energies

Article
Regenerative Intelligent Brake Control for
Electric Motorcycles
Juan Jesús Castillo Aguilar * ID , Javier Pérez Fernández, Juan María Velasco García and
Juan Antonio Cabrera Carrillo ID
Department of Mechanical Engineering, University of Málaga, 29071 Málaga, Spain; javierperez@uma.es (J.P.F.);
juanmav@uma.es (J.M.V.G.); jcabrera@uma.es (J.A.C.C.)
* Correspondence: juancas@uma.es; Tel.: +34-951-952-372; Fax: +34-951-952-605

Received: 17 August 2017; Accepted: 13 October 2017; Published: 20 October 2017

Abstract: Vehicle models whose propulsion system is based on electric motors are increasing in
number within the automobile industry. They will soon become a reliable alternative to vehicles
with conventional propulsion systems. The main advantages of this type of vehicles are the
non-emission of polluting gases and noise and the effectiveness of electric motors compared to
combustion engines. Some of the disadvantages that electric vehicle manufacturers still have to solve
are their low autonomy due to inefficient energy storage systems, vehicle cost, which is still too
high, and reducing the recharging time. Current regenerative systems in motorcycles are designed
with a low fixed maximum regeneration rate in order not to cause the rear wheel to slip when
braking with the regenerative brake no matter what the road condition is. These types of systems
do not make use of all the available regeneration power, since more importance is placed on safety
when braking. An optimized regenerative braking strategy for two-wheeled vehicles is described
is this work. This system is designed to recover the maximum energy in braking processes while
maintaining the vehicle’s stability. In order to develop the previously described regenerative control,
tyre forces, vehicle speed and road adhesion are obtained by means of an estimation algorithm.
A based-on-fuzzy-logic algorithm is programmed to carry out an optimized control with this
information. This system recuperates maximum braking power without compromising the rear
wheel slip and safety. Simulations show that the system optimizes energy regeneration on every
surface compared to a constant regeneration strategy.

Keywords: regeneration; electric vehicles; antilock brake system (ABS); fuzzy logic

1. Introduction
A large number of countries are adopting policies to increase the number of electric vehicles
in their fleets. The aim of these policies is to improve air quality by reducing pollutant emissions,
especially in urban areas [1]. In this environment, the relationship between braking time and the
vehicle moving time is very high. This indicates that there is a large amount of energy wasted as heat.
This wasted kinetic energy can be recovered in electric vehicles by means of a regenerative braking
system [2–4]. In large cities, with frequent stops and intersections, the energy dissipated in braking
can reach 34% of total traction energy [3]. In traditional vehicles, that energy is wasted as heat through
the brake friction. For this reason, regeneration or energy recovery systems are being widely used in
electric vehicles.
Electric vehicles have the ability to easily implement traction and/or braking control systems on
the drive wheels. Since the motor is already part of the system, it is only necessary to implement a
minimum of software and hardware to control that torque in the wheel. The great advantage is that
electric motors can also be used as power generators. When a positive torque is required, the motor

Energies 2017, 10, 1648; doi:10.3390/en10101648 www.mdpi.com/journal/energies

Energies 2017, 10, 1648 2 of 16

consumes battery power and when a negative torque is applied, motor power is delivered to the
batteries. Nowadays, traction and braking controls are widely implemented in vehicles, with the aim of
reducing braking distance or providing better manoeuvrability in emergency situations. Some of these
systems are the Antilock Brake System (ABS) or the Traction Control System (TCS). These systems
have evolved from their origin, using increasingly sophisticated algorithms and complex control
architectures. Fuzzy logic [5,6], sliding control [7–9], control by artificial neural networks [10,11] and
nonlinear control [12,13] are examples of the most used control methods. These systems try to optimize
the longitudinal and lateral force in the tire, obtaining the maximum available force in the wheel-road
contact during braking and traction processes.
At present, regeneration systems in commercial electric vehicles do not take advantage of the
characteristics mentioned above. A maximum level of regeneration is fixed for safety or battery
capacity reasons. Therefore, they do not take advantage of the maximum available longitudinal force
in tire-road contact. This way, these systems do not get the highest possible regenerated energy.
In addition, most of the research on regeneration systems has focused on four-wheeled vehicles,
although some authors have addressed this issue for two-wheeled vehicles. Most of them only deal
with the electrical part of the system and do not take into account vehicle-road interaction [14–17].
Robinson and Singh [18] developed a control technique that applies a braking torque to the rear wheel,
regenerating that braking energy from the rear wheel to the battery system. Moreover, the system
controls the slip to prevent the locking of the rear wheel during braking. Lin et al. [19] studied an ABS
braking control for electric scooters based on regenerative braking.
In this work the design of a regenerative braking system for electric motorcycles is carried out
using control and estimation of the road adherence algorithms. The system recovers the maximum
energy during the braking process with the rear wheel of the motorcycle, always guaranteeing the
safety of the rider thanks to the incorporation of a wheel slip control system. As we have mentioned,
both energy saving and safety are fundamental subjects of study in vehicles. The system proposed here
aims to improve the safety of electric vehicles as well as save energy in a novel way. It makes use of a
fuzzy control that estimates the road adhesion and determines the optimal regenerative braking torque
without causing the wheel to slip. The proposed system will reduce the use of the traditional rear
friction brake depending on road conditions thanks to the high braking torque that can be achieved
with the electric motor.
This article begins with the description of the dynamic model of the electric motor used for
simulation in Section 2. The method to estimate the speed of the vehicle as well as the detection
of the type of road is included in Section 3. The regenerative control system is described in
Section 4. The performance of the whole system is verified through simulations included in Section 5.
Finally, the conclusions are presented in Section 6.

2. Electric Motor Model

Modelling of the electric motor is required to know the torque on the rear wheel. The model is
used in the simulations of the regeneration system.
Our prototype motorcycle has a Permanent Magnet Alternate Current (PMAC) motor, which is
powered by a controller. This controller is also powered from the battery system. The battery system
is able to absorb the energy recovered during the regeneration. In our case, the maximum current
regenerated by the system is limited to 150 A to prevent heating of the batteries. The maximum torque
that can be applied is 80 Nm. The curve relating the torque to the angular speed of rotation of the
motor is shown in Figure 1. These curves have been obtained by means of a dynamometric test by
varying the angular speed of the motor and the position of the accelerator. As shown in Figure 1,
a negative torque that brakes the vehicle is applied to the rear wheel of the motorcycle when the motor
is in regeneration mode.
Energies 2017, 10, 1648 3 of 16
Energies 2017, 10, 1648 3 of 16
Energies 2017,
Energies 2017, 10,
10, 1648
1648 33 of
of 16
16

Figure 1. Torque curves of the electric motor.

Figure1.
Figure
Figure 1. Torque
1. Torque curves
Torque curves of
curves of the
the electric
the electric motor.
electricmotor.
motor.
As it
As it can
can bebe seen,
seen, the
the curves
curves areare proportional
proportional to to the
the Throttle
Throttle or
or Regeneration
Regeneration Rate Rate (TR).
(TR). In
In this
this
AsAs
paper,itTRit can
can = bebe
[−1,
seen,the
seen,
1],
thecurves
where
curvesare
−1
areproportional
means
proportional to
100%
to the
regeneration the Throttle
Throttle
and 1
or
orRegeneration
means Regeneration
100%
Rate
Rate(TR).
acceleration. (TR).In In
Previous
this
this
paper, TR
paper, TR == [−1,
[−1, 1],
1], where
where −1 −1 means
means 100%100% regeneration
regeneration and and 11 means
means 100%
100% acceleration.
acceleration. Previous
Previous
paper, TR
curves =
have
curves have [ −
have been1,
been1], where
modelled
been modelled −
modelled using 1 means
using a 100%
Look-up
using aa Look-up regeneration
table.
Look-up table. The
table. The rear
The rear and 1
wheel
rear wheel means
torque
wheel torque
torque is100%
is acceleration.
obtained
is obtained
obtained by by means
by means Previous
means ofof the
of the
the
curves
following
curves expression:
have expression:
following been modelled using a Look-up table. The rear wheel torque is obtained by means of the
following expression:
following expression: = ∙∙ (( ,, )) (1)
T= =
= ϕ·∙f ((ωm ,,TR)) (1)
(1) (1)
where
where T T is the
T is the torque
the torque
torque on on the
on the rear
the rear wheel,
rear wheel,
wheel, is the
is the transmission
the transmission ratio
transmission ratio between
ratio between
between the the motor
the motor and and the
and the rear
the rear
rear
where
where T isωis
the ϕ isisthe between themotor
wheel, m is torque on the
the angular rear of
speed wheel,
the motor, TRtransmission
is the Throttleratio
or Regeneration motor
rate andand
f(∙) the
is therear
wheel,
wheel, ω
ω m is the angular speed of the motor, TR is the Throttle or Regeneration rate and f(∙) is the
m is the angular speed of the motor, TR is the Throttle or Regeneration rate and f(∙) is the
wheel, ω m is the
adjustment angular
function speed
of the
the of the
look-up motor, TR is the Throttle or Regeneration rate and f (·) is the
table.
adjustment
adjustment function
function of
of look-up
thelook-up
look-up table.
table.
adjustment
The function
controller of the
installed on the table.
electric motorcycle has has been
been modelled
modelled using
using aa step
step input
input (Figure
(Figure
The controller
The controller installed
installed on on the
the electric
electric motorcycle
motorcycle has been modelled using a step input (Figure
2). The
The
2). The response
controller
The response
response is is approximated
installed on
is approximated
approximated tothe to a second
electric order
motorcycle
to aa second
second order system
has
order system according
been modelled
system according
according toto the
to the expression
using a step
the expression shown
input
expression shown in the
(Figure
shown in in the 2).
2). the
figure.
Thefigure.
response is approximated to a second order system according to the expression shown in the figure.
figure.

1
1
0.007s 22 +10.134 s + 1
+ 0.134
0.007ss 2 +
0.007 + 11
0.134 ss +

Figure 2. Second order response model of the electric motor.

Figure 2.Second
Figure2.2.
Figure Second order
Second order response
order modelof
model
response model ofthe
of theelectric
the electricmotor.
electric motor.
motor.

The actual
The actual responses
responses of
of the
the motor
motor have
have been
been compared
compared toto the
the programmed
programmed model.
model. Two
Two step
step
The actual
The actualresponses
responsesofof
thethemotor
motor have
have been
been compared
compared to to the
theprogrammed
programmedmodel.model. Two
Two step
step
inputs have
inputs have been
been used.
used. The
The results
results are
are shown
shown in
in Figure
Figure 3.
3. As
As it
it can
can be
be seen,
seen, the
the model
model response
response
inputs
inputs have
have beenused.
been used. Theresults
results are
are shown
shown in Figure
Figure 3.
3. As
Asititcan
canbebeseen,
seen,the
themodel response
matches
matches with
with measuredThe
measured one appropriately.
one appropriately. model response
matches
matches with
with measured
measured one
one appropriately.
appropriately.

Figure 3. Real and simulated response to step input.

Figure 3.
Figure 3. Real
Real and
and simulated
simulated response
response to
to step input.
input.
Figure 3. Real and simulated response tostep
step input.
Energies 2017, 10, 1648 4 of 16

3. Estimation of Road Type and Vehicle Parameters

Energies 2017, 10, 1648 4 of 16
The slip on each of the wheels is used to determine the road type. The slip is defined with the
following equation in the case of braking:
3. Estimation of Road Type and Vehicle Parameters
ω·r
1−
The slip on each of the wheels is used sto=determine the road type. The slip is defined with the (2)
vx
following equation in the case of braking:
where ω is the angular velocity of the wheel, r the radius ∙ of the tire and vx the longitudinal velocity of
the vehicle. =1− (2)
An Extended Kalman Filter (EKF) [5] based on a model that simulates the straight line behavior
where ω is the angular velocity of the wheel, r the radius of the tire and vx the longitudinal velocity
of the motorcycle is used in order to determine the speed of the vehicle. This model is described next.
of the vehicle.
In Figure 4 the basic geometry of a motorcycle is represented schematically. The equations that
An Extended Kalman Filter (EKF) [5] based on a model that simulates the straight line behavior
govern the dynamics of the vehicle are:
of the motorcycle is used in order to determine the speed of the vehicle. This model is described next.
In Figure 4 the basic geometry of a motorcycle
.. . . is represented
. 2 schematically. The equations that
Ma x = M( x + θ z) = Fxr − C x (3)
govern the dynamics of the vehicle are:
.. ..
Maz ==M(z −+θ z + g=
) = N−f + Nr (4)
(3)
..
Iy=θ = N−
r lr − + = Fx z+
Nf l f − (4)
(5)
=T = ϕ−·Km · I − (5)
(6)

The equations defining the front and rear = ∙ ∙

suspensions can be simplified as: (6)
The equations defining the front and rear suspensions can . be simplified
. as:
Nr = −kr ·(z − L0 + lr ·θ ) − cr ·(z + lr ·θ ) (7)
=− ·( − + · )− · + · (7)
  . .
N f==−−k ·f · z−− L0++ l f· ·θ −− c f ··(z +
+ l f ·θ· ) (8)
(8)
The parameters and variables used in this model are
are shown
shown in
in Table
Table1.
1.

θ

Figure 4. Simplified motorcycle model.

Figure 4. Simplified motorcycle model.

Table 1. Motorcycle model parameters.

Description Description
M Mass J Moment of inertia of the wheel
ax Longitudinal acceleration Iy Inertia on the Y axis
az Vertical Acceleration Nf Normal front force
x Longitudinal displacement Nr Normal rear force
z Vertical displacement lf Front half length
θ Pitch angle lr Rear half-length
Rr Rear tire radius kf Stiffness of front suspension
T Torque applied on rear wheel kr Stiffness of rear suspension
Energies 2017, 10, 1648 5 of 16

Table 1. Motorcycle model parameters.

Description Description
M Mass J Moment of inertia of the wheel
ax Longitudinal acceleration Iy Inertia on the Y axis
az Vertical Acceleration Nf Normal front force
x Longitudinal displacement Nr Normal rear force
z Vertical displacement lf Front half length
θ Pitch angle lr Rear half-length
Rr
Energies 2017, 10, 1648 Rear tire radius kf Stiffness of front suspension 5 of 16
T Torque applied on rear wheel kr Stiffness of rear suspension
CC Aerodynamic
Aerodynamic coefficient
coefficient cfcf Front
Front Damping
Damping
F Longitudinal rear wheel force c Rear damping
Fxr xr Longitudinal rear wheel force cr r Rear damping
ω Rear wheel angular speed L0 Suspension length
ωr r Rear wheel angular speed L0 Suspension length
Km Torque-Intensity Ratio I Intensity
Km Torque-Intensity Ratio I Intensity

The
The control
control model
model to
to be
be used
used in
in this
this work
work is
is shown
shown in
in Figure
Figure5.
5.

θ θ

Figure 5.
Figure 5. Motorcycle
Motorcycle regenerative
regenerative control
control scheme.
scheme.

As it can be seen, the regeneration control block (4) has the optimum slip (sopt) as input.
As it can be seen, the regeneration control block (4) has the optimum slip (sopt) as input. Optimum
Optimum slip is defined as the slip value that produces the maximum longitudinal force during
slip is defined as the slip value that produces the maximum longitudinal force during braking or
braking or traction. This is the desired slip for the motorcycle to have the highest regeneration rate.
traction. This is the desired slip for the motorcycle to have the highest regeneration rate. In addition,
In addition, control block 4 also has the current slip (s) on the rear wheel of the motorcycle as input.
control block 4 also has the current slip (s) on the rear wheel of the motorcycle as input. The output
The output of this block is the throttle or regeneration rate. In our case, this rate is an input to the
of this block is the throttle or regeneration rate. In our case, this rate is an input to the motor of
motor of the motorcycle.
the motorcycle.
A parameter estimation algorithm is needed to obtain the inputs to the regeneration control,
A parameter estimation algorithm is needed to obtain the inputs to the regeneration control,
since these cannot be measured with sensors in the vehicle. This is carried out in blocks (2) and (3) of
since these cannot be measured with sensors in the vehicle. This is carried out in blocks (2) and (3) of
Figure 5.
Figure 5.
The linear speed of the motorcycle, the longitudinal force and the coefficient of friction in the
The linear speed of the motorcycle, the longitudinal force and the coefficient of friction in the rear
rear wheel are estimated first. To this end, an Extended Kalman filter as proposed in [5] is used.
wheel are estimated first. To this end, an Extended Kalman filter as proposed in [5] is used.
The EKF is defined according to the following equations:
The EKF is defined according to the following equations:
= ( )+ (9)
x k = f k −1 ( x k −1 ) + w k −1 (9)
=ℎ ( )+ (10)
where wk and vk represent the noise of the model and the measures respectively, xk is the state vector
and jk is the measurement vector. The state vector is:

( )= , , , , , (11)

and the measurement vector is:

Energies 2017, 10, 1648 6 of 16

jk = hk ( xk ) + vk (10)

where wk and vk represent the noise of the model and the measures respectively, xk is the state vector
and jk is the measurement vector. The state vector is:
h iT
x (k) = Fx,r , Nr , N f , v x , T (11)

and the measurement vector is: . T

j(k) = [ a x , az , θ, I ] (12)

A first order random walk is used to estimate the longitudinal and vertical forces. Therefore,
the model equations are included next:

Fxr (k ) = Fxr (k − 1) (13)

Nr (k ) = Nr (k − 1) (14)

N f ( k ) = N f ( k − 1) (15)

Fxr (k − 1) − C ·v x 2 (k − 1)
 
v x (k) = v x (k − 1) + ∆t· − θ ·vz (16)
M
T ( k ) = T ( k − 1) (17)

Finally, the measurement equations are the following:

Fxr (k) − C ·v x 2 (k)

a x (k) = (18)
M

N f (k) + Nr (k) − M· g
az (k ) = (19)
M
.. Nr (k)·lr − N f (k)·l f − Fxr (k)·z(k)
θ (k) = (20)
Iy
T (k)
I (k) = (21)
Km
The EKF algorithm estimates the longitudinal and vertical forces and the longitudinal speed of
the motorcycle in addition to the rear wheel torque with the equations developed above.
An estimate of the slip and the coefficient of friction can be made according to the
following equations:
ωr (k)· Rr
sest (k) = 1 − (22)
v x (k)
Fxr (k)
µest (k) = (23)
Nr (k)
Figure 6 shows, by way of example, a simulation performed with BikeSim® (Ann Arbor, MI, USA)
where the parameters necessary for the determination of the slip and the coefficient of adhesion are
estimated. The test consists of the following sections: 40 m of high adhesion surface, 40 m of low
adhesion surface and a third section with high adhesion again. The initial speed is 120 km/h. In these
simulations the signal from the sensors has been altered with zero-mean random noise. A subscript “k”
in the legend indicates that it is an estimated magnitude. The parameters estimated by means of the
EKF fit quite accurately to the values provided by the simulation software. It can be verified that the
speed (vx ), front vertical force (Fzf ), rear vertical force (Fzr ), rear traction force (Ft ) and brake torque (T)
are estimated correctly.
 adhesion surface and a third section with high adhesion again. The initial speed is 120 km/h. In these
simulations the signal from the sensors has been altered with zero-mean random noise. A subscript
“k” in the legend indicates that it is an estimated magnitude. The parameters estimated by means of
the EKF fit quite accurately to the values provided by the simulation software. It can be verified that
the speed
Energies (vx1648
2017, 10, ), front vertical force (Fzf), rear vertical force (Fzr), rear traction force (Ft) and brake torque
7 of 16
(T) are estimated correctly.

Figure6.6.Estimation
Figure Estimation of
of speed,
speed, tractor
tractoreffort
effortand
andvertical
verticalforces.
forces.
Energies 2017, 10, 1648 7 of 16
Energies 2017, 10, 1648 7 of 16
Finally,
Finally,the
thefollowing figureshows
following figure showsthethe results
results obtained
obtained in atest
in a real real test carried
carried out with out with the
the vehicle
vehicle described
Finally,
described in the in Appendix
following
Appendix A.estimation
A. figure
This This estimation
shows the results algorithm
obtained
algorithm inverification
a realtest
verification testwas test
carriedwas
out performed
performed withwith with a
the vehicle
a 20%
20%described
constant in Appendix
regeneration A. This
rate. estimation
Figure 7 algorithm
includes the verification
actual speed (vtest
x ), was performed
measured
constant regeneration rate. Figure 7 includes the actual speed (vx), measured by a high frequency by a with
high a 20%
frequency
constant
Global regeneration rate. Figure 7 includes the actual speed (v x), measured
Global Positioning System-based speed sensors and the measured brake torque. It can be seen thatthe
Positioning System-based speed sensors and the measured brake torque. by
It a
can high
be frequency
seen that
Global
speed
the speedPositioning
is perfectly
is perfectlySystem-based
estimated.
estimated. speed
The brake
The sensors
torque
brake and
is also
torque the measured
is correctly
also estimated
correctly brakeexcept
estimated torque.inItthe
except can be final
in final
the seen that
partpart
of the
the
test, speed
of where is perfectly
the test,a where
higheraerrorestimated.
higheris error The
observed. brake
is observed. torque is also correctly estimated except in the final part
of the test, where a higher error is observed.

Figure7.7.Estimation
Figure Estimation of
of speed
speed and
andbrake
braketorque.
torque.
Figure 7. Estimation of speed and brake torque.
These values are used to determine the road type and to obtain the optimal slip, as shown in
These
These
Figure 8. Avalues
values
controlare
areused
based onto
used to determine
determine
fuzzy the
logic isthe roadtype
road
programmedtypeand
toand to to obtain
obtain
estimate thethethe optimal
optimal
road type. slip,slip,
The asofshown
as shown
ouput in
this
Figure
in block
Figure 8.
8.A A control
controlbased on
based fuzzy
on logic
fuzzy is
logic programmed
is programmed to estimate
to the
estimateroad type.
the roadThe ouput
type.
is an index called Road Condition Index (RCI), related to the adherence of the road. Finally, the Theof this
ouput
block
of optimalis an
this blockslipindex
isisan called
index Road
obtained calledCondition
through Road Index
Condition
a Neural (RCI),
Network. related
Index to the
(RCI), adherence
related to theofadherence
the road. Finally,
of the the
road.
optimal slip is obtained through a Neural Network.
Finally, the optimal slip is obtained through a Neural Network.

Figure 8. Control of road type and estimation of optimum slip.

Figure8.8.Control
Figure Controlof
ofroad
road type
type and
and estimation
estimation of
ofoptimum
optimumslip.
slip.
The road type estimation block has the slip (s), the longitudinal coefficient of friction (μ) and the
The road
variation of thetype estimation
longitudinal block haswith
coefficient the slip (s), the
respect longitudinal
to the slip (dμ/ds)coefficient
as input. Itofproduces
friction (μ)
anand the
output
variation of the longitudinal coefficient with respect to the slip (dμ/ds) as input. It produces
value indicating the type of road on which the vehicle circulates. This value is within the range (0– an output
valuewhere
1.2), indicating the type aofroad
1.2 represents roadwith
on which the vehicle
the highest circulates.
coefficient This value
of adhesion and 0isrepresents
within thetherange (0–
lowest
1.2),
one. where 1.2 represents a road with the highest coefficient of adhesion and 0 represents the lowest
one. The input and output variables have been fuzzyfied by means of the membership functions
The input and output variables have been fuzzyfied by means of the membership functions
Energies 2017, 10, 1648 8 of 16

The road type estimation block has the slip (s), the longitudinal coefficient of friction (µ) and the
variation of the longitudinal coefficient with respect to the slip (dµ/ds) as input. It produces an output
value indicating the type of road on which the vehicle circulates. This value is within the range (0–1.2),
where 1.2 represents a road with the highest coefficient of adhesion and 0 represents the lowest one.
The input and output variables have been fuzzyfied by means of the membership functions
Energies 2017, 10, 1648 8 of 16
indicated in Figure 9. In addition, the rules defined for the fuzzy control are presented in Table 2.

Figure 9.
Figure 9. Membership
Membership functions
functions for
for input
input and
and output
output variables
variables of
of the
the road
road type
type detection
detection block.
block.

Finally,asasshown
Finally, shown in in Figure
Figure 8, once
8, once the road
the road type
type has hasobtained
been been obtained by index,
by the RCI the RCI theindex,
optimumthe
optimum
slip slip is calculated.
is calculated. In this
In this work, thework, the optimum
optimum slip has
slip has been been estimated
estimated by meansbyofmeans of a trained
a trained neural
neural network using the Burkhardt model [20]. The network has the RCI
network using the Burkhardt model [20]. The network has the RCI index, obtained from theindex, obtained from the
fuzzy
fuzzy block, and the wheel slip angle as inputs. The net provides the optimum slip for
block, and the wheel slip angle as inputs. The net provides the optimum slip for each surface from each surface
from these
these inputs.
inputs. In ourIncase,
our the
case, theangle
slip slip angle has taken
has been been taken as null.
as null.

Table 2.
Table 2. Road
Road control
control rules.
rules.

Rule Number Friction Coefficient Slip dμ/ds RCI

Rule Number 1 Friction Coefficient
MPR Slip
mid dµ/ds VSRF
--- RCI
1 2 MPR MPR mid
high --- — ZRF VSRF
2 3 MPR MPR high
zero Lmud — MRF ZRF
3 4 MPR MPR zero
zero Lmud LRF
Mmud MRF
4 5 MPR MPR zero
zero Mmud LRF
Hmud LRF
5 MPR zero Hmud LRF
6 PR mid --- SRF
6 PR mid — SRF
7 PR high --- VSRF
7 PR high — VSRF
8 8 PR PR zero
zero Lmud
Lmud LRF LRF
9 9 PR PR zero
zero Mmud
Mmud LRF LRF
10 10 PR PR zero
zero Hmud
Hmud VLRF VLRF
11 11 RM RM mid
mid --- — MRF MRF
12 12 RM RM high
high --- — MRF MRF
13 13 RM RM zero
zero Lmud LRF
Lmud LRF
14 14 RM RM zero
zero MmudVLRF
Mmud VLRF
15 15 RM RM zero
zero Hmud HRF
Hmud HRF
16 RN mid --- VLRF
17 RN high --- LRF
18 RN zero Lmud VLRF
19 RN zero Mmud HRF
20 RN zero Hmud ERF
21 MUR mid --- HRF
Energies 2017, 10, 1648 9 of 16

Table 2. Cont.

Rule Number Friction Coefficient Slip dµ/ds RCI

16 RN mid — VLRF
17 RN high — LRF
18 RN zero Lmud VLRF
19 RN zero Mmud HRF
20 RN zero Hmud ERF
21 MUR mid — HRF
22 MUR high — ERF
23
Energies 2017, 10, 1648 MUR zero Lmud HRF 9 of 16
24 MUR zero Mmud ERF
25 25 MUR MUR zero
zero Hmud
HmudERF ERF

4. Regenerative Control
4. Regenerative Control
The last step in the control scheme is to obtain the level of regeneration that is desired in order
The last step
to obtain in the control
the maximum scheme
energy is to in
recovered obtain the level
the braking of regeneration
processes. thataiscontrol
To this end, desired in order
based on to
obtainfuzzy
the maximum
logic has been developed (Figure 10). This fuzzy block has the slip error and the variation offuzzy
energy recovered in the braking processes. To this end, a control based on
logic has beenindeveloped
the error (Figure
time as inputs, 10).are
which This fuzzy
defined as:block has the slip error and the variation of the error
in time as inputs, which are defined as: ( )= ( )− ( ) (24)
( () t−
e∆(t() )==sopt ) −( s(−t)1) (25) (24)
∆ ∆
As it can be seen, the slip error∆e t) t) −
e(as
is (defined t − 1)
thee(difference between the optimal slip and the
= (25)
actual slip at each instant of time. Once ∆tthe error and∆tthe error variation have been calculated, the
percentage
As it can beofseen,
regeneration to be applied
the slip error according
is defined to the next block
as the difference is obtained.
between The output
the optimal variable
slip and the actual
of the fuzzy block, called Δ, is a multiplying factor of the regeneration rate established in the previous
slip at each instant of time. Once the error and the error variation have been calculated, the percentage
time interval. Thus, the rate at time t is obtained from:
of regeneration to be applied according to the next block is obtained. The output variable of the fuzzy
( + 1) = ∆ ∙ rate
block, called ∆, is a multiplying factor of the regeneration ( ) established in the previous time(26)
interval.
Thus, the The at time t is obtained
rateregeneration from:
rate can never be less than a certain minimum value. This is so because if a null
value were reached at an instant of time, the system would not be able to raise the rate again. On the
other hand, the rate can never exceedTR ( t + 1) = ∆
a maximum · TR(This
value. t) maximum value is imposed by the (26)
maximum bearable charge current of the battery system.

Figure 10. Membership functions for the input and output variables of the regeneration rate block.
Figure 10. Membership functions for the input and output variables of the regeneration rate block.
5. Simulations
The vehicle dynamics simulation software BikeSim® was used for this purpose. Simulations
were carried out on Simulink® (Natick, MA, USA), using BikeSim® as S-function. BikeSim® was
incorporated to the model to simulate vehicle behaviour. The model has the outputs from the control
system as inputs and provided the measures obtained in the vehicle. These simulations allow
Energies 2017, 10, 1648 10 of 16

The regeneration rate can never be less than a certain minimum value. This is so because if a null
value were reached at an instant of time, the system would not be able to raise the rate again. On the
other hand, the rate can never exceed a maximum value. This maximum value is imposed by the
maximum bearable charge current of the battery system.

5. Simulations
The vehicle dynamics simulation software BikeSim® was used for this purpose. Simulations
Energiescarried
were 2017, 10,outSimulink® (Natick, MA, USA), using BikeSim® as S-function. BikeSim®10 was
1648on of 16

incorporated to the model to simulate vehicle behaviour. The model has the outputs from the
Table 3. Motorcycle model parameters.
control system as inputs and provided the measures obtained in the vehicle. These simulations
allow evaluating the potential and feasibility of the proposed
Description Value regeneration algorithm. Table 3 shows
the main characteristicsTotal
of the vehicle used in the simulations.
mass 275 kg
Wheel radius 0.32 m
inertia3.ofMotorcycle
Moment ofTable the wheel model parameters.
0.484 kg∙m2
Distance from the COG * to the front axle 0.86 m
Distance from the COG Description Value
* to the rear axle 0.67 m
Height of gravity centre
Total mass 0.4
275mkg
Front
Wheelarea
radius 0.6
0.32mm2

Aerodynamic coefficient
Moment of inertia of the wheel 0.484 kg·m2
0.55
Distance fromtransmission
Motor-wheel the COG * toratiothe front axle 0.86 m
1:6.4
Distance from the*COGCOG*=toCentre
the rear
of axle
Gravity 0.67 m
Height of gravity centre 0.4 m
Front area 0.6 m2
5.1. Low Adhesion Condition Simulation
Aerodynamic coefficient 0.55
In the first place, aMotor-wheel transmission
transition between a lowratio 1:6.4low adhesion surface has been
adhesion and a very
simulated. The test begins at the low adhesion * COG = Centre of Gravity.
surface. After 20 m the surface is changed to a very
low adhesion surface. The initial speed is 60 km/h. The test ends when the motorcycle reaches 20
5.1.
km/h.Low Adhesion Condition
Maximum regeneration Simulation
has been limited to 50%. This rate of regeneration corresponds to an
intensity
In theoffirst
148place,
A, less than 150 between
a transition A of the amaximum
low adhesioncharge
andcurrent
a very lowrecommended by thehas
adhesion surface battery
been
manufacturer.
simulated. The test begins at the low adhesion surface. After 20 m the surface is changed to a very low
In Figure
adhesion 11a,The
surface. the initial
outputspeed
provided
is 60by the road
km/h. Theestimation algorithm
test ends when and the neural
the motorcycle network
reaches can
20 km/h.
be observed. The system is able to detect the change of surface and adapt the optimum
Maximum regeneration has been limited to 50%. This rate of regeneration corresponds to an intensity slip to the
new conditions.
of 148 A, less than 150 A of the maximum charge current recommended by the battery manufacturer.
The
In output
Figure of the
11a, the output
fuzzy block that by
provided provides
the roadtheestimation
regeneration rate is and
algorithm shownthein Figure
neural 11b. It can
network can
be verified how, when changing surface, the system adapts the rate in order to obtain
be observed. The system is able to detect the change of surface and adapt the optimum slip to the the maximum
possible
new regeneration on each surface.
conditions.

(a) (b)

Figure 11.
Figure 11. Estimation
Estimation of road type
of road type and
and optimum
optimum slip
slip (a)
(a) and
and regeneration
regeneration rate
rate (b).
(b).

Finally, the slip and velocity evolution throughout the test is shown in Figure 12. The slip tends
to reach the optimum slip, adapting its value after the change of surface.
Energies 2017, 10, 1648 11 of 16

The output of the fuzzy block that provides the regeneration rate is shown in Figure 11b. It can
be verified how, when changing surface, the system adapts the rate in order to obtain the maximum
possible regeneration on each surface.
Finally, the slip and velocity evolution throughout the test is shown in Figure 12. The slip tends to
Energiesthe
reach optimum
2017, 10, 1648 slip, adapting its value after the change of surface. 11 of 16
Energies 2017, 10, 1648 11 of 16

Figure 12. Optimum slip, real slip and speed.

Figure 12. Optimum
Figure 12. Optimum slip,
slip, real
real slip
slip and
and speed.
speed.
5.2. High to Low Adhesion Transition Simulation
5.2.
5.2. High to Low Adhesion
Adhesion Transition
TransitionSimulation
Simulation
Secondly, a transition between a high adhesion surface and a low adhesion surface has been
Secondly,
simulated. The aatest
Secondly, transition
consistsbetween
transition between aa high
high adhesion
of the following sections:surface
adhesion 40 m ofand
surface and aa low
high low adhesion
adhesion
adhesion surface
surface
surface, 40 mhas
ofbeen
has been
low
simulated.
simulated.
adhesion The test
surface andconsists
a third of the following
section with highsections:
adhesion40 40 highinitial
m ofThe
again. adhesion
speedsurface, 40 m Figure
is 120 km/h. of low
low
adhesion
adhesion surface
surface and
and a third
a section
third with
section high
with adhesion
high again.
adhesion The
again. initial
The speed
initial is
speed 120is km/h.
120
13a shows the type of surface and the optimal slip. As in the previous case, the estimation algorithms Figure
km/h. 13a
Figure
shows the
13a shows
quickly type of
the surface
detect surface
type of surface and the optimal slip. As in the previous case, the estimation algorithms
changes.and the optimal slip. As in the previous case, the estimation algorithms
quickly
quickly detect
detect surface
surface changes.
changes.

(a) (b)
(a) (b)
Figure 13. Estimation of road type and optimum slip (a) and regeneration rate (b).
Figure 13.
Figure 13. Estimation
Estimation of
of road
road type
type and
and optimum
optimum slip
slip (a)
(a) and
and regeneration
regeneration rate
rate (b).
(b).
Figure 13b shows the output of the fuzzy block which provides the regeneration rate. It is
Figure
observed that13btheshows
rate ofthe output of is
regeneration the fuzzy block
maintained which
at the provides
maximum in the high
regeneration
adhesionrate. It is
surface.
Figure 13b shows the output of the fuzzy block which provides the regeneration rate. It is
observed
This is duethat
to thethefact
ratethat,
of regeneration
even with thatis maintained
regenerationatpercentage,
the maximum in in
a slip thethe
high
rearadhesion
axle closesurface.
to the
observed that the rate of regeneration is maintained at the maximum in the high adhesion surface.
This is due
optimum to is
slip the fact
not that, even
achieved. with
Once thethat
lowregeneration percentage,
adhesion surface a slip
has been in thethe
reached, rearregeneration
axle close torate
the
This is due to the fact that, even with that regeneration percentage, a slip in the rear axle close to the
optimum slip is not achieved. Once the low adhesion surface has been reached,
decreases in order to adjust the slip to the new adhesion conditions (Figure 14). the regeneration rate
optimum slip is not achieved. Once the low adhesion surface has been reached, the regeneration rate
decreases in order to adjust the slip to the new adhesion conditions (Figure 14).
decreases in order to adjust the slip to the new adhesion conditions (Figure 14).
(km/h)
(km/h)
SlipSlip

Speed
Speed
 Figure 13b shows the output of the fuzzy block which provides the regeneration rate. It is
observed that the rate of regeneration is maintained at the maximum in the high adhesion surface.
This is due to the fact that, even with that regeneration percentage, a slip in the rear axle close to the
Energies 2017,
optimum 10, 1648
slip 12 rate
is not achieved. Once the low adhesion surface has been reached, the regeneration of 16
decreases in order to adjust the slip to the new adhesion conditions (Figure 14).

Speed (km/h)
Slip

Figure
Figure 14.
14. Optimum
Optimum slip,
slip, real
real slip
slip and
and speed.
speed.

5.3. Controls Comparison

Simulations have been carried out to verify that the new control manages to optimize regenerated
energy. For this purpose, tests in which the vehicle starts at a speed of 80 km/h have been carried out.
The test lasts five seconds. At the end of that time, the velocity achieved and the energy recovered are
determined (Table 4). The constant regeneration rate has been set at 10%.
It is observed how the proposed control recovers more energy in all cases. This difference becomes
more evident the greater the adhesion of the surface is. With low adhesion the difference is less because
the optimal rate is very close to the 10% established as a constant. In addition, the proposed control
does not lead to the blocking of the wheels under any circumstances.

Table 4. Comparison of controls.

Surface/Control Vo (km/h) Vf (km/h) Energy (Wh)

High adhesion/Regen. cte. (10%) 80 55.52 4.53
High adhesion/Controlled regeneration 80 22.4 15.93
Medium adhesion/Regen. cte. (10%) 80 55.54 4.53
Medium adhesion/Controlled regeneration 80 36.5 11.37
Low adhesion/Regen. cte. (10%) 80 55.54 4.53
Low adhesion/Controlled regeneration 80 56.1 4.54

5.4. Regenerative vs. Conventional Brake Comparison

Finally, emergency braking processes have been simulated. The following simulations allow
evaluating the viability of substituting the conventional rear friction brake with the regenerative brake.
The first test consists of the following sections: 40 m of high adhesion surface and 40 m of low adhesion
surface. The initial speed is 120 km/h. The test is similar to the previous cases except that, in this
case, the front brake is also being actuated. The pressure in the front axle is controlled by an anti-lock
system with slip and wheel angular speed control. The proposed system detects the change in the
adhesion conditions (Figure 15a), adapting the regeneration rate to the adhesion conditions of each
road (Figure 15b). The regeneration rate is kept at its maximum value on the high adhesion surface.
In this zone, the slip on the rear wheels approaches the optimal slip slowly. Once in the low adhesion
surface, the regeneration rate is changed to adapt the slip to the optimal one provided by the estimation
algorithm (Figure 16).
 in this case, the front brake is also being actuated. The pressure in the front axle is controlled by an
anti-lock system with slip and wheel angular speed control. The proposed system detects the change
in the adhesion conditions (Figure 15a), adapting the regeneration rate to the adhesion conditions of
each road (Figure 15b). The regeneration rate is kept at its maximum value on the high adhesion
surface. In this zone, the slip on the rear wheels approaches the optimal slip slowly. Once in the low
adhesion
Energies 2017, surface, the regeneration rate is changed to adapt the slip to the optimal one provided by 13 of 16
10, 1648
the estimation algorithm (Figure 16).

(a) (b)

Figure 15. Estimation of road type and optimum slip (a) and regeneration rate (b).
Figure 15. Estimation of road type and optimum slip (a) and regeneration rate (b). 13 of 16
Energies 2017, 10, 1648

Energies 2017, 10, 1648 13 of 16

Figure 16. Optimum slip, real slip and speed.

Figure 16. Optimum slip, real slip and speed.
Figure
Two additional simulations have16.been
Optimum slip, real
conducted. Theslip and speed.
initial and final speeds are 80 and 20 km/h
Tworespectively
additional in simulations
both cases. Inhave the first
been case, an emergency
conducted. Thebraking is conducted
initial and on a high
final speeds are 80 adhesion
and 20 km/h
TwoThe
surface. additional
rear brakesimulations
torque is have been conducted.
generated with a The initialantilock
conventional and finalsystem
speeds(ABS)
are 80and
andwith20 km/h
respectively in both cases.
respectively in bothsystem
In thethe
cases. In
first case, an emergencybraking braking is conducted highthe
on aadhesion adhesion
regenerative brake (REG).first case, are
Results an emergency
shown in Figure 17a. is conducted
It can be seenon a that
highthe brake
surface.surface.
The rear brake torque is generated with a conventional ABS and with the regenerative the brake
distance The rear when
is lower brake thetorque is generated
vehicle is equipped with a conventional
with antilock
the conventional ABSsystem
system.(ABS)
This and
is duewith
to the
system (REG). Results
regenerative brakeare shown
system in
(REG).Figure 17a.
Results areIt can
shown be seen
in that
Figure the
17a. brake
It can
fact that the regeneration rate is limited to 50%. With this maximum rate, the regenerative brake bedistance
seen that is lower
the when the
brake
distance
vehicle is equipped
torque is lower
cannot with when
brake the the vehicle
conventional
as efficiently is equipped with the
ABS system.system
as the conventional conventional
This does. ABS
is dueSimilarly, system.
to the fact This
that the
a second is due to
test regeneration
is has beenthe rate
facttothat
is limited 50%.
simulated the
on regeneration
With
a lowthis
adhesion ratesurface
maximum is limited
rate, to
(seethe 50%. WithInthis
regenerative
Figure 17b). maximum
thisbrake torque
condition, rate, the regenerative
thecannot brakebrake
regenerative brake
as efficiently
can as
torque cannot
produce enoughbrake as efficiently
brake torque as cause
to the conventional
the rear systemtodoes.
wheel reachSimilarly,
the a second
optimal test
slip. is hasthese
Under been
the conventional system does. Similarly, a second test is has been simulated on a low adhesion surface
simulated on a the
circumstances, low regenerative
adhesion surface brake (see Figure 17b).
decelerates theInvehicle
this condition,
faster than the regenerative
the conventionalbrakeABS can
(see Figure 17b).enough
produce In thisbrake
condition,
torque the
to regenerative
cause the rear brake
wheel can
to produce
reach the enough
optimal brake
slip. Under torque
theseto cause
systems thanks to a better control of the rear wheel slip.
the rearcircumstances,
wheel to reach the optimal slip. Under these circumstances, the regenerative
the regenerative brake decelerates the vehicle faster than the conventional ABS brake decelerates
the vehicle faster
systems thantothe
thanks conventional
a better control of the ABSrearsystems
wheel slip.thanks to a better control of the rear wheel slip.

(a) (b)

Figure 17. Slip(a)

and vehicle speed. High adhesion surface (a); Low adhesion
(b) surface (b).

Figure 17. Slip and vehicle speed. High adhesion surface (a); Low adhesion surface (b).
Finally,17.
Table
Figure Slip5and
summarizes the results
vehicle speed. High obtained
adhesionin the previous
surface (a); Lowtests. The last
adhesion column
surface (b).also
includes the results of tests in which the maximum regeneration rate has been increased to 80%. The
Finally, Table
regenerative control5 performs
summarizes theon
better results obtained
the low in the
adhesion previous
surface due totests. The last
a better and column also
more stable
includesofthe
control theresults of tests
rear wheel inOn
slip. which the maximum
the contrary, regeneration
the regenerative rate has
control been increased
in which to 80%. rate
the regeneration The
regenerative
is control
limited to 50% performs
cannot betterthe
substitute onconventional
the low adhesion
brake surface
on highdue to a better
adhesion roadsand more
since the stable
brake
control ofand
distance the deceleration
rear wheel slip.
areOn the contrary,
worse. However, theifregenerative
the maximum control in which rate
regeneration the regeneration
is increased, rate
the
regenerative brake can generate a higher rear brake torque, which makes it perform since
is limited to 50% cannot substitute the conventional brake on high adhesion roads betterthe brake
than the
distance and brake.
conventional deceleration are worse. However, if the maximum regeneration rate is increased, the
regenerative brake can generate a higher rear brake torque, which makes it perform better than the
Energies 2017, 10, 1648 14 of 16

Finally, Table 5 summarizes the results obtained in the previous tests. The last column also
includes the results of tests in which the maximum regeneration rate has been increased to 80%.
The regenerative control performs better on the low adhesion surface due to a better and more stable
control of the rear wheel slip. On the contrary, the regenerative control in which the regeneration
rate is limited to 50% cannot substitute the conventional brake on high adhesion roads since the
brake distance and deceleration are worse. However, if the maximum regeneration rate is increased,
the regenerative brake can generate a higher rear brake torque, which makes it perform better than the
conventional brake.
These simulations show that the proposed system can be installed in a real vehicle to cooperate
with or in substitution of the conventional friction brake. The system could control the slip no
matter which surface the vehicle is moving on if the maximum regeneration rate were increased.
However, batteries with a higher admissible charge current or ultracapacitors are necessary if the
maximum regeneration rate is increased [14,16].

Table 5. Comparison ABS vs. Regenerative braking.

Regenerative Regenerative
Surface Control ABS
Reg. Max. = 50% Reg. Max. = 80%
Time (s) 1.39 1.42 1.40
High adhesion Distance (m) 19.54 20.24 20.85
Mean deceleration (m/s2 ) 11.95 11.71 11.86
Time (s) 3.90 3.76 3.80
Low adhesion Distance (m) 53.51 52.19 52.62
Mean deceleration (m/s2 ) 4.27 4.44 4.38

6. Conclusions
Algorithms for the estimation of parameters and the control necessary to optimize the regeneration
in two-wheeled vehicles have been developed in this paper. The parameter estimation is necessary to
be able to determine the speed of the vehicle and the road adhesion conditions. The control algorithm
allows one, once the previous data is known, to optimize the rate of regeneration in the rear wheel of a
motorcycle. Generally, regeneration in two-wheeled vehicles is usually limited to a minimum value,
around 5–10% of motor power. This pursues two objectives: on one hand, to limit the charge current in
the battery to protect it. Secondly, the possibility of locking the rear wheel in any adhesion condition is
reduced. If this occurs, ABS control is required.
With the proposal developed in this work, the rate of regeneration on any surface is optimized,
blocking the wheel is prevented by always working in the vicinity of the optimal slip. Protection of the
battery is achieved by limiting the intensity in those cases in which the regeneration could exceed a
value of risk for its integrity. Simulations show that the system optimizes the regeneration rate on every
surface and that the energy recovered surpasses the one obtained with constant regeneration rates.
Finally, despite the fact that this paper is mainly focused on recovering the maximum energy with
the regenerative brake, we have shown that the regenerative brake could substitute the rear friction
brake provided higher regeneration rates can be achieved. High regeneration rates require batteries
with a higher admissible charge current or ultracapacitors on vehicles. However, there is still a great
deal of research to be carried out. Experimental tests have to be carried out to verify the aptitude of this
proposal, especially in high adherence conditions. The work reported here is exploratory. Future works
will include tests on different surfaces, with different maneuvers and with higher regeneration rates.

Acknowledgments: This work is partly supported by the Ministry of Economy, Industry and Competitiveness
under grant TRA2015-67920-R and partly by the University of Málaga.
Author Contributions: Juan Jesús Castillo Aguilar performed the simulations and conceived and designed
the experiments. Javier Pérez Fernández and Juan María Velasco García performed most experiments and
measurements and contributed to the programming of the estimations algorithm. Juan Antonio Cabrera Carrillo
value of risk for its integrity. Simulations show that the system optimizes the regeneration rate on every
surface and that the energy recovered surpasses the one obtained with constant regeneration rates.
Finally, despite the fact that this paper is mainly focused on recovering the maximum energy
with the regenerative brake, we have shown that the regenerative brake could substitute the rear
friction brake provided higher regeneration rates can be achieved. High regeneration rates require
Energies 2017, 10, 1648 15 of 16
batteries with a higher admissible charge current or ultracapacitors on vehicles. However, there is
still a great deal of research to be carried out. Experimental tests have to be carried out to verify the
wrote the
aptitude of introduction, helped
this proposal, to analyse
especially inthe dataadherence
high and provided some useful
conditions. Thesuggestions in the writing
work reported here isand
revision of the text.
exploratory. Future works will include tests on different surfaces, with different maneuvers and with
Conflicts
higher The authors declare no conflict of interest.
of Interest:rates.
regeneration

Appendix
Appendix A.A. Experimental
Experimental Vehicle
Vehicle
TheThe motorcycle
motorcycle used
used in in
thethe tests
tests is shown
is shown below.
below. The
The table
table includes
includes thethe main
main characteristics
characteristics of of
thethe vehicle.
vehicle.

Component
Component Parameter
Parameter Description
Description
Motorcycle weight 135 Kg
Chassis Type Steel Tubular
Height of gravity centre 621 mm
Distance between axis 1370 mm
Vehicle
Wheel radius 300 mm
Distance from the COG to the front axle 670 mm
Front tire 95/70 R 17
Rear tire 115/70 R 17
Brand Heinzmann PMS 150
Type Axial Flux Permanent Magnet
Maximum speed 6000 rpm
Electric motor
Maximum torque 80 Nm
Torque constant (Km ) 0.145 Nm/A
Maximum power 34.1 KW (46.36 CV)
Battery Type LiPo
Cell layout 26S5P
Total capacity 4.8 KWh
Battery
Rated Voltage 96 V
Maximum discharge current 1250 A
Maximum load current 300 A

References
1. Ferrero, E.; Alessandrini, S.; Balanzino, A. Impact of the electric vehicles on the air pollution from a highway.
Appl. Energy 2016, 169, 450–459. [CrossRef]
2. Ye, M.; Bai, Z.F.; Cao, B.G. Energy recovery for battery electric vehicles. Proc. Inst. Mech. Eng. Part D J.
Automob. Eng. 2008, 222, 1827–1839. [CrossRef]
3. Tehrani, M.; Hairi-Yazdi, R.; Haghpanah-Jahromi, B.; Esfahanian, V.; Amiri, M.; Jafari, R. Design of an
anti-lock regenerative braking system for a series hybrid electric vehicle. Int. J. Automot. Eng. 2011, 1, 16–20.
4. Gao, Y.; Chen, L.; Ehsani, M. Investigation of the effectiveness of regenerative braking for EV and HEV.
SAE Tech. Pap. 1999. [CrossRef]
5. Cabrera, J.A.; Castillo, J.J.; Carabias, E.; Ortiz, A. Evolutionary Optimization of a Motorcycle Traction Control
System Based on Fuzzy Logic. IEEE Trans. Fuzzy Syst. 2015, 23, 1594–1607. [CrossRef]
Energies 2017, 10, 1648 16 of 16

6. Lin, H.; Song, C. Design of a fuzzy logic controller for ABS of Electric Vehicle based on AMESim and
Simulink. In Proceedings of the 2011 International Conference on Transportation, Mechanical, and Electrical
Engineering (TMEE), Changchun, China, 16–18 December 2011; pp. 779–782.
7. Choi, S.; Cho, D.W. Design of Nonlinear Sliding Mode Controller with Pulse Width Modulation for Vehicular
Slip Ratio Control. Veh. Syst. Dyn. 2011, 36, 57–72. [CrossRef]
8. Harifi, A.; Aghagolzadeh, A.; Alizadeh, G.; Sadeghi, M. Designing a sliding mode controller for slip control
of antilock brake systems. Transp. Res. Part C 2008, 16, 731–741. [CrossRef]
9. Habibi, M.; Yazdizadeh, A. A novel Fuzzy-Sliding Mode Controller for Antilock Braking System.
In Proceedings of the 2010 2nd International Conference on Advanced Computer Control (ICACC), Shenyang,
China, 27–29 March 2010; Volume 4, pp. 110–114.
10. Topalov, A.; Oniz, Y.; Kayacan, E.; Kaynak, O. Neuro-fuzzy control of antilock braking system using sliding
mode incremental learning algorithm. Neurocomputing 2011, 74, 1883–1893. [CrossRef]
11. Cirovic, V.; Aleksendric, D.; Mladenovic, D. Braking torque control using recurrent neural networks. Proc. Inst.
Mech. Eng. Part D J. Automob. Eng. 2012, 226, 754–766. [CrossRef]
12. Gerard, M.; Pasillas-Lépine, W.; de Vries, E.; Verhaegen, M.; Cho, D.W. Improvements to a five-phase ABS
algorithm for experimental validation. Veh. Syst. Dyn. 2012, 50, 1585–1611. [CrossRef]
13. Tanelli, M.; Astolfi, A.; Savaresi, S. Robust nonlinear output feedback control for brake by wire control
systems. Automatica 2008, 44, 1078–1087. [CrossRef]
14. Sun, L. Systematic experimental evaluation of a novel regenerative braking system on DC motor drive
using super-capacitors. In Proceedings of the 2014 IEEE Conference and Expo Transportation Electrification
Asia-Pacific (ITEC Asia-Pacific), Beijing, China, 31 August–3 September 2014; pp. 1–6.
15. Guarisco, M.; Bouriot, B.; Ravey, A.; Bouquain, D. Powertrain energy management for Hybrid Electric
Scooter. In Proceedings of the 2014 IEEE Transportation Electrification Conference and Expo (ITEC),
Dearborn, NI, USA, 15–18 June 2014; pp. 1–4.
16. Andorinha, N.; Sousa, S.; Antunes, V. A new control method of an electric scooter with a hybrid
ultracapacitor/battery energy storage topology. In Proceedings of the 2nd Experiment@ International
Conference (exp.at’13), Coimbra, Portugal, 18–20 September 2013; pp. 11–16.
17. Lai, Y.S.; Lin, Y.K. Design and implementation of digital-controlled bi-directional converter for scooter
applications. In Proceedings of the 2013 IEEE 10th International Conference on Power Electronics and Drive
Systems (PEDS), Kitakyushu, Japan, 22–25 April 2013; pp. 271–276.
18. Robinson, J.; Singh, T. eABS: Regenerative Anti-Lock Braking for Electric Motorcycles. SAE Int. J. Passeng.
Cars Mech. Syst. 2013, 6, 1484–1492. [CrossRef]
19. Lin, W.C.; Lin, C.L.; Hsu, P.M.; Wu, M.T. Realization of anti-lock braking strategy for electric scooters.
IEEE Trans. Ind. Electron. 2014, 61, 2826–2833. [CrossRef]
20. Castillo, J.J.; Cabrera, J.A.; Guerra, A.J.; Simón, A. A Novel Electrohydraulic Brake System with Tire–Road
Friction Estimation and Continuous Brake Pressure Control. IEEE Trans. Ind. Electron. 2016, 63, 1863–1875.
[CrossRef]

© 2017 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access
article distributed under the terms and conditions of the Creative Commons Attribution
(CC BY) license (http://creativecommons.org/licenses/by/4.0/).