Engineering Structures 152 (2017) 750–757

Contents lists available at ScienceDirect

Engineering Structures
journal homepage: www.elsevier.com/locate/engstruct

Enhancing the teaching of structural dynamics using additive

manufacturing
Lawrence Virgin
School of Engineering, Duke University, Durham, NC 27708, USA

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

Article history: This paper provides a companion study to a previous paper by the same author (Virgin, 2017). In that
Received 24 July 2017 paper, 3D printing was used to provide a hands-on experience for students of (linear) structural analysis
Revised 16 August 2017 based on the lateral stiffness of plane frames. In this paper, a related set of structural plane frames is
Accepted 25 September 2017
investigated in terms of their natural frequencies, perhaps the fundamental feature in structural dynam-
Available online 5 October 2017
ics. Again, a 3D printer is used to provide a variety of parameter variations, and the extent to which cer-
tain simplifying analytical assumptions are justified is assessed.
Keywords:
Ó 2017 Elsevier Ltd. All rights reserved.
Structural dynamics
Flexural stiffness
Thermoplastic
3D printing
Education

1. Introduction tional behavior of rectangular portal frames can be described in

Rectangular, plane, portal frames offer compelling pedagogical
opportunities in the context of learning structural dynamics. Their
relatively simple geometric form allows a variety of comparative
studies to be made, over and above single structural members.
Additive manufacturing is an ideal mechanism for producing nom-
inally similar structures to within relatively high specifications,
and the widespread availability of 3D printers means that it is pos-
sible to incorporate an accessible practical aspect to otherwise the-
oretical studies in structural dynamics. Obtaining the natural
frequencies of plane frames is a relatively simple task, especially
for frames consisting of slender members, such that the behavior
is dominated by elastic flexure [2]. Analysis is especially simple
for frames undergoing sway motion - often the case for practical
geometries associated with buildings in which floors are relatively
stiff in comparison with columns, for example.
The analysis for free vibrations tends to be a little more involved
than for stiffness, since mass effects need to be considered. Fur-
thermore, in terms of modal analysis, the lowest few frequencies
may be important even in linear regimes. Part of the reason for this
is that under external excitation, any of the lowest modes can be
excited and dominate subsequent dynamic behavior, especially in
systems with relatively low damping. However, in terms of practi-
cal testing the extraction of frequencies is a standard procedure,
once a time series has been measured. We shall see that the vibra-
E-mail address: l.virgin@duke.edu
terms of various parameter ratios, essentially relating the distribu-
tion of stiffness and mass between the beam and columns, and pro-
viding an accessible means of allowing a schematic parametric
study to be conducted.
Understanding the dynamic behavior of structures has also
proved to be useful in damage assessment [3] and finite element
model updating [4,5], and of course earthquake engineering is a
natural context for this type of structural dynamics [6].
1.1. Some basic modeling
A spring (of stiffness coefficient K) attached to a mass (M), is
modeled, via Newton’s second law, by a second-order ordinary dif-
ferential equation:
€ þ KX ¼ 0;
MX ð1Þ
where the overdot means differentiation with respect to time, i.e.,
X_  dX=dt, and given some non-trivial initial deflection Xð0Þ (away
from equilibrium at X ¼ 0) the mass will oscillate according to
XðtÞ ¼ Xð0Þ cos xn t ð2Þ
in which the key response characteristic is the natural frequency:
pffiffiffiffiffiffiffiffiffiffiffi
xn ¼ K=M : ð3Þ
This harmonic motion reflects the continual exchange of energy
between the potential energy stored in the spring ðV ¼ 12 KX 2 Þ and
kinetic energy associated with the moving mass ðT ¼ 1 MX_ 2 Þ.
2

https://doi.org/10.1016/j.engstruct.2017.09.052
0141-0296/Ó 2017 Elsevier Ltd. All rights reserved.
 L. Virgin / Engineering Structures 152 (2017) 750–757
This spring-mass system and its natural frequency are repre-
sentative of a wider class of structural system in which the stiff-
ness and mass properties are typically associated with elastic
deformation and distributed mass. Thus, simplistically, we see that
the stiffer a structure, the higher the frequency of vibration, and
the greater the mass, the lower the natural frequency. All mechan-
ical and structural systems possess some form of energy dissipa-
tion or damping, thus oscillations have a tendency to decay in
the absence of external excitation. However, in similarity with
many practical systems (including the experimental systems to
be described later) this damping will be relatively small, and will
not have any significant effect on natural frequencies.
Given a rectangular, planar, portal frame, it is instructive to
assess how natural frequencies depend on a range of parameters
including material properties, boundary conditions, and geometry.
It is the latter that will provide the primary focus of interest in this
paper. For this type of continuous system there are an infinite
number of natural frequencies, but it is often the case that they
can be modeled by a very much lower-order model, and sometimes
using a single-degree-of-freedom (SDOF) model. Consider a square
portal frame, in which each of the three members is made of the
same material and has the same length, as shown in Fig. 1(a). How-
ever, the cross-beam is much stiffer than the columns. The bottom
of each vertical member is fixed to a rigid base and the top ends
join the cross-beam with moment-transmitting joints. This system
can be modeled as a SDOF system with the lateral deflection X
completely describing the overall deflection of the system: pure
sidesway.
The simplified model can be developed using the well-known
concept of an equivalent (lumped) spring and mass. Details of this
approach can be found in a number of texts, for example [7], but
essentially use is made of equivalent potential and kinetic energies.
The deflected shape of the frame under consideration depends
entirely on the columns: there are many functions that can
describe their shape (independently of magnitude). For example,
the following polynomial for lateral deflection w satisfies the
boundary conditions of zero rotation at both ends of the structural
member (w0 ¼ 0 at x ¼ 0 and L) but allowing one end to translate
(sway) relative to the other [8]:
 3  
2x 2x
wðxÞ ¼ 1  3 1 þ 2:
L L
It is also well-established that for a beam-like structural component
RL
the strain energy stored in flexure is V ¼ 12 EI 0 ðw00 Þ2 dx, where
w ¼ wðxÞ is the deflected shape, and EI is the flexural rigidity, i.e.,
the product of Young’s modulus E and the second moment of area
I. Evaluating this expression using Eq. (4), and equating it with
Fig. 1. Rectangular (portal) plane frames. (a) the pure sway case, (b) the more general case.
751
the potential energy of a spring, we obtain an equivalent stiffness
K e ¼ 12EI=L3 (for a single column).
Similarly, the effective mass associated with the sway motion of
a beam (column) can be obtained using the kinetic energy for the
RL 2 R
continuous system T ¼ 12 qA 0 w _ 2 dx, where q is the
_ dx ¼ 12 m 0L w
density and m is the mass, both per unit length. We can then
equate this with the point-mass kinetic energy T ¼ 12 M e v 2 ¼
1
2
M e X_ 2 , and again using Eq. (4) as the shape function we obtain
M e ¼ 0:37 mL. We shall see later that these values are associated
with the (1,1) elements in the beam stiffness and consistent mass
matrices. It is not surprising that only a portion of the total mass
contributes, given the fact that one end is not moving.
Thus, for a column vibrating with pure-sway motion as shown
pffiffiffiffiffiffiffiffiffiffiffiffiffiffi
in Fig. 1(a), we have a natural frequency xn ¼ K e =M e ¼
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
12EI=0:37qAL4 ¼ 5:69 EI=mL4 , where m ¼ qA is the mass per
unit length. The coefficient is much lower than that for the (much
stiffer) clamped-clamped beam with no sway (22.37), which is
associated with a symmetric vibration mode [9]. Solving the gov-
erning partial differential equation [8], results in a lowest natural
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
frequency x1 ¼ 5:59 EI=mL4 ; the slight difference with the
lumped model results from Eq. (4) not exactly describing the shape
of the column during the motion.
In the context of a rectangular frame in pure sway, we thus have
columns that supply the sway stiffness and a rigid bar that supplies
the mass (in addition to a smaller amount of effective mass con-
tributed by the columns). For a rectangular portal frame we thus
have [8]
" #1=2
1 12RððEIÞc Þ
f1 ¼ ; ð5Þ
2p L3 ðM þ 0:37RM c Þ
in which M is the total mass of the beam, and the subscript c refers
to the column. For frames made of the same material throughout
the Young’s modulus E and density q provide a simple square root
pffiffiffiffiffiffiffiffiffi
scaling, f / E=q. We see that unless the mass of the columns is
negligible (relative to the mass of the beam) the natural frequency
depends in a non-simple way on the properties of the columns. But,
ð4Þ suppose the mass of the columns is very small compared to the
mass of the beam: M  RM c . Then we can more easily isolate the
effect of varying a single parameter. For example, with all other
things held equal, we have f / ðLÞ3=2 for a changing column length.
We also see that the natural frequency is independent of the width
b (since both the stiffness and mass scale linearly with b and thus
cancel). This is a slightly counter-intuitive result but also applies
752 L. Virgin / Engineering Structures 152 (2017) 750–757

 
to a prismatic cantilever, for example. We shall see the extent to ½K  x2 ½M ¼ 0: ð9Þ
which certain simplifying assumptions might be valid later.
In the current study the frame members are slender, so that For the uniform, plane, portal frame, the structure stiffness and con-
behavior is predominantly in flexure (negligible shear deforma- sistent mass matrices are given explicitly by [11,13,14]

2 3
ðAL2 =I þ 12Þ 0 6L AL2 =I 0 0
6 7
6 ðAL2 =I þ 12Þ 6L 0 12 6L 7
6 7
6 2
6L 2 7
36 8L 0 2L 7
K ¼ EI=L 6 7 ð10Þ
6 ðAL2 =I þ 12Þ 0 6L 7
6 7
6 7
4 sym: ðAL2 =I þ 12Þ 6L 5
8L2

tion). For example, reference [10] introduces a parameter and

c ¼ AL2 =I for which coupling with axial vibration is generally neg-
ligible if c > 104 , which is always the case for our frames - the least
slender column element has c  3  104 . Basically the same mea-
sure, popularly used in buckling, is the slenderness ratio SR ¼ L=r,
pffiffiffiffiffiffiffi
in which r ¼ I=A is the radius of gyration, and in the following
material the least slender column corresponds to SR  340, which
is considered relatively slender and leads to behavior dominated
by flexure [11].
The rest of this paper seeks to establish natural frequencies and
their geometric parameter dependence, and extending this to the
multi-degree-of-freedom (MDOF) model in which the beam might
be relatively stiff or flexible including when the sway approxima-
tion is no longer appropriate [12–15]. Our goal here is to exploit
versatile 3D printing for deepening an appreciation for structural
dynamics, and specifically in terms of printing a range of elastic,
relatively slender, plane frames. This practical component can be
used to enhance the understanding of matrix methods in particu-
lar. Additionally, with relatively simple loading and the given
boundary conditions it is then possible to assess the role of geom-
etry and its influence on certain dynamic properties, specifically
frequencies of the frames using measured data.
2. The case of a uniform portal frame
As a convenient starting point, we consider the case in which
the cross-beam has the same properties (length, area, second
moment of area, density) as the columns, a case we shall refer to
as the uniform square portal frame. Reference [1] showed the lat-
eral stiffness in this case to be approximately 0.7 times the stiffness
of the pure sway case. However, for frequency we also have a
changing mass in addition to the changing stiffness, and the anal-
ysis is no longer achieved using a SDOF model.
Consider the square portal frame as shown in Fig. 1(b), with
LB ¼ LC ¼ L; AB ¼ AC ¼ A; ðEIÞB ¼ ðEIÞC ¼ EI. The stiffness method
can be used to extract natural frequencies from this system by
solving a generalized eigenvalue problem. For a MDOF model we
generalize Eq. (1) using matrix notation (in which the standard
stiffness and mass element matrices can be found in appendix A):
€ þ ½KX ¼ 0;
½MX
with a standard harmonic solution of the form
XðtÞ ¼ Xeixt :
Upon substitution of Eq. (7) into Eq. (6) we obtain
 
½K  x2 ½M X ¼ 0;
and for non-trivial solutions we solve the determinant
2 3
296 0 22L 70 0 0
6 296 22L 0 13L 7
54
6 7
6 7
6 8L2
0 13L 3L2 7
M ¼ qAL=4206
6
7 ð11Þ
6 296 0 22L 7
7
6 7
4 sym: 296 22L 5
8L2
respectively, where the elements are conventionally ordered
ðX 1 ; Y 1 ; h1 ; X 2 ; Y 2 ; h2 Þ, see Fig. 1(b). It is then a simple step to solve
for the natural frequencies (in rad/s) and vibration mode shapes
using standard routines in Mathematica [16] or Matlab [17].
For example, let’s assume an overall frame dimension (relevant
to the physical dimensions to be 3D printed later) of L ¼ 0:1 m, and
a rectangular cross-section area ðb  dÞ ¼ ð0:01 m  0:002 m), giv-
ing an area A ¼ 20  106 m2, second moment of area
12 4 2 2 3
I ¼ 6:67  10 m , and thus AL =I  ðL=rÞ ¼ 30  10 , where r
is the radius of gyration. As mentioned earlier, these parameters
relate to a geometry that may be considered relatively flexible,
and this facilitates relatively easy measurements. Also, for pur-
poses of physical relevance for the 3D printed frames, we assume
a Young’s modulus E ¼ 2:1 GPa and a material density of
q ¼ 1040 kg/m3 (specific gravity 1.04); that is, the thermoplastic
material is roughly 100 times less stiff than steel. With these
parameters, solving the generalized eigenvalue problem associated
with the stiffness method (assuming one element per structural
member) we obtain a lowest natural frequency of 41.9 Hz. It is
interesting to note that the second lowest frequency (associated
with a symmetric mode of vibration) is 197.4 Hz. Scrutinizing
Eqs. (10) and (11) we confirm that the frame width b (into the
page) does not influence the frequencies since it enters into the
3
common factors I ¼ bd =12 and A ¼ bd in both ½K and ½M equally.
The interested reader can also observe the potentially beneficial
role played by neglecting axial deformation and exploiting symme-
try, i.e., where Y 1 ¼ Y 2 ¼ 0, or h1 ¼ h2 . This is an exercise in model
reduction and related to the rank reduction of a matrix in linear
algebra [18,19]. In addition, singular value decomposition (SVD)
can be used to identify the dominant degrees of freedom. However,
the need for this kind of simplification is less pronounced with the
ð6Þ sheer convenience of Mathematica [16] and Matlab [17], unless the
structure under consideration has an exceedingly high number of
degrees of freedom.
ð7Þ
3. The more general case and experiments
ð8Þ We now consider the much wider class of structures in which
the column and beam properties are not so narrowly restricted,
and how measurements are taken from the printed specimens. It
 L. Virgin / Engineering Structures 152 (2017) 750–757
Fig. 2. Printed frames. Each frame clamped at the base and vibration initiated in sway and measured at the ‘top’ end.
is instructive to consider a variety of geometries, in which assess-
ing the effect of changing a single parameter is clearly useful. Fig. 2
shows the range of plane frames (the test matrix) produced using
the 3D printer. They differ according to column lengths, beam
thicknesses and overall widths, thus providing a range of parame-
ter combinations. The columns lengths are 100 mm and 150 mm;
the beam lengths 50 mm and 100 mm; the beam thickness 2 mm
and 8 mm; and the overall width 10 mm and 20 mm. These dimen-
sions were chosen mostly based on providing frequencies that
could be adequately measured using the capabilities of a standard
proximity laser.
The 3D printer used in this study (Stratsys) uses ABS thermo-
plastic [20] for which we take E ¼ 2:1 GPa and q ¼ 1:04 g/cm3,
and the frames are printed into an integral rigid base providing
clamped boundary conditions at the base of each column [21].
Motion was initiated by applying an initial deflection and then
releasing (an impact can also be used), and the subsequent time
series was measured using a proximity laser (with a sampling rate
of 750 Hz). These data were then subject to spectral analysis and
the fundamental frequency extracted. Repeat tests were conducted
in order to establish statistical variations.
Fig. 3 shows the general arrangement of the experimental set-
up. In this particular case the frame is a two-story variation on
the basic theme and will be presented in a later section. However,
laser
Fig. 3. The experimental test set-up. Frames are clamped at the base, subject to a
sudden disturbance, with the motion measured at the corner.
753
we observe how data are acquired using the laser at the level of the
frame beam. The small tabs visible at the story levels were added
to facilitate initial perturbations (including their utility in related
stiffness measurements [1]) - they add a negligible amount to
the stiffness and a little to the mass. All the data acquired were
from free-decay time series ðt; XÞ, from which the natural fre-
quency was extracted using well-established signal processing,
specifically the Fast Fourier Transform (FFT) within Matlab.
An example measurement is shown in Fig. 4(a). The frame was
given a small lateral displacement at the top corner and released.
The ensuing vibration was measured using the laser proximity sen-
sor [22] at the opposite corner and results in a slowly decaying
(lightly-damped) harmonic oscillation. The initial displacement is
not important as long as it is ‘small’ (in this case the we have
X 1  7 mm for a column length of 100 mm). The FFT in part (b)
confirms the frequency of oscillation at 54.4 Hz (the value reported
later in Table 1 for frame A). A brief discussion of the effects of
energy dissipation can be found in Appendix B.
4. Results
For this kind of parametric study the main emphasis is on the
effect of geometrical changes. As such, the exact values of E and
q are less of an issue because all the frames are made of the same
material, printed in the same manner, using the same printer, and
tested under nominally the same conditions [21].
Table 1 contains the outcome from both theoretical and exper-
imental studies. A total of eight (single bay, single story) frames
was printed, tested, and analyzed. For each frame, four frequencies
were obtained using: the stiffness method (column 6); a SDOF
sway analysis (column 7, where appropriate); Solidworks [23] (col-
umn 8); and the experimentally measured value (column 9).
This table contains a wealth of information. We first address the
validity of the pure sway approximation. Frames A, D, and H have
the thicker cross beam: four times the thickness of the beams in
the other frames, and hence 64 times the second moment of area
of the columns. In these cases the pure sway assumption provides
an excellent result when compared with the stiffness method. Also,
for these sway frames we can make comparisons: earlier we
observed that f / ðLÞ3=2 , and thus we would expect frame D to
have a frequency that is approximately 1:53=2  0:54 times the
frequency of frame A (which Table 1 confirms).
Comparing frames F and G confirms the independence of fre-
quency on the overall width dimension ðbÞ. In general, we observe
754 L. Virgin / Engineering Structures 152 (2017) 750–757

Fig. 4. A sample of the acquired experimental data (taken from frame A). (a) free decay time series, (b) corresponding frequency content (FFT). A is in arbitrary units.

Table 1
Frame geometries (with reference to Figs. 1 and 2). In all cases dC ¼ 2 mm. Frequencies in Hz.

Frame # dB (mm) b (mm) LC (mm) LB (mm) f 1 (theo) f (sway) f 1 (sw) f 1 (exp)

A 8 20 100 50 38.5 38.6 37.3 43.6
B 2 20 150 50 26.1 – 24.9 26.9
C 2 20 100 50 53.2 – 50.85 54.4
D 8 20 150 50 19.7 19.7 19.1 22.3
E 2 10 100 50 53.2 – 50.3 52.6
F 2 10 100 100 41.9 – 39.9 39.6
G 2 20 100 100 41.9 – 40.3 41.3
H 8 20 100 100 29.3 29.4 29.1 32.0

the anticipated trend that the taller the frame, the lower the natu-
ral frequency since the stiffness is proportional to EI=L3 [1].
The experimentally measured frequencies agree well with the
results from the stiffness method, especially for the cases in which
the pure sway assumption is not valid (i.e., the 2 mm thick cross-
beam cases). Interestingly, for the thicker cross-beam cases the
stiffness method consistently gives slightly lower frequencies than
experiment (about 10%). But again, given the inherent uncertain-
ties in E and q, and in the thickness dC that tended to consistently
be a little greater than the 2 mm specified to the printer (and these
errors propagate), it is encouraging to observe the consistent scal-
ing within the experimental data.
4.1. Multi-column frames
A simple modification to the rectangular frame that can easily
be made using the 3D printer is the addition of further columns.
Just as an example, consider the frame H. If we add one, and then
four, more columns between the outer two (see Fig. 5) then we
might expect an increase in stiffness and hence natural frequency.
Eq. (5) suggests a frequency increase from 29.3 Hz to 34.7 Hz
( 18%), with most of the difference arising from the stiffening
effect of the additional column. The fundamental frequency was
measured at 39.2 Hz, (and a Solidworks FEA gave 34.4 Hz), thus
confirming the stiffening effect of the additional column. It does
add a little mass, but not much. With six total columns (Fig. 5(c))
the sway frequency now increases to 44.4 Hz ( 51% greater than
the two-column baseline case), with a measured value of 50.4 Hz
(and a Solidworks result of 44.3 Hz). The columns need not have
the same cross-sectional properties, it is their cumulative lateral
stiffness and mass that enters into Eq. (5) for the sway-
dominated frame. Even in the case in which the cross-beams are
not especially stiff relative to the columns, the additional columns
still have a tendency to stiffen the cross-beam by providing inter-
mediate support when the overall beam length is maintained. For
an arbitrary arrangement of frame members we resort back to the
conventional stiffness method. An early example of the computa-
tion of frequencies (including excellent experimental agreement)
can be found for square multi-bay frames in [24,25]. They found
that, all other things held equal, such a frame with four equal bays,
for example, would exhibit a lowest natural frequency about 4.5%
less than the two equal bay case, and about 10.5% less than the sin-
gle bay square frame; thus additional mass (and wider column
spacing) makes a larger contribution to the frequency than in the
pure sway case. See Appendix C for a further brief discussion.

Fig. 5. Additional columns. (a) The standard, single bay frame, (b) with one additional column, (c) a 3D-printed frame with four additional columns.
 L. Virgin / Engineering Structures 152 (2017) 750–757 755

5. Two-story sway frames
We now consider two-story frames. We shall again initially
focus on the sway case in which the flexural stiffnesses of the
cross-beams are much larger than those of the columns. In this
case we effectively have a two-DOF system, and extracting the fre-
quencies and mode shapes we can then compare with a 3D printed
test specimen. If we consider a two-story sway frame with the total
mass split evenly between both beams, and the columns of equal
stiffness, then a simple analysis leads to the determinant:
   
24EI 2 1 1 0 

 3 1 1  x M 0 1  ¼ 0
2
L discrepancy between the experimental data and the theoretical
Solving this system gives the eigenvalues x21 ðML3 =EIÞ ¼ 9:167 and
x22 ðML3 =EIÞ ¼ 62:83, with the second frequency about 2.62 times
higher than the first, and the corresponding eigenvectors
ð/1 ; /2 Þ ¼ ð0:525731; 0:850651Þ and ð0:850651; 0:525731Þ.
Sketches of the mode shapes (eigenvectors) are shown in Fig. 6(a)
and (b).
The left side of Fig. 7 shows a photographic image of two
3D-printed two-story frames, with the following dimensions:
 (a): LC ¼ 0:075 m, LB ¼ 0:05 m; b ¼ 0:02 m; dB ¼ 0:008 m.
 (b): LC ¼ LB ¼ 0:1 m; b ¼ 0:02 m; dB ¼ 0:002 m.
Using the appropriate values for the frame in part (a) the 2-DOF
sway analysis (Eq. (12)) gives frequencies f 1 ¼ 43:0 Hz and
f 2 ¼ 112:7 Hz. Both frames were tested experimentally: displaced
horizontally and then released, generating free-decay time series.
A variety of initial conditions was used to promote the participa-
tion of the second mode, i.e., the two stories were initially dis-
placed in opposite directions (see Fig. 6(b)). In contrast to the
time series for the essentially SDOF system in Fig. 4, the two-
story data is initially not necessarily a simple harmonic. A second
mode participates in the motion, and despite the fact that it tends
to damp-out more quickly, its contribution can still be detected in
the frequency spectrum as a second (smaller) peak, see Fig. 7(d).
The measured natural frequencies were f 1 ¼ 43:7 Hz and
f 2 ¼ 121:4 Hz. A Solidworks FEA analysis [23] provided the fre-
quencies f 1 ¼ 38:1 Hz and f 2 ¼ 101:48 Hz. Although there is some
ð12Þ
results, they both exhibit the 2.6–2.7 ratio between the first two
frequencies. Clearly the 2 DOF sway analysis is more appropriate
for the geometry in part (a). In Fig. 7(b) the frame is closer to the
uniform geometry encountered in Section 2, i.e., not adequately
modeled as a sway frame. The lowest two natural frequencies for
this frame were measured at 20.7 Hz and 68.4 Hz, where the FFT
was based on data acquired from both story levels (with almost
no difference, as expected). A corresponding Solidworks FEA anal-
ysis [23] gave 19.1 Hz and 62.1 Hz for the lowest two modes, and
they are shown visually in Fig. 6(c) and (d), in which some beam
flexure can be observed. It is also relatively easy to investigate
the issue of orthogonality in multiple DOF (and continuous) sys-
tems, see appendix D. For more complicated geometries, e.g., space
frames, and especially in nonlinear analysis, the frequencies are
not necessarily so well-separated and may even change position
with parameter changes.

(a) (b) (c) (d)

Fig. 6. The modes of vibration of two-story frames. (a) Lowest mode, pure sway, (b) second mode, sway mode, (c) lowest mode, general case, (d) second mode, general case.

Fig. 7. Images of two-story frames: (a) the sway case, with thick beams, (b) the uniform case, (c) a measured time series from the frame in part (b), (d) corresponding FFT.
756 L. Virgin / Engineering Structures 152 (2017) 750–757

6. Discussion and other practical considerations
The natural frequencies of rectangular portal frames depend on
a number of factors, including the mass of the beam relative to that
of the columns; the flexural rigidity of the beams relative to that of
the columns; and the length of the beams relative to that of the
columns. The limiting case of rigid beams leads to a very simple
approximate model based on a pure side-sway mode of oscillation,
essentially described by a single DOF, or n-DOF for an n-story
frame.
The 3D printer is an accessible means of producing these
frames, and physical testing is straightforward. A comparative
exercise is easily achieved to confirm the role played by various
geometric parameters. The material is elastic within certain ranges
of deformation, although a little care must be taken with 3D-
printer thermoplastic since its mechanical properties may not be
quite as uniform or homogeneous as desired, but since the main
point of the material here is comparative, the issue of high-
fidelity material characterization is not crucial.
Small tabs can be printed at the frame corners to facilitate load-
ing and initial displacement. They add a little mass but not stiff-
ness. Similarly small fillets are used to round-off corners to
mitigate the possible effects of stress concentrations.
The measuring device (an OPTO NCDT 1302 proximity laser was
used in this study [22]) is not especially sophisticated, however the
measurement rate of 750 Hz meant that the natural frequencies of
the frames needed to be somewhat less than this limit (see Nyquist
frequency in [28]). That is, the frames couldn’t be too stiff other-
wise the frequency would be too high for the laser. There are a
number of issues associated with signal processing, including alias-
ing, windowing, etc., and these come into play especially for more
complicated structures, and the interested reader is referred to ref-
erence [7] for more details.
As mentioned in the companion paper to this one [1], it is a sim-
ple matter to vary the geometry according to student ID numbers
column (stiffness and mass) elements, and utilizes a variety of
techniques from linear algebra, mainly solving the generalized
eigenvalue problem. 3D-printed plane frames can provide a bene-
ficial practical appreciation, especially in terms of how the geome-
try affects lateral stiffness and vibration frequencies, and the
extent to which simplifying assumptions, e.g., pure sway, are justi-
fied. Making simple geometric changes to a basic configuration
enables a direct comparative study to be conducted. Furthermore,
a comparison between theory and experimental data is easily
achieved, based on simultaneously testing and analyzing simple
frame structures. The sway case is especially easy to analyze, but
for moderately complex structures it is still straightforward to
compare structural behavior of a physical specimen with the
dynamic stiffness method and any number of FEA packages, but
especially Solidworks since the printed specimens are drawn
directly from Solidworks data files.
In summary, the material described in this paper, at least based
on anecdotal evidence and preliminary student feedback, suggests
that a deeper appreciation of structural behavior can be developed
using this dual theoretical-experimental approach.
Acknowledgement
The author appreciates the assistance of Joe Giliberto in exper-
imental aspects of this paper.
Appendix A. Beam element stiffness and mass matrices in
global coordinates
For completeness, the element stiffness matrix ½K and the con-
sistent mass matrix ½M in global coordinates for a prismatic
beam/column are included here [13,26]. A member (in the plane)
is oriented from the local (horizontal) to the global axes via a rota-
tion h with c ¼ cos h and s ¼ sin h:

2 3
ðEA=LÞc2 ðEA=LÞcs ðEA=LÞc2 ðEA=LÞcs
6 7
6 þð12EI=L3 Þs2 ð12EI=L3 Þcs ð6EI=L2 Þs ð12EI=L3 Þs2 þð12EI=L3 Þcs ð6EI=L2 Þs 7
6 7
6 7
6 ðEA=LÞs2 ð6EI=L2 Þc ðEA=LÞcs ðEA=LÞs2 ð6EI=LÞc 7
6 7
6 þð12EI=L3 Þc2 þð12EI=L3 Þcs ð12EI=L3 Þc2 7
6 7
6 7
6 4EI=L ð6EI=L2 Þs ð6EI=LÞc 2EI=L 7
½K ¼ 6
6
7
7 ð13Þ
6 ðEA=LÞc2 ðEA=LÞcs ð6EI=L2 Þs 7
6 7
6 þð12EI=L3 Þs2 ð12EI=L3 Þcs 7
6 7
6 7
6 sym: ðEA=LÞs2 ð6EI=LÞc 7
6 7
6 7
4 þð12EI=L3 Þc2 5
4EI=L

2 3
in order to produce a somewhat random variety of geometries, 140c2 70c2
within reasonable limits. Furthermore, students can be asked to 6 þ156s2 16cs 22Ls þ54s2 16cs 13Ls 7
6 7
produce frames that exhibit a required frequency within certain 6 7
6 140s 2
70s 2 7
constraints, with weight an obvious restriction especially useful 6 7
6 þ156c2 22Lc 16cs þ54c2 13Lc 7
in the aerospace context. It is also possible to harmonically excite 6 7
6 7
these frames to examine resonance. However, this involves the use qAL 6
6 4L2 13Ls 13Lc 3L2 77
½M ¼
of shakers, and goes beyond the relatively simple nature of the 420 6
6 140c2 7
7
6 7
experiments described here. 6 þ156s2 16cs 22Ls 7
6 7
6 140s2 7
6 sym: 7
6 7
7. Concluding remarks 4 þ156c2 22Lc 5
4L2
The teaching of structural dynamics for frame-type structures ð14Þ
typically relies on matrix methods, incorporating beam or beam-
 L. Virgin / Engineering Structures 152 (2017) 750–757
Appendix B. A Note on damping
Energy dissipation, or damping, in mechanical/structural sys-
tems is inevitable. For both modeling and analysis reasons it is usu-
ally assumed that any damping forces are directly proportional to
velocity: linear viscous damping F d ¼ C x. _ In terms of a damping
ratio f  C=C c , where C c ¼ 2M xn , we obtain a damped natural fre-
pffiffiffiffiffiffiffiffiffiffiffiffiffi
quency xd ¼ xn 1  f2 for a single-DOF system. However, Fig. 4
(a) displays a very slowly decaying oscillation, and use can be
made of the logarithmic decrement approach [9]:
 References
1 A0
f¼ ln ; ð15Þ
2p n An
where A0 is an initial amplitude, and An is the amplitude n cycles
later. Using the free decay data results in an estimate of f  0:007,
or 0:7%, and thus there is no effective difference between the
damped and undamped natural frequencies for the systems studied
here. Damping tends to be greater for higher modes.
Appendix C. An ‘Exact’ analysis
Despite the ubiquity of the stiffness method/FEA, it is possible
to obtain frequencies and mode shapes for portal frames based
on their governing partial differential equations of motion
[24,25,27]. This is a challenging undertaking, given the non-
simple boundary conditions at the ‘free’ corners. However, follow-
ing references [8,10] the lowest few roots of the characteristic
equation can be obtained and tabulated according to three non-
dimensional groupings: mC =mB ; ðEC IC Þ=ðEB IB Þ; and LC =LB . If we take
frame A as an example, these ratios correspond to 1, 1, and 2,
respectively. Using the tables in [10] and adopting trilinear inter-
polation results in a frequency of 50.9 Hz. This compares very well
with the Solidworks result, for example. Solidworks uses the same
STL file for printing and conducting FEA and typically uses many
elements for each structural member. For example, the analysis
in [10] shows that using 4 elements per structural member gives
essentially the same results as solving the characteristic equation.
Appendix D. Orthogonality
For systems with more than a single DOF, the approach is effec-
tively a part of modal analysis [28], i.e., the system can be trans-
formed into modal (principal) coordinates using the eigenvalues
and eigenvectors. The equations of motion are then uncoupled:
the vibration modes are orthogonal to the mass and stiffness
matrices. This can be verified from /Ti ½M/j ¼ 0 and /Ti ½K/j ¼ 0.
The vibration modes are of arbitrary magnitude but can be subject
to mass ortho-normalization for convenience, choosing
757
/T1 ½M/1 ¼ 1:0, for example, and this leads to /Ti ½K/i ¼ x2i . Orthog-
onality also applies in continuous (infinite-dimensional) vibration
problems, and forms an important component in system identifica-
tion, for example [29]. We also mention the role of damping in this
context. Providing the damping is assumed to be of the form
½C ¼ a½M þ b½K, in which a and b are real, scalar constants, i.e.,
proportional damping, then the damped equations of motion can
still be decoupled. However, we again emphasize that in the sys-
tems described in this paper the damping is extremely light.
[1] Virgin LN. Enhancing the teaching of linear structural analysis using additive
manufacturing. Eng Struct 2017;150:135–42.
[2] Timoshenko SP. Strength of materials. New Jersey: Van Nostrand; 1977.
[3] Law SS, Zhu XQ. Damage models and algorithms for assessment of structures
under operating conditions. CRC Press; 2009.
[4] Baruch M. Correction of stiffness matrix through vibration tests. AIAA J
1982;20:441–2.
[5] Friswell MI, Mottershead JE. Finite element model updating in structural
dynamics. Dordrecht: Kluwer Academic Publishers.; 1995.
[6] Chopra AK. Dynamics of structures. Prentice Hall; 1995.
[7] Palm WJ. Modeling, analysis and control of dynamic systems. Wiley; 1983.
[8] Blevins RD. Formulas for natural frequency and mode shape. Florida: Krieger
Publishing Company; 1979.
[9] Thomson WT. Theory of vibration with applications. Prentice Hall; 1981.
[10] Yang TY, Sun CT. Axial-flexural vibration of frameworks using finite-element
approach. J Acoust Soc Am 1973;53:137–46.
[11] Virgin LN. Vibration of axially-loaded structures. Cambridge University Press;
2007.
[12] Rubinstein MF. Matrix computer analysis of structures. Prentice-Hall; 1966.
[13] Przemieniecki JS. Theory of matrix structural analysis. Dover; 1968.
[14] Weaver W, Gere JM. Matrix analysis of framed structures. 3rd ed. NY: van
Nostrand Reinhold; 1990.
[15] McGuire W, Gallagher RH, Ziemian RD. Matrix structural analysis. Wiley;
1999.
[16] Mathematica 11.0, Wolfram Research Inc., Champaign, Illinois; 2016.
[17] MATLAB and Statistics Toolbox Release, The MathWorks, Inc., Natick,
Massachusetts; 2012.
[18] Strang G. Introduction to linear algebra. Wellesley-Cambridge Press; 2009.
[19] Kassimali A. Matrix analysis of structures. Cengage Learning; 2012.
[20] <http://www.stratasys.com/materials/fdm/absplus>.
[21] Virgin LN. On the flexural stiffness of 3D printer thermoplastic. Int J Mech Eng
Educ 2017;45:59–75.
[22] <http://www.micro-epsilon.com/>.
[23] <http://www.solidworks.com/>.
[24] Rieger NF, McCallion H. The natural frequencies of portal frames – I. Int J Mech
Sci 1965;7:253–61.
[25] Rieger NF, McCallion H. The natural frequencies of portal frames – II. Int J Mech
Sci 1965;7:263–72.
[26] Archer JS. Consistent mass matrix for distributed mass system. J Struct
Division ASCE 1963;89:161–77.
[27] Bishop RED. The analysis of vibrating systems which embody beams in flexure.
Proc Inst Mech Eng 1955;169:1031–50.
[28] Ewins DJ. Modal testing: theory and practice. Research Studies Press; 1984.
[29] Yang JN, Lei Y, Pan S, Huang N. System identification of linear structures based
on Hilbert-Huang spectral analysis. Part 1: Normal modes. Earthquake Eng
Struct Dynam 2003;32:1443–67.