Académique Documents
Professionnel Documents
Culture Documents
seate
A Dynamic System Model of an were
betwi
thora
E. L. Wang Off-Road Cyclist tions
onai
Assistant Professor,
were
University of Nevada, Reno, To optimize the performance of off-road bicycle suspension systems, a dynamic model seat,
Department of Mechanical Engineering, of the bicycle/rider system would be usefuL. This paper takes a major step toward
Reno, NV 89557
off-n
this goal by developing a dynamic system model of the cyclist. To develop the cyclist strai!
model, a series of four vibrational tests utilizing random inputs was conducted on minii
seven experienced off-road cyclists. This allowed the transfer functions for the arms hand
M. L. Hull and legs to be determined. To reproduce the essential features (i.e., resonance peaks) the h
Professor,
of the experimental transfer functions, the system model included elements represent- In
University of California, Davis,
Department of Mechanical Engineering,
ing the visceral mass along with the arms and legs. Through simulations, the frequency bar i
Davis, CA 95616 responses of the system model of the rider in each of the four ~sts were comp~ted. This
Optimal stifess and damping parameter values for each subject were determined betw
by minimizing the diference between the experimental and simulation results. Good play,
agreement between experimental and simulation results indicates that modeling the tion.
rider as a lumped parameter system with linear springs and dampers is possible. n
theF
For
rigid
Furthermore, Minetti and Belli (19) found that neglecting the used
Introduction lowe
motion of the visceral mass can cause serious errors in locomo-
The recent growth in popularity of off-road cycling, com- tion studies.
Fc
bined with the desire to increase comfort and control, has led Thus, it appears from the reviewed literature that it is possible their
to an increasing use of suspension systems in off-road bicycles. to model pars of the body such as the ars, legs, and visceral the F
To optimize. the performance of suspension systems when tra- mass using traditional lumped parameter models with linear distri
versing rough terrain, a two-dimensional dynamic system model elements. The objective of the work reported in this aricle was the s
of a bicycle and rider would be usefuL. To achieve this, however, to develop a dynamc system model of an off-road cyclist and the s
an accurate dynamic description of the rider is necessar. identify the parameter values that resulted in the best match D~
Based on previous studies of the ars (1-3), legs (4-6), between simulated and experimentally measured frequency re- vibn
and whole body (7 -12), it appears that modeling the rider as sponse functions for seven subjects. menl
a lumped parameter system with linear springs and dampers is
possible. While nonlinearities have been found to exist (13- Methods
15), the nonlinearties may be neglected if the model is re-
strcted to a single body configuration, loading condition, and The model development involved three distinct stages: vibra-
small range of motion. Thus, most of the previous studies are tory testing to determne transfer funçtions for the ars and
not directly relevant to this study because the flexion angles legs, development of computer-based dynamic system models,
used and joint moments are not typical of off-road cycling. and an optimization on stiffness and damping parameter values.
Since the ars and legs provide significant vibration isolation Vibratory Tests.. To determne the frequency response
for an off-road cyclist, they should be included as model compo- functions (FRs) of the ars and legs experimentally, a series
nents. Wong and Hull (2) determined that for an on-road cy- offour vibratory tests was conducted on seven experienced off-
cling position, the ars could be modeled as a spring and road cyclists. Their masses ranged from 72.7 to. 79.5 kg (avg
damper in parallel from the shoulders to the hands. Similarly, = 76.3 kg, std dev = 2.2 kg) and their heights ranged from
Greene and McMahon ( 4 ) found that the legs could be modeled 1.5 to 1.83 m (avg = 1.80 m, std dev = 0.03 m). The first
as a spring and damper in parallel from the hip to the foot. Of two tests were used to determne the FRFs of the arms in flexion
the above-mentioned previous studies dealing with the arms and and the second two were used to determne the FRs of the
legs, these are the only two that had even remotely similar body legs in flexion. Each subject assumed his normal riding position
configurations and load lèvels to those seen in off-road cycling. (i.e., flexion angles were not specified).
Wilczynski and Hull (16) used the studies of Wong and Hull Depending on whether the ars or legs were being tested,
(2) and Greene and McMahon (4) to create a dynamic system either a set of handlebars or a set of bicycle crank ars and
model of the rider. Their simulations showed that the range of pedals was mounted on the ram of a servohydraulic actua~or.
motion of the.imbs during off-road cycling is sufficiently small Although the inputs to the bicycle/rider system can be as high
to justify the use of linear elements. as 20 Hz (16), preliminar tests showed no significant rider
Along with the ars and legs, several researchers have found dynamics above 10 Hz. Thus, the ram was displaced with a
the motion of the visceral mass to be important. Zagorski et al. random signal (white noise) bandlimited to 12 Hz using a low-
(17) and Coermann (7) found that the internal organs had a pass filter ( 120 db/ decade rolloff). The peak-to-pe~. amplitude
natural frequency of 2-5 Hz and 3 Hz, respectively, for seated was set to approximately 12 mmto produce arealistic off-road
subjects. Garg and Ross (8) found the internal organs to have riding level of comfort (as judged by the subjects).
a natural frequency of about 2 Hz for a standing subject. Muk-
-
Three riding positions were tested: seated, standing, and
sian and Nash (18) used a nonlinear model of a seated human downhilL. In the seated position approximately 30 percent of
and found that the thoracic organs resonated at about 3 Hz. the rider's mass was supported by the hands (the rest was
distributed over the seat and pedals). In the standing position
only the hands (30 percent) and feet (70 percent) supported Fig. 1
Contributed by the Bioengineering Division for publication in the JOURAL the rider's mass. Finally, in the downhill position, the rider's posit
OF BIOMECHANICAL ENGINEERIG. Manuscript received by the Bioengineering twee
Division March 27,1995; revised manuscript received September 16, 1996. Asso- feet supported most (approximately 90 percent) of the rider's
barsl
ciate Technical Editor: A. G. Erdman. weight with the hands supporting the rest.
//Q:
The
.. ..
Sel!ent
upper ar
Prooert
lengt (m)
ma (kg)
Ave (std devl
0.32
4,19
(0,01)
(0.08)
lower leg
Prooert
lengt (m)
ma (kg)
inerta (kg-m2)
lengt (m)
Ave (st devl
0.32 (0.1)
2.S5 (0.07)
0.016 (0.00)
0.002 (0,00)
0.015 0.001
0,44 (0.01)
the pai
of the
times :
The
compa
showe
upper leg lengt (m)
8.0:5 (0.28) ma (kg) 3.13 (0,09) displa:
ma (kg)
inertallo-m2 0.144 (0.005) inerta r1ø~m2, 0.045 (0.01) for ev(
tors lengt (m) 0.53 (0.02) hand mass ekel 0,96 (0.03) riding
ma (kg) 32.79 (0.92) foot mass ekgl 1.9 (0.04) The
inerta eke.m'i 2.221 (0.141) viscera mas (kel 11.5 (0.33) tent be
the avi
Notes: Mass and inerta for an segments is for both left and right segments combined.
Fig. 2 Dynamic system model representing the tests involving the arms. Inerta for ar segments is about segment principa axes. values
Inerta for al.other segments is about axs perpndicuar to saggta plane.
The input, x(t), represents the motion of the handlebars, and the output, Ma and inerta for leg segments is for right only (left is ased to be the sa):
betwei
y(t), represents the motion of the reflective. marker. Note that for clarity Torso ma and inerta include the bead. tremel
the arms are shown in a position different than that in which they were the lai
modeled. Solid circles represent revolute joints.
The
be see
order Runge- Kutta routine. The input displacements used in genen
Because of the random input, random data analysis tech- the simulations were the actual displacement data taken during the pi
niques were applied (22). An estimate of the amplitude ratio, the vibration tests. subjee
AR ,at frequency f was defined as The same data reduction routine that was used to calculate simila
the amplitude ratios for the experimental data was used on the from i
AR(f) = I ~Xy(f) I (1 ) output data from the simulations. The difference between the The
Sxx(f) experimental and simulation results formed the basis for the puter
cost function: discre
where Sxy(f) is the cross-spectral density function of the input N
and output, and Sxx(f) is the autospectral density function of CF = i. (ARexp(¡;) - ARmodel(f¡ ))2 (2)
the input. SXy(f) and Sxx(f) were calculated by averaging 45 i=l
subrecords of 2 seconds each (leading to 0.5 Hz resolution). where ARexp and ARmodel are the estimates of the experimental.
For all of the results presented, a normalized random error of and model amplitude ratios respectively,f¡ is the ith frequency,
less than 10 percent was maintained. and N is the total number of discrete frequencies for which the
To determne the amplitude ratio for the models, the equa- normalized random error was less than 10 percent. However,
tions of motion for each model were developed using Kane's because the estimate given by Eq. (I) does not estimate the
method (23). The equations were then integrated using a fourth- "valleys" of the FRF well (22), for the cases where a definite
"valley" existed, the lowest frequency between the two major
resonance peaks and its two neighboring frequencies were ex-
cluded from the cost function. Ths exclusion prevented "val-
(1
leys" from erroneously enlarging the cost function. The cost
function given in Eg. (2) was then minimized using Powell's
Results
The optimal values for the stiffness and damping parameters .
for the ars, legs, and visceral mass are presented in Table 2.
The stiffness and damping values for the hip (downhill case
only) are not shown because the parameter optimization did
not cause changes from initial values (300 N-m/rad and 50 N-
. x(t) 1
Table 2 Stiffness and damping parameter values
an
stdin
Bl daing
sub"ee 1m -fm
i 14.68 659
Fig. 3 Dynamic system model representing the tests involving the legs. 2 14.20 104
The input, x(t), represents the motion of the crank spindle, and the out- 3 )5.20 843
8.30 775
4
put, y(t), represents the motion of the reflective marker. For clarity, the , 8.93 728
torsional hip spring and damper used in the downhil position are not 6 12.98 532
7 14.70 916
shown. Solid circles represent revolute joints.
JOUI
250 I Vol. 119, AUGUST 1997 Transactions of the ASME
r
i;
netric para me- m-slrad). Furthermore, manual adjustment of these values by The discrepancy was apparent in all the subjects that displayed
a factor of ten had no effect on the results. a prominent second peak.
The varation between subjects was quite large for most of Although the amplitude ratio plots are not shown for all the
Avo (std dev) the parameter values (Table 2). For example, for the stiffness subjects, the frequency at which the resonance peaks occurred
0.32 (0.01) of the legs in the standing position, the largest value was 2.5 was fairly consistent. For the legs in the downhill position, the
2.S5 (0.07) times larger than the smallest value. first peak ranged froID 1 to 2.5 Hz. For the legs in the standing
;;
0.016 (0.000) The damping for the ars increased for every subject when position, the peak always occurred between 2 and 4 Hz. For
~
0.002 (0.00)
comparng the seated and the standing cases. The stiffness the arms in the seated position, the peaks occurred between 2
O.OIS lO.oon
showed a similar trend for six of the seven subjects. The legs and 3.5 Hz. For the arms in the standing position, the peak was
0." (0.01)
3.13 (0.09) displayed a consistent trend of decreasing stiffness and damping between 2.5 and 3.5 Hz. Finally, for the visceral mass, the peak
0.045 (0.001) for every subject when comparing the standing and the downhil vared from 6 to 8.5 Hz.
I:'
0." (0.03) L;
riding positions.
U. (0.04) fi The stiffness values for the visceral mass were fairly consis- Discussion
.l~
II.4S (0.33) "
,,' tent both within a subject (four cases) and among subjects, with In creating the dynamic models, several assumptions were
i':
its combined. i ~, the average being about 19 kN /m. Within a subject the stiffness made. The most basic assumption was that the system could be
I.,;: values vared a maximum of 13 percent (relative to maximum) modeled using linear elements. While the elements were linear,
lplane.
10 be the sae). I' between test cases. The damping values, however, were ex- nonlinearities in the model could exist due to geometrical fac-
ß tremely inconsistent (both within and between subjects) with tors.
Li
,I the largest value being nearly 20 times that of the smallest. Preliminary tests using a swept sine wave to determine the
tl The large varation in the damping of the visceral mass can amplitude ratios indicated another source of nonlinearities. For
be seen iu the plots of the amplitude ratios (Figs. 4 and 5). In a single-component sine wave, the rider was able to adapt his
ients used in 11 general, subjects could be divided into two categories based on response by varing the muscle forces without varying flexion
taken during .l the prominence of the second resonance peak (five of seven angles (through co-contraction). To minimize the possibility of
"1 subjects exhibited prominent second peaks). Because of the adaptation, random inputs were used. Qualitatively, each subject
i to calculate stated that the inputs felt very similar to actual off-road cycling.
similarty of the results, those for only two of the subjects (one
s used on the ~ from each category) are shown. This nonlinear behavior of muscles also prevents a meaningful
: between the The agreement between the experimental data and the com- comparison of the results presented in this study to the results
basis for the of the previous studies by Wong and Hull (2) and Greene and
~ puter simulations was quite good (Figs. 4 and 5). The largest
discrepancy was in the prediction of the second peak (Fig. 4). McMahon (4).
(2)
I
experimental Arms (seated position) Arms (standing position)
th frequency, 1.2 0.9
for which the ~
:nt. However, 0.8
1.0
. estimate the
iere a definite
0
~= 0.8
~0= 0.7
Fig.4 Experimental (points) and simulation (solid line) amplitude ratios for all four test cases for Subject
7. These plots are representative of those subjects for which a prominent second peak was evident. Only
the amplitude ratios for which the normalized random error was less than 10 percent are reported.