Documente Academic
Documente Profesional
Documente Cultură
A
3
A
4
H
31
H
32
H
41
H
42
S
1
S
2
(11.1)
The off diagonal terms H
32
and H
41
are the coupling terms. It is important to underline that the system matrix [H] is
always measured with a cyclist riding on the bike. It should also be noted that a typical road excitation input is used to
measure [H].
Fig. 11.1 Road bike simulator
and the measurement points
( )
ij
H
S
2
(w)
A
3
(w)
A
4
(w)
S
1
(w)
Fig. 11.2 MIMO representation
of the system showing the direct
path and the coupled transfer path
11 A Laboratory Technique to Compare Road Bike Dynamic Comfort 101
The inverse matrix Y D H
1
can be calculated frequency by frequency. All the calculation are done with MATLAB.
The inverse problem allows calculating in the frequency domain the system inputs using the outputs as indicated in (11.2)
S
1
S
2
Y
13
Y
14
Y
23
Y
24
A
3
A
4
(11.2)
The reproduction system (bicycle and cyclist) impulse responses y
ij
.t / can be obtained by calculating the inverse Fourier
transform of the corresponding term Y
ij
. One can then calculate the time signal s
1
.t / and s
2
.t / by summing the contribution
of the time signal output a
3
.t / and a
4
.t / convoluted (*) by the respective impulse responses as shown in (11.3) and (11.4).
s
1
.t / D y
13
.t / a
3
.t / Cy
14
.t / a
4
.t / (11.3)
s
2
.t / D y
23
.t / a
3
.t / Cy
24
.t / a
4
.t / (11.4)
The time functions y
ij
.t / being the impulse responses, each term of (11.3) and (11.4) corresponds to a ltering process
of acceleration signals using a Finite Impulse Response (FIR) lter with coefcients provided by the time functions y
ij
.t /.
11.3 Experimental Validation
The reproduction procedure developed in this work is implemented as follows:
(i) Acceleration signals a
3
.t / and a
4
.t / measured simultaneously on several different bikes being tested. These signals can
be obtained on the road or in laboratory conditions with a cyclist riding the bike. The phase relationship between these
two signals must be maintained.
(ii) The selected reproduction bike is installed on the simulator. With a cyclist sitting on the bike, the systemis characterized
by measuring its system matrix [H].
(iii) Using the previously measured quantity in step (i), the shakers excitation signals s
1
.t / and s
2
.t / are calculated using
(11.3) and (11.4).
Having a set of s
1
.t / and s
2
.t / for several bikes, it becomes possible to implement comparison perception tests such as
split testing.
11.3.1 Validation Using the Reproduction Bike Only
The quality of the reproduction was rst investigated using data measured with the reproduction bike only. Three different
tests were done to obtain this data:
Test (a) H
31
and H
41
measured simultaneously when the front shaker only is activated.
Test (b) H
32
and H
42
measured simultaneously when the front shaker only is activated.
Test (c) s
1
.t /, s
2
.t /, a
1
.t / and a
2
.t / measured simultaneously when both shakers are activated.
The same road signal was used for both shakers and with the same cyclist on the bike.
Figure 11.3 shows the measured system matrix amplitude [H] of the reproduction bike with a cyclist on the bike.
The system matrix measurement was repeated several times in a row to evaluate measurement variability. Between two
consecutive tests, the cyclist was asked to dismount. Figure 11.4 shows the FRFs H
31
and H
42
. The vertical axis is not
expressed in dB for this gure. At each frequency, the ratio between the minimum and maximum amplitude of all the curves
was calculated. The maximum ratio for all the frequency range 0100 Hz is 1.25 (1.9 dB) for H
31
and 1.06 (0.5 dB) for H
42
.
Figure 11.5 shows the matrix [Y] which is simply the inverse of [H] calculated frequency by frequency. All calculations
were done in MATLAB with no ltering or matrix regulation.
To calculate the FIR lter coefcients y
ij
.t /, the inverse Fourier transform of each term Y
ij
was calculated. Prior to this
calculation, the DC value and the rst frequency components were zeroed using a simple cosine taper high pass lter with
a cutoff frequency set at 3 Hz. The frequency band of interest is 3100 Hz. A low pass lter with a cutoff frequency set to
150 Hz was also used to remove the high frequency content. The sampling frequency was set at 8,192 Hz and the FFT record
length was 2 s. The numerical impulse response duration was 2 s yielding to a 16,384 coefcients FIR lter.
102 Y. Champoux et al.
20 40 60 80 100
0
20
40
a
|
H
3
1
|
d
B
r
e
1
(
m
/
s
2
)
/
V
20 40 60 80 100
0
20
40
b
|
H
3
2
|
d
B
r
e
1
(
m
/
s
2
)
/
V
20 40 60 80 100
0
20
40
c
Frequency (Hz)
|
H
4
1
|
d
B
r
e
1
(
m
/
s
2
)
/
V
20 40 60 80 100
0
20
40
d
Frequency (Hz)
|
H
4
2
|
d
B
r
e
1
(
m
/
s
2
)
/
V
Fig. 11.3 System matrix amplitude [H] of the reproduction bike. (a) H
31
, (b) H
32
, (c) H
41
, (d) H
42
. A cyclist is riding the bike
0 10 20 30 40 50 60 70 80 90 100
0
50
100
150
a
b
H
3
1
(
m
/
s
2
)
/
V
0 10 20 30 40 50 60 70 80 90 100
0
50
100
150
Frequecy (Hz)
H
4
2
(
m
/
s
2
)
/
V
Fig. 11.4 Measurement variability for H
31
and H
42
(a) H
31
ve tests; (b) H
42
three tests
The lter coefcients were rearranged to obtain an impulse response with evanescent ends to minimize the truncation
errors. Figure 11.6 shows the calculated four impulse response functions.
This rst validation test used measurements obtained only from the reproduction bike. The input and output signals were
measured simultaneously as indicated previously in the description of Test (c). Using the output signals a
3
.t / and a
4
.t / in
(11.3) and (11.4), it was possible to solve the inverse problem and calculate the inputs s
1
.t / and s
2
.t /.
The rst validation test consists of comparing the calculated and the measured inputs. Figure 11.7 allows us to compare the
measured and the calculated inputs signal s
1
.t / for a typical short time segment. The PSD of the measured and reproduced
shaker input signals G
s1s1
and G
s2s2
are compared in Fig. 11.8. The mean discrepancy between the calculated and the
measured PSD is 0.2 dB and the maximum discrepancy is 2.3 dB. The ratio between the measured and the calculated total
power for s
1
.t / is 0.98 and 0.94 for s
2
.t /.
11 A Laboratory Technique to Compare Road Bike Dynamic Comfort 103
20 40 60 80 100
-60
-40
-20
0
a
Y
1
3
d
B
r
e
1
V
/
(
m
/
s
2
)
20 40 60 80 100
-60
-40
-20
0
b
Y
1
4
d
B
r
e
1
V
/
(
m
/
s
2
)
20 40 60 80 100
-60
-40
-20
0
c
Y
2
3
d
B
r
e
1
V
/
(
m
/
s
2
)
Frequency (Hz)
20 40 60 80 100
-60
-40
-20
0
d
Y
2
4
d
B
r
e
1
V
/
(
m
/
s
2
)
Frequency (Hz)
Fig. 11.5 Inverse cyclist-bike
system matrix [Y] of the
reproduction bike. (a) Y
13
, (b)
Y
14
, (c) Y
23
, (d) Y
24
-1 -0.5 0 0.5 1
0
10
20
a
y
1
3
(
t
)
(
V
/
(
m
/
s
2
)
/
s
)
-1 -0.5 0 0.5 1
-4
-2
0
2
b
y
1
4
(
t
)
(
V
/
(
m
/
s
2
)
/
s
)
-1 -0.5 0 0.5 1
-4
-2
0
2
c
y
2
3
(
t
)
(
V
/
(
m
/
s
2
)
/
s
)
Time (s)
-1 -0.5 0 0.5 1
0
10
20
d
y
2
4
(
t
)
(
V
/
(
m
/
s
2
)
/
s
)
Time (s)
Fig. 11.6 Impulse response
(lter coefcients) for the four
terms of [Y]. (a) y
13
, (b) y
14
, (c)
y
23
, (d) y
24
11.3.2 Reproduction Quality Assessment
To explore howthe proposed approach of using a reproduction bike succeed to reproduce the vibrational behavior of different
bikes, a road bike named the target bike of a different model and brand than the reproduction bike was selected and tested.
The target bike was installed on the simulator and the same shaker road input signals were used. The seatpost and saddle
acceleration levels a
3t
.t / and a
4t
.t / were measured. These signals and their respective frequency spectrum A
3t
and A
4t
are the target signals. Using the procedure presented previously, the signals s
1t
.t / and s
2t
.t / were calculated. These signals
were then used to drive the shakers, and then the reproduced output acceleration signals a
3r.t /
and a
4r.t /
were measured.
Comparing the target and the reproduced signals is the ultimate way to check the validity of the approach. An accurate
reproduction would show a perfect match between reproduced and target signals.
Figure 11.9 shows the comparison of the reproduced and the target acceleration PSD at the stem and at the saddle. A third
curve (thick line) shows the measured PSD when the original road signals were used. The shape change of the curves from
the thick line to the solid line demonstrates how the reproduction bike outputs are modied by changing the input signals.
104 Y. Champoux et al.
20.6 20.7 20.8 20.9 21 21.1 21.2 21.3 21.4
-2
-1.5
-1
-0.5
0
0.5
1
1.5
2
s
1
(
t
)
(
V
)
Time (s)
Fig. 11.7 Comparison of the
measured shaker input signal
s
1
.t / with the calculated signal.
Solid line: measured signal;
dotted line: calculated signal
exciter
10 20 30 40 50 60 70 80 90 100
-60
-40
-20
0
a
G
s
1
s
1
d
B
r
e
1
(
m
/
s
2
)
2
/
H
z
G
s
2
s
2
d
B
r
e
1
(
m
/
s
2
)
2
/
H
z
10 20 30 40 50 60 70 80 90 100
-60
-40
-20
0
b
Frequency
Fig. 11.8 Comparison of the
measured and calculated PSD for
the shaker input signals (a) Front
exciter G
s1s1
(b) rear shaker G
s2s2
11.4 Discussion
The problem investigated in this study is somewhat similar to what is known as crosstalk cancellation in the audio research
community. Crosstalk cancellation is often applied to binaural sound reproduction at a single listeners ears using two or
more loudspeakers [912]. For audio applications, crosstalk cancellation is somewhat complicated by: (i) the propagation
time between sources and receivers and (ii) the full audio bandwidth. In the case studied in this paper, the mostly non-
resonant and highly damped behavior of the reproduction bike is sufcient simplies the derivation of the inverse, crosstalk
cancellation, lter and simple inverse processing.
The identication of natural frequency using Fig. 11.3 is difcult because there is no well-dened peak in the FRFs.
Modal analysis showed that the peak and valley in H
31
around 30 Hz is associated with the front and back cantilever beam
mode of the bicycle fork and front wheel.
Several parameters can inuence cyclist-bike system behavior [3]. In this paper, a cyclist sits on the bike during all tests
for the following reasons: (i) the human body part vibrational behavior in contact with a vibrating structure is recognized
to have some non-linearity; (ii) the normal stiffness of road bike tire changes drastically with prealod. Testing at running
11 A Laboratory Technique to Compare Road Bike Dynamic Comfort 105
10 20 30 40 50 60 70 80 90 100
-10
-5
0
5
10
15
a
G
a
3
a
3
d
B
r
e
1
(
m
/
s
2
)
/
H
z
10 20 30 40 50 60 70 80 90 100
-10
-5
0
5
10
15
b
G
a
4
a
4
d
B
r
e
1
(
m
/
s
2
)
/
H
z
Fig. 11.9 Comparison of the
target and measured acceleration
PSD at the stem and the saddle.
(a) Saddle acceleration PSD
G
a3a3
, (b) Seatpost acceleration
PSD G
a4a4
Solid line: replicated
PSD, dotted line: target PSD,
Thick line: reproduction bike
PSD output when shakers are
driven by the original road signal
preload levels imposed by the cyclist allows a reduction of the nonlinear effect. One of the main elements that must be taken
into account when testing bicycle vibration is the cyclists posture [3] which is related to the way the cyclist leans on the
bike, leans onto the handlebar, sits on the saddle and contracts his muscles. These elements have an important effect on the
vibrational behavior of the cyclistbicycle system. For studying bike design modications in relation to its dynamic behavior
and comfort, the inherent variability of the cyclist must be taken into account, and testing techniques must be developed to
minimize its impact on test variability. Figure 11.4 shows measurement variability of the matrix [H].
The inverse matrix system [Y] is directly calculated in MATLAB from [H]. No regulation of the matrix is required. To
obtain the impulse response, it is necessary to get rid of the DC component and the few rst frequency lines for all Y
ij
.
Because a piezoelectric accelerometer is used, the rst few lines of the FRFs for [H] are not measured correctly and the
amplitude is close to zero. Consequently, the corresponding lines of Y
ij
have very high amplitude and must be removed.
Figure 11.6 shows the impulse responses. No time windowing was necessary and the use of 16,384 coefcients seems to be
adequate so that both ends of each impulse response asymptotically converge toward zero amplitude.
Figures 11.7 and 11.8 provide a rst indication of the quality of the reproduction. Because in this case the measurements
were done on the same bike, one would expect that the measured and the calculated shaker excitation signals would be
identical. Indeed, Figs. 11.7 and 11.8 show that the discrepancies between the target and the measured values are relatively
small. The system matrix and the reproduction measurement require three different tests. The intervariability of the cyclists
position and posture could explain most of the discrepancies.
The last validation test results are shown in Fig. 11.9. The target signals (stem and saddle PSDs) are compared with the
reproduced PSDs. The response of the reproduction bike (thick line) was denitively reshaped by the use of a different shaker
input signals for reproducing the target bike outputs. A frequency span of 5100 Hz is used to plot the graphs. The capacity of
hand and buttock to perceive vibration varies with frequency as described in the ISO standards [13, 14]. Taking the hand and
buttock sensitivity into account and the fact that most of the vibrational power shown in G
a3a3
and G
a4a4
is below 55 Hz, it
would be justied to judge the reproduction quality between 5 and 50 Hz. The reproduction quality at the saddle is better than
the one at the stem. This is expected because typically two-third of the cyclists mass is on the saddle and the intervariability
of a subject is lower at the saddle. In Fig.11.9a the replicated amplitude is overestimated around 23 Hz when compared to
the target signal. The results are taken from a single measurement and in a certain way this could be considered as a worst
case scenario. In any case, the frequency spectrum shapes of both the target and the reproduced signals are very similar.
11.5 Conclusion
For perception studies of comfort in a laboratory, a commercial bike was used to develop a reproduction system. The use
of vibrational output signals to calculate the required inputs for driving the shakers is essentially an inverse problem. It
was shown in this paper that inverting the system is quite straightforward and no specic matrix or system regulation is
required. The capacity of the reproduction system to reproduce vibration accelerations at the stem and saddle was examined
106 Y. Champoux et al.
and it was shown that the frequency spectrum shapes are well reproduced. Nonetheless, some amplitude discrepancies
were observed, mainly at the stem. This approach requires that a cyclist sit on the bike during all tests as well as when
characterizing the system. The reproduction system includes both the cyclist and the bike. Practical testing considerations
related to nonlinearities were raised to justify including a cyclist. The current challenge to studying comfort is related to the
fact that the cyclist has a strong inuence on the vibrational response of the bike cyclist system and a change in position or
posture can introduce important variability. The results presented in this paper are based on a single measurement. Repeating
all measurements over several days will prove to be worthwhile. It will also be interesting to consider the effect of different
cyclists. Finally, we also need to ask if, in order to maximize the quality of the reproduction, it would be necessary to repeat
all measurements, including the system matrix characterization, for each cyclist.
Acknowledgements The authors gratefully acknowledge nancial support from the National Science and Engineering Council of Canada
(NSERC) and the participation of Cerv elo and Vroomen-White Design.
References
1. Richard S, Champoux Y (2004) Modal analysis of a road bikes front components. In: Proceedings of IMAC XXII, Dearborn, Feb 2004
2. Champoux Y, Richard SE, Drouet J (2007) Bicycle structural dynamics. Sound Vib 41(7):1624
3. L epine J, Champoux Y, Drouet JM (2013) Inuence of test conditions in comfort ranking of road bicycle wheels. Paper accepted for the IMAC
XXXI, Garden Grove, Feb 2013
4. Gibson J (1979) The ecological approach to visual perception, 1986 ed. Houghton Mifin, Boston
5. Gaver W (1993) What in the world do we hear? An ecological approach to auditory event perception. Ecol Psychol 5(1):129
6. Guastavino C, Katz B, Polack JD, Levitin D, Dubois D (2005) Ecological validity of soundscape reproduction. Acust United Acta Acust
91(2):333341
7. Guastavino C (2009) Validit e ecologique des dispositifs exp erimentaux. In: Dubois D (ed) Le Sentir et le Dire. Concepts et m ethodes en
psychologie et linguistique cognitives. LHarmattan (Coll. Sciences Cognitives), Paris, pp 229248
8. Brassard F (2010) D eveloppement dun simulateur de vibration pour v elo de route. Master degree thesis, Universit e de Sherbrooke, Sherbrooke
9. Nelson PA, Rose JFW (2005) Errors in two-point sound reproduction. J Acoust Soc Am 118:193204
10. Kirkeby O, Nelson PA (1999) Digital lter design for inversion problems in sound reproduction. J Audio Eng Soc 47(7/8):583595
11. Norcross SG, Soulodre GA, Lavoie MC (2004) Subjective investigations of inverse ltering. J Audio Eng Soc 52(10):10031028
12. Lentz T (2006) Dynamic crosstalk cancellation for binaural synthesis in virtual reality environments. J Audio Eng Soc 54(4):283294
13. ISO 26311 (1997) Mechanical vibration and shock evaluation of human exposure to whole-body vibrationPart 1: general requirements
14. ISO 53491 (2001) Mechanical vibration measurement and evaluation of human exposure to hand-transmitted vibrationPart 1: general
requirements