Precision Engineering 42 (2015) 346–351

Contents lists available at ScienceDirect

Precision Engineering
journal homepage: www.elsevier.com/locate/precision

Technical note

Entangled structures as high cycle compression springs

Folkers E. Rojas ∗ , Alexander H. Slocum
Department of Mechanical Engineering, Massachusetts Institute of Technology, Cambridge, MA 02139, USA

a r t i c l e i n f o a b s t r a c t

Article history: Entangled structures, such as steel wool, can be used as inexpensive, high cycle, low stiffness, thin profile
Received 2 February 2015 compressive springs where uniform pressure on a surface is required particularly in elevated temperature
Accepted 20 April 2015 and/or harsh environments. Mechanical compression tests were performed on a variety of steel wool sam-
Available online 27 May 2015
ples to determine the stress–strain curve behavior over high cycles. After initial conditioning cycles, good
repeatability can be obtained with hysteresis dependent on strain. The results show a nonlinear behavior
Keywords:
over large strains (>10%) and reasonable linear behavior for strains less than 10%. The properties of an
Entangled structures
entangled structure spring can be selected to achieve the desired stiffness for a particular application.
Foams
Fibrous metals © 2015 Elsevier Inc. All rights reserved.
Springs

1. Introduction created by the entangled wires [11,14]. While compressive

properties and pseudo-elastic hysteresis behaviors have been char-
The mechanics of a random array of interwoven fibers, i.e. entan- acterized for ETWMs, a high cycle analysis of the nonlinear elastic
gled wires such as found in steel wool, have been studied for region of ETWMs is needed to make the design process more deter-
applications such as damping, isolating, or filtering. In semiconduc- ministic.
tor test applications “Fuzz Buttons,” a compacted random array of Prior work has shown that entangled structures have a
wire, have been used as temporary electrical contacts for “low sig- three-stage stress–strain behavior under compressive loads: (1)
nal distortion, high frequency, and low insertion force, planarity, nonlinear elastic deformation, (2) strain-hardening, and (3) den-
and shock/vibration resistance” [1]. Contact testers made from sifying [10]. Here we build on this work by studying the higher
micro entangled structures have been used by the semi-conductor cycle compliance of entangled structures under uniaxial compres-
industry to test Through Silicon Via (TSV) and micro-bumps arrays sion for the use as an inexpensive and repeatable low-moderate
[2,3]. The characterization of entangled wires is also very important stiffness thin profile large contact area spring. This type of spring
for micro-electronics, micro fluidics, micro sensors and actuators has wide application in presses and rectilinear format batteries.
that are based on entangled silica nano-wire structures [4]. The This work also examines the effect of densification on the appar-
generation of the silica nanowire entangled arrays are grown, and ent Young’s modulus for entangled structures for comparison with
not created via a mechanical means and they often appear as entan- existing analytical models. Analogies are made with low density
gled [4,5]. Entangled structures are also used by the space industry materials such as foams, composites, and elastomers in order to
to isolate instruments or to damp vibrations [6,7]. help develop modeling tools.
Entangled structures have also been evaluated as compati-
ble replacements for porous aluminum as they have been shown
2. Experimental procedures
to have good toughness and strength [8,9]. Entangled titanium
wire materials (ETWM) and porous titanium foams have been
2.1. Sample characterization
considered for tissue reconstruction, orthopedic implants, and
bone repairs [10–14]. The challenge for the medical industry
One method for generating an entangled metal wire structure
has been to match the mechanical properties of entangled wire
involves wrapping the wire around a rod to create a loose steel
structures to similar mechanical properties of bone which has
wire bundle. The unconsolidated deformed wire is placed into a set
been attempted by altering the effective porosity of the structure
of molds where it is compressed and sintered to remove internal
stresses [8,10,15]. Common steel wool, which is much more compli-
ant and inexpensive, is made by shaving small strands from straight
∗ Corresponding author. Tel.: +1 603 513 8997; fax: +1 603 225 3102. steel wire and consolidating the strands into an entangled struc-
E-mail address: [email protected] (F.E. Rojas). ture using a needle punch machine to generate a mat of wool. The

http://dx.doi.org/10.1016/j.precisioneng.2015.04.013
0141-6359/© 2015 Elsevier Inc. All rights reserved.
 F.E. Rojas, A.H. Slocum / Precision Engineering 42 (2015) 346–351 347

Table 1 show a settling/shifting effect in the stress–strain curve over first

Characterization of samples tested.
100 cycles in the non-linear elastic region. Van Wyk alluded to this
For cyclic study For densification and cyclic study shifting effect in his 1964 paper: “the curve shifts with each succes-
3
Initial size (mm ) 52 × 52 × 8.5 61 × 63 × 36 sive cycle of compression and release, but finally attains a steady
Mass (g) 5.83 6.14 position” [17]. After the initial settling the data was repeatable. The
discrepancy is attributed to the initial settling and distortion of
the entangled structure in the process toward reaching stability.
pads are held together by the entangling nature of the process that A 1000 cycle run was thus used to remove any long term settling of
produces a semi-aligned structure with the fibers in the pad.1 Prior the samples. As a check, a low stiffness coil spring was substituted
work on characterizing steel wool pads was done in 2006 by Masse for the sample to ensure that testing machine was operating as
et al.; however, only 60 cycles were used to derive the stress–strain expected, and the result confirmed the observation of the entangled
curve [16]. Their work showed that the stress–density curves had structure behavior.
a power-law relation [16]. In the present work, we are interested The coarse sample was cycled on the testing machine at a fre-
in the cyclic performance of steel wool as a spring that acts over a quency of 0.5 Hz and at a constant strain rate of 24%, 2 mm/sec over
wide area subject to thousands of cycles. Applications include pro- a sample height of 8.5 mm. The fine sample was also cycled at a
viding a uniform pressure over the surface of an object as might be 0.5 Hz, but with a smaller strain rate as the stiffness was higher
needed in a press or a rectilinear format battery. than the coarse sample. The densification sample was also tested
To study high cycle behavior, two types of samples were initially at a constant strain rate with sample thickness starting from 36 mm
evaluated: coarse (Grade #3) and fine (Grade #0) steel wool with and decreasing to 22 mm by 2 mm increments. For the initial data
wire diameters of 90 !m and 25 !m respectively were used. Test set the steel wool is cycled from 36 mm down to 34 mm. The first
samples were cut from non-sintered steel wool pads.2 The man- thousand cycles are discarded from the data, to establish the steady
ufacturing method of the wool pad creates an inherent preferred state behavior. An additional two thousand cycles are performed
orientation of a planar cross pattern. and used to analyze the steady state behavior in that region. With
Hydrostatic weighting is typically used for determining the the completion of the initial data set, the following data set starts
porosity of a sample [10], in this case the mass of the sample was from a 34 mm thickness and is cycled down to 32 mm. The same
measured using an AG204 Delta Range Mettler Toledo scale with process is carried out to all following data sets until reaching a final
an accuracy of 0.1 mg/1 mg, at room temperature. The volume of thickness of 22 mm.
a sample was measured with calipers accounting for an error of
about 1 mm on each length measurement. The samples were cut 3. Results
from 7-in. diameter pads into squares as summarized in Table 1.
The objective of the densification experiment was to explore The data collected from the testing machine is shown in terms
the repeatability over a range of compression states to determine of vertical displacement and the compressive load measured by the
effect of density on stiffness. The densification study is done with a load cell gauge. The data is combined with the geometry and sample
highly porous steel wool sample (Grade 1 – purchased as a 1 pound properties to generate engineering stress–strain curves, as shown
bag from McMaster-Carr3 ) that is incrementally compressed while in Fig. 3.
testing long term cyclic behavior at each stage. The sample used
for the densification experiment does not contain the planar cross 3.1. Cyclic testing
pattern of the wool pads. The sample is incrementally compressed
until more than fifty percent strain has been achieved. At each state The coarse sample was loaded for >11k cycles, with an initial 1k
the sample is not strained more than ten percent at any given time. cycle break-in period for allowing the fibers to settle into a steady
state region. Note that this type of break-in period is also required
2.2. Testing procedure for steel cables [18]. As shown in Fig. 4, every two thousand samples,
the loading curve is repeatable and below the initial curve of the
The stress–strain behavior of the samples was performed using break-in period. The first 1k sample (i.e. break in) was where it
an Admet testing machine (Model 5601QP2 – Serial 0611091 reaches steady state. Then the system was set to do a 2k run, when
eXpert) with an Interference load cell, rated to 120 N (Serial # done the same preload was affirmed and then a run was done for
708336). As shown in Fig. 1, the experimental setup consists of the next 2k sample (i.e. sample 2001 is the first value). Notice that
a base support with a pocket region for the samples that prevents the break in curve is higher than all of the other curves. Each data
horizontal shifting of the sample during the cycling process. A com- set following the break in consists of 2000 cycles. At the conclusion
pressive plate provides an even load, measured by the load cell, of each data set the sample is reset to the starting preload in the
across the sample. The testing machine is controlled by a pro- event that drifting in the testing machine took place during cycling.
grammed displacement cycle. Only the first data point of each data set is plotted to illustrate
repeatability. After collecting 11k cycles, the average of all the first
2.3. Cyclic testing data sets was calculated and plotted. Future work will calculate the
mean and error of all of the data points combined.
The preliminary tests, shown in Fig. 2, show the loading for a
fine and course sample with ∼97.7% porosity. The fine steel wool 3.2. Young’s modulus as densification
sample has a stiffer stress–strain response compared to the coarse
sample, due to the limitation of the strain gauge only the course The sample used for the densification is a homogeneous uncon-
sample is tested extensively over a wider range of strains. The tests solidated coarse steel wire wool with a diameter of 90 !m. The goal
was to demonstrate that a small change in the porosity of the struc-
ture can result in a change in the effective stiffness. The cycle stroke
1
was set to 2 mm, which limited the strain from 5.5% (at 36 mm) to
See for example http://www.youtube.com/watch?v=iaDzLP3TT I.
2
Manufacturer of Wool pads: International Steel Wool Inc. (http://
9.1% (at 22 mm), to the authors surprise the stiffness was linear
steelwooldirect.com/) for the 7 inch Tex-Wool floor pads grade 0 and 3. in these regions. Fig. 5 shows the response starting from a 36 mm
3
Grade 1 Steel wool – 1 pound bag from McMaster-Carr Item Number 7363T24. thick sample undergoing a minimum of 3000 cycle test, for clarity
348 F.E. Rojas, A.H. Slocum / Precision Engineering 42 (2015) 346–351

Fig. 1. System used to measure the cyclic stress–strain relation of the samples.

Fig. 2. Coarse versus fine characterization. Height of fine sample is 13.7 mm and coarse sample 14.0 mm. The data shown is for a complete cycle of load/unload. The stroke
refers to the compression of the sample.

Fig. 3. Typical loading and unloading data from the tester and engineering stress–strain representation for the course sample.

only the loading portion of the graph is shown. The curves shown 4. Discussion
in Fig. 5 are the final sample of the compression test.
The nonlinear section in the initial one percent strain region 4.1. Models
is attributed to overcoming static friction between the fibers. It is
hypothesized that as the strands begin to move pass each other, and There are four predominant models used to predict the mechan-
friction transitions from static to dynamic, they achieve a steady ical behavior of entangled or porous structures: (1) Gibson–Ashby
state stiffness as observed. Fig. 6 illustrates how the preloaded sam- model, (2) Nielson model, (3) Baudequin relation, and (4) van Wyk
ples stack and the effective stiffness is obtained as a function of the model. Other prior work includes using finite element models to
strain at each sample thickness. directly simulate the deformation of “dry” fiber networks, where
 F.E. Rojas, A.H. Slocum / Precision Engineering 42 (2015) 346–351 349

which will later be compared to the experimental data to establish

applicability for the entangled structures tested herein.
The Gibson–Ashby equation is used to affirm that the applied
stress is lower than the collapse stress of the entangled structure,
where it is recognized here that this is an approximation since
the Gibson–Ashby equation is for porous structures [20]; however,
the Gibson–Ashby equation [20] can also be used to calculate the
Young’s modulus as well as the Yield strength for entangled media
[8]:

∗
!pl ! "∗ "n1
= C1 · (1)
!ys "s

∗
Epl ! "∗ "n2
= C2 · (2)
Fig. 4. Coarse steel wool sample loading behavior with initial thickness 8.5 mm Eys "s
and cyclically compressed to thickness 6.5 mm, and loaded at 2 mm/sec at a cycle
frequency of 0.5 Hz. For clarity the data shown is for the loading portion of the test.
The average is calculated taking the first sample of every thousand data set. The where !pl ∗ /! is the ratio of the plastic collapse stress of the porous
ys
cross bars on the average line represent a standard deviation. material to the solid material yield stress, "*/"s is the ratio of the
sample density to the solid density, and Epl ∗ /E is the elastic Young’s
ys
strands are assumed to be bonded together with elastic springs and modulus ratio of the porous structure to that of a solid. The values
no subsequent sliding takes place [19]. As shown by Rubshtein et C1 , C2 , n1 , and n2 are constants related to the porous structure. C1 is
al, the predicted Young’s modulus can differ for the varying models typically about equal to 0.3 and n1 is approximately 1.5 for a variety
[12]. Thus, the models are compared below for varying parameters, of foams [21]. Note that the constants are empirically determined.

Fig. 5. Densification structure stress–strain curve of the loading section at each height.

Fig. 6. (L) Stacked cycling response for densification sample. (R) Stiffness measured for the array of strain densification tests.
350 F.E. Rojas, A.H. Slocum / Precision Engineering 42 (2015) 346–351

Fig. 7. Analytical model comparison between experimental data and analytical model.

The Nielsen model has also been used to calculate the expected is used to correlate the “resistance to compression” (i.e. Stiffness)
Young’s modulus for a porous elastic material as a function of poros- to physical values:
ity [22].
2
˛ · Y · m3
(1 − #) A= (7)
∗
E = Es · (3) "3
1 + (1/ϕ − 1)ϕ

where # is the porosity, and % is a geometric factor as a function of where ˛ is a constant (to be found experimentally), m is the mass,
area and perimeter and " is the density of the wool.
A comparison of the models to the experimental data shows that
4 · & · Area the Gibson–Ashby model best matches the data collected. While the
ϕ= (4)
P2 trend of the Nielson model shows that with increasing porosity the
stiffness decreases, the model over estimates the expected Young’s
The Baudequin power law relation for glass wool, where the
modulus by at least by two orders of magnitude; the effect of the
stress is proportional to the strain to the −3/2 power, has been
geometric area is not significant enough to compensate for the
supported by experimental data [23]. Since the material properties
discrepancy. The Nielson model has a trend to the second power,
of glass wool are different than those of steel wool, the relation may
which does not match the data accurately. Baudequin et al.’s rela-
not be applicable but only experimental data will tell.
tion, derived for glass wool, between compressive stress and how it
! ∝ (ε∗ − ε)−3/2 (5) is related to the strain to a −3/2 power did not match the measured
our data [23]. The observed behavior has the strain to a positive 3/2
The van Wyk model consists of relating the global behavior of power, which matched the data. The trend predicted by the van
the sample to beam bending dynamics at the micro scale [17]. The Wyk model does not match the data.
derivation is similar to the work done by van Wyk and Pawlak et al. As shown in Fig. 7 the Gibson–Ashby model, with n2 = 1.5 and
[17,24]. Eq. (6) is the basic equation used for the Wyk derivation. c2 = 4e−4, reasonably predicts the stiffness observed from the den-
sification samples. The Nielsen model yields a higher Young’s
k · Y · Ic
dF = · dy (6) modulus of the materials without capturing the nonlinear behav-
S3 ior. The Baudequin model has an inverse stiffness–strain relation
where k is an unknown factor due to the variability in the geometry to that observed in the data, and as the strain increases the models
of contact and that the points of contact are not regularly spaced, S predict a drop in the stiffness. As shown in Fig. 8, the percentage
is the mean distance between the fibers. The resulting equation (7) error of the models ranges between about ±15%.

Fig. 8. Percentage error of analytical models.

 F.E. Rojas, A.H. Slocum / Precision Engineering 42 (2015) 346–351 351

5. Conclusion

The high cycling of entangled structures about a set preloaded

thickness has been demonstrated to show their potential use as a
thin low stiffness spring acting over a wide area. It was also shown
that highly porous structures can exhibit linear stiffness for low
strains. One advantage of this material is its ability to conform to an
irregular surface while still allowing for the mechanical properties
to be altered by changing the porosity. Hence entangled structures
can be used as an alternative to foam pads for applying pressure to
an object. In particular, at elevated temperatures or in aggressive
environments, stainless steel wool has particular promise as a thin
profile area spring.
Fig. 9. Strain versus stiffness behavior for the densification experiment with an
overlay of the Gibson–Ashby model.
References
4.2. Elastic hysteresis [1] Custom interconnects – fuzz button performance; 2014. Available from: http://
www.custominterconnects.com/ (23.06.14).
The separation between the loading and unloading curves [2] Yaglioglu O, Eldridge B. Direct connection and testing of TSV and microbump
devices using nanopierce contactor for 3D-IC integration. In: 30th IEEE VLSI
that allude to hysteresis, a known characteristic of entangled
test symposium. 2012.
structures, is attributed to the friction between wire elements [3] Yaglioglu O [Doctor of Philosophy] Carbon nanotube based electromechanical
sliding across each other during loading. Another hypothesis is probes. Mechanical Engineering Massachusetts Institute of Technology; 2007.
[4] Woldegiorgis A, Jansson K, Roerrade J. Fabrication of silica nano wires on the
the non-linearity comes from contacts being created during the
internal perimeter of narrow bore fused silica tubing by non-isothermal etch-
compression process [23]. The curved shape of the stress–strain ing. J Mater Sci 2005:583–9.
curves indicates the complex mechanics that take place during [5] Yan C, Zhang T, Lee PS. Flow assisted synthesis of highly ordered silica nanowire
the elastic stage, whereas the porosity decreases hysteresis is arrays. Appl Phys A: Mater Sci Process 2009;94:763–6.
[6] Castagnede B, Moussatov A, Lafarge D, Saeid M. Low frequency in situ metrol-
smaller. The findings confirmed the power-law relation of the ogy of absorption and dispersion of sound absorbing porous materials based
stress–strain curve; however, it also showed an entry/break-in on high power ultrasonic non-linearly demodulated waves. Appl Acoust
time period where the samples’ properties change as the num- 2008;69:634–48.
[7] Sun F, Chen H, Wu J, Feng K. Sound absorbing characteristics of fibrous metal
ber of cycles increase. When repeatability is needed, the drifting materials at high temperatures. Appl Acoust 2010;71:221–35.
that occurs upon initial use must be removed by cyclically load- [8] Tan Q, Liu P, Du C, Wu L, He G. Mechanical behaviors of quasi-ordered entangled
ing (break-in period) the samples over 100 times prior to establish aluminum alloy wire material. Mater Sci Eng A 2009;527:38–44.
[9] Tan Q, He G. 3D entangled wire reinforced metallic composites. Mater Sci Eng
steady state behavior and then using them in a preloaded state. A 2012;546:233–8.
Although the friction/slip between the fibers is irreversible the [10] Liu P, Tan Q, Wu L, He G. Compressive and pseudo-elastic hysteresis behavior
long term cycle loading behavior appears reasonably repeatable, of entangled titanium wire materials. Mater Sci Eng A 2010:3301–9.
[11] Barbas A, Bonnet AS, Lipinski P, Pesci R, Dubois G. Development and mechanical
allowing for use as a constant force (pressure over a wide area
characterization of porous titanium bone substitutes. J Mech Behav Biomed
), low-moderate stiffness spring. Mater 2012:34–44.
[12] Rubshtein AP, Trakhtenberg IS, Makarova EB, Triphonova EB, Bliznets DG,
Yakovenkova LI, Vladimirov AB. Porous material based on spongy titanium
4.3. Porosity dependence
granules: structure, mechanical properties, and osseointegration. Mater Sci Eng
C 2014:363–9.
Similar to tensile tests performed by others [10,15] the com- [13] Chen YJ, Feng B, Zhu YP, Weng J, Wang JX, Lu X. Fabrication of porous titanium
pressive properties of entangled structures strongly depend on implants with biomechanical compatibility. Mater Lett 2009:2659–61.
[14] Krishna BV, Bose S, Bandyopadhyay A. Low stiffness porous Ti structures for
the porosity of the sample, where larger porosity results in lower load bearing implants. Acta Biomater 2007;3:997–1006.
Young’s modulus. Other research has examined entangled steel [15] Liu P, He G, Wu L. Uniaxial tensile stress–strain behavior of entangled steel wire
wire materials using stereomicroscopes and scanning electron material. Mater Sci Eng A 2009;509:69–75.
[16] Masse JP, Salvo L, Rodney D, Brechet Y, Bouaziz O. Influence of relative density
microspores to gain insight into the tensile dynamics of an entan- on the architecture and mechanical behaviour of a steel metallic wool. Scr Mater
gled structure [15]. The high cycle tests results collected herein 2006;54:1379–83.
showed that the complex intertwined structure has modest hys- [17] Wyk CM v. Note on the compressibility of wool. J Text Inst 1946:T285–92.
[18] Werkmeister JB, Slocum AH. Theoretical and experimental determina-
teresis and thus can be used as a spring especially when the strain tion of capstan drive stiffness. J Int Soc Precis Eng Nanotechnol 2007
about any particular preload is kept less than ±10%. January;31:55–67.
[19] Kulachenko A, Uesaka T. Direct simulation of fiber network deformation and
failure. Mech Mater 2012:1–14.
4.4. Densification
[20] Gibson LJ, Ashby MF. The mechanics of three-dimensional cellular materials.
Proc R Soc Lond Ser A 1982:43–59.
Preloading the samples (densification) to a fixed thickness and [21] He G, Liu P, Tan Q. Porous titanium materials with entangled wire struc-
ture for load-bearing biomedical applications. J Mech Behav Biomed Mater
then cycling them by a fixed amount about the preloaded thick-
2012:16–31.
ness revealed a linear region for highly porous structures where the [22] Nielsen LF. Elasticity and damping of porous materials and impregnated mate-
stiffness was constant and repeatable and can be characterized as rials. J Am Ceram Soc 1984;67:93–8.
a function of porosity. Thus it is foreseeable to vary preload, geom- [23] Baudequin M. Non-linear elastic behavior of light fibrous materials. Eur Phys J
B 1999;12:157–62.
etry, and materials to acquire a desired linear stiffness for highly [24] Pawlak JJ, Keller DS. The compressive response of a stratified fibrous structure.
porous materials used as springs acting over a wide area (Fig. 9). Mech Mater 2005;37:1132–42.