Professional Documents
Culture Documents
D = 0.
2) N = 0, D = 0 :
D takes the value
D(t) = D(t t)
so that N/
D = inf.
3) Other than above:
D = D remain unchanged.
For the example of the instantaneous drive efciency (case
6) it is N = P
mech
, D = P
in
, and
D =
P
in
.
III. DRIVE SCENARIO SIMULATION TECHNIQUE
A. Test vehicle description
For the experimental investigation, an electric bicycle with
a brushed dc-motor installed in the front hub, a controller, a
thumb throttle, and a battery pack is used (Fig. 7). This bicycle
is a commercially available bicycle that has been available in
the laboratory. All experiments were carried out using this test
vehicle. The electric hub-motor in the front wheel is not used
during the measurements, yet, using this bicycle, the actual set
up of an electric bicycle is represented.
Fig. 7. Electric bicycle test set-up used for the experimental investigation.
The torque and speed are directly measured in the hub of
the rear wheel of the test bicycle, using a Power Tap
R
hub
(Fig. 8) [9]. The measurement information is transmitted to
the CPU through the receiver. For all measurements, the tire
pressure was kept at 50,...,60 psi, which is typical for bicycles
that are used for leisure and commuting and that are commonly
not re-inated before each ride. The anemometer that can be
seen in Fig. 7 is not used for the measurements discussed in
this paper.
B. Data recording
The riding proles were recorded (measured) in terms of
power, P
PT
, torque, T
PT
, and ground speed,
PT
. The sampling
interval is set at its minimum time, t
s
= 1.26s. Table I shows
a sample data set.
1576
Fig. 8. Power Tap
R
hub [9] used for the experimental investigation.
TABLE I
SAMPLE DATA SET FROM POWER TAP
R
Time Torque Speed Power Distance
Minutes Nm km/h W km
0 0 0 0 0
0.021 6.1 4.1 21 0.006
0.042 4.5 6.5 24 0.008
0.063 3.7 6.5 20 0.01
C. Data processing and model extension
The measurements of the riding interval proles as de-
scribed above (Section III-B) are inputs to the model using the
Simulink
R
signal builder block. In addition to the torque T
PT
and the speed
PT
, the speed difference
PT
and acceleration
(d/dt)
PT
=
PT
/t
s
between two sampling intervals are
used for this purpose.
1) Command torque and load torque: The model contains
the load and command torques T
load
and T
cmd
as inputs (Figs. 3
and 4). However, only the torque produced by the rider is
available in the form of T
PT
, along with the speed
PT
. As
this torque shall be produced by the electric motor, it is used
as input for the torque command, T
cmd
= T
PT
= T
cmd,PT
.
The load torque T
load
needs to be derived from the measure-
ment data at each sampling interval. Two different approaches
are taken:
1) Delay of T
load
when compared with T
command
:
The load torque at time t, T
load
(t), is approximated
by the command torque of the previous time interval,
T
load
(t +t
s
) = T
cmd
(t), or, T
load
(t) = T
cmd
(t t
s
).
2) Approximation of T
load
using the mechanical equation:
The load torque T
load
is derived using the mechanical
equation (4) that becomes
T
load,PT
= T
cmd,PT
B
PT
+J
PT
t
s
. (5)
Both approaches have been implemented. For this purpose,
the motor and controller models (Figs. 3 and 4) have been
modied to use the measured data T
cmd,PT
and, in the case
of approach no. 2,
PT
and (d/dt)
PT
as inputs. From these,
the new parameters T
load,PT
, D
PT
, V
cmd,PT
, and I
cmd,PT
are
derived. Here, the
sim
, and recorded (measured) speed,
PT
, are not in agreement.
IV. RESULTS
A. Riding interval proles
The test-bicycle described above (Section III-A) was used
to obtain riding interval proles as follows: The measure-
ments were designed to obtain road data of real-life appli-
cations. To this aim, riding interval proles, proles no. 1
through 4, of four different riders and with intervals of 15 to 25
minutes were recorded, where the bicycle was used for a short
leisurely ride, grocery shopping, or commuting (Table II).
TABLE II
CHARACTERISTICS OF RECORDED RIDING PROFILES
Prole Rider Riding P
mech,max
P
mech,ave
T
max
T
ave
no. weight interval
kg min W W Nm Nm
1 50 18 204.0 35.6 27.9 4.7
2 75 16 389.1 133.9 40.8 8.2
3 85 22 368.6 66.3 26.4 5.9
4 95 25 857.0 179.0 50.2 9.9
Exemplarily, the measured torque versus time and power
versus time characteristics of riding proles 1 and 4 are
shown in Figs. 1215. They illustrate the spread of the prole
characteristics, and thus requirements on the drive, as riding
proles 1 and 4 are those with the lowest and highest torque
and power demands respectively (It should be noted that the
maximum speed of the riding prole 4 exceeds the speed limit
for low-speed electric bicycles according to U.S. law, which
is 20mph.). The characteristics of riding proles 2 and 3 are
in-between these two extremes.
B. Simulation results
1) Parameters: The four rides were analyzed using the
developed model and example-case values of a dc-drive sys-
tem. The base example-case values were adapted from the
commercially available electric bicycle that was also used as
test vehicle for the riding prole measurements (Section III-A,
Fig. 7). The bicycle has a 24V brushed dc-hub motor and a
battery system of two 12V, 12Ahr lead acid batteries.
In this work, we seek to identify the limits of the drives
and the simulation techniques and not to optimize one single
drive conguration for one given riding prole. Therefore, we
select by intention a comparatively large value of the arma-
ture resistance to account for low-cost motors and additional
resistance of the connections. The parameters of the motor
0
10
20
30
40
50
0 5 10 15 20 25
Time [min]
Torque [Nm]
Fig. 12. Measured torque versus time of riding prole no. 1 (same scales
as Fig. 14 by intention), T
max
= 27.9Nm, T
ave
= 4.7Nm.
0
100
200
300
400
500
600
700
800
900
0 5 10 15 20 25
Time [min]
Power [W]
Fig. 13. Measured power versus time of riding prole no. 1 (same scales as
Fig. 15 by intention), P
max
= 204.0W, P
ave
= 35.6W.
0
10
20
30
40
50
0 5 10 15 20 25
Time [min]
Torque [Nm]
Fig. 14. Measured torque versus time of riding prole no. 4, T
max
=
50.2Nm, T
ave
= 9.9Nm.
are set to R
a
= 1, L
a
= 1mH, k
v
= k
t
= 1Nm/A, and
R
batt
= 2 12m. The parameters of the PI-controller were
designed to achieve 0.06s rise time, 5% overshoot,0.2s settling
time, and 30rad/sec bandwidth (K
p
= 100, K
i
= 0.3). We set
J = 10kg/m
2
according to [10].
1578
0
100
200
300
400
500
600
700
800
900
0 5 10 15 20 25
Time [min]
Power [W]
Fig. 15. Measured power versus time of riding prole no. 4, P
max
= 857.0W,
P
ave
= 179.0W.
Using these base example-case values of the different
parameters as starting point, we vary selected parameters
throughout the analysis. When doing so, the abilities of the
modeling technique are analyzed using the performance of
the low-cost example-case drive for illustration. For reasons
of comparison, only the rst 840s (14min) of the rides are
simulated. The results illustrate well how the performances of
a given drive system depends on the characteristic of a given
riding prole.
2) Speed: A comparison of the simulated and the recorded
(measured) speed shows the limitations given by the simplied
single-wheel model of the mechanical system: Exemplarily,
Fig. 16 shows an extract of riding prole no. 1. The simplied
model does not reproduce the speed (dynamic behavior) of
the drive correctly, notably for fast speed slopes, but the drive
itself is able to produce the required torque as can be seen
from Fig. 17.
It is important to recall thatas a result of the way the
recorded data are used as inputs to the modelthe command
signals the motor receives are completely decoupled from
inaccuracies of the speed simulation (Section III-C.2). Corre-
sponding simulations were also carried out for the other riding
proles, with different values of the lumped inertia as small
as J = 0.1kgm/s
2
, and of the controller, leading to similar
results. In order to correct these inaccuracies, more complex
models of the mechanical system, such as those suggested in
[10] or [11], would need to be implemented and adapted to
work with the measurement data as inputs.
As a result of the limits of the speed simulation, the output
efciency needs to be calculated via the modied performance
submodule, using the recorded (measured) speed
PT
and the
electric torque T
e
(Fig. 11).
3) Torque: Operating points can occur where the drive
cannot produce the command torque due limitations imposed
by the current or the back-EMF. Whenever the command
torque exceeds the torque limit of the drive at its current
speed, deep speed dips can be seen in the simulated speed
sim
when compared with the recorded (measured) one,
PT
.
For illustration, the simulation of riding prole no. 1 is
50 100 150 200
0
5
10
15
Time [s]
S
p
e
e
d
[
r
a
d
/
s
]
w
PT
w
sim
Fig. 16. Recorded and simulated speed w
PT
and w
sim
of riding prole no. 1
(40s-230s): The simplied model does not reproduce the speed (dynamic
behavior) of the drive correctly, notably for fast speed slopes, but the drive
itself is able to produce the required torque as can be seen from Fig. 17.
50 100 150 200
10
5
0
5
10
15
20
25
30
Time [s]
T
o
r
q
u
e
[
N
m
]
T
cmd,PT
T
e
Fig. 17. Command and electrical torques T
cmd,PT
and T
e
of riding pro-
le no. 1 (40s-230s): The drive is able to produce the required torque.
shown in Fig. 18, where several of such dips can be seen (at
approximately 30s, 250s, 400s, 440s, 480s, 615s, and 720s).
The inability of the drive to produce the command torque can
further be seen in Fig. 19, where the electric torque does not
reach the command torque whenever a speed dip can be seen
in Fig. 18.
These limits of the drive are even more evident with the
simulations of the riding prole no. 4. (Figs. 20 and 21).
If not the recorded (measured) speed
PT
but the simulated
one
sim
were used as inputs into the feedback loops, and in a
real case, such decreases in speed would cause a decrease of
back-EMF and thus increase of current and torque. However,
as the motor does not receive the simulated, but the higher
recorded (measured) speed as inputs in the simulations, the
simulated torque remains low, further decreasing the simulated
speed. Therefore, drawing incorrect conclusions from these
rapid drops in the simulated speed drops must be avoided.
1579
0 200 400 600 800
0
5
10
15
Time [s]
S
p
e
e
d
[
r
a
d
/
s
]
w
PT
w
sim
Fig. 18. Recorded and simulated speed w
PT
and w
sim
of riding prole no. 1
(rst 840s); at approximately 30s, 250s, 400s, 440s, 480s, 615s, and 720s, the
command torque exceeds the torque limit of the drive at the current speed.
0 200 400 600 800
10
5
0
5
10
15
20
25
30
Time [s]
T
o
r
q
u
e
[
N
m
]
T
cmd,PT
T
e
Fig. 19. Command and electrical torques T
cmd,PT
and T
e
of riding pro-
le no. 1 (rst 840s); at approximately 30s, 250s, 400s, 440s, 480s, 615s, and
720s, the command torque exceeds the torque limit of the drive at the current
speed.
Of course, such speed simulation results with speed dips
where the motor cannot produce the command torque can even
less be used for energy consumption calculations than those
discussed above (Section IV-B.2). Therefore, the modied
performance submodule (Fig. 11) needs to be used.
The orders of magnitudes are illustrated with the following
simple approximation: With torque and back-EMF constant of
k
t
= k
v
= 1Nm/A, the back-EMF is 10V at 10rad/s speed.
Neglecting a possible current limit of the machine, 14V are left
for the voltage drop at the resistance(s), giving approximately
14Nm torque.
4) Efciencies: First, we consider the following two com-
puted values for each of the four riding proles: (i) The battery
energy output, which is the energy input to the drive, W
in
,
and is calculated from the battery voltage and current V
batt
and I
batt
(Fig. 6.1). (ii) The drive output energy W
out,
which
is calculated from the recorded (measured) speed
PT
and the
electric torque T
e
(Fig. 11) (Table III).
0 50 100 150 200
30
20
10
0
10
20
Time [s]
S
p
e
e
d
[
r
a
d
/
s
]
w
PT
w
sim
Fig. 20. Recorded and simulated speed w
PT
and w
sim
of riding prole no. 4
(rst 200s): The drive is not able to produce the required torque. The
discrepancy is that large that the simulated speed becomes negative.
0 200 400 600 800
10
0
10
20
30
40
50
Time [s]
T
o
r
q
u
e
[
N
m
]
T
cmd,PT
T
e
Fig. 21. Command and electrical torques T
cmd,PT
and T
e
of riding pro-
le no. 4 (rst 840s): The drive is not able to produce the required torque.
TABLE III
RESULTS OF DRIVE SCENARIO SIMULATIONS
Prole W
in
W
out,
W
out,
/W
in
W
prole
W
out,
/W
prole
no. [Wh] [Wh] [Wh]
1 29.36 9.36 0.31 12.03 0.78
2 34.48 15.45 0.45 16.36 0.95
3 37.03 22.37 0.60 24.93 0.90
4 30.00 17.29 0.58 54.76 0.32
From these two values, the ratio W
out,
/W
in
can be calcu-
lated, which has the form of an equivalent efciency. This
quantity does not consider if the drive is able to produce the
required torque, but it only considers the torque the motor can
deliver. As both the input and the output energy are derived
from the recorded speed, these computations are not affected
by the inaccuracies of the simulated speed. Here, relative large
amounts of time during which the drive is operated at high
speed translates into higher values of this ratio for the different
riding proles.
1580
Next, the energy requirement of the riding proles, W
prole
,
each for the rst 840s, are considered (Table III), from which
a second ratio, W
out,
/W
prole
, can be derived. From this
quantity conclusions on how much of the riding prole can
be produced by the drive can be drawn.
In the model, the armature and battery currents I
a
and I
batt
can be both positive and negative. As a result, the simulations
include the possibility of generating mode. This becomes
notably obvious with riding prole no. 4. At operating points
where the back-EMF that is computed from the recorded (mea-
sured) speed
PT
exceeds 24rad/s, the back-EMF is larger than
the maximum command voltage, the armature current becomes
negative (Fig. 10), eventually translating into a negative battery
current I
batt
(Figs. 4 and 5) that recharges the battery.
By comparing W
out
and W
prole
, conclusions on how much
of the riding prole can be produced by the drive can be drawn.
These results do not change with a change of J, as the
different energy computations and the command signals are
decoupled from the simulated speed. Furthermore, the simu-
lations with different values of the controller, K
p
= 2 and
K
i
= 0.5, were carried out, leading to the same results.
Regarding the energy requirements of the different riding
proles in a more general way, most proles could be sup-
plied from one or 1.5 laptop-size batteries, when assuming a
conservative estimate of 30% overall efciency.
V. CONCLUSIONS
Different technical performance criteria of electric bicycle
drive systems for given operating cycles can be evaluated
via a model implementation that uses riding proles based
on actual road tests. Such analysis can contribute to de-
signing better, custom-designed electric bicycles and thereby
overcoming common issues such as high cost and weight.
The implementation shows how the characteristic of a given
drive depends on the characteristic of the riding prole. The
presented tool can be used to develop more general answers to
operating areas of drives with respect to the different torque-
speed combinations and their derivations with respect to time
as they can occur with electric bicycles.
APPENDIX
Appendix A: List of abbreviations
Acronym Denition
DC Direct Current
EMF Electro-Magnetic Force
PI Proportional-Integral
PWM Pulse-Width Modulation
Appendix B: List of symbols
Name Description
B rolling resistance
I
a
armature current
I
batt
battery current
I
cmd
current resulting from command duty
I
cmd,PT
current resulting from command duty,
derived from measured values
J inertia
List of symbols continued
Name Description
k
v
back-EMF constant
k
t
torque constant
L
a
armature inductance
P
in
input power
P
mech
mechanical power
t time
R
a
armature resistance
R
b
battery resistance
T
cmd
command torque
T
cmd,PT
command torque, derived from recorded values
T
PT
recorded (measured) torque
T
e
electric torque
T
load
load torque
T
load,
load torque, derived from recorded values
T
net
net torque
T
PT
measured torque
V
batt
battery voltage
V
c
dc-link voltage
V
cmd
command voltage
V
cmd,PT
command voltage, derived from recorded values
W
batt
battery remaining energy
W
batt0
battery capacity
W
in
energy input
W
mech
mechanical energy
speed
PT
recorded (measured) speed
REFERENCES
[1] B. Kumar and H. Oman, Power control for battery-electric bicycles,
Proceedings of NAECON 93 - National Aerospace and Electronics
Conference, vol. 1, pp. 428-434, Dayton, OH, May 24-28, 1993.
[2] W.C. Morchin, Battery-powered electric bicycles, Proceedings of
Northcon 94, pp. 269-274, Seattle, WA, October 11-13, 1994.
[3] E.A. Lomonova, A.J.A. Vandenput, J. Rubacek, B. dHerripon, and
G. Roovers, Development of an improved electrically assisted bicycle,
Proceedings of 37th IEEE IAS Annual Meeting, pp. 384-389, Pittsburgh,
PA, October 13-18, 2002.
[4] F.E. Jamerson, Electric bikes worldwide 2002: with electric scooters &
neighborhood EVs, Naples, Fla: Electric Battery Bicycle Co, 2002.
[5] A. Muetze, A.G. Jack, and B.C. Mecrow, Brushless-dc motor using soft
magnetic composites as a direct drive in an electric bicycle, Proceedings
of the 9th European Conference on Power Electronics and Applications
(EPE), paper no. 350, Graz, 2001.
[6] D.G. Wilson, J. Papadopoulos, and F.R. Whitt, Bicycling science,
Cambridge, Mass: MIT Press, 2004.
[7] P. Fairley, Chinas cyclists take charge: electric bicycles are selling by
the millions despite efforts to ban them, IEEE Sprectrum, vol. 42, no. 6,
pp. 54-69, June 2005.
[8] A. Muetze and Y.C. Tan, Performance evaluation of electric bicylces,
Proceedings of 40th IEEE IAS Annual Meeting, pp. 2865-2872, Hong
Kong, October 2-6, 2005.
[9] Power Tap
R
is by Graber Products, Inc., 5253 Verona Road, Madison,
WI USA, http://www.cycle-ops.com.
[10] S. Lee and W. Ham, Self stabilizing strategy in tracking control of
unmanned electric bicycle with mass balance, Proceedings of IEEE
Intellingent Robots and Systems (RSJ), vol. 3, pp. 2200-2205, September
30 - October 4, Lausanne, 2002.
[11] L. Li, F.Y. Wang, and Q. Zhou, Integrated longitudinal and lateral
tire/road friction modeling and monitoring for vehicle motion control,
IEEE Transactions on Intelligent Transportation Systems, vol. 7, no. 1,
pp. 1-19, March 2006.
1581