You are on page 1of 9

Rotational and frictional dynamics of the slamming of a door

Pascal Klein, Andreas Müller, Sebastian Gröber, Alexander Molz, and Jochen Kuhn

Citation: American Journal of Physics 85, 30 (2017); doi: 10.1119/1.4964134


View online: https://doi.org/10.1119/1.4964134
View Table of Contents: https://aapt.scitation.org/toc/ajp/85/1
Published by the American Association of Physics Teachers

ARTICLES YOU MAY BE INTERESTED IN

Sliding down an arbitrary curve in the presence of friction


American Journal of Physics 85, 108 (2017); https://doi.org/10.1119/1.4966628

The motion of a ball moving down a circular path


American Journal of Physics 85, 124 (2017); https://doi.org/10.1119/1.4972177

Why trains stay on tracks


American Journal of Physics 85, 178 (2017); https://doi.org/10.1119/1.4973370

Weight of an hourglass—Theory and experiment in quantitative comparison


American Journal of Physics 85, 98 (2017); https://doi.org/10.1119/1.4973527

Life under a black sun


American Journal of Physics 85, 14 (2017); https://doi.org/10.1119/1.4966905

The physics of juggling a spinning ping-pong ball


American Journal of Physics 84, 936 (2016); https://doi.org/10.1119/1.4964104
Rotational and frictional dynamics of the slamming of a door
Pascal Kleina)
Department of Physics, University of Technology, Erwin-Schr€
odinger St. 46, 67663 Kaiserslautern, Germany
€ller
Andreas Mu
Section de Physique et Institut Universitaire de Formation des Enseignants, Universit
e de Genève,
Quai Ernest-Ansermet 24, 1205 Geneva, Switzerland
€ber, Alexander Molz, and Jochen Kuhn
Sebastian Gro
Department of Physics, University of Technology, Erwin-Schr€
odinger St. 46, 67663 Kaiserslautern, Germany
(Received 6 October 2015; accepted 18 September 2016)
A theoretical and experimental investigation of the rotational dynamics, including friction, of a
slamming door is presented. Based on existing work regarding different damping models for
rotational and oscillatory motions, we examine different forms for the (angular) velocity
dependence (xn, n ¼ 0, 1, 2) of the frictional force. An analytic solution is given when all three
friction terms are present and several solutions for specific cases known from the literature are
reproduced. The motion of a door is investigated experimentally using a smartphone, and the data
are compared with the theoretical results. A laboratory experiment under more controlled
conditions is conducted to gain a deeper understanding of the movement of a slammed door. Our
findings provide quantitative evidence that damping models involving quadratic air drag are most
appropriate for the slamming of a door. Examining this everyday example of a physical
phenomenon increases student motivation, because they can relate it to their own personal
experience. VC 2017 American Association of Physics Teachers.
[http://dx.doi.org/10.1119/1.4964134]

I. INTRODUCTION prone to air drag. While the velocity increases with the radial
distance from the axis of rotation, is it reasonable to assume
Each of us opens and closes several doors every day (Fig. 1). both Stokes friction (valid for small velocities/low Reynolds
Have you ever wondered how the motion of a closed could be numbers) and Newtonian friction (large velocities/high
characterized in terms of physics? After giving the door an ini- Reynolds numbers). Hence, we might expect that all three
tial angular velocity x0, everyday experience and qualitative types of friction occur simultaneously. Second, as the door
experiments indicate that angular velocity xðtÞ decreases while approaches the door frame, there is an accumulation of air
the door is rotating as friction occurs at the bearings and air that causes turbulence, so that air drag becomes dependent
drag counters the rotational motion. In this paper, we discuss on geometry. Third, there is a small space between the door
the theoretical description of this friction phenomenon and and floor whereas the top and side edges typically have
its experimental investigation using an acceleration sensor. none. Such an asymmetry could affect the air flow around
We also explain why it is worth considering this vivid example the door.
of damped rotational motion from both a theoretical and meth-
odological perspective. For didactic purposes, this example is B. Methodology of investigation
particularly suitable as students can relate it to their everyday
experience. First, we will derive an analytical expression xðtÞ given
the simultaneous presence of the aforementioned types of
A. Friction models friction and test it against the experimental data of a
slammed door. In this regard, we will explore whether the
Recent papers discussing damped oscillatory or rota- door frame has significant influence on the rotational data.
tional motion confirm continuing educational interest in We will also compare the data to less complex friction mod-
this subject.1–7 The theoretical description of damped rota- els to quantify how much accuracy is gained or lost (if any)
tional motion depends on basic friction models involving by removing or adding a particular friction term. From a
different forms of (angular) velocity dependence, e.g., / measurement perspective, smartphone acceleration sensors
x2 (Newtonian friction), / x (Stokes friction), and / x0 ¼ are very suitable for monitoring acceleration data in gen-
a ¼ const: (dry friction). Table I provides an overview of eral,8 and rotational data in particular;7,9 they are easy to
the recent literature, including the different experimental operate, comparatively inexpensive, and they possess suffi-
setups and the assumed frictional models. These authors cient measurement frequencies (up to 100 Hz) for measuring
most often applied a simplified theoretical framework with the slamming of a door (which only lasts for a few seconds).
a single dominant friction term, while neglecting other Taped onto a door, as shown in Fig. 1, data collection and
influences. As Mungan showed in the particular case of a data transfer are straightforward. Furthermore, the mobile
rotating disc, the model validity can increase if two friction character of the smartphone makes the slammed door experi-
models are assumed to occur simultaneously.2 ment accessible for many people. But in addition to investi-
Considering the friction effects of the slammed door, the gating the slammed door, we will use a low-cost laboratory
situation is not simple. First, the bearings induce a (assumed) experiment with better control of the physical parameters to
constant frictional torque and the large surface of the door is validate our findings. By doing so, the measurement range

30 Am. J. Phys. 85 (1), January 2017 http://aapt.org/ajp C 2017 American Association of Physics Teachers
V 30
ð ðh ðd ðw 
2
I ¼ qdoor r? dV ¼ qdoor dz x2 þ y2 dx dy
V 0 0 0
ðd ðw ðd ðw !
¼ hqdoor x2 dx dy þ y2 dx dy
0 0 0 0
1
¼ hqdoor ðdw3 þ wd 3 Þ
3
1
 hqdoor dw3 ¼ mw2 : (1)
3
The frictional torque sf, as explained above, will be
assumed to consist of constant, linear, and quadratic friction
pieces, as in

sf ¼ a þ bx þ cx2 ; (2)

with a, b, and c are constants greater than or equal to zero.10


The rotational version of Newton’s second law connects sf
with an angular acceleration dx=dt by
dx
sf ¼ I ; (3)
dt
where the minus sign indicates that the frictional torque
counters the rotational motion. Combining Eqs. (2) and (3),
we obtain an inhomogeneous, non-linear, first-order differen-
tial equation. Assuming an initial condition xð0Þ ¼ x0 , the
solution to this equation (see the Appendix) is given by

2x0 c þ b  c tanðct=2IÞ b
xDSN ðtÞ ¼   ; (4)
Fig. 1. A short-lasting, impulsive torque results in an angular momentum 2c 1 þ ð2x0 c þ bÞtanðct=2IÞ=c 2c
and thus causes the door to rotate about the z-axis. A smartphone accelera-
tion sensor monitors the rotational data. pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
where c ¼ 4ac  b2 . The subscript DSN indicates that the
solution is valid for a friction model including constant
can be extended and we can estimate the influence of the (dry), linear (Stokes), and quadratic (Newtonian) terms. This
thin air gap at the bottom of the door. analytical expression of xðtÞ will lead to the special cases
reported in the literature if particular friction terms are
removed (see the Appendix).
II. THEORETICAL BACKGROUND Table II summarizes all possible special cases, in particu-
A. Frictional torque on a rotating plate lar considering only dry friction (a 6¼ 0 and b ¼ c ¼ 0) or
neglecting dry friction (a ¼ 0, b 6¼ 0, and c 6¼ 0). The objec-
We treat the door like a thin plate (mass m ¼ qdoor V, tive is to determine which of these functions is best suited
width w, height h, thickness d  w) rotating around the z-axis, for describing our experimental results. In order to answer
as shown in Fig. 1. The mass density of the door, qdoor , is this question, we compare the fit quality among different
assumed to be constant throughout the door. Using Cartesian models statistically as well as relying on validity
coordinates x, y, z, the moment of inertia I is given by considerations.

Table I. Recent literature about different damping mechanisms. The model column gives the (angular) velocity dependence of the frictional torque assumed by
the authors as: Constant/Dry (D), Stokes (S), and Newton (N).

Authors Experimental setup Model Measurement technique

Simbach and Priest3 Mass with rod pivoted at a variable resistor D Potentiometer (LABVIEW)
Eadkhong et al.1 Thin cylindrical plate rotating around surface normal S Digital camera (video analysis)
Zonetti et al.4 Mass with rod pivoted at a variable resistor D Potentiometer
… Pendulum immersed in water S …
Mungan2 Identical to Eadkhong et al. DþS Force apparatus
Mungan and Lipscombe5 Circular disc on a thin rod N LabPro interface
Wang et al.6 Wooden board supported by a metallic rod D Range finder
… Magnetically damped aluminum plate S …
… Large area sail N …
Hochberg et al.7 Identical to Eadkhong et al. D Smartphone

31 Am. J. Phys., Vol. 85, No. 1, January 2017 Klein et al. 31


Table II. Analytical expressions for the angular velocity for several different • p1 ¼ p2: The model that fits the data best (smallest SSE)
friction models (the most general model is sf ¼ a þ bx þ cx2 ). The sub- is usually preferred without further statistical analysis.
scripts D, S, and N denote Dry, Stokes, and Newtonian models of friction, Instead, the fit parameters are evaluated with respect to
respectively.
scientific appropriateness. This case occurs when the
one-parameter models D, S, and N, respectively, are
Friction model Solution for xðtÞ
compared.
bx0
a¼0 xSN ðtÞ ¼ • p1 6¼ p2 : As it turns out, this case does not need to be con-
cx0  ðx0 c þ bÞ expðbt=IÞ
pffiffiffiffiffiffiffi pffiffiffiffiffi sidered here. A possible method to compare separate mod-
x0  a=c tanð ac t=IÞ els with a different number of parameters is based on the
b¼0 xDN ðtÞ ¼ pffiffiffiffiffiffiffi pffiffiffiffiffi
1 þ x0 c=a tanð ac t=IÞ Akaike information criterion (AIC).12
c¼0 xDS ðtÞ ¼ ðx0 þ a=bÞ expðbt=IÞ  a=b
x0
a ¼ 0, b ¼ 0 xN ðtÞ ¼ 3. Validity check
1 þ cx0 t=I
a ¼ 0, c ¼ 0 xS ðtÞ ¼ x0 expðbt=IÞ Apart from statistical considerations, we judge the validity
b ¼ 0, c ¼ 0 xD ðtÞ ¼ x0  at=I of the estimated fit parameters according to the physical
dimensions of the experiment. Independent of fit quality fac-
tors, we reject a model if the estimated model parameters
B. Approaches to comparing models make no scientific sense.
As motivated above, we use different theoretical models
to fit the experimental data. We consider different quality III. THE SLAMMED DOOR EXPERIMENT
factors (coefficient of determination R2, sum of square errors A. Experimental setup
SSE) to choose the optimal fit function. In order to judge if a
specific model is in fact significantly better than another A smartphone is taped to the door a distance r from the
model, statistical tests must be performed. Such tests are of axis of rotation (see Fig. 1). In our setup, we use an iPod
particular importance if the models being compared are touch and the app SparkVue13 to read out the acceleration ax
nested in each other. in x-direction, which is equivalent to the radial component of
total acceleration given that the smartphone rotates with the
door. These acceleration data are used to calculate the angu-
1. Comparing nested models lar velocity
Models are called nested if one model is a simpler case of rffiffiffiffiffiffiffiffiffiffi
another model. For example, the N model (sf ¼ cx2 ) is ax ðtÞ
x ðt Þ ¼ : (6)
nested in the SN model (sf ¼ bx þ cx2 ), which is nested in r
the DSN model (sf ¼ a þ bx þ cx2 ). Obviously, the unre-
Before starting the measurement, vertical alignment of the
stricted DSN model will fit experimental data at least as well
smartphone has to be assured, which is indicated by ax  0.
as the restricted DN (or the even more restricted N) model,
Then, the door is opened as wide as possible, the measure-
but the question is whether the unrestricted model in fact
ment is started, and the door is closed with some initial
provides a significantly better fit (statistically speaking) to
angular velocity x0 (hereafter referred to as “spin”). The
the data. This problem is approached by using an F-test.
measurement is repeated for large and small spin values.
Given n data points, a model with p2 parameters (model 2), Although one could control x0 using more sophisticated
and a model with p1 < p2 parameters (model 1) that is nested means, within the scope of a hands-on educational experi-
in model 2, the F-ratio quantifies the relationship between ment, it seems adequate to determine x0 post hoc by reading
the relative increase in sum-of-squares due to error of the fit off its value from the data.
(SSE) and the relative increase in degrees of freedom In our setup, we have r ¼ 0.95 m, while the physical
(df ¼ n  p þ 1) when going from the more complicated to parameters of the door are m ¼ 50 kg, h ¼ 2 m, w ¼ 1 m, and
the simpler model.11 Mathematically, the F-ratio is given by d ¼ 5 cm. Using Eq. (1) we then obtain I ¼ 16.67 kg m2 (the
smartphone’s moment of inertia is neglected).
ðSSE1  SSE2 Þ=SSE2
F¼ The error Dx is obtained by applying error propagation to
ðdf1  df2 Þ=df2 Eq. (6), giving
ðSSE1  SSE2 Þ=ðp2  p1 Þ    
¼  @x   
SSE2 =ðn  p2 þ 1Þ
: (5)
Dx ¼  Dax þ  @x Dr (7)
@ax   @r 
If we assume that model 2 does not provide a significantly Dax þ x2 Dr
better fit than model 1 (null hypothesis), F follows an F- ¼ : (8)
distribution with ðp2  p1 ; n  p2 þ 1Þ degrees of freedom. 2xr
The null hypothesis is rejected if the F-value is greater than To determine the sensitivity of the acceleration sensor, and
some critical value of the F-distribution (typically 0.05). thus its measurement accuracy, we positioned the smart-
phone on a desk, left it untouched and read out the sensor
2. Comparing separate (non-nested) models data for 20 s. We averaged the data and found r ¼ 0:02 m/s2
as the standard deviation, which will be considered as the
If the models being compared are not nested, the F-test is measurement accuracy Dax . The uncertainty of r is assumed
not valid and two cases can be distinguished as follows: to be Dr ¼ 0:001 m, the accuracy of our meter stick.

32 Am. J. Phys., Vol. 85, No. 1, January 2017 Klein et al. 32


Fig. 2. Angular velocity as a function of time with a large (left) and a small (right) initial spin, with error bars according to Eq. (8). In the first case, the initial
angular acceleration Dx=Dt is roughly 2:6 s1 =0:25 s  10 s2 (large initial spin), while in the second case it is 0:7 s1 =0:5 s  1:4 s2 (small initial spin).
Embedded graphs show the data fits according to theory.

B. Results and analysis (4) From (c)–(d): The angular velocity continues to
decrease, but at a faster rate than in the previous phase.
In the following, we analyze the experimental data (as pre- This behavior is caused by the influence of air drag
sented in Fig. 2 and Table III) from three points of view. between the door and the door frame and will be dis-
First, we discuss the empirical xðtÞ data qualitatively and cussed below.
relate them to the main phases of the motion process. (5) After (d): The door impacts the frame, which causes
Second, the free rotational motion data are fitted to the theo- some measurement noise. The door is at rest again.
retical solutions discussed above. We identify the dominant
factor in each case and compare the quality factors among Taking a closer look at phase (4), we notice that the angu-
the fit curves. Third, we estimate the validity of the parame- lar velocity decreases at a faster rate just before the door hits
ters found in the second step. the frame. In this position, air is accumulated in front of the
door causing turbulence, which increases the frictional tor-
1. Qualitative discussion of measurement data que (here the drag coefficient becomes dependent on geome-
try). Considering that the complete rotational motion from
Observing the measurement data in Fig. 2, we can identify (b) to (d) would cause a systematic error due to the influence
the following sections: of the door frame, we exclude phase (4) from the quantitative
(1) Before (a): The door is initially at rest. analysis and evaluate the motion only during phase (3), from
(2) From (a)–(b): The angular velocity suddenly increases (b)–(c).
from x ¼ 0 to x0. Note that we have shifted the time At first glance, it appears that the rotational data look quite
axis so that xðt ¼ 0Þ ¼ x0 . linear, which would lead to an assumption that the door is
(3) From (b)–(c): The angular velocity xðtÞ decreases due to exposed to a constant frictional torque. However, through
friction forces. The fit lines presented as embedded parameter estimations and comparisons of fit quality factors,
graphs in Fig. 2 refer to this section. we will show that this is not the case.

Table III. Fit parameters based on different frictional torque models for two impulses of different strength (corresponding to left and right graphs in Fig. 2,
respectively). The quality factors for the fits are given in columns R2 and SSE.

Frictional torque model a/I (s2) b/I (s1) c/I R2 SSE (103 )

Large spin (x0 ¼ 2:6 s1)


Dry (D) 0.58 – – 0.991 5.0
Stokes (S) – 0.26 – 0.995 2.9
Dry and Stokes (DS) <104 0.26 – 0.995 2.9
Newton (N) – – 0.11 0.997 2.0
Dry and Newton (DN) <109 – 0.11 0.997 2.0
Stokes and Newton (SN) – <104 0.11 0.997 2.0
Dry, Stokes, and Newton (DSN) <105 <104 0.11 0.997 2.0
Small spin (x0 ¼ 0:7 s1)
Dry (D) 0.06 – – 0.840 34
Stokes (S) – 0.10 – 0.872 28
Dry and Stokes (DS) <109 0.10 – 0.872 28
Newton (N) – – 0.19 0.900 21
Dry and Newton (DN) <109 – 0.19 0.900 21
Stokes and Newton (SN) – <108 0.19 0.900 21
Dry, Stokes, and Newton (DSN) <109 <109 0.19 0.900 21

33 Am. J. Phys., Vol. 85, No. 1, January 2017 Klein et al. 33


2. Comparing fit quality of different models coefficient (steel-steel) and g is the gravitational field
strength. Putting in values from Table IV, we obtain a=I 
The relevant portions of the two data sets presented in Fig. 2 102 s2, which is in reasonable agreement with values
have been fitted by various forms of Eq. (4), as given in Table
found in the literature when the assumption of dry friction is
II. The initial angular velocities were set to x0;l ¼ 2:6 s1 and
reasonable2,4,7 (considering the physical dimensions of the
x0;s ¼ 0:7 s1 for the large and small initial spins, respec-
rotating objects the authors used, as shown in Table I).
tively, and the parameters a/I, b/I, and c/I were determined via
However, in our setup, the door’s moment of inertia is
the MATLAB Curve Fitting Toolbox.14,15 Considering the large
initial spin case, all fit quality factors R2 presented in Table III I ¼ 16.67 kg m2, hence the dry frictional torque has an abso-
are close to 1, indicating that every model fits the data well. In lute value of 9.74 N m for the large-initial-spin experiment
contrast, the data from the small spin case show a greater mea- (0.93 N m for the small-initial-spin experiment), which
surement uncertainty, which can be explained by the inverse seems unreasonably large. Despite the deceptively good fit
x-dependence of the measurement error given in Eq. (8). In quality, the model of dry friction is shown to be inapplicable
both experiments, there is only a small difference between the by this result.
D, S, and N models (differences occur in the order of 103 ). Next, for the Stokes friction parameter, our reasoning is as
However, comparing the D model with the DN model accord- follows. First, in order to obtain an estimate, we replace veloc-
ing to Eq. (5), we find that the DN model explains the data sig- ity and torque with their maximum values across the door’s
nificantly better than the D model: FL ð1; 19Þ ¼ 27:0, dimensions: v ¼ xr < xw and sf ;Stokes ¼ rFStokes < wFStokes ,
p < 0.001 for the large (L) spin case, and FS ð1; 36Þ ¼ 21:7, respectively. Second, we recall Stokes law, FStokes
p < 0.001 for the small (S) spin case, respectively. Likewise, ¼ 6pgre xw, where re is the so-called aerodynamic radius, a
the SN model explains the data better than the S model in both kind of equivalent radius for an object that is not spherical
cases: FL ð1; 19Þ ¼ 8:1, p ¼ 0.010 and FS ð1; 36Þ ¼ 11:7, (used, e.g., in atmospheric science).16 Taking re < w and
p ¼ 0.002. As there is no difference in fit quality between the combining the preceding estimates, one obtains
nested Newtonian models (N, DN/SN, DSN), we can conclude
that the simplest Newtonian model (N) is also (significantly) b sf ;Stokes 6pgw3
¼ < : (10)
better suited to describe the data as the Dry or Stokes model. I Ix 1
mw2
The reason why the more complicated Newtonian models pro- 3
vide no better data fits than the restricted Newtonian model is
analyzed in the following. Inserting the values from Table IV, one obtains
In order to estimate the influence of each friction coeffi- b=I  105 s1, which does not agree with the fit parameters
cient, we estimate the minimum and maximum angular obtained from the Stokes fit xD ðtÞ in Table III (b=I ¼ 0:25
velocities obtained in the experiment and apply them and b=I ¼ 0:10).
against the parameter estimates presented in Table III. For Finally, we estimate the Newtonian friction parameter
the large initial spin case, we obtain a measurement range following a reasoning analogous to the preceding case.
from x0 ¼ xmax ¼ 2:6 s1 to approximately xmin ¼ 2:0 s1 Again we replace velocity and torque with their maximum
(see Fig. 2). It is easy to see that ðb=IÞxmax=min is several values across the door’s dimensions. We then recall the
orders of magnitude larger than a/I for the constant plus Newtonian drag force equation, FN ¼ ð1=2ÞcD qv2 (with
linear friction (DS) model. Similarly, ðc=IÞx2max=min is A ¼ hw and the air drag coefficient cD), leading first to
always several orders of magnitude larger than a/I and c ¼ sf ;Newton =x2 < ð1=2ÞcD qAw4 , and then to
ðb=IÞxmax=min whenever quadratic friction (N) is included
in the model. In other words, the friction term with the 1 3
highest power dominates the other terms. Similar results c 2 cD qAw 3 cD wqA
< ¼ : (11)
hold true for the small initial spin case. Thus, the model I 1 2 m
mw2
with the best data fit is provided by the simple Newtonian 3
model. If Stokes friction or dry friction (or both) are taken
into account, the fit quality does not improve and the Using values from Table IV, we find c=I  101 , in good
parameters a and b are empirically estimated to be negligi- agreement with the parameter estimates of our Newtonian
ble. As a result of both statistical and analytical arguments, models (N, D þ N; N þ S, and D þ N þ S; see Table III).
we can conclude that the simple Newtonian model is accu- Summing up our results, we conclude that only Newtonian
rate and sufficient for describing the frictional torque acting models of frictional torque are both valid and accurate for
on the slammed door.
Table IV. Quantities and parameters used for order of magnitudes estimation
3. Validity of fit parameters of the slammed door experiment.

We now provide order-of-magnitude estimates to the fric- Quantity/parameter Estimated/measured value


tion parameters of our slammed door if we assume the valid-
ity of the Dry, Stokes, and Newtonian models, respectively. Dynamic viscosity coefficient g in air 1:7  105 Pa s
For the dry friction parameter, we have Air density q 1.3 kg m3
Air drag coefficient cD 1–2 for a rectangular plate
a lsl mgw Sliding friction coefficient lsl 0.01 (steel-steel, lubricated)
¼ ; (9)
I 1 Door width w 1m
mw2
3 Door height h 1m
Door mass m 50 kg
where we have made use of Eq. (1) and written Door moment of inertia I 16.7 kg m2
a ¼ sf ;dry ¼ wFdry ¼ wlsl gm, where lsl is the sliding friction

34 Am. J. Phys., Vol. 85, No. 1, January 2017 Klein et al. 34


describing the rotational motion of a slammed door. That
being said, we note that the slammed door experiment suf-
fers from two shortcomings. First, as stated above, the exper-
iment has a very short measurement range. The difference in
angular position when the door rotates is at best 180 and the
angular velocity data only range from x0 to approximately
0:75x0 (Fig. 2). We will overcome this shortcoming by per-
forming a more traditional experiment that allows measure-
ments over a longer time scale. Second, the parameter
estimates for c/I do not agree between the large- and small-
spin situations, indicating a possible x-dependency of the
friction parameter c. The most likely explanation for this
result is the fact that the air drag coefficient cD is indeed not
a constant but a function of Reynolds number. Our models
are incomplete as they do not account for dynamic changes
in air flow (changing Reynolds number) while the (initial)
angular velocity varies. Hence, we observe different parame-
ter estimates that depend on the initial condition x0. While
theoretical model reconciliation would go beyond the scope
of this paper, we see educational value in pointing out model
limitations.

IV. TRANSITION TO A CLASSICAL LABORATORY


EXPERIMENT
A. Experimental setup and results
We use a thin square plate (mass m ¼ 0.35 kg, width and
height w ¼ h ¼ 0:3 m) with the center of the baseline as the
pivot representing an experimental model for the door (see
Fig. 3, top left). By using this setup, we assume that rotation
around the center of the baseline and around the edge are Fig. 3. Upper left: The laboratory slammed door experimental setup consist-
ing of a thin square plate spinning around the center of the baseline. In addi-
physical equivalent. We justify this assumption with the fact tion to the data acquisition device, a second “dummy” smartphone is used as
that both situations are described by the same relationships a counterweight. Upper right: A cardboard baseplate is mounted beneath the
between the physical quantities of Eq. (3). The rotational plate to model the floor in the real slamming door experiment. Lower:
motion is explored in the same manner as above, by spinning Measurement data from both experimental setups with data fits according to
the plate and measuring the decrease of angular velocity xN ðtÞ in Table II.
with the smartphone acceleration sensor. We also mount a
cardboard baseplate beneath the rotating plate (Fig. 3, top In both experiments, the fit parameters c=I  0:04 and
right), resembling a floor or door frame in order to come 0.02, respectively, are consistent across the Newtonian mod-
closer to a real slamming door and to explore whether the els and share the same order of magnitude (cf. Table V). It is
model lacks validity due to this modification or not. counterintuitive that the angular velocity decreases more
The experimental results, as shown in the bottom of Fig. slowly if the baseplate is mounted. The difference of the c/I
3, show good agreement with the Newtonian model fit. In parameters of about 50% indicates a non-linear relationship
the following, we compare the different friction models with between “air circulated around edges” and the air drag coef-
respect to fit quality and parameter validity, as before. ficient cD. Going into further detail would exceed the scope
of this paper, but the experimental setup could easily be
B. Analysis and discussion modified to investigate this effect empirically.
Finally, we conduct an order-of-magnitude estimate for
The fit parameters are shown in Table V. In the first exper- the c/I parameter similar to the procedure used in Sec. III B.
imental setup (no baseplate), the data indicate that the Taking into account the two smartphones (one used as a
Newtonian models provide the best data fits. Comparing the counterweight), which are no longer negligible, the total
D model with the DN model and comparing the S model moment of inertia of the rotating system becomes
with the SN model reveals a statistically significant improve-
ment of model quality. In contrast to the findings of the real 
1 1
slamming door experiment, the nested Newtonian models I ¼ mw2 þ 2 miPod w2iPod þ miPod r 02
12 12
differ from each other: the DN and SN models provide a sig-
nificantly better fit than the restricted N model (p < 0.001). ¼ 6:6  103 kg m2 ; (12)
However, the unrestricted DSN model leads to no further
improvement as compared to the DN model. The same find- where we have used the physical dimensions of an iPod
ing holds true when the baseplate is added to the experimen- touch (wiPod ¼ 5:5 cm, miPod ¼ 0:127 kg) and the distance
tal setup. Considering the maximum and minimum angular r0 ¼ 0:125 m from the axis of rotation. According to Eq.
velocities obtained in both experiments, we again observe (11), we obtain c=I < 0:07 cD , which is a reasonable upper
that the term of highest order dominates the lower terms. bound for our experimental results in both cases (cD  1).

35 Am. J. Phys., Vol. 85, No. 1, January 2017 Klein et al. 35


Table V. Fit parameters based on different frictional torque models for the laboratory experiments with and without a baseplate (corresponding to Fig. 3, upper
right and left, respectively). The quality factors for the fits are given in columns R2 and SSE.

Frictional torque model a/I (s2) b/I (s1) c/I R2 SSE

Without baseplate, x0 ¼ 12:66 s1


Dry (D) 1.61 – – 0.443 2814
Stokes (S) – 0.25 – 0.924 385.3
Dry and Stokes (DS) <109 0.25 - 0.924 385.3
Newton (N) – – 0.04 0.997 13.94
Dry and Newton (DN) 0.06 – 0.04 0.998 10.39
Stokes and Newton (SN) – 0.02 0.04 0.998 10.85
Dry, Stokes, and Newton (DSN) 0.06 <109 0.04 0.998 10.39
With baseplate, x0 ¼ 13:10 s1
Dry (D) 1.20 – – 0.851 601.2
Stokes (S) – 0.14 – 0.974 103.5
Dry and Stokes (DS) <109 0.13 – 0.974 103.5
Newton (N) – – 0.02 0.992 32.09
Dry and Newton (DN) 0.13 - 0.02 0.994 25.03
Stokes and Newton (SN) – 0.03 0.02 0.994 25.16
Dry, Stokes, and Newton (DSN) 0.13 <109 0.02 0.994 25.03

V. CONCLUSION Special thanks go to Henning Fouckhardt and Hans J€urgen


Korsch for critically proof-reading the manuscript. Furthermore,
In this paper, we characterized the common phenomenon of the authors thank the Wilfried-and-Ingrid-Kuhn Foundation for
a slammed door in terms of physics by discussing the theory of generous financial support.
damped rotational motion and its experimental investigation
with an acceleration sensor. We constructed a theoretical
model to describe the decrease of angular velocity when APPENDIX: DERIVATION OF EQUATION (4)
the rotating door is exposed to different forms of friction. In the following, we first sketch how to obtain Eq. (4)
Using simple measurement techniques (smartphone accelera- from the first-order, non-linear, ordinary differential
tion sensors), it was shown that friction models containing a equation
Newtonian friction term (containing an x2 -term) more accu-
rately describe the motion of the door than either Stokes or dry dx
friction models, as long as the door is not influenced by the I ¼ a þ bx þ cx2 ; (A1)
dt
door frame. Moreover, the simple Newtonian model (neglect-
ing dry and Stokes friction) is nearly as accurate as assuming and then outline the mathematical ideas for obtaining the
the simultaneous presence of all friction terms. We confirmed equations in Table II from Eq. (4); interested readers may fill
our findings in a laboratory experiment and showed that the in the gaps themselves.
thin gap between the floor and the door has no influence on the We begin by separating variables and integrating both
validity of the model, although it does alter air drag. sides of Eq. (A1) to obtain
The experiment described here, performed with smart-
ð ð
phones and freely available software, is particularly suitable dx
as a laboratory activity for university undergraduates as it I ¼ dt: (A2)
a þ bx þ cx2
allows students to quantitatively determine the effects of dif-
ferent types of friction. Furthermore, the slamming door
To evaluate the integral on the left-hand side, we complete
experiment could also be applied to a mechanics course as a
the square
possible “take-home” experiment. It is a conceptual and
experimental approach for analyzing a real-world phenome-  2
b22 b pffiffiffi
non and for questioning the validity of simple model assump- a þ bx þ cx ¼ a  þ p ffiffi
ffi þ cx (A3)
tions. The transition to a lab experiment as part of a 4c 2 c
discipline-specific method allows for a better control of the pffiffiffi pffiffiffi pffiffiffi
physical parameters, which is otherwise difficult to obtain and and substitute u ¼ b=2 c þ c x (to that du ¼ c dx),
surely not accessible in its quantitative aspects. More gener- which leads to
ally, from an educational point of view, the proposed activity ð ð
dx 1 du
allows students to improve their experimental skills, in partic- 2
¼ p ffiffi
ffi : (A4)
ular, those related to data collection and analysis, and may a þ bx þ cx c ða  b =4cÞ þ u2
2

increase their interest in exploring everyday phenomena.


Next, we factor out a  b2 =4c from the denominator and
substitute
ACKNOWLEDGMENTS
u
The authors gratefully acknowledge the contributions to this s ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi (A5)
work made by Patrik Vogt (Freiburg University of Education). a  b2 =4c

36 Am. J. Phys., Vol. 85, No. 1, January 2017 Klein et al. 36


pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
and Now if we assume b ¼ 6 0, then the quantity c ¼ b2 4ac can
become complex. Hence, tanc is p holomorphic
ffiffiffiffiffiffiffi and we need to
du recall the following identities (i ¼ 1):
ds ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ; (A6)
a  b2 =4c
i tanðzÞ ¼ tanhðizÞ; (A15)
which leads to z
e e z

ð tanhðzÞ ¼ ; (A16)
dx ez þ ez
a þ bx þ cx2 and
ð
1 du
¼ pffiffiffi (A7) 1 þ tanhðzÞ
c ða  b =4cÞ½u =ða  b2 =4cÞ þ 1
2 2
¼ e2z : (A17)
ð 1  tanhðzÞ
2 ds
¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 2 : (A8)
4ac  b s þ 12 If we first assume a ¼ 0, then c ¼ ib and we obtain xSN ðtÞ
from Table II. Setting c ¼ 0 in xSN ðtÞ then leads to xS ðtÞ.
The integral can now be evaluated to give tan1 s. Finally, the solution xDS ðtÞ when only c ¼ 0 can be derived
Substituting back for s and u, Eq. (A2) becomes in a similar manner.
 a)
2I b þ 2cx Electronic mail: pklein@physik.uni-kl.de
 pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi tan1 pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ¼ t þ k; (A9) 1
T. Eadkhong, R. Rajsadorn, R. Jannual, and S. Danworaphong,
4ac  b2 4ac  b2 “Rotational dynamics with tracker,” Eur. J. Phys. 33, 615–62 (2012).
2
C. E. Mungan, “Frictional torque on a rotating disc,” Eur. J. Phys. 33,
where k is a constant.
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ffi Solving Eq. (A9) for xðtÞ and setting 1119–1123 (2012).
3
c ¼ 4ac  b2 then gives J. C. Simbach and J. Priest, “Another look at a damped physical pendu-
lum,” Am. J. Phys. 73, 1079–1080 (2005).
 4
L. F. C. Zonetti, A. S. S. Camargo, J. Sartori, D. F. de Sousa, and L. A. O.
b c ct
xðtÞ ¼  þ tan k  : (A10) Nunes, “A demonstration of dry and viscous damping of an oscillating
2c 2c 2I pendulum,” Eur. J. Phys 20, 85–88 (1998).
5
C. E. Mungan and T. C. Lipscombe, “Frictional torque on a rotating disc,”
Finally, we use the initial condition xð0Þ ¼ x0 to deduce Eur. J. Phys. 34, 1243–1253 (2013).
6
that X. Wang, C. Schmitt, and M. Payne, “Oscillations with three damping
effects,” Eur. J. Phys. 23, 155–164 (2002).
 7
K. Hochberg, S. Gr€ ober, J. Kuhn, and A. M€ uller, “The spinning disc:
2cx0 þ b
k ¼ tan1 ; (A11) Studying radial acceleration and its damping process with smartphones’
c acceleration sensor,” Phys. Educ. 49(2), 137–140 (2014).
8
J. Kuhn, “Relevant information about using a mobile phone acceleration
which leads to 9
sensor in physics experiments,” Am. J. Phys. 82, 94 (2014).
P. Vogt and J. Kuhn, “Analyzing radial acceleration with a smartphone

 acceleration sensor,” Phys. Teach. 51, 182–183 (2013).
b c 2cx0 þ b ct
xDSN ðtÞ ¼  þ tan tan1  (A12) 10
Note that we can restrict our considerations to scalars because all vector
2c 2c c 2I quantities such as torque, angular velocity, or frictional forces are assumed
to be either parallel or pairwise perpendicular to each other. Furthermore,
2x0 c þ b  c tanðct=2IÞ b note that sf is a composite of the friction at the hinges (constant torque)
¼   ; (A13) and air drag (linear and quadratic air drag).
2c 1 þ ð2x0 c þ bÞtanðct=2IÞ=c 2c 11
N. R. Draper, H. Smith, and E. Pownell, Applied Regression Analysis
(Wiley, New York, 1966), Vol. 3.
12
where we have made use of identity H. Akaike, “Likelihood of a model and information criteria,”
J. Econometrics 16, 3–14 (1982).
13
tanðxÞ þ tanð yÞ More information on SPARKvue software can be found at <http://
tanðx þ yÞ ¼ : (A14) www.pasco.com/sparkvue>.
1  tanðxÞtanðyÞ 14
While the determination of x0 from the graph seems appropriate in our
case (cf. Sec. III A), this procedure can negatively affect the reliability of
Now let’s consider the pspecial cases listed in Table II. If our conclusions; a lot of variance on the coefficients a, b, and c will be
ffiffiffiffiffi consumed by freedom on x0.
we take b ¼ 0, then c ¼ 2 ac and xDN ðtÞ from Table II fol- 15
The MATLAB Curve Fitting Toolbox (MathWorks, Inc.) <http://de.math-
lows immediately. If we further assume that either a ! 0 or works.com/products/curvefitting/>.
c ! 0, it follows that c  1, so that tanðct=2IÞ  ct=2I, and 16
S. N. Pandis and J. H. Seinfeld, Atmospheric Chemistry and Physics:
we find either xN ðtÞ or xD ðtÞ from Table II, respectively. From Air Pollution to Climate Change, 4th ed. (Wiley, New York, 2006).

37 Am. J. Phys., Vol. 85, No. 1, January 2017 Klein et al. 37

You might also like