12th IFToMM World Congress, Besançon (France), June18-21, 2007

Optimization and comparison of passive, active, and semi-active

vehicle suspension systems
S. Segla1 S. Reich2
Technical University of Kosice Technical University of Kosice
Kosice, Slovakia Kosice, Slovakia

Abstract — In the vehicle suspension design the optimal to be fixed to the axle.
suspension should fulfil the following basic requirements: the The equations of motion (assuming permanent road-tyre
ride comfort, reduction of dynamic road-tyre forces, and contact) have the following forms
reduction of relative motions between the vehicle bodies.
Controllable suspension systems are required to improve the
compromise between the conflicting demands. In the paper the m3 engine mass
performances of the passive suspensions are optimized and
compared with the performances of. active and semi-active z3
suspensions (quarter car models). b3 engine suspension
k3
Keywords: vehicle, suspension, optimization
m2 sprung mass
z2
I. Introduction b2
k2 main suspension
The problem of designing vehicle suspension systems is
complex owing to the necessity to meet several conflicting m1 axle mass
demands. The passanger comfort should be as high as z1
possible, while the relative motion between the body of tyre elasticity
the vehicle (the sprung mass) and the axle with the wheels k1
(the unsprung mass) must be limited to a reasonable value. road irregularity
The same demand holds for the dynamic road-tyre force. y
In this paper first possibilities of improvement of the
passive suspensions by nontraditional configurations [2], Fig. 1. Three-mass basic system (case A)
[3], [4], [12] are shown. Configurations with additional
dampers mounted between the engine and the axle, a m3
separate mass (split axle) acting as a vibration absorber,
and relaxation damping are optimized and compared. z3
b3 k4
To demonstrate the potential of controlable suspensions, k3
two basic active and semiactive controlable suspensions
are fully optimized. Their dynamic properties are com- m2
pared with those of the optimized passive suspensios.
z2
The Optimization Toolbox for Use with Matlab [5] is
used to solve the optimization problems. k2 b2 b4

m1
II. Passive suspensions
z1
Figures 1 and 2 show dynamic models of the three-mass
suspensions: case A (figure 1) is the basic one and case B k1
(figure 2) with additional springs and dampers between
the engine and the axle (see also figure 3). Figure 4 shows y
the dynamic model of the three mass suspension with a
vibration absorber – case C (its mathematical model has Fig. 2. Three-mass system (case B)
the same form as the case B). This idea is based on the
assumption, that not all masses (brakes, differentials) have Case A:
m1&z&1 − b2 ( z& 2 − z&1 ) + k1 ( z1 − y ) − k 2 ( z 2 − z1 ) = 0,
1
m2 &z&2 + b2 ( z& 2 − z&1 ) − b3 ( z&3 − z& 2 ) + k 2 ( z 2 − z1 ) − k 3 ( z 3 − z 2 ) = 0,
E-mail: [email protected]
2
E-mail: [email protected] m3 &z&3 + b3 ( z&3 − z& 2 ) + k 3 ( z 3 − z 2 ) = 0.
12th IFToMM World Congress, Besançon (France), June18-21, 2007

Case B: Case E:

m1&z&1 − b2 ( z& 2 − z&1 ) − b4 ( z&3 − z&1 ) + k1 ( z1 − y ) − k 2 ( z 2 − z1 ) − m1&z&1 + k1 ( z1 − y ) − k 2 ( z 2 − z1 ) − k3 ( z3 − z1 ) = 0,

− k 4 ( z 3 − z1 ) = 0, (m2 + m3 ) &z&2 + b2 ( z&2 − z&3 ) + k 2 ( z 2 − z1 ) = 0,
m2 &z&2 + b2 ( z& 2 − z&1 ) − b3 ( z&3 − z& 2 ) + k 2 ( z 2 − z1 ) − k 3 ( z 3 − z 2 ) = 0, k3 ( z3 − z1 ) − b2 ( z&2 − z&3 ) = 0.
m3 &z&3 + b3 ( z&3 − z& 2 ) + b4 ( z&3 − z&1 ) + k 3 ( z 3 − z 2 ) + k 4 ( z3 − z1 ) = 0.

m2
z2
vehicle body
k2 b2

additional damper m1
z1
engine
k1
wheel

y
Fig. 3. Half of the car model (case B)
Fig. 5. Basic two-mass suspension (case D)
m2 sprung mass
z2 m2
k3 b3 z2
b2
k2
k2 b2 z3
m3 dynamic absorber k3
relaxation spring
k4 b4 z3
m1
m1 unsprung mass z1
z1
k1 tyre elasticity
k1
road
y
y
Fig. 4. Suspension with a vibration absorber (case C)
Fig. 6. Two-mass suspension with relaxation damping (case E)

Figure 5 shows the basic two-mass suspension (case D).

The dynamic model of the two-mass suspension with III. Active suspension
relaxation damping (figure 6) involves the additional The fully active suspension involves replacing the
spring k3 mounted between between the damper b2 and conventional suspension elements with a (hydraulic)
actuator that is controlled by a high frequency response
the unsprung mass m1 .
servo valve. The force demand signal (usually generated
The equations of motion of the two-mass suspensions in a microprocessor) is governed by a control law that is
are as follows normally obtained by application of various forms of the
Case D: optimal control theory. The major drawback of this kind
of suspension is that it requires an external power supply.
m1&z&1 − b2 ( z&2 − z&1 ) + k1 ( z1 − y ) − k 2 ( z 2 − z1 ) = 0, The following control law is used to govern the actuator
(m2 + m3 ) &z&2 + b2 ( z&2 − z&1 ) + k 2 ( z 2 − z1 ) = 0. force
12th IFToMM World Congress, Besançon (France), June18-21, 2007

Fact = bsky z& 2 , Fsad = bsad ( z&2 − z&1 ) ,

where bsky is so called sky-hook damping coefficient (see  bmax , if ba > bmax ,
the fictitious sky-hook damper in figure 7). 
bsad =  ba , if bmin < ba < bmax ,
 b , if b < b ,
 min a min

bsky bsky z&2

ba = , if z&2 (z& 2 − z&1 ) > 0, otherwise ba = bmin .
z&2 − z&1

m2 The semi-active damper is controlled in such a way,

that it tries to have the same effect on the sprung mass as
z2 the fictitious sky-hook damper. However, its damping
Fact force is minimum, when its effect is oriented opposite to
k2 the effect of the fictitious sky-hook damper. The equations
of motion are the same as those for the active suspension,
but Fact must be replaced by Fsad.
m1
V. Road excitation
z1
k1 The suspensios above are subjected to the deterministic
excitation [9], [10] determined by a circular bump, figure
9, of the height h = 0.05 m, and length L = 3 m.
y
y L
Fig. 7. Active suspension (case F)

The equations of notion of this system are

h
m1&z&1 + k1 ( z1 − y ) − k 2 ( z 2 − z1 ) − Fact = 0,
x
m2 &z&2 + k 2 ( z 2 − z1 ) + Fact = 0.

Fig. 9. Circular bumb

IV. Semi-active suspension

m2 VI. Formulation of the optimization problem

First appropriate design variables that effect the ride
z2 comfort, dynamic road-tyre forces, and also relative
motions between the vehicle bodies must be chosen. In the
k2 bsad cases above these are damping coefficients, spring
stifnesses and control variables. The objective function
expressing the vertical acceleration of the sprung mass
m1 (associated with the vehicle comfort) has the following
form [6], [7], [10], [13]
z1
k1 tmax N
f op = ∫ &z&2 (t ) dt ≅ ∑
i =1
&z&2i ∆ ti ,
t =0

y
where &z&2i denotes acceleration of the sprung mass at time
Fig. 8. Semi-active suspension (case G) ti and ∆ t i is the integration step. It is apparent that
minimization of f op leads to minimization of the area
The semi-active algorithm is derived from the sky-hook
damping. The semi-active force is under the curve of the sprung mass acceleration [8]
(negative area is taken as positive).
12th IFToMM World Congress, Besançon (France), June18-21, 2007

Further, the following costraints are taken into account x min ≤ x ≤ x max .
1. Constraint on the relative motion between the Fmincon uses a sequential quadratic programming based
sprung and unsprung masses on the derivation of simultaneous nonlinear equations
using Kuhn-Tucker conditions to the Lagrangian of the
z 2 (t ) − z1 (t ) ≤ z 21,max , t ∈ 〈0, t max 〉 . constrained optimization problems. A quadratic program-
ming subproblem is solved at each iteration.
2. Constraint on the dynamic road-tyre force If an optimization problem with mixed continuous/in-
teger design variables is to be solved (e.g. the design var-
iable representing a tyre with several different radial stiff-
k1 ( z 2 (t ) − y (t )) ≤ f dyn,max , t ∈ 〈0, t max 〉 .
nesses) then the program GOOD presented e.g. in [11, 12]
can be used.
3. Constraint on the relative motion between the
sprung mass and engine VIII. Optimization results
z3 (t ) − z 2 (t ) ≤ z32,max , t ∈ 〈 0, t max 〉 . The values of the objective functions for cases A and B
(table I) show that the additional damper b4 has a positive
effect on the ride comfort (the spring stifness is zero).
In the above inequalities z 21, max means the permissible However, the improvement is very small – about 4 %. On
allowance of the relative motion between the sprung and the other hand, the improvement for the ride of the vehicle
unsprung masses, z 32, max is the permissible allowance of on along a stochastically uneven road [1] was about 25 %.
Tables II and III show that using dynamic absorber (of
the relative motion between the sprung mass and engine,
mass m3 = 15 kg) has again a slight positive effect
and f dyn,max is the permissible allowance of the dynamic
compared with the basic two-mass system D (the
road-tyre force. improvement of about 5 %), but the value of the objective
The following variables are specified: function for case E with relaxation damping is slightly
v = 20 m/s – the velocity of the ride, worse.
m1 = 55 kg – the unsprung mass, For the both cases C and E it was shown in [1], that they
m2 = 400 kg – the sprung mass, had positive effect on the ride comfort with improvements
up to 46.6 % (corresponding mass of the vibration
m3 = 100 kg – the engine mass, absorber was 32 kg), and 21.2 % for relaxation damping.
z 21, max = 0.08 m – the permissible allowance of the
relative motion between the sprung and unsprung masses, Values
z 32, max = 0.04 m – the permissible allowance of the relative
motion between the sprung mass and engine,
Variable Units Case A Case B
f dyn,max = k B f stat , k B = 0.8 – the permissible allowance (three mass (additional
of the dynamic road-tyre force, basic system) dampers)
f stat = g ( m1 + m2 + m3 ) – the static road-tyre force.
k1 N/m 225157.0 225196.0
VII. Method of optimization
The Optimization Toolbox for Use with MATLAB [5] k2 N/m 40000 40000
was used to solve the optimization problems formulated in
b2 Ns/m 1687.5 1687.0
chapter VI. The minimization routine fmincon finds the
constrained minimum of the objective function f op for- k3 N/m 52527.9 54909.1
mulated in the form of a time integral of the sprung mass
acceleration. The function returning the value of the ob- b3 Ns/m 3500 335.2
jective function uses an explicit Runge-Kutta (4, 5) for-
k4 N/m – 0
mula [14] on a time interval t ∈ 〈 0, 2 s〉 . Figures 10 to 12
show that the final time is sufficient to take into account b4 Ns/m – 687.6
all important dynamic properties of the systems.
The second function returns the value at all constraints bsky Ns/m – –
at the current vector of the design variables x. The
constrained minimization routine, fmincon, is then invoked. fop m2/s4 0.988 0.949
It defines a set of lower and upper bounds of the design
variables so that the solution xopt is always in the range TABLE I. The optimization results – cases A, B
12th IFToMM World Congress, Besançon (France), June18-21, 2007

Values Values
Variable Units Case C Case D Variable Units Case D Case G
(vibration (basic two (basic passive (active
absorber) mass system) suspension) suspension)
k1 N/m 262713.7 220004.3 k1 N/m 220004.3 218408.2
k2 N/m 40000 40624.2
k2 N/m 40000 40000
b2 Ns/m 3047.6 –
b2 Ns/m 2969.1 3047.6
bsky Ns/m – 14909.9
k3 N/m 34972.8 –
fop m2/s4 0.984 0.615
b3 Ns/m 668.5 –
TABLE V. The optimization results – cases D, G
k4 N/m 0 –
Tables IV and V show the comparison of the two-mass
fop m2/s4 0.936 0.984 basic passive suspension with the active and also semi-
active ones. It is apparent that the active and also semi-
active suspension systems are superior to the passive one.
TABLE II. The optimization results – cases C, D
The improvements in the values of the objective function
are significant – 41.5 % (active suspension system) and
Values 37.5 % (semi-active suspension system).
Variable Units Case E Case D
(vibration (basic two
absorber) mass system)
k1 N/m 290000 220004.3

k2 N/m 40000 40000

b2 Ns/m 2522.0 3047.6

k3 N/m 450000 –

b3 Ns/m – –

k4 N/m – –

fop m2/s4 1.007 0.984

Fig. 10. Time history of the sprung mass acceleration &z&2 (t ) on time
TABLE III. The optimization results – cases E, D

Values
Variable Units Case D Case F
(basic passive (active
suspension) suspension)
k1 N/m 220004.3 200000

k2 N/m 40000 40000

b2 Ns/m 3047.6 –

bsky Ns/m – 15000

fop m2/s4 0.984 0.576

TABLE IV. The optimization results – cases D, F Fig. 11. Time history of the relative motion z 2 (t ) − z1 (t ) on time
12th IFToMM World Congress, Besançon (France), June18-21, 2007

Figures 10, 11, and 12 show time histories of the sprung suspension systems (Part II - Three-mass systems). Vehicle System
Dynamics, Vol. 3: 55-98, 1974.
mass acceleration &z&2 (t ) , relative motion between the [5] Optimization Toolbox for Use with Matlab - User's Guide. The
sprung and unsprung masses z 2 (t ) − z1 (t ) , and dynamic Mathworks Inc., Natick, 2000.
[6] Zmindak M., Saga M., Tvaruzek J., Husar S. Optimization of
road-tyre force f dyn (t ), respectively. Mechanical Systems. EDIS ZU in Zilina, 2000.
[7] Dekys V., Saga M., Zmindak M. Some aspects of structural
optimization by finite element method. In Int. Scient. Conf.
Innovation and utility in the Visegrad four, pages 605-610,
Nyiregyhaza, October 13-15, 2005.
[8] Siemieniako F., Peszynski K. Automatyka w Przykladach i
Zadaniach. Wydawnictwo Politechniky Bialostockiej, 2005.
[9] Mitschke M. Dynamik der Kraftfahrzeuge, Band B - Schwingungen.
Springer-Verlag, 1984.
[10] Nigam N.C., Narayanan S. Application of Random Vibration.
Springer-Verlag, 1994.
[11] Kalker-Kalkman C.M. A general multi-objective optimization
program for mixed continuous/integer variables based on genetic
algorithms. In Progress in Industrial Mathematics at ECMI 96,
page 348, Stuttgart, 1997.
[12] Segla S., Kalker-Kalkman C.M. Optimization of the two-wheel
trailer suspension parameters. Engineering Mechanics, 3:177-186,
1996.
[13] Bocko J., Muller J. Evolution and structural shape optimization.In
CTU Workshop 96, pages 15-23, Brno, June, 14-16, 1996.
[14] Dormand J.R., Prince P.J. A family of embedded Runge-Kutta for-
mulae. J. Comp. Appl. Math., 6:19-26, 1980.
[15] Rao S.S. Engineering Optimization. Theory and Practice. Third
Fig. 12. Time history of the dynamic road-tyre force f dyn (t ) on time Edition. John Wiley & Sons, 1996.

VIII. Conclusion
In the paper optimization and comparison of passive,
active, and semi-active vehicle suspension systems was
presented. It was shown, that not only the active, but also
semi-active suspension system is able to improve the ride
comfort significantly compared with the classical passive
suspension systems.
Using the vibration absorber (split axle) and additional
dampers between the engine and axle have improved the
ride comfort only slightly. Using the relaxation damping
even slightly worsened the ride comfort compared with
the basic two-mass passive suspension system. However,
in [1] it was shown that these results were much better in
the case of the ride on along an stochastically uneven
road.

Acknowledgement
The author gratefully acknowledges the support by the
Scientific Grant Agency of the Ministry of Education of
the Slovak Republic (project No. 1/3156/06).

References
[1] Segla S. Possible improvements in ride comfort of automobile
suspension systems by additional spring-damping elements. In 9th
Scientific Seminar on Developments In Machinery Design and
Control, pages 87-88 (full paper on CD), Bydgoszcz-Muszyna,
September 11-13, 2005.
[2] Ryba D. Semi-active damping with an electromagnetic force
generator. Vehicle System Dynamics, Vol. 22: 79-95, 1993.
[3] Ryba D. Improvements in dynamic characteristics of automobile
suspension systems (Part I - Two-mass systems). Vehicle System
Dynamics, Vol. 3: 17-46, 1974.
[4] Ryba D. Improvements in dynamic characteristics of automobile