 3808 in the
C3D8R model). The analyses using
the CPE6M and
C3D10M elements use a mass scaling
factor of 2758. Figure 9 and Figure 10 show the distribution of the equivalent plastic strain of the plate for the
two-dimensional and three-dimensional cases, respectively. The results are in
reasonably good agreement with other element formulations. However, the
CPU costs are higher since the modified
triangular and tetrahedral elements use more than one integration point in each
element and the stable time increment size is somewhat smaller than in analyses
that use reduced-integration elements with the same node count. For the mesh
refinements used in this problem, the
CPE6M model takes about twice the
CPU time as the
CPE4R model, while the
C3D10M model takes about 5.75
times the CPU time as the
C3D8R model.
Three-dimensional case (using
C3D8R elements) with a
mass scaling factor of 2758, an analytical rigid surface of
TYPE=REVOLUTION,
and the CENTROID
kinematic and
STIFFNESS
hourglass section control options.
Three-dimensional case (using
C3D8R elements) with a
mass scaling factor of 2758, rigid elements, and the
CENTROID
kinematic and
STIFFNESS
hourglass section control options.
Three-dimensional case (using
C3D8R elements) with a
mass scaling factor of 2758, rigid elements, and the
ORTHOGONAL
kinematic and
COMBINED
hourglass section control options.
Three-dimensional model (using
C3D8R elements) with a
mass scaling factor of 2758, an analytical rigid surface of
TYPE=REVOLUTION,
and the
ORTHOGONAL
kinematic and
COMBINED
hourglass section control options.
Three-dimensional model (using
C3D8R elements) with a
mass scaling factor of 2758, an analytical rigid surface of
TYPE=REVOLUTION,
and the
ORTHOGONAL
kinematic and
ENHANCED
hourglass section control options.
Three-dimensional model (using
C3D8R elements) with a
mass scaling factor of 2758, an analytical rigid surface of
TYPE=REVOLUTION,
and the CENTROID
kinematic and
ENHANCED
hourglass section control options.
Three-dimensional model (using
C3D8R elements) with a
mass scaling factor of 2758, an analytical rigid surface of
TYPE=REVOLUTION,
and the default section control options.
Three-dimensional case (using
C3D8R elements) with
automatic mass scaling, an analytical rigid surface of
TYPE=REVOLUTION,
and the CENTROID
kinematic and
STIFFNESS
hourglass section control options.
Three-dimensional model (using
C3D8R elements) with a
mass scaling factor of 2758, an analytical rigid surface of
TYPE=CYLINDER,
and the default section control options.
Additional mesh refinement case (using
C3D8R elements) with
contact mass scaling activated included for the sole purpose of
testing the performance of the code.
References
Kobayashi, S., S. I. Oh, and T. Altan, Metal Forming and the Finite Element Method, Oxford University Press, 1989.
Kopp, R., and P. M. Dohmen, “Simulation und Planung von Walzprozessen mit Hilfe der Finite-Elemente-Methode (FEM),” Stahl U. Eisen, no. 7, pp. 131–136, 1990.
Tables
Table 1. Analysis cases and relative CPU costs. (The
two-dimensional explicit analyses all use
CPE4R elements and stiffness
hourglass control. The three-dimensional explicit analyses use
C3D8R elements and the centroidal
kinematic and stiffness hourglass section controls.)
Analysis Type
Mass Scaling
Factor
Effective Roll Surface
Speed (m/sec)
Relative
CPU Time
Explicit, plane strain
110.3
11.2
4.99
Explicit, plane strain
2758.5
56.1
1.00
Explicit, plane strain
68961.8
280.5
0.21
Implicit, plane strain
quasi-static
13.4
Explicit, 3D
2758.5
56.1
13.8
Explicit, 3D
automatic
~96
9.5
Table 2. Explicit section controls tested (mass scaling factor=2758.5).
CPE4R and
C3D8R elements are employed for
the two-dimensional and three-dimensional cases, respectively. Spread values are
reported for the half-model at node 24015.
Analysis Type
Section
Controls
Relative
CPU Time
Spread (mm)
Kinematic
Hourglass
Explicit, plane strain
n/a
stiffness
1.00
n/a
Explicit, plane strain
n/a
relax stiffness
1.11
n/a
Explicit, plane strain
n/a
combined
1.04
n/a
Explicit, plane strain
n/a
enhanced
1.02
n/a
Explicit, 3D
average strain
relax stiffness
20.8
2.06
Explicit, 3D
orthogonal
combined
17.1
2.07
Explicit, 3D
centroidal
stiffness
13.8
2.10
Explicit, 3D
centroidal
enhanced
14.8
2.10
Explicit, 3D
orthogonal
enhanced
17.3
2.10
Figures
Figure 1. Equivalent plastic strain for the plane strain case
(CPE4R) with stiffness hourglass
control (mass scaling factor=110.3). Figure 2. Equivalent plastic strain for the plane strain case
(CPE4R) with stiffness hourglass
control (mass scaling factor=2758.5). Figure 3. Equivalent plastic strain for the plane strain case
(CPE4R) with stiffness hourglass
control (mass scaling factor=68961.8). Figure 4. Shear stress for the plane strain case
(CPE4R) with stiffness hourglass
control (mass scaling factor=110.3). Figure 5. Shear stress for the plane strain case
(CPE4R) with stiffness hourglass
control (mass scaling factor=2758.5). Figure 6. Shear stress for the plane strain case
(CPE4R) with stiffness hourglass
control (mass scaling factor=68961.8). Figure 7. Equivalent plastic strain for the three-dimensional case
(C3D8R) using the centroidal
kinematic and stiffness hourglass section controls (mass scaling
factor=2758.5). Figure 8. Equivalent plastic strain for the three-dimensional case
(C3D8R) using the average strain
kinematic and relax stiffness hourglass section controls (mass scaling
factor=2758.5). Figure 9. Equivalent plastic strain for the plane strain case
(CPE6M) (mass scaling
factor=2758.5). Figure 10. Equivalent plastic strain for the three-dimensional case
(C3D10M) (mass scaling
factor=2758.5).