:B04-0001

Generation of Random Road Profiles

Feng Tyan and Yu-Fen Hong Shun-Hsu Tu and Wes S. Jeng
Computational Dynamics and Control Lab Sky Leading Corportation
Dept. of Aerospace Engineering Chupei, Hsin Chu County, Taiwan 30204, R. O. C.
Tamkang University
Tamshui, Taipei County, Taiwan 25147, R. O. C.

ITRI Project: 5353C46000

Abstract In this work we review two of the most com- road profile is given in section IV. The amplitude of each
monly adopted methods, namely shaping filter and sinusoidal sinusoidal function is re-derived for completeness here.
approximation, for generating one-dimensional random road Finally, two examples are given to demonstrate the result-
profiles, that are used in the simulation of a quarter car
(or half car) vehicle suspension system control. For the ing grade B profiles generated by the two aforementioned
shaping filter method, it is found that the time constant of methods respectively.
the first order transfer function generating the road profile
is independent of the grade of road. While for the sinusoidal II. C LASSIFICATION OF ROAD P ROFILES
approximation method, a detail derivation of the amplitude The road profile can be represented by a PSD function.
of each sinusoidal function is re-derived for completeness. The power spectral densities of roads show a characteristic
keywords: random process, shaping filter, sinusoidal ap- drop in magnitude with the wave number. To determine
proximation the power spectral density function, or PSD, it is necessary
to measure the surface profile with respect to a reference
I. I NTRODUCTION plane. Random road profiles can be approximated by a
Profiles taken along a lateral line show the supereleva- PSD in the form of
 w  w
tion and crown of the road design, plus rutting and other n
distress. Longitudinal profiles show the design grade, () = (0 ) or (n) = (n0 ) ,
0 n0
roughness, and texture (see Fig. 1) [1]. In this paper, (2.1)
we focus on the longitudinal profiles, which classifi- where
Whatcations Isareabased Profile?
on the International Organization for = 2 L in rad/m denotes the angular spatial fre-
Standardization (ISO 8606). The ISO has proposed road quency, L is the wavelength,
A roughness
profile is a classification
two-dimensional using
slice the power
of the road spectral along an 0 , (0 ) in m2 /(rad/m) describes the values of
density
surface, taken
imaginary line.
(PSD) values as shown in Fig. 2, and Table I, II [2]. Paved the psd at the reference wave number 0 = 1rad/m,
Profiles
roads taken
are along a lateral
generally line show to
considered the be
superelevation
among road and crown of the road n = is the spatial frequency, n = 0.1cycle/m,
classes 2 0
design, plus rutting and other distress. Longitudinal profiles show the design grade,
A to D.and texture. In this book, we will focus on longitudinal profiles.
roughness, w is the waviness, for most of the road surface, w =
2.
TABLE I
ROAD ROUGHNESS VALUES C LASSIFIED BY ISO

degree of roughness (n0 )(106 m2 /(cycle/m))

where n0 = 0.1cycle/m
road class lower limit geometric mean upper limit
Longitudinal
Profiles A (very good) 16 32
B (good) 32 64 128
Lateral C (average) 128 256 512
Profile D (poor) 512 1,024 2,048
E (very poor) 2,048 4,096 8,192
Fig. 1.or ground
A profile of a road, pavement, Road Profile
can be measured along any continuous
imaginary line on the surface. If a measurement is repeated, the same profile can only
be expected if the same imaginary line is followed. (To obtain repeatable measures, it For a rough and quick estimation of the roughness
helpsThis
to makepaper
the lineislessorganized
imaginary byasusing paint or tape
follows. In tosection
mark it physically.)
II we quality, the following guidance is given:
You can take
briefly many profiles
introduce for a road, each
the classification alongprofiles
of road differentbased
a line. 1) new roadway layers, such as, for example, asphalt
onIt is
thepossible
PSD to ofmeasure
road profile. Then
the profile for athe concept
curved of generating
line. Normally, the expectation or concrete layers, can be assumed to have a good
forrandom
a road isprofiles
that the line
by isshaping
a constant distance
filter from the centerline
is re-visited. or some other
In the mean or even a very good roughness quality;
reference that follows the road geometry. Frequently, profile is measured along two
time,
lines the time
per lane, one in constant of theForassociated
each wheeltrack. greater detailtransfer function
any number of lines can be 2) old roadway layers which are not maintained may
is derived. After that, a sinusoidal approximation of the
measured. be classified as having a medium roughness ;
The width of the line is not standard.
The width is usually defined by the type of instrument used. For example,
measures made with a laser system may cover a slice of the road just a few
millimeters thick, while measures made with an ultrasonic system may cover a thicker
slice of several centimeters. The effect of profile width is not yet understood. CSME-1373
However, it is harder to exactly repeat a profile measure if the line for the profile is
very thin.
 prEN 1991-2:2002 (E)

where XT (t) is a truncated version of a stationary process

X(t), Z T
F [XT (t)] , X(t)e d.
T

Then S() is the Fourier transform (if it exists) of R( ),

that is Z
S() , R( )e d, (3.2)

and conversely
Z
1
R( ) = S()e+ d. (3.3)
2
Remark 3.1: If X(t) is real, then the auto-correlation
function R( ) is real and even, that is, R( ) = R( ), it
follows that
Z
S() = 2 R( ) cos( )d, (3.4)
0
and Z
1
R( ) = S() cos( )d. (3.5)
0
Key
3
To avoid negative wave numbers, usually a one-sided psd
1 Displacement power spectral density, Gd (n) [m ]
2
Fig. 2. Road Surface Classification (ISO 8608). The axes surrounding
Wavelength, [m]
is defined with
3
the frame are defined as 1: displacement
Displacement power spectral density, Gd ()
psd, (n)[m3 ], 2: wavelength,
[m3]
(
4 [m],
Spatial3:frequency,
displacement psd, ()[m3 ], 4: spatial frequency, n[cycle/m],
n [cycles/m] 2S(), for 0,
5 5:Angular
angular spatial
spatial frequency,
frequency, [rad/m].
[rad/m] () = (3.6)
Figure B.1 Road surface classification (ISO 8608) 0, for < 0.

TABLE
The limit values of Gd and Gv for the first 5 road surface Therefore,
IIclasses in terms of n and are given in Tables we obtain
B.1 and B.2, respectively.
D EGREE OF ROUGHNESS E XPRESSED IN TERMS OF
Z
1
Table B.1 Degree of roughness expressed in terms of spatial frequency units, n R( ) = () cos( )d. (3.7)
2 0
degree Degree
of roughness ( )(106 m3 )
of roughness 0
Road Pavement a -6 where 0 = 1rad/m
Gd (n0) [10 m] Gv (n) [10-6 m] A. Road Profiles in Spatial and Temporal Domain
class road class
quality Lower limit lower limitmeangeometric
Geometric mean upper
Upper limit limitmean
Geometric
A Very A good
(very good)
--- 16 32 1 6,3 2
It is well known that the amount of road excitation
B Good B (good) 32 64 2 128 4 25,38 imposed at the vehicle tire depends on two factors:
C Medium
D Poor
C (average)128
512
256
1024
8 512
2048
16 101,1
32
404,3 1) the road roughness which is a function of the road
E Very (poor) 2048
D poor 4096 32 8192 64 128
1617,0
a E (very poor)
n0=0,1 cycle/m 128 256 512 roughness coefficient,
2) the vehicle velocity V .
Let s be the path variable. By introducing the wavelength
3) roadway layers consisting of cobblestones or similar 2
material may be classified as medium (average) or , = (3.8)
127
bad (poor, very poor). and assuming that s = 0 at t = 0, the term s can be
III. S HAPING F ILTER written as
2 V
s = s = 2 t = t, (3.9)
Before we proceed, some of the fundamental theorems
which will be used later are given in the following [3, 4]. where (rad/sec) is the angular velocity in time domain,
Theorem 3.1 (Parsevals formula): If X() = we end up with
F[x(t)] = A()ej() is the Fourier transform of X(t), V = . (3.10)
then
Z Z Z Hence, in the time domain the excitation frequency is
1 1 V
|x(t)|2 dt = A2 ()d = |X()|2 d. given by f = 2 = . For most of the vehicles
2 2 the rigid body vibrations are in between f = 0.5Hz
(3.1)
to f = 15Hz. This range is covered by waves which
Theorem 3.2 (Wiener-Khintchine Theorem): Let R( )
satisfy the conditions 0.5Hz v 15Hz. For a given
be an autocorrelation function of X(t), and define the
wavelength, lets say = 4m, the rigid body vibration
corresponding power spectral density function as:
  of a vehicle are excited if the velocity of the vehicle is
1 2 varied from Vmin = 0.5Hz 4m = 2m/sec = 7.2km/h to
S() , lim E |F[XT (t)]| ,
T 2T Vmax = 15Hz 4m = 60m/sec = 216km/h. Hence, to

CSME-1374
achieve an excitation in the whole frequency range with and in the steady state the covariance of road irregularities
moderate vehicle velocities profiles with different varying is
wavelengths are needed. 2 w
When a vehicle is moving along the road with velocity lim E[zR (t)] = = 2 . (3.16)
t 2V
V , the excitation frequency of the road input (rad/sec)
becomes = V . The mean squared value of road sur- Remark 3.2: Recall that the variance of a discrete
face roughness, that is the total area of the power spectral stochastic process zRk and its continuous counter part zR
density function, does not change with the velocity of a is related by:
vehicle. Let () represents the power spectral density
of road input with respect to displacement excitation E[zRk zR T T
k ] = E[zR zR ] t, (3.17)
frequency. Therefore we have the following relation:
where t is the sampling time.
()d = ()d, (3.11)

which in turn yields the relationship between () and B. The Determination of Road Profile Variance
()
1 The most commonly used standard for pavement rough-
() = () . (3.12) ness is the proposed ISO 8608 [2]. It adopts the follow-
V
ing standard formulation to describe pavement roughness
Henceforth, we have PSD:
V
() = (0 )20

. (3.13) (0 )2 , for 0 1 ,
2

1 2
() = (0 )
This indicates that the road profile can be obtained from 0
, for 1 < N ,

integrating a white noise (i.e. a random walk) in time

0, for N < ,
domain. While to prevent standard deviation from going (3.18)
up with time as the integration period is increased. In [5, where the reference values of psd at 0 = 1(rad/m),
6] the road roughness PSD distribution is modified as (0 ), are given by ISO 8608 as shown in the Table III.
2V 2 The ISO suggested that 1 = 0.02(rad/m), and N =
() = , (3.14) 6(rad/m) [9]. As a result the variance of the random road
2 + 2 V 2
profile can be approximated by
where
Z  
2 denotes the road roughness variance and 1 (0 ) 2 1
2 = ()d = 5.05(0 ).
V the vehicle speed, whereas 2 0 2 1 N
depends on the type of road surface.
Since the spectral density of the road profile can be But for most of the literature, it was chosen that 2 =
factored as 4(0 ).
From (3.14), the one-sided PSD of road profile can be
2V 2 written in spatial frequency as
() = = H()w H T (),
(V + )(V )
2 2
where () = . (3.19)
2 + 2
H() , V 1+ is the frequency response function
of the shaping filter, To determine , we simply use the relationship
w , 2V 2 is the spectral density of a white noise
process. 2 2 2 4(0 )
(0 ) = = , (3.20)
Hence, if the vehicle runs with constant velocity ds
dt = V ,
20 + 2 20 + 2
then the road profile signal, zR (t), whose PSD is given
by (3.14), may be obtained as the output of a linear filter which yields = 0.127(rad/m). Note that in this case,
expressed by the differential equation [7, 8] is independent of the (0 ), that is the road class. While
in [5], the author adopted
d
zR (t) = V zR (t) + w(t), (3.15) 1) = 0.15m1 , 2 = 9mm2 , V = 10 50m/s that
dt
correspond to an asphalt road profile,
where w(t) is a white noise process with the spectral 2) = 0.45m1 , 2 = 300mm2 , V = 5 30m/s in
density w . It can be shown that the case of a paved road.
Z t
For the other class of road, the standard deviation of
zR (t) = eV t zR (0) + eV (t ) w( )d,
0
the corresponding road class can be found in Table III.

CSME-1375
 TABLE III
we have
ROAD ROUGHNESS S TANDARD D EVIATION Z
Jij = Ai sin(i s i )Aj sin(j s j )ds
Road Class (103 m) (0 )(106 m3 ), 0 = 1 (rad/m)
A (very good) 2 1 0.127 1 Ai Aj
B (good) 4 4 0.127 = sin(ij s ij )
C (average) 8 16 0.127 2 ij
D (poor) 16 64 0.127 1 Ai Aj
E (very poor) 32 256 0.127 + sin(i+j s i+j ),
2 i+j
where
IV. S INUSOIDAL A PPROXIMATION ij , i j , ij , i j .
Henceforth, we have
If the vehicle is assumed to travel with a constant
speed V over a given road segment with length L, a N N N
1 X X 1 XX X

random profile of a single track can be approximated by 2 = lim [Jii ]2 X + lim [Jij ]2 X
X X 2 X X 2
a superposition of N ( ) sine waves [4, 10] i=1 i=1 j=1
i6=j
N X
N
(
N
X X Ai Aj sin(ij X
2)
zR (s) = Ai sin(i s i ), (4.1) = cos(ij ) lim
i=1 j=1
ij X X
i=1
i6=j
where the amplitude Ai are defines as follows, ) N
Ai Aj sin(i+j X
2)
X 1
r + cos(i+j ) lim + A2i
i+j X X i=1
2
Ai = (i ) , i = 1, . . . , N, (4.2)
N
X 1
= A2i . (4.3)
N 1
in which , N 1 (rad/sec), and the phase angles i=1
2
i , i = 1, . . . , N are treated as random variables, follow-
Note that the sine and cosine terms are limited to values
ing a uniform distribution in the interval [0, 2).
of zero.
Proof: The random process generated by (4.1) can
Next, we determine the estimate of covariance from
be shown that it has zero mean as follows,
PSD. From (3.7) we know that
N Z N
1
X
E[zR (s)] = Ai E[sin(i s i )],
X
2 = R()|=0 = ()d (i ) .
i=1 2 0 i=1
2
N
Ai {sin(i s)E[cos i ] cos(i s)E[sin i ]} It is obvious that the variance of a sinusoidal approxima-
X
=
i=1
tion to a random road profile can be obtained by letting
= 0.
r

Ai = (i ) , i = 1, . . . , N, (4.4)

The variance of the sinusoidal representation is then given
by where the wave numbers i are chosen to lie at N equal
intervals .
Z L2 "X N
#
Remark 4.1: From the PSD defined by (2.1) or (3.18),
1
2 = lim Ai sin(i s i ) and the amplitude Ai given by (4.4), we have
L L L
2
i=1 1
N Ai , for i  1,

X
Aj sin(j s j ) ds i

j=1 which is similar to the property of Fourier series coeffi-

cient of a piecewise continuously function that itself has
For the case i = j: discontinuity [11].
Z Remark 4.2: It is easy to see that the road profile can
Jii = A2i sin2 (i s i )ds also be generated in the time domain as
N
A2i 1 X
= [i s i sin(2i s 2i )]. zR (t) = An sin(n0 t n ), (4.5)
2i 2
n=1
For the case i 6= j: use trigonometric relationship, where the fundamental temporal frequency
1 1 2
sin(x) sin(y) = cos(x y) cos(x + y), 0 , V , , ,
2 2 L

CSME-1376
 Grade B, =0.004
and r 0.02

An = (n ) , n = 1, . . . , N. 0.01

zR (m)
V. N UMERICAL E XAMPLES 0

Example 5.1: Figure 3 depicts the SIMULINK model 0.01

of the first order filter give by ( 3.15). A typical output of a 0.02

0 20 40 60 80 100 120 140 160 180
road profile generated from the filter is shown in Figure 4. s (m)
0
10
z_R

PSD () m2/m1
road profile
5
10
1
s
Random Integrator r_R
Number
10
10 2 1 0 1 2
10 10 10 10 10
K V
frequency (rad/m)
alpha velocity

Fig. 5. zR and the PSD generated from the sinusoidal approximation

V spatial (4.1)
Clock Gain1 spatial coordinate

Fig. 3. First order linear system (3.15) Simulink model

of the related PSD, which is similar to the property of
Fourier series coefficient.

Grade B, V=60(km/h), =0.004

ACKNOWLEDGEMENTS
0.02
This research was supported by the Industrial Technol-
0.01 ogy Research Institute through Grant ITRI 5353C46000.
The authors acknowledge the support of Department of In-
zR (m)

0
dustrial Technology of the Ministry of Economic Affairs,
0.01 R.O.C., through Science Technology program.
0.02
0 1 2 3 4 5 6 7 8 9 10
R EFERENCES
time (sec)

10
0 [1] M. W. Sayers and S. M. Karamihas, The little
book of profiling, basic information about measuring
PSD (f) m2/m1

and interpreting road profiles, The Regent of the

10
5
University of Michigan, Tech. Rep., September 1998.
[2] ISO, Reporting vehicle road surface irregularities,
Technical Report, ISO, ISO/TC108/SC2/WG4 N57,
10
10 1 0 1 2 1982.
10 10 10 10
frequency (Hz) [3] A. Papoulis, The Fourier Integral and Its Applica-
tions. New York: McGraw-Hill, June 1962.
Fig. 4. zR and the PSD generated from the shaping filter (3.15)
[4] Y. K. Wen and P. Gu, Description and simulation
of nonstationary processes based on hilbert spectra,
Example 5.2: A typical grade B road surface created Journal of Engineering Mechanics, vol. 130, no. 8,
by sinusoidal approximation is shown in Figure 5. pp. 942951, 2004.
[5] A. Hac, Suspension optimization of a 2-dof vehicle
VI. C ONCLUSIONS model using a stochastic optimal control technique,
Two of the most commonly adopted methods, namely Journal of Sound and Vibration, vol. 100, no. 3, pp.
shaping filter and sinusoidal approximation, for generating 343357, 1985.
random road profiles. For the shaping filter we found [6] A. Giua, M. Melas, C. Seatzu, and G. Usai, Design
that the time constant of the associated first-order sys- of a predictive semiactive suspension system, Vehicle
tem transfer function is independent of the road profile System Dynamics, vol. 41, no. 4, pp. 277300, 2004.
grade. However, most of the literature suggest that the [7] G. Corriga, S. Sanna, and G. Usai, An optimal
time constant is road profile or even vehicle traveling tandem active-passive suspension system for road
speed dependent. In the sinusoidal approximation, for long vehicles with minimum power consumption, IEEE
enough load profile, we confirmed that the amplitude of Trnasactions on Industrial Electronics, vol. 38, no. 3,
each sinusoidal function is proportional to the square root pp. 210216, June 1991.

CSME-1377
 [8] L. J. Zhang, C. M. Lee, and Y. S. Wang, A study
on nonstationary random vibration of a vehicle in
time, International Journal of Automotive Technol-
ogy, vol. 3, no. 3, pp. 101109, 2002.
[9] L. Sun and T. W. Kennedy, Spectral analysis and
parametric study of stochastic pavement loads, Jour-
nal of Engineering Mechanics, vol. 128, no. 3, pp.
318327, 2002.
[10] G. Verros, S. Natsiavas, and C. Papadimitriou, De-
sign optimization of quarter-car models with passive
and semi-active suspensions under random road ex-
citation, Journal of Vibration and Control, vol. 11,
pp. 581606, 2005.
[11] M. D. Greenberg, Advanced Engineering Mathe-
matics, 2nd ed. Upper Saddle River, New Jersey:
Prentice Hall, 1998.

,
,
()
,

CSME-1378