Anda di halaman 1dari 6

SCIENTIFIC PROCEEDINGS 2014, Faculty of Mechanical Engineering, SUT in Bratislava

Vol. 22, 2014, pp.26‐31, DOI:10.2478/stu‐2014‐0005 

MULTI-OBJECTIVE OPTIMIZATION OF VEHICLE SUSPENSION PARAMETERS


CONSIDERING VARIOUS ROAD CLASSES
Ferdinand Havelka 1, Miloš Musil27

Keywords: half-car, vehicle suspension, optimization, random search method

Abstract

Vehicle suspension optimization for various road classes travelled at different velocities is
performed. Road excitation is modeled using a first order shaping filter. A half-car model is
adopted to simulate the vehicle’s vertical dynamics. The excitation time delay between the rear
and the front tire is modeled using Pade approximation. Suspension parameters are optimized
using a random search method with respect to “comfort” and “sporty driving” considering
the design constraints of the suspension and road holding and maximum suspension travel
constraints. Optimal suspension parameters suitable for various road classes and vehicle
velocities have been chosen.

1 INTRODUCTION
Passenger cars travel over random surfaces of different types during their operation.
In terms of vertical dynamics, it is desirable to design vehicle suspension in such a way as to
ensure permanent contact of the tires with the road (road holding criterion), to keep the suspension
deflections within acceptable limits (suspension rattle space criterion) while minimizing variances
of frequency-weighted accelerations in the space designed for passengers (ride comfort criterion)
[1], [4], [5]. When meeting the first two conditions – road holding and suspension rattle space –
the suspension can be tuned either more to comfort or sporty driving. Tuning the suspension
for comfort corresponds to minimizing the variances of frequency-weighted accelerations
in the space designed for passengers, while setting the suspension for sporty driving corresponds
to a minimization of the dynamic tire forces. The above concept of suspension settings has been
chosen by several car manufactures (BMW, Mercedes, Porsche and others) also when using active
or semi-active suspensions. The driver can usually choose “comfort”, “sport” or “normal”
(compromise) settings, and smoothly switch between those modes.
Optimizing the suspension means finding such a combination of the suspension parameters that
the chosen criterion (sport or comfort) is minimal when meeting the boundary conditions (road
holding and suspension rattle space) and respecting the suspension design constraints. Finding
the optimal parameters of the vehicle suspension under the above criteria is a boundary
optimization problem. Classical derivative methods are inadequate for multivariable multi-
objective optimization problems with constraints. Therefore, global optimization approaches such

                                                            
71
  Facultyof Mechanical Engineering, STU in Bratislava, Námestie slobody 17, 812 31 Bratislava 1,
ferdinand.havelka@stuba.sk
2Faculty of Mechanical Engineering, STU in Bratislava, Námestie slobody 17, 812 31 Bratislava 1,

milos.musil@stuba.sk
SCIENTIFIC PROCEEDINGS 2014, Faculty of Mechanical Engineering, SUT in Bratislava
Vol. 22, 2014, pp.26‐31, DOI:10.2478/stu‐2014‐0005 

as genetic algorithms, simulated annealing method and generalized pattern search algorithms are
used in the scientific literature [5], [6].

2 ROAD EXCITATION MODELLING


Road excitation is modelled using a first order filter [1] such that its power spectral
density corresponds to the road surfaces defined in the ISO 8608 standard [2] – figure 1.a).

w  t    vw  t     t  (1)

where α is the coefficient depending on road type, v is the vehicle velocity and ξ(t) is a white noise
process with constant power spectral density (PSD)

   2 v w2 (2)

where σw2 is variance of the road irregularities. Then, the PSD of the road excitation is

2av w2
 w    (3)
a v  2
2 2

3 VEHICLE VERTICAL DYNAMICS MODELLING


A half-car model is used for simulating the vehicle vertical dynamics – figure 1 b.).

a.) b.)
Figure 1: a.) PSD of roads of classes A – E for various vehicle velocities, ISO 8608 vs. shaping
filter (3); b.) Half-car model

The equations of motion are in the form


ywf  btf  y wf  w f   bsf  y wf  y b  lsf b   ktf  ywf  w f   k sf  ywf  yb  lsf b   u f
mwf 
ywr  btr  y wr  w r   bsr  y wr  y b  lsrb   ktr  ywr  wr   k sr  ywr  yb  lsrb   ur
mwr 
yb  bsf  y b  lsf b  y wf   bsr  y b  lsrb  y wr   k sf  yb  lsf b  ywf   k sr  yb  lsrb  ywr   u f  ur (4)
mb 
I bb  bsf  y b  lsf b  y wf  lsf  bsr  y b  lsrb  y wr  lsr  k sf  yb  lsf b  ywf  lsf
 k sr  yb  lsrb  ywr  lsr  u f lsf  ur lsr
SCIENTIFIC PROCEEDINGS 2014, Faculty of Mechanical Engineering, SUT in Bratislava
Vol. 22, 2014, pp.26‐31, DOI:10.2478/stu‐2014‐0005 

As a result of the optimization process, optimal values of stiffness and damping of front and rear
suspensions will be found for various road types travelled at a given range of velocities.

4 MODELLING OF THE EXCITATION TIME DELAY BETWEEN THE REAR AND


THE FRONT TIRE
If the drive of the vehicle is direct, the rear tire is driven by the same signal as the front
one located on the same side but with some time delay. It can be mathematically expressed as
wr  t   w f  t   d  (5)
where wf and wr are the road excitations under the front and rear tire respectively, τd = lr/v
represents excitation time delay between the rear and front tire and lr is the wheelbase. In practice,
Pade approximation [3] is used for modelling the time delay, which can be expressed
in the Laplace area in the form
m

a s i
i

H pd  s   i 0
n
(6)
1   bj s j
j 1

where ai and bj are factors dependent on the time delay τd and on a chosen order of the Pade
approximant. The 10–th order Pade approximant is chosen in the simulations – figure 2 a.).

a.) b.)
Figure 2: a.) Modelling of the time delay between the excitation of the front and the rear
tire for E-class road and travelling vehicle velocity 20 km/h; b.) Comparison of the frequency
weighting curve Wk and shaping filter (7)

5 FREQUENCY WEIGHTING OF ACCELERATIONS


Standard ISO 2631-1 [4] deals with the evaluation of human exposure to whole-body
vibration. It examines the effect of vibration on the human body in three perpendicular axes.
A frequency weighting curve is defined for a particular direction of the applied vibration and
the position of the human body. In the vertical direction for the "sitting" position, the acceleration
is weighted by frequency curve Wk – figure 2 b.).

In the scientific literature, low order filters are proposed to approximate the Wk curve. They are
in the form
SCIENTIFIC PROCEEDINGS 2014, Faculty of Mechanical Engineering, SUT in Bratislava
Vol. 22, 2014, pp.26‐31, DOI:10.2478/stu‐2014‐0005 

c s i
i

Wk  s   i 0
m
(7)
s m 1
 djs j
j 0

In this paper, 5-th order filter is utilized. Its factors are listed in the table below.

Table 1: Factors of shaping filter (7)


1. 2. 3. 4. 5.
ci 87.72 1138 11336 5453 5509
dj 92.6854 2549.83 25969 81057 79783

In this paper, accelerations in the space designed for the passengers are frequency-weighted using
the mentioned shaping filter (7).

6 SUSPENSION OPTIMIZATION USING MODIFIED RANDOM SEARCH METHOD


Finding the optimal parameters of vehicle suspension is a boundary optimization
problem. In this paper, a modified random search method is adopted for the suspension parameter
optimization. The essence of this method, when optimizing n parameters, is to generate randomly
and uniformly distributed values of the optimized parameters, in the n-th dimensional space inside
an n dimensional hypersphere, then to find the value of the objective function for each
combination of optimized parameters and to choose the combination with a minimal value
of appropriate objective function. The center of the hypersphere is subsequently moved
to the place with a minimal value of the objective function.
The optimized parameters are stiffness (ks) and damping (bs) of the front and rear
suspensions. Their values are bounded from below and from above

k sf   ksf min , k sf max  , k sr   ksr min , k sr max 


(8)
bsf   bsf min , bsf max  , bsr   bsr min , bsr max 

When tuning the suspension to “compromise”, the scalar objective function contains the addition
of the sum of mutual weighted standard deviations of frequency-weighted accelerations (σa w) and
the sum of mutual weighted maximum dynamic tire forces (Ft dyn max)
2 2
J   qa i  a w i   q f j Ft dyn max j 
k ,k , b , b
sf sr sf sr
 min (9)
i 1 j 1

where qa i a qf i are the appropriate weights. For the weights applies


2 2
qa i , q f j   0,1   qa i   q f j  1
i 1 j 1
(10)

2 2
If q
i 1
ai  1 , the suspension is tuned to “comfort”, if q
j 1
f j  1 , the suspension is tuned

to “sporty driving”. Assessment of the validity of the road holding and suspension travel criterions
is performed by nonlinear constraint functions given as the following inequalities
SCIENTIFIC PROCEEDINGS 2014, Faculty of Mechanical Engineering, SUT in Bratislava
Vol. 22, 2014, pp.26‐31, DOI:10.2478/stu‐2014‐0005 

2
Ft dyn max j ( ksf ,ksr , bsf , bsr )  Ft stat j  0, ys max j ( ksf ,ksr , bsf , bsr )  ys allow j  0 (11)
3
Values of optimized parameters which do not satisfy the conditions (11) are excluded. Standard
deviation of accelerations and forces are calculated using the Lyapunov equation. The three-sigma
rule (99.7 % probability) is utilized for evaluation of maximum dynamic tire forces.
The optimal suspension parameters are sought for road classes A-D for various travelling
velocities. For class A and D, the maximum velocities are chosen as 250 km/h and 40 km/h
respectively. Results of the suspension optimization process are shown in figure 3.

Figure 3 : Optimal stiffness and damping of front and rear suspensions for road classes A-D and
for various travelling velocities when tuned to “comfort” and “sport”

From the results of the suspension optimization process it follows that:


1. Optimal front suspension stiffness is identical for all road classes and all vehicle velocities
ksf = 20 000 N/m.
2. Optimal rear suspension stiffness is also almost identical for all road classes and all vehicle
velocities. Rear suspension stiffness was chosen as ksr = 21 000 N/m.
3. Optimal front suspension damping when tuned to sporty drive does not vary much, ranging
around 2 000 N.s/m. When tuned to comfort, it fluctuates slightly, but when driving on a less
quality road of class D, it gradually increases with vehicle velocity (up to 1 120 N.s/m),
in order to maintain the road holding criterion. A similar argument also applies for the rear
suspension damping.
4. The bands limited by the blue and the red curves represent the values of allowable suspension
damping for the appropriate road class. Intersection of all bands represents the values
of allowable suspension damping valid for all road classes. Optimal suspension damping must be
SCIENTIFIC PROCEEDINGS 2014, Faculty of Mechanical Engineering, SUT in Bratislava
Vol. 22, 2014, pp.26‐31, DOI:10.2478/stu‐2014‐0005 

greater than the maximum value of the damping when tuned to comfort – blue curve – for all
road classes and for all velocities (as if it is lower, at least at the highest velocity when travelling
on a road of class D, it is a question of road holding criterion failure) and should be lower than
the minimum value of damping when tuned to sporty drive – red curve. The total allowable
range of damping is defined by the blue and red dashed curves. As optimal following values
of front and rear suspension damping were selected: bsf = 1 500 N.s/m, bsr = 1 900 N.s/m.

3 CONCLUSION
In this paper vehicle suspension optimization for various road classes travelled
at different velocities was performed. Suspension parameters were optimized using a modified
random search method with respect to “comfort” and “sporty driving” considering the road
holding and maximum suspension travel constraints and the design constraints of the suspension
parameters. From the set of allowable suspension parameters were selected the optimal front and
rear suspension stiffness and damping suitable for common road classes and vehicle velocities.

ACKNOWLEDGEMENTS
This work was supported by Vega grant 1/0197/12.

REFERENCES
[1] Hac, A., “Suspension Optimization of a 2 - DOF Vehicle Model Using a Stochastic
Optimal Control Technique”, Journal of Sound and Vibration, vol. 100, no. 3, 1985

[2] STN ISO 8608: Mechanické kmitanie. Profily povrchu cesty. Zaznamenávanie
nameraných údajov., 2000

[3] Hać A., Youn I.: “Optimal Design of Active and Semi-Active Suspensions Including
Time Delays and Preview”, Journal of Vibration and Acoustics, vol. 115, no. 4, pp.
498-508, 1993

[4] STN ISO 2631-1: Mechanické kmitanie a otrasy. Hodnotenie expozície človeka kmitaniu
na celé telo., 1999

[5] Gobbi M., Levi F. & Mastinu G.: “Multi-objective stochastic optimization
of the suspension system of road vehicles”, Journal of Sound and Vibration, vol. 293,
no. 4 - 5, pp. 1055 - 1072, 2006

[6] Hók V.: “Virtuálny model systému zavesenia kolesa pre prenos budenia do karosérie”,
Dizertačná práca, Bratislava: Slovenská technická univerzita v Bratislave, Strojnícka
fakulta, Ústav aplikovanej mechaniky a mechatroniky, 2009

Anda mungkin juga menyukai