FEA Coil Spring 2007
FEA Coil Spring 2007
FEA Coil Spring 2007
Steven M. Cron
Michelin Americas Research & Development Corp.
Greenville, South Carolina
One of the many powerful capabilities of Abaqus is its ability to easily model a structure
embedded inside of a host structure. This technique has been used to analyze an RFID chip and
helical antenna embedded in a rubber patch for use in tire applications. The main objective of the
analysis of the chip/antenna/patch assembly was to determine the stress concentration in the
antenna caused by its interaction with the relatively rigid chip. In the course of conducting this
analysis, a high degree of sensitivity was observed in the interaction between the embedded
helical antenna and the host rubber patch. This paper will show how adequate refinement of the
host structure becomes crucial if accurate results are to be obtained with the embedded element
technique. It will also present results of the antenna stress concentration analysis.
Keywords: Embedded Element, Mesh Refinement, RFID, Coil Spring.
1. Introduction
The use of RFID chips is now commonplace in a broad range of applications. To facilitate
communication with the RFID device, an antenna is sometimes required. In the case of RFID
chips used in tires, one proposed solution for the antenna is to employ a coil spring which, along
with the RFID chip, is embedded in a rubber patch.
Abaqus provides a powerful tool for analyzing such structures with the Embedded Element
technique wherein a mesh representing a given structure can be embedded within an independent
mesh representing a host structure. In the course of using this technique to analyze stresses in an
embedded RFID coil spring antenna, a significant degree of non-physical stiffening was observed,
depending on the host mesh refinement. This paper will explore the parameters that influenced
this observed stiffening and provide some ideas for how to assess the quality of results when
modeling complex structures embedded in relatively low modulus host structures. A brief
summary of the stress analysis results for a practical implementation will also be presented.
2.1 Strategy
The original motivation for this effort was to understand the stress concentrations produced in a
steel coil spring when used as an antenna soldered to an RFID chip assembly and embedded in a
deformable rubber host structure. The analysis strategy employed was to first analyze a simple
structure involving a representative spring model embedded in a representative host model. This
allowed for a quick assessment of the effect of various parameters influencing the interactions
caused by the embedding constraint.
A test spring model that yields reasonably accurate stress and stiffness results will first be
presented. Subsequently, a simple cylindrical rubber test model will be evaluated. The next series
of steps will involve analysis of the combined structure where the spring mesh will be held
constant and the host mesh varied.
The spring test model is shown in Figure 1. Note that for a practical implementation the antenna
would be much longer, up to 50 mm.
Catia V5 was used to generate the geometry and Abaqus CAE was used to import the geometry (in
STEP format) and generate the finite element model. The mesh is comprised of linear hexahedral
elements with reduced integration (C3D8R).
To evaluate the quality of the spring model, the structure was stretched by 0.1 mm (2.7%) along
its centerline The reader should note that the load was applied along the spring axis using beam
connectors which allowed the spring to both elongate and rotate freely.
Figure 2 shows a section of the deformed spring with the peak shear stress highlighted.
As indicated, the peak shear stress computed with Abaqus was approximately 190 MPa. To check
the quality of the model, the Abaqus result was compared to the theoretical value given by Shigley
[1]:
Gd 2 ⎛D ⎞
τ max = ε ⎜ + 0.5 ⎟ = 191.3MPa (1)
PD 3π ⎝d ⎠
Where,
τ = Shear stress
Gd 4
K= = 6.68 N / mm (2)
8N a D3
Where Na is equal to the number of “active” coils which for this configuration is 4.5. The
predicted stiffness computed with Abaqus is 6.13 N/mm.
Given the level of approximation involved, the agreement between the theoretical values and
numerically computed values seems reasonable.
With a satisfactory model for the spring in hand, we next turn our attention to modeling of the host
rubber test structure into which the antenna model was embedded. The simple cylindrical test
model shown in Figure 3 was used to evaluate the interactions between the spring mesh and the
rubber mesh. The cylinder diameter is 1.5 mm and the length is just slightly longer than the total
spring at 3.95 mm.
Since our intention was to model the spring embedded in an incompressible rubber host, the
elements in this initial model are linear (Q1) hybrid hexahedral elements. The rubber material
constitutive behavior was modeled with the Mooney-Rivlin law with C10 = 0.5 MPa and C01 =
0.0. This will give a Young’s modulus of approximately 3 MPa.
Before embedding the spring, this model was also stretched by 0.1 mm (2.5%) along its axis to
establish a baseline stiffness. Under this loading condition, the finite element model predicted a
stiffness of 1.29 N/mm vs. a theoretical stiffness of 1.34 N/mm based on the Young’s modulus
given above. It should be noted that the model was loaded in uniaxial tension by applying a U1 =
0 condition on the left end and U1 = 0.l mm on the right end. Transverse deformations were not
constrained except to eliminate the rigid body translations.
At this point we can say that, individually, the models for the spring and cylinder are reasonably
accurate for the prediction of stress and load under prescribed deformation conditions. It would
also seem safe to make an initial estimate of the combined stiffness of the model shown in Figure
4 to be the sum of the two individual FE derived stiffnesses: 6.13 N/mm + 1.29 N/mm = 7.42
N/mm. This estimate assumes that the two structures act as springs in parallel.
Upon application of a 0.1mm elongation to the combined model of Figure 4, however, a very
different result was obtained. In fact the predicted stiffness was 578 N/mm! The deformed model
is shown in Figure 5. Note that the spring model was embedded in the host using the default
Abaqus/Standard embedded constraint (*EMBEDDED ELEMENT).
At this point we must conclude that either a) there is a huge error introduced by the interaction
between the host and embedded models or b) we are observing a non-obvious physical interaction
between the spring and the incompressible host material. Most likely the observed stiffening was
erroneous.
Our interest at this point was to explore the causes of this stiffening behavior and to satisfy
ourselves that we had an accurate result. To that end we explored the two most likely causes for
Clearly the overall stiffness is strongly dependant on this mesh refinement. With 18 elements
along the cylinder axis, the difference between the value computed with the FE model and the
summation estimate of 7.42 N/mm becomes relatively small.
We can now easily check the effect of element order by simply converting the models used to
generate the Q1 stiffnesses in Figure 6 to quadratic (Q2) elements. Figure 7 shows the Q1 results
Again, the use of Q2 elements greatly reduces the tendency of the model to stiffen even using the
relatively coarse 12-element mesh.
To this point we have only addressed refinement of the host mesh along the length of the cylinder.
For completeness, a quick check of the effect of cross-section mesh refinement was also made.
Figure 8 shows the cross-section mesh of the reference model (of Figure 4) and a refined test
mesh.
Table 1 compares the stiffnesses over the range of lengthwise mesh refinements for the two cross-
section meshes shown above.
12 578 773
15 114 159
18 14.6 12.2
24 10.9 9.73
30 10.1 9.15
What is interesting to note here is that the adverse effect on stiffness is exacerbated by the finer
cross-section mesh when combined with the coarser lengthwise meshes.
The last effect that should be explored is that of changing the geometry of the spring, in particular
the number of coil turns per unit length. For this test we embedded an equal-length spring of 2.5
turns vs. the reference (of Figure 4) spring with 4.5 turns. All other parameters were held
constant. The new combined model is shown in Figure 9. From Equation 2, the theoretical spring
stiffness is 12 N/mm. Adding the baseline rubber cylinder stiffness of 1.29 N/mm gives us an
estimated stiffness of 13.3 N/mm for the combined structure.
What then is the critical parameter that controls this stiffening behavior? One hypothesis would
be that the number of elements spanning the gap between spring coils must be above some
minimum threshold level. Looking back to Figure 6 we see that a huge reduction in artificial
stiffening was observed (for the Q1 models) when the number of elements along the length was at
or above 18. If we divide 18 by the number of turns, 4.5, we get a value of 4 elements per spring
coil. We can test this hypothesis with the 2.5 turn model of Figure 9. Given that the spring has
2.5 turns we would expect to see a large rise in artificial stiffening with any number of elements
along the length less than 10. In fact, when the number of elements along the length is reduced to
8, the stiffness value jumps to 50 N/mm as expected. In the limit, as the spring becomes a straight
rod, the sensitivity to lengthwise mesh refinement disappears.
While a “4 element rule” might be satisfactory for modeling embedded springs, a more general
diagnostic technique would certainly be more useful for assessing the quality of results where the
embedded structure is of an arbitrary shape. One possible technique would be to look at elements
grouped by maximum distortion. For our purposes, the maximum shear strain is taken as a
measure of total element distortion. Thus, the first step would be to calculate the maximum shear
strain (max principal strain – min principal strain) for each element using the “Create Field Output
from Field” function within Abaqus/CAE. The second step would be to use Display Groups to
eliminate from the display all elements above an arbitrary maximum shear strain level. As this
arbitrary level is reduced, more and more host elements will disappear. At some point, elements
from the embedded body will also begin to disappear. At the level where embedded elements
begin to disappear, the following assessment can be made. If the remaining host elements are
3.1 Introduction
Confident that we understand how to model a spring embedded in a rubber cylinder, we can move
on to look at the results for an actual application. Our objective will be to determine the stress
concentration factor seen by a steel coil spring antenna when it is soldered to an RFID chip
assembly, embedded into a generic patch of rubber, and then stretched by 4%. Symmetry is
assumed so that only half of the structure need be modeled.
The half-model of the chip, solder and base is shown in Figure 13. Figure 14 shows the same
model with the antenna model in its position embedded in the solder joint.
Chip
3.0 mm Solder
Base
1.0 mm
0.25 mm
4.0 mm
5.0 mm
Figure 14. Chip, Solder and Base Model with Antenna Embedded
This model is intended to constrain the deformation of the coil antenna and provide a reasonable
representation of the discontinuity caused in the patch rubber. The elements of the
chip/solder/base are linear hexahedral (C3D8R) and the material is taken to be linear elastic with a
modulus of 3000 MPa.
The combined model of the patch, chip/solder/base assembly, and antenna coil is shown below in
Figures 15, 16, and 17. It should be apparent that the model represents half of the entire structure.
As such, symmetry boundary conditions were enforced as indicated in Figure 15.
Figure 16. Zoom of Combined Model of Substrate, Patch, Coil and Chip
The mesh of the substrate region and the coarsely meshed portion of the patch were comprised of
C3D20RH elements. The fine mesh into which the chip assembly and antenna were embedded
was comprised of C3D20H elements. The patch rubber was again modeled as a Mooney-Rivlin
material with C10=0.5. The substrate was modeled as a linear elastic material with a modulus of
3000 MPa. The model contained approximately 259,000 d.o.f.
4. Results
As mentioned above, the intended deformation of the patch rubber structure was to be 4%
elongation. To evenly distribute this deformation into the patch, the relatively stiff substrate
material was stretched along the axis of the spring such that the entire structure experienced a
nominal imposed elongation of 4%. Figure 18 shows the strain in the 11-direction (LE11).
In Figure 19 we isolate just the portion of the patch rubber surrounding the chip assembly and the
antenna and see the strain field at this level.
Figure 19. 11-Strain in Patch Rubber Surrounding Chip Assembly and Antenna
The peak shear stress is approximately 530 MPa. Referring back to Equation 1, we can calculate
the theoretical shear stress at 4% spring elongation to be 286.9 MPa. Furthermore, we can say that
the stress concentration due to the antenna and chip being embedded in the patch rubber is roughly
530/286.9 = 1.85.
5. Conclusions
The first, and probably most significant finding of this work is the fact that modeling of embedded
structures that have complex geometries must be done with some care. It has been shown that
enormous amounts of artificial stiffness can be generated when host mesh refinement (or element
order) is too low. Remedies for this behavior include increasing the order of the host elements as
well as increasing mesh refinement. A proposed technique for evaluating the quality of results has
also been presented based on a qualitative measure of the degree to which elements of relatively
low distortion are grouped around the embedded structure.
The analysis of the chip/solder/base assembly embedded in a low modulus rubber patch indicates
that there is significant effect on the patch rubber strain field in the vicinity of connection between
the antenna and the RFID chip. For a nominal elongation of 4% the net effect on the stress
concentration in the antenna has been found to be 1.85.
6. References
1. Shigley, J.E., “Mechanical Engineering Design”, Third Ed., McGraw Hill, 1977.