Control of Motorcycle

Steering Instabilities
SIMOS EVANGELOU, DAVID J.N. LIMEBEER, ROBIN S. SHARP, and MALCOLM C. SMITH

PASSIVE MECHANICAL COMPENSATORS

INCORPORATING INERTERS

dvances in modeling bicycle and motor-

A cycle dynamics are providing

improved understanding of the prin-
cipal modes of motion under
straight-running and steady-
state cornering conditions; see, for exam-
ple, [1] and the references therein. These
studies show that under certain operat-
ing conditions, some of the machine’s
modes can be lightly damped or even
unstable. The most important of these
modes are wobble and weave. Wobble
is a steering oscillation that is reminis-
cent of the caster shimmy that occurs in
the front wheels of a supermarket cart,
while weave is a fishtailing-type motion
involving roll and yaw. The frequency of
the wobble mode is of the order 8 Hz,
while the weave frequency is about 3 Hz,
where the exact figures depend on the speed
and machine parameter values.
Modern high-performance motorcycles
often employ a steering damper producing a
moment that opposes the angular velocity of the
steering assembly relative to the main frame. For
machines with a stiff front frame, a steering damper is
required to stabilize the wobble mode at high speeds,
while older, more flexible machines may require a steering
damper at intermediate speeds. Despite its benefits for wobble-
mode performance, a steering damper has a destabilizing effect on the © HARLEY-DAVIDSON
®
2001 DIGITAL PRESS KIT
weave mode. As a result of this conflict, only a narrow range of damper coeffi-
cient values may be usable [2], [3].
Establishing damper settings that provide an optimal compromise between wobble- and
weave-mode damping is an issue of considerable interest and is a focus of the present article.

78 IEEE CONTROL SYSTEMS MAGAZINE » OCTOBER 2006 1066-033X/06/$20.00©2006IEEE

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY DELHI. Downloaded on January 05,2022 at 07:14:42 UTC from IEEE Xplore. Restrictions apply.
 In particular, we replace the conventional steering damper The approach described here is underpinned by classi-
with a network of interconnected mechanical components cal passivity ideas from circuit theory [4] as well as analo-
that retains the virtue of the damper, while improving the gies between electrical and mechanical networks [5]; see
weave-mode performance. The improved performance is “Passive Circuit Synthesis.” In the standard electrical-
due to the fact that the network introduces phase compen- mechanical current-force analogy, an inductor corresponds
sation between the relative angular velocity of the steering to a spring, while a resistor represents a damper. To com-
system and the resulting steering torque. plete this analogy, a mechanical component is needed to

Passive Circuit Synthesis

O
ne of the principal motivations for the introduction of the point impedance or admittance of a network comprised of only
inerter in [5] is the synthesis of passive mechanical networks. resistors, capacitors, and inductors [28]. The result can be trans-
The fact that the mass element, together with the spring and lated over to the mechanical setting as follows [5].
damper, is insufficient to realize the totality of passive mechani-
cal impedances can be seen using the force-current analogy THEOREM
between mechanical and electrical circuits. In this analogy, force Let Y (s) be a positive-real rational function. Then there exists
and current are the through variables and velocity and voltage a two-terminal mechanical network that consists of a finite
are the across variables. Moreover, the terminals of mechanical interconnection of springs, dampers, and inerters and whose
and electrical elements are in one-to-one correspondence. For admittance equals Y (s).
the mechanical elements, the spring and damper have two inde- This result allows the optimization of system properties with-
pendently movable terminals, whereas the terminals of the mass out fixing the network structure in advance. For example, it can
are its center of mass and a fixed point in an inertial frame be shown [3] that the biquadratic real rational function
(mechanical ground). The mass is therefore analogous to a
a2 s2 + a1 s + a0
grounded capacitor. In contrast, the inerter is a two-terminal Y (s) = ,
d2 s2 + d1 s + d0
device, analagous to an ungrounded capacitor, with both termi-
nals freely and independently movable. where a2 , a1 , a0 , d2 , d1 , d0 are nonnegative and at least one of d0 ,
Figure A shows a table of element correspondences in the d1 , d2 is positive, is positive real if and only if
force-current analogy with the inerter replacing the mass element.
 2

The admittance Y (s) is the ratio of through to across quantities, a1 d1 ≥ a0 d2 − d0 a2 .
where s is the standard Laplace transform variable. For mechani-
cal networks in rotational form, the
through and across variables are
torque and angular velocity, respec-
tively. It should be mentioned that the Mechanical Electrical
reciprocal definition of admittance for
F F i i
mechanical elements is commonly Y(s) = ks Y(s) = 1
v2 v1 Ls
found in the literature, for example in v2 v1
[7], [8], and [25]. For further back- dF = k(v − v ) Spring di = 1 (v − v )
2 1 Inductor
ground on network analogies, see [5], dt dt L 2 1

[24], and [26]. F F i i

The theory of passive circuits has Y(s) = bs Y(s) = Cs
v2 v1
been widely studied in the electrical v2 v1
d(v2 − v1) d(v2 − v1)
engineering literature [4], [27]. A linear F=b Inerter i=C Capacitor
dt dt
time-invariant two-terminal network
that possesses a real rational imped- F F i i
Y(s) = c Y(s) = 1
ance or admittance function is passive v2 v1 R
v2 v1
if and only if the impedance or admit-
F = c (v2 − v1) Damper i = 1 (v2 − v1) Resistor
tance is positive real; see [4] and [27]. R
A celebrated result in electrical circuit
synthesis, proved by Bott and Duffin, FIGURE A Electrical and mechanical circuit symbols and correspondences. In the force-
says that a rational, positive real func- current analogy, forces substitute for currents, and velocities substitute for voltages. The
tion can be realized as the driving- admittance Y(s) maps velocity and voltage into force and current, respectively.

OCTOBER 2006 « IEEE CONTROL SYSTEMS MAGAZINE 79

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY DELHI. Downloaded on January 05,2022 at 07:14:42 UTC from IEEE Xplore. Restrictions apply.
 represent a capacitor. Although a mass is analogous to a where ω1 and ω2 are the angular velocities of the bodies,
capacitor with one terminal grounded, a new device is and b is the inertance in kg-m2 . As in [5], this use of the
required to represent a capacitor in general. The suitable term inertance is consistent with its usage in acoustics [6]
masslike element is the inerter [5], which can be either a but is the reciprocal of the traditional usage in mechanical
translational or rotational device. vibrations [7], [8], where it represents a transfer function
In its rotational form, the inerter generates a resisting from force to acceleration and is synonymous with the
moment M between two hinged bodies that is proportional term accelerance in that field.
to the relative angular acceleration between them. Mathe- In control systems terms the inerter is used to pro-
matically, duce phase lead. The effect of the inerter, which con-
tains a flywheel of modest mass and dimensions, is
M = b(ω̇1 − ω̇2 ), amplified by high-ratio gearing. This gearing can be
realized by an epicyclic arrangement
[9] or by a harmonic drive [10]. It is
possible to generate high inertance
1.4
values with relatively low-mass com-
ponents. More information about the
1.2
p10 mechanical synthesis of inerters can
be found in “Inerter.”
Vertical Displacement (m)

0.8 p3 Twist Axis MOTORCYCLE MODEL

p9 The motorcycle model used in this arti-
p2
0.6
cle is based on the Suzuki GSX-R1000, a
p13 stiff-framed high-performance produc-
p8
0.4 p4 tion vehicle. This motorcycle is fitted
p14 with a standard telescopic front fork
p11 p5
p7 p22 p6
0.2 p19 suspension system and a swinging-
p21 Steer Axis arm-based rear suspension that incor-
p20 p1
0 porates a single spring-damper unit
linked mechanically to the swing-arm,
−0.4 −0.2 0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 a monoshock suspension; see Figure 1.
Longitudinal Displacement (m)
In Figure 1, each constituent mass is
represented by a red disk of diameter
FIGURE 1 Schematic of the GSX-R1000 motorcycle model. The scaled motorcycle and
rider model shows the machine layout with each mass depicted as a proportionally scaled
proportional to the corresponding
red disk. The motorcycle is shown in its nominal configuration with key points used in the mass. The model has a tree structure,
multibody description labeled as p. except for one kinematically closed
loop, which is associated with the
monoshock description. The freedom
associated with each body is shown in
Inertial Frame; n Figure 2. The symbolic multibody soft-
ware system AUTOSIM [11] is used to
assist the model-building process.
Main; Translate (x, y, z); Rotate (z, y, x)
A detailed description of the model
and its parameter set can be found in
Upper_Body; ff_Twist; Rotate x´ [12]; only a summary is given here. The
Swing_Arm; Rotate y
Rotate x main frame, which is allowed unre-
stricted motion, is pin connected to the
ff_Steer; Rotate z´
Link 1; Rotate y steering system, the rider’s upper
Rear_Wheel; Rotate y
No-Movement body, and the rear swing-arm. The tor-
ff_Sus; Translate z´ sional compliance of the frame near the
Link 2; Rotate y
steering head is modeled by including
Front_Wheel; Rotate y´ a twist degree of freedom; this twist
occurs about an axis that is perpendic-
FIGURE 2 Model parent-child kinematic dependencies [15]. The machine and rider multi-
ular to the steer axis. The lower part of
body hierarchy show the relevant bodies and their degrees of freedom. Reference axis the front forks and the front wheel are
directions accord with SAE vehicle dynamics standards [16]. free to translate along the fork line.

80 IEEE CONTROL SYSTEMS MAGAZINE » OCTOBER 2006

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY DELHI. Downloaded on January 05,2022 at 07:14:42 UTC from IEEE Xplore. Restrictions apply.
 Inerter

T
he inerter is a two-terminal
device with the property that Pinions
an equal and opposite force
applied at the terminals is propor-
tional to the relative acceleration
between them. Mathematically, Rack
the inerter obeys the force-veloci-
ty law F = b(v̇1 − v̇2 ), where the
constant of proportionality b is
Terminal 2 Flywheel Terminal 1
called the inertance and has the
units of kilograms. In its rotational
FIGURE B Schematic of the inerter principle. A rack and pinion gearing arrangement drives a
form, the inerter obeys a
rotating flywheel; the device’s inertance is a function of the gear ratio and the flywheel inertia.
moment-angular velocity law
M = b(ω̇1 − ω̇2 ); in this case the
inertance is measured in kg-m2 . To be practically useful, the of inertia [5]. In general, if the device gives rise to a flywheel rota-
device must have a small mass. Also, the device must have rea- tion of α radians per meter of relative displacement between the t
sonable overall dimensions and must be able to function in all erminals, then b = Jα 2 , where J is the flywheel’s moment of in-
spatial orientations. ertia and the remaining inertial effects are neglected. For a rota-
One way in which a translational inerter can be constructed is tional inerter in which there is a gear ratio of n between rotations
illustrated in Figure B. The device comprises a rack-and-pinion of the terminals and a flywheel with moment of inertia J, it follows
mechanism, with the rack constrained to translate relative to the that b = Jn2 . Several prototype devices according to [9] have
housing. For such devices the value b of the inertance is easy to been built and tested in the Engineering Department of Cam-
compute in terms of the gear ratios and the flywheel’s moment bridge University; see for example figures C and D.

FIGURE C A prototype inerter. This translational inerter employs a FIGURE D Prototype motorcycle steering compensator. This rota-
ballscrew to convert the linear motion of the plunger into rotation- tional device employs an epicyclic gear box connected to a fly-
al motion of the flywheel (not shown). The device has a mass of wheel through a fluid coupling to realize a series inerter-damper.
approximately 1 kg and an inertance in the range 60–200 kg The mass is approximately 1.5 kg, while the inertance and damp-
depending on the size of the flywheel. The device was designed ing are approximately 0.25 kg-m2 and 10 N-m-s/rad, respectively.
by N.E. Houghton and manufactured in the Cambridge University The device was designed by N.E. Houghton and manufactured in
Engineering Department. the Cambridge University Engineering Department.

Each road wheel is axisymmetric and is allowed to spin. nominal static equilibrium states, and the wheel camber
Each tire is represented as having a finite width, and the angles are used in magic formula tire models [1], [12], [13]
ground contact points are allowed to migrate both circum- to compute the forces and aligning moments generated by
ferentially and laterally. This migration process is modeled the tires. Magic formula methods comprise empirical for-
by viewing the lowest point on each tire as the center of mulas and parameters that describe the forces and
the road-tire contact patch. The motion states of the tire moments developed, as measured in laboratories, as func-
contact centers, the compression of the tires from their tions of the operating conditions. Important operating

OCTOBER 2006 « IEEE CONTROL SYSTEMS MAGAZINE 81

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY DELHI. Downloaded on January 05,2022 at 07:14:42 UTC from IEEE Xplore. Restrictions apply.
 machine behavior, or evaluating the
machine’s trim states, is obtained.
Wobble is a steering oscillation that is reminiscent Once a trim condition is reached, dis-
of the caster shimmy that occurs in the front wheels turbances such as road profiling can be
introduced. When the model is config-
of a supermarket cart, while weave is a fishtailing- ured in its linear mode, the system is
symbolically linearized for small per-
type motion involving roll and yaw. turbations about a general trim condi-
tion. The linearized models generated
by AUTOSIM take a state-space form
and a MATLAB M file is produced. Each
variables include the normal load, longitudinal and lateral of the state-space matrices is parameterized to correspond
slip, and wheel camber (or inclination) angle. As a result of to the trim condition being studied. The symbolic MAT-
the lateral compliance of the tires’ carcasses, lateral force LAB file can be used to generate standard plots such as
and moment variations are not generated instantaneously. Nyquist and root locus diagrams.
The time constants associated with the force and moment When modeling the motorcycle in a conventional man-
production process or, more precisely, the relaxation ner, the steering damper is included as an integral part of
lengths, are functions of speed [14]. The suspension the machine. This normally low-profile component gener-
springs and dampers are treated as linear, although sus- ates a steering torque that is proportional to steering veloci-
pension and steering limit stops are included. ty. To allow a more general relationship between the
Two feedback controllers are used in the model. The steering torque and the steering velocity, the steering
first is a fixed-gain proportional-plus-integral speed con- damper is separated from the remainder of the system and
troller that generates the rear-wheel drive torque. The sec- represented as an external feedback loop as shown in Fig-
ond is a speed-adaptive proportional-integral-derivative ure 3 [2], [3], [17]. The nominal motorcycle-damper combi-
steering torque controller that responds to the lean angle nation will be referred to as the standard machine, while the
error. With the aid of these controllers, the machine can motorcycle without a steering damper will be referred to as
track prescribed speed and lean-angle trajectories and, in the basic machine. A conventional steering damper is repre-
particular, can be run to any feasible trim state to deter- sented as a pure gain in Figure 3. More general mechanical
mine equilibrium configurations. The controller gains are
set to zero for the uncontrolled motorcycle.
The AUTOSIM model file can be configured in a linear
or nonlinear format. In the nonlinear configuration, a 0.15
numerical simulation program for studying transient
0.1

0.05
Imaginary Axis

0 A C
d(s) δ(s)
P (s) −0.05

−0.1
Ts(s) sδ(s)
−0.15
−0.3 −0.25 −0.2 −0.15 −0.1 −0.05 0 0.05
K(s)
Real Axis

FIGURE 4 Nyquist diagram of the straight-running basic machine.

FIGURE 3 Linearized motorcycle model and steering compensator. The diagram corresponds to the single-loop system with steering
The linearized motorcycle model is denoted P(s), while the steering torque as input and with steering system angular velocity as output;
compensator is shown as K (s) in the feedback loop. The input d(s) the machine’s forward speed is 75 m/s. The frequency associated
represents vertical road-displacement forcing, while the output δ(s) with crossing point A, the wobble-mode resonant frequency, is 47.6
is the steering angle and Ts(s) is the steer torque. The road forcing rad/s. The frequency associated with the cusp B is 33.8 rad/s, and
signal, which is a scalar variable, is applied to both wheels. The that at crossing point C, the weave-mode resonant frequency, is
rear-wheel forcing signal is a delayed version of that at the front; the 28.4 rad/s. If the damper coefficient is selected so that the point
time delay is given by τ = w/v in which w is the wheelbase and v is −1/c is at A, the motorcycle oscillates at the wobble-mode frequen-
the forward speed. The steering compensator K (s) is constant in cy. If the damper coefficient is selected so that the point −1/c is at
the case of a conventional steering damper. C, the machine oscillates at the weave-mode frequency.

82 IEEE CONTROL SYSTEMS MAGAZINE » OCTOBER 2006

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY DELHI. Downloaded on January 05,2022 at 07:14:42 UTC from IEEE Xplore. Restrictions apply.
 networks are represented by K(s) in which s is the Laplace es. The damper has almost no influence on the frequency
variable. The transfer function K(s) from angular velocity (imaginary part) associated with either of these modes.
to torque is called the admittance function of a mechanical Replacing the damper with an inerter, for which
network following the convention of [24, pp. 45, 46, 326] K(s) = bs, leads to the root locus diagram of Figure 6.
but in contrast to other uses in mechanical engi-
neering [6, p. 238] and [8, p. 207].

60
CHARACTERISTICS OF THE
NEAR-STANDARD MACHINE Wobble
Figure 4 shows a Nyquist diagram of the lin- 50
earized model of the basic machine in a straight-
running equilibrium condition at 75 m/s. This 40
diagram can be used to study the effect of chang-

Imaginary Axis
ing the steering damper coefficient. The model
has two right-half-plane poles for this operating 30
condition because the wobble mode is unstable.
Since the steering damper is represented as a 20 Weave
pure gain in the feedback loop of Figure 3, it fol-
lows from the Nyquist criterion that stability of
10
the standard machine requires two counterclock-
wise encirclements of the point −1/c, where c is
the steering damper constant. If the steering 0
damping is set at a low value such that the point −18 −16 −14 −12 −10 −8 −6 −4 −2 0 2 4
−1/c is located at A, the machine is on the stabili- Real Axis
ty boundary and thus oscillates at 47.6 rad/s,
which is the wobble-mode frequency. If the FIGURE 5 Influence of the steering damper on the root loci of the straight-running
motorcycle. Speed, which is the varied parameter, is increased from 5 m/s (
) to
steering damping is now increased, two counter-
75 m/s (
). The × locus is associated with the nominal machine damping value
clockwise encirclements of the point −1/c result, of 6.944 N-m-s/rad, the ◦ locus is associated with a steering damping value of
and the motorcycle is stable. If the steering 3.94 N-m-s/rad, and the + locus with a steering damping value of 0.94 N-m-
damping is increased further so that the point s/rad.
−1/c coincides with C, the machine oscillates at
28.4 rad/s and the weave mode is on the stability
boundary. Further increases in the steering 60
damping render the machine unstable since the
Wobble
point −1/c is not encircled. The nominal steering
50
damper coefficient is 6.944 N-m-s/rad, thereby
locating the point −1/c at −0.144, which is
approximately midway between A and C. 40
Imaginary Axis

When the speed increases from 75 m/s to 85

m/s, the interval on the negative-real axis 30
associated with stabilizing damper parameter
values moves to the left [3]. This shift moves Weave
the weave-mode crossing point toward the 20
point −1/c associated with the nominal damp-
ing coefficient, and the wobble-mode crossing 10
point moves away from it. As a result, this
speed increase reduces the weave-mode damp-
0
ing factor and increases the wobble-mode −15 −10 −5 0 5
damping factor. Real Axis
Figure 5 shows straight-running root loci
where speed is the varied parameter. When the FIGURE 6 Influence of a steering inerter on the root loci of the straight-running
steering damper coefficient is reduced from its motorcycle. Speed, which is the varied parameter, is increased from 5 m/s (
)
to 75 m/s (
). The × locus is associated with the basic machine, the ◦ locus
nominal (standard machine) value, the wobble represents the basic machine fitted with a steering inertance of 0.1 kg-m2 , and
mode becomes unstable at high speed, while the + locus represents the basic machine fitted with a steering inertance of 0.2
the high-speed weave-mode damping increas- kg-m2 .

OCTOBER 2006 « IEEE CONTROL SYSTEMS MAGAZINE 83

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY DELHI. Downloaded on January 05,2022 at 07:14:42 UTC from IEEE Xplore. Restrictions apply.
 The dynamic characteristics of high-performance motorcycles can be improved
by replacing the conventional steering damper with a passive mechanical
steering compensator.

Although the inerter stabilizes the weave mode at high quency range (5–9 Hz) and inerter-like at the lower fre-
speeds, it has a detrimental effect on the wobble-mode quencies associated with the weave mode (2-3 Hz) might
damping. The reduction in the wobble-mode frequency is be beneficial. Over the still lower frequency range used by
caused by the effective increase in the steering system’s the rider (0–0.5 Hz) [19], [20], for balancing and path-fol-
moment of inertia. With regard to the damping of the lowing control, the network must be high pass to allow
weave and wobble modes, the inerter and damper intro- unhindered rider steering action. Spring-like properties are
duce opposite trends. undesirable as are high values of steady-state gain.
It can be shown that the series connection of an inerter
COMPENSATION NETWORK and damper has the admittance function scb/(sb + c),
where b and c are the inertance and damping coefficient
Desired Network Properties respectively. The frequency response of this admittance
The results of the previous section suggest that a mechani- function is similar to that of an inerter (respectively,
cal network that is damper-like over the wobble-mode fre- damper) at frequencies below (respectively, above) the
break frequency c/b. This frequency response function can
be beneficial for simultaneous control of wobble and
b weave. A related network is the resonant filter consisting
c k of the series connection of a damper, an inerter and a
spring (see Figure 7). This network has the admittance
s
Y(s) = , (1)
s2 + sk/c + k/b
FIGURE 7 Series resonant filter. This series connection of a
damper, inerter, and spring, constituting a resonant filter, provides where b, c, and k denote the inertance, damping coefficient,
frequency-localized damping influence.
and spring stiffness, respectively.

10
ζ = 0.4
0
Magnitude (dB)

1
−10 2 1.5

−20 1
−30
Imaginary Axis

0.5
−40
(a) 0
90
−0.5
45
Phase (°)

−1
0
−1.5
−45

−90 −3.5 −3 −2.5 −2 −1.5 −1 −0.5 0 0.5 1

10−1 100 101 Real Axis
Frequency (rad/s)
(b) FIGURE 9 Nyquist diagram of the straight-running motorcycle with a
forward speed of 75 m/s. The solid line represents the standard
FIGURE 8 Frequency-response characteristics of the series resonant machine, while the dashed line corresponds to the compensated
filter network with the resonant frequency normalized to ωn = 1. system using the series resonant filter shown in Figure 7 with
Three values of damping ratio ζ are illustrated. design values ωn = 50 rad/s, ζ = 0.4, and k = 500 N-m/rad.

84 IEEE CONTROL SYSTEMS MAGAZINE » OCTOBER 2006

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY DELHI. Downloaded on January 05,2022 at 07:14:42 UTC from IEEE Xplore. Restrictions apply.
 Frequency Response Design wobble and the weave modes under all operating condi-
It is instructive to develop initial design guidelines for the tions are achieved. The improvement in the high-roll
network parameters using classical frequency-response ideas. angle (45◦ ) case is worthy of particular note [3].
The network admittance (1) can be rewritten in the form
OPTIMIZATION
s
Y(s) = k , We now optimize the parameters of the series resonant
s2 + 2ζ ωn s + ωn2
filter. The chosen performance criterion reflects the role
in which played by road-displacement disturbances in stability-
√ related road traffic accidents [21] in the form of an H∞

bk
ωn = k/b , ζ = . (2) response measure, together with a penalty on the close-
2c
ness of approach of the open-loop Nyquist locus to the
It can be seen from the frequency-response characteristics point −1.
of the network shown in Figure 8 that this network acts The objective function considered is
like an inerter at low frequencies and introduces damping
in the vicinity of ωn , which must be tuned to the wobble- Jf = max{max{J1 , γ J2 }}, (3)

mode frequency. The damping ratio ζ is a design parame-
ter that determines the sharpness of the magnitude peak where
and the rate of change of phase with frequen-
cy. Smaller values of ζ give a more rapid
transition from inerterlike to damperlike behav-
ior. Larger values of ζ provide a wider range 70
of frequencies where damperlike behavior per- Wobble
sists. The peak value of the magnitude char- 60
acteristic is achieved at ωn and takes the
value of c, the damper constant. It can be seen 50
in (2) that ωn and ζ fully specify the phase
Imaginary Axis

characteristics of the network. 40

For an initial trial design the parameters
ωn = 50 rad/s and ζ = 0.4 are selected. Prior 30
to fixing a value of the spring constant, it is Weave
observed that k = 320 N-m/rad places the
20
point −1 in the middle of the stable k-value
range in the 75 m/s straight-running condi-
10
tion. However, since this value of k does not
produce adequate wobble-mode damping
0
performance at high lean angles, the spring −18 −16 −14 −12 −10 −8 −6 −4 −2 0 2
stiffness is increased to k = 500. This change Real Axis
improves the wobble-mode damping global-
ly but at the expense of the weave mode. FIGURE 10 Root loci for the compensated motorcycle. Four values of roll angle are
◦ ◦ ◦
Back substitution gives inerter and damper illustrated: straight running (×), 15 (◦), 30 (+), and 45 (♦). The speed is varied
from 7 m/s (
) to 75 m/s (). The machine is fitted with the series resonant network
parameter values of b = 0.2 kg-m 2 and
with the parameter values b = 0.2, c = 10, and k = 500.
c = 10 N-m-s/rad, respectively. The influ-
ence of this particular choice of parameters
is illustrated in Figure 9. It may be observed that the net-
work moves the negative-axis crossing point associated TABLE 1 Design parameters, objective functions, and
with weave-mode instability toward the origin and the worst-case configurations obtained by optimizing the
frequency-domain index (3) for the conventional steering
crossing point linked to wobble to the left of the diagram. damper and the series resonant filter.
The network opens up the interval over which two coun-
terclockwise encirclements can be achieved. The root
Maximum
locus plot that results from this mechanical network with Compensator Parameters Jf deg m/s rad/s
parameter values b = 0.2 kg-m2 , c = 10 N-m-s/rad, and Conventional c = 8.0695 109.8412 45 9 52.97
k = 500 N-m/rad is shown in Figure 10. Although the damper
design is based on a single high-speed straight-running Series resonant k = 594.08 40.576 45 7 49.95
filter c = 13.716
linearized model, compared with the standard machine,
b = 0.24252
substantial improvements in the damping of both the

OCTOBER 2006 « IEEE CONTROL SYSTEMS MAGAZINE 85

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY DELHI. Downloaded on January 05,2022 at 07:14:42 UTC from IEEE Xplore. Restrictions apply.


The set  of linear motorcycle models used in (3) con-

P11 ( jωi)

J1 = max


(4) tains linearized models corresponding to trim roll angles
ωi 1 − K( jωi)P22 ( jωi)

of 0, 3, 6, . . . , 45◦ and trim speeds of 7, 9, 11, . . . , 75 m/s.
and The subindex J1 in (4) is reminiscent of the L∞ -norm of the


transfer function between road-displacement forcing and

1

J2 = max


. (5) steer angle (see the “Motorcycle Model” section and Figure
ωi 1 − K( jωi)P22 ( jωi)

3), while J2 in (5) is the reciprocal of the distance of closest
approach between the Nyquist locus and the
point −1. J2 effectively penalizes high values
of the classical sensitivity function [22]. The
subindex J2 is weighted in Jf by the constant γ ,
70 set by trial to a value 16. When evaluating Jf , a
Wobble 100-point angular frequency sequence in geo-
60 metric progression, from ω = 101.3 rad/s to
ω = 101.85 rad/s is used; this range includes
50 all of the peaks in Jf . The MATLAB sequen-
tial quadratic programming algorithm fmin-
Imaginary Axis

40 con [23] is used for optimization. The

algorithm is initialized using the parameters
30 obtained from the results described in the
Weave “Frequency Response Design” section.
20
RESULTS
10 The results of optimizing with the frequency
response index (3) are given in Table 1 for
0
the conventional damper and the series reso-
−18 −16 −14 −12 −10 −8 −6 −4 −2 0 2 nant filter. In addition to the optimal para-
Real Axis meter values, the table provides the
minimum values achieved for Jf , the trim
FIGURE 11 Root loci for the motorcycle compensated by the optimized series res- condition at which the minimum is achieved,
onant filter with parameters as in Table 1. Four values of roll angle are illustrated, and the worst-case frequencies associated
namely, straight running (×), 15◦ (◦), 30◦ (+), and 45◦ (♦). The speed is varied
with the limiting value of the index. The
between 7 m/s (
) and 75 m/s ().
wobble mode dictates the lowest achievable
value of Jf . For the filter, the closest approach
between the Nyquist plot and the point −1
exceeds 16/40.576 = 0.394, where 16 is the
value of γ used in (3).
Figure 11 shows the root locus of the
30 motorcycle’s key modes for a wide range of
25 speeds and roll angles using the optimized
20 series resonant filter network. The network
achieves improved damping ratios for each
dB

15
10 of the machine’s lightly damped modes.
Figure 12 shows the road-forcing
5
response J1 in (4) with the optimized series
0
75 resonant filter network installed; the trim
60 state ranges over the motorcycle’s operating
40 45 regime. As expected, under straight-running
40 30
20 20 conditions, the road-forcing response is
Speed (m/s) 10 Roll Angle (°)
7 3 zero. In common with each of the other net-
works, the highest gain values, which occur
at low speeds and high roll angles, corre-
FIGURE 12 Road-forcing gain as a function of operating condition. The speed is
varied between 7 and 75 m/s, and the roll angle between 3◦ and 45◦ . The motor- spond to the excitation of the wobble mode.
cycle is fitted with the frequency-response-optimized series resonant compensa- High values of road-forcing-to-steer-angle
tion network. gain also occur under high-speed, low-roll-

86 IEEE CONTROL SYSTEMS MAGAZINE » OCTOBER 2006

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY DELHI. Downloaded on January 05,2022 at 07:14:42 UTC from IEEE Xplore. Restrictions apply.
 45 25

40
20
35

Magnitude (dB)
15
Magnitude (dB)

30
25
10
20
15 5
10
0
5
0 −5
101 102 101 102
Frequency (rad/s) Frequency (rad/s)

FIGURE 13 Bode magnitude plot of steer angle response to road- FIGURE 14 Bode magnitude plot of steer angle response to road-
displacement forcing (0 dB = 1 rad/m). The machine is operating displacement forcing (0 dB = 1 rad/m). The machine is operating
at a forward speed of 15 m/s and a roll angle of 45◦ . The solid line at a forward speed of 75 m/s and a roll angle of 15◦ . The solid line
represents the standard machine, while the dashed line repre- represents the standard machine, while the dashed line repre-
sents the machine with the optimized series resonant filter net- sents the machine with the optimized series resonant filter net-
work compensator. work compensator.

angle conditions (see 75 m/s and 15◦ roll angle), which mization. The results show substantial global perfor-
correspond to excitation of the weave mode. mance improvements as compared with conventional
The road-forcing characteristics of the machine are steering dampers.
also illustrated by the frequency-response plots shown The broad issue of the practical implementation of pas-
in Figures 13 and 14. As is evident from the first of sive mechanical compensators is the subject of ongoing
these figures, the series resonant filter network decreas- research, including the fabrication of integrated mechani-
es the peak wobble-mode road-forcing gain by approxi- cal networks as illustrated in Figure 7. Additional issues
mately 15 dB. For the trim condition considered in such as the selection of optimal gear ratios for the inerters
Figure 13, the wobble mode is particularly vulnerable to and the correct dimensioning of the components so that
road-displacement forcing. Figure 14 corresponds to a they have sufficient working life are also important consid-
high-speed trim condition in which the weave mode is erations. From a packaging perspective, a steering com-
correspondingly challenged. In this case, the network pensator needs to be acceptably small and light. Ideally,
reduces the weave-mode peak by approximately 3 dB as the compensator will fit in the space normally occupied by
compared to the standard machine. a conventional steering damper.

CONCLUSIONS AUTHOR INFORMATION

This article shows that the dynamic characteristics of high- Simos Evangelou received the B.A./M.Eng. degree in elec-
performance motorcycles can be improved by replacing trical and information sciences from the University of
the conventional steering damper with a passive mechani- Cambridge, United Kingdom, in 1999 and the Ph.D. degree
cal steering compensator. The steering compensators are in control engineering from Imperial College London, in
mechanical networks comprising springs, dampers, and 2004. In 2006 he was appointed a lecturer at Imperial Col-
inerters. The compensators show the potential to signifi- lege London, for the departments of mechanical and elec-
cantly improve the damping of both wobble and weave trical and electronic engineering. Previously he worked as
modes simultaneously. These networks exploit the inert- a research associate from 2004 to 2005. His research inter-
er’s ability to provide phase advance. The compensator’s ests include the modeling, stability analysis and control of
role can be interpreted as that of a feedback element in a single-track vehicles, the application of control theory to
control systems structure, despite its consisting of passive mechanical systems, dynamic system modeling, and con-
mechanical elements. trol systems theory.
The design methodology adopted uses Nyquist fre- David J.N. Limebeer ([email protected]) received
quency response ideas, root-locus analysis, and loop- the B.Sc. degree in electrical engineering from the Univer-
shaping design to obtain a preliminary choice of sity of Witwatersrand, Johannesburg, in 1974, the M.Sc.
parameters, which are then refined by numerical opti- and Ph.D. degrees in electrical engineering from the University

OCTOBER 2006 « IEEE CONTROL SYSTEMS MAGAZINE 87

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY DELHI. Downloaded on January 05,2022 at 07:14:42 UTC from IEEE Xplore. Restrictions apply.
 of Natal, Durban, South Africa, in 1977 and 1980, respec- control system design, frequency response methods, H∞
tively, and the D.Sc. degree from the University of Lon- optimization, nonlinear systems, active suspension, and
don in 1992. He has been with Imperial College London, mechanical systems. He was a corecipient of the 1992 and
since 1984, where he is currently the head of the Depart- 1999 George Axelby Outstanding Paper Awards, both
ment of Electrical and Electronic Engineering. He has pub- times for joint work with Dr Tryphon T. Georgiou. He is a
lished over 100 papers and a textbook on robust control Fellow of the IEEE.
theory. Three of his papers have been awarded prizes,
including the 1983 O. Hugo Schuck Award. He is a past
REFERENCES
editor of Automatica and a past associate editor of Systems [1] D.J.N. Limebeer and R.S. Sharp, “Bicycles, motorcycles, and models,”
and Control Letters and the International Journal of Robust and IEEE Control Syst. Mag., vol. 26, no. 5, pp. 34–61, 2006.
Nonlinear Control. He is a Fellow of the IEEE, the IEE, the [2] S. Evangelou, D.J.N. Limebeer, R.S. Sharp, and M.C. Smith, “Steering
compensation for high-performance motorcycles,” in Proc. 43rd CDC, Par-
Royal Academy of Engineering, and the City and Guilds adise Island, Bahamas, 14–17 Dec. 2004, pp. 749–754.
Institute. His research interests include control system [3] S. Evangelou, D.J.N. Limebeer, R.S. Sharp, and M.C. Smith, “Mechanical
design, frequency response methods, H-infinity optimiza- steering compensators for high-performance motorcycles,” J. Appl. Mech., to
be published.
tion, and mechanical systems. He is qualified as an IAM [4] R.W. Newcomb, Linear Multiport Synthesis. New York: McGraw Hill,
senior motorcycle instructor and received a RoSPA certifi- 1966.
cate for advanced motorcycling. He can be contacted at [5] M.C. Smith, “Synthesis of mechanical networks: The inerter,” IEEE Trans.
Automat. Contr., vol. 47, no. 10, pp. 1648–1662, 2002.
Imperial College London, Department of Electrical and [6] C.L. Morfey, Dictionary of Acoustics. New York: Academic, 2001.
Electronic Engineering, Exhibition Road, London SW7 [7] D.E. Newland, Mechanical Vibration Analysis and Computation. White
2AZ U.K. Plains, NY: Longman, 1989.
[8] D.J. Inman, Engineering Vibration, 2nd ed. Englewood Cliffs, NJ: Prentice
Robin S. Sharp is professorial research fellow in the Hall, 2001.
Department of Electrical and Electronic Engineering at [9] M.C. Smith, “Force-controlling mechanical device,” patent pending, Int.
Imperial College London. He is a member of the Dynami- pub. WO 03/005142, 4 July 2001, [Online]. Available: http://v3.espacenet.
com/textdoc?DB= EPODOC&IDX=WO03005142&F=0
cal Systems and Mechatronics Working Group of the [10] Harmonic drive. [Online]. Available: http://www.hdsystemsinc.com
International Union of Theoretical and Applied Mechan- [11] Autosim 2.5+ Reference Manual, Mechanical Simulation Corp., Ann
ics, the editorial board of Vehicle System Dynamics, and the Arbor, MI, 1998 [Online]. Available:http://www.carsim.com
[12] R.S. Sharp, S. Evangelou, and D.J.N. Limebeer, “Advances in the model-
editorial advisory board of Multibody System Dynamics. He ling of motorcycle dynamics,” Multibody Syst. Dynamics, vol. 12, no. 3, pp.
was first vice-president and secretary general of the Inter- 251–283, 2004.
national Association for Vehicle System Dynamics, editor- [13] H.B. Pacejka, Tyre and Vehicle Dynamics. Oxford, U.K.: Butterworth
Heinemann, 2002.
ial panel member and book review editor for The [14] E.J.H. de Vries and H.B. Pacejka, “The effect of tyre modeling on the sta-
Proceedings of the Institution of Mechanical Engineers, Journal bility analysis of a motorcycle,” in Proc. AVEC’98, Nagoya: SAE Japan, 1998,
of Mechanical Engineering Science, Part C, and editorial pp. 355–360.
[15] R.S. Sharp, S. Evangelou, and D.J.N. Limebeer, “Multibody aspects of
panel member of The Proceedings of the Institution of motorcycle modelling with special reference to autosim,” in Advances in
Mechanical Engineers, Part D, Journal of Automobile Engi- Computational Multibody Systems, J.G. Ambrósio, Ed. Dordrecht, The Nether-
neering. From 1990–2002, he was professor of automotive lands: Springer-Verlag, 2005, pp. 45–68.
[16] Society of Automotive Engineers, “Vehicle dynamics terminology,” in
product engineering at Cranfield University. He was a Proc. SAE J670e, Warrendale, PA, 1976.
visiting associate research scientist at the University of [17] C. Papageorgiou and M.C. Smith, “Positive real synthesis using matrix
Michigan Transportation Research Institute, Ann Arbor. inequalities for mechanical networks: Application to vehicle suspension,” in
IEEE Trans. on Contr. Syst. Tech, vol. 14, no. 3, pp. 423–435, 2006.
His research covers topics in automotive dynamics and [18] R.C. Dorf and R.H. Bishop, Modern Control Systems. Englewood Cliffs,
control as well as in control and stability of single-track NJ: Prentice Hall, 2001.
vehicles, unmanned air vehicles, and the application of [19] A. Aoki, “Experimental study on motorcycle steering performance,”
SAE paper 790265, 1979.
optimal preview and learning control to road vehicle dri- [20] M. Sugizaki and A. Hasegawa, “Experimental analysis of transient
ving/riding. response of motorcycle rider systems,” SAE paper 881783, 1988.
Malcolm C. Smith received the B.A. (M.A.) degree in [21] D.J.N. Limebeer, R.S. Sharp, and S. Evangelou, “Motorcycle steering
oscillations due to road profiling,” J Appl. Mech., vol. 69, no. 6, pp.
mathematics, the M.Phil degree in control engineering and 724–739, 2002.
operational research, and the Ph.D. degree in control engi- [22] M. Green and D.J.N. Limebeer, Linear Robust Control. Englewood Cliffs,
neering, all from the University of Cambridge, U.K. He NJ: Prentice Hall, 1995.
[23] The Mathworks Inc., MATLAB 6 Reference Manual, 2000 [Online]. Avail-
was subsequently a research fellow at the German Aero- able: http://www.mathworks.com
space Centre, a visiting assistant professor and research [24] J.L. Shearer, A.T. Murphy, and H.H. Richardson, Introduction to System
fellow with the Department of Electrical Engineering at Dynamics. Reading, MA: Addison-Wesley, 1967.
[25] E.L. Hixson, “Mechanical impedance,” in Shock and Vibration Handbook, 2nd
McGill University, and an assistant professor with the ed., C.M. Harris and C.E. Crede, Eds. New York: McGraw-Hill, 1976, chap. 10.
Department of Electrical Engineering at the Ohio State [26] A.G.J. MacFarlane, Dynamical System Models. London: Harrap, 1970.
University. In 1990 he returned to the University of Cam- [27] B.D.O. Anderson and S. Vongpanitlerd, Network Analysis and Synthesis.
Englwood Cliffs, NJ: Prentice Hall, 1973.
bridge where he is currently a professor. He is a fellow of [28] M.E.V. Valkenburg, Introduction to Modern Network Synthesis. New York:
Gonville and Caius College. His research interests include Wiley, 1960.

88 IEEE CONTROL SYSTEMS MAGAZINE » OCTOBER 2006

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY DELHI. Downloaded on January 05,2022 at 07:14:42 UTC from IEEE Xplore. Restrictions apply.