Professional Documents
Culture Documents
Optimization of Speed Control Hump Profiles: Journal of Transportation Engineering May 2009
Optimization of Speed Control Hump Profiles: Journal of Transportation Engineering May 2009
net/publication/245307076
CITATIONS READS
8 1,831
2 authors:
Some of the authors of this publication are also working on these related projects:
All content following this page was uploaded by Selahattin Caglar Baslamisli on 11 January 2015.
Abstract: In this study, the goal attainment method is implemented for the optimization of speed control hump profiles. Basic dimensions
for a number of hump profile functions are optimized for single vehicles classified in five categories and for a specified distribution of
these vehicles. Objective functions are selected as combinations of the longitudinal and vertical acceleration components at pre-specified
Downloaded from ascelibrary.org by Middle East Technical University on 10/21/14. Copyright ASCE. For personal use only; all rights reserved.
points on the vehicle body. The main objective is to minimize vehicle response functions below the hump crossing speed limit and to
maximize them above the speed limit. Typical characteristics of a series of two axle vehicles, described by the half car model, are used
and the performances of various hump profiles are assessed.
DOI: 10.1061/共ASCE兲TE.1943-5436.0000002
CE Database subject headings: Traffic speed; Optimization; Traffic management.
Introduction tion of its suspension. The end result was an optimal circular
hump that could effectively punish high-speed crossings of light
Speed control humps and bumps are common means of regulating vehicles equipped with semiactive suspensions, while substan-
the speed of traffic flow. Speed humps are characterized by much tially improving the ride comfort of heavy vehicles at crossing
longer lengths as shown in Fig. 1 and provide for better ride speeds below the prescribed hump crossing speed.
comfort compared to speed bumps, as long as speed limits are Pedersen 共1998兲 proposed that the objective be a ride as pleas-
obeyed. Drivers know by experience that, at high crossing speeds, ant as possible while going over the hump below the speed limit
humps cause large vehicle body pitching motion, large suspension and a disturbing ride when going too fast. The objective function
travel, and may further result in wheels losing contact with the
was selected as the maximum acceleration that the driver’s head
road surface. Speed control bumps, on the other hand, offer a
would experience in a ride over the hump. The speed limit was
harsh effect on rides at low crossing speeds and may lose their
selected as 30 km/ h and the desired level of acceleration upon
effectiveness at higher crossing speeds.
Currently used humps are designed basically by experimenta- exceeding the speed limit was arbitrarily fixed at 40 m / s2. The
tion as proposed by Fwa and Tan 共1992兲 and Smith and Giese optimization procedure included two kinds of vehicles, namely a
共1997兲. Surprisingly, limited theoretical work can be found in the sports utility vehicle and a generic car, and hump lengths were
literature on the optimization of speed control hump profiles. fixed at 4 m. The shape of the humps was controlled by ampli-
Maemori and Ando 共1987兲 claimed that both speed humps and tudes of basis orthogonal functions. The author concluded that
speed bumps caused excessive shocks for drivers of heavy ve- significant improvement could be made on standard hump de-
hicles and that semiactive suspensions of light vehicles could im- signs.
prove ride characteristics and decrease the shocks experienced by Khorshid and Alfares 共2004兲 considered the shape optimiza-
the drivers of these vehicles at crossing speeds far above the tion of Watts 共a symetrical hump with sinusoidal or harmonic
speed limit. Their aim was to determine the optimal circular hump rising and returning parts兲, flat topped 共a flat topped hump with
dimensions in order to reduce the excessive shocks experienced sinusoidal or harmonic rising and returning parts兲, and polyno-
by drivers of heavy vehicles passing current humps at normal mial humps following basically the same approach as presented
speed and to compensate for the too few shocks experienced by by Pedersen 共1998兲, i.e., shaping of the driver’s head acceleration
light vehicles with semiactive suspensions passing over the same
versus speed curve. A set of 42 optimal designs were found by
humps at excessive speeds. The maximum acceleration of the
combining different rise/return profiles for the three types of
heavy vehicle driver’s seat and that of the light vehicle driver’s
humps above stated. Again, the writers concluded that significant
seat were set as a multiobjective function for optimization of the
hump. The maximum acceleration of the light vehicle driver’s improvement could be made on standard hump designs
seat was selected as a single objective function for the optimiza- It should be noted that in all the above stated studies, the
writers considered optimization of hump profiles using at most
1 two individual vehicles. In the present study, several objective
Instructor, Mechanical Engineering Dept., Hacettepe Univ., Ankara
06800, Turkey. functions, i.e., horizontal and vertical components of accelera-
2
Professor, Mechanical Engineering Dept., Middle East Technical tions of points selected on the vehicle body, are implemented in a
Univ., Ankara 06531, Turkey 共corresponding author兲. E-mail: unlusoy@ multiobjective optimization procedure, namely the goal attain-
metu.edu.tr ment method 共GAM兲 first proposed by Gembicki 共1974兲, to allow
Note. Discussion open until October 1, 2009. Separate discussions them to follow initially determined ideal curves as closely as
must be submitted for individual papers. The manuscript for this paper
possible. Efforts have been made to extend the number and range
was submitted for review and possible publication on March 4, 2008;
approved on September 5, 2008. This paper is part of the Journal of of vehicles considered in order to provide a method applicable to
Transportation Engineering, Vol. 135, No. 5, May 1, 2009. ©ASCE, the variety of vehicles in traffic for the optimization of speed
ISSN 0733-947X/2009/5-260–269/$25.00. control hump profiles.
Fig. 1. Speed hump versus speed bump 共1兲 defining the horizontal distance between the driver’s seat and
the center of gravity of the vehicle body and suspension charac-
teristics of the driver’s seat; and 共2兲 an anatomical model for the
Vehicle Model and Specifications driver. To relieve problems of generalization introduced by such
Downloaded from ascelibrary.org by Middle East Technical University on 10/21/14. Copyright ASCE. For personal use only; all rights reserved.
冤 冥再 冎
0 0
0 0 yf
兵F其 = 关H兴兵y其 = 共3兲
k3 0 yr
0 k3
冤 冥
M 0 0 0
0 I 0 0
关M兴 = 共5兲
0 0 m1 0
0 0 0 m2
冤 冥
c1 + c2 − ac1 + bc2 − c1 − c2
− ac1 + bc2 a2c1 + b2c2 ac1 − bc2
关C兴 = 共6兲
− c1 ac1 c1 0
Fig. 2. Half-car model − c2 − bc2 0 c2
冤 冥
k1 + k2 − ak1 + bk2 − k1 − k2 the rear suspension; 共2兲 pitch and bounce frequencies should be
2 2 close; and 共3兲 neither frequency should be greater than 1.3 Hz for
− ak1 + bk2 a k1 + b k2 ak1 − bk2
关K兴 = 共7兲 passenger cars.
− k1 ak1 k1 + k3 0 An iterative method was implemented to achieve the conve-
− k2 − bk2 0 k2 + k4 nient ranges of frequencies and deflections for each vehicle. In
Table 1, Vehicle A represents a medium size passenger car, Ve-
hicle B a small size passenger car, Vehicle C a large size passen-
Selection of Vehicle Categories ger car, Vehicle D a small size commercial vehicle, and Vehicle E
a large size commercial vehicle.
The range of vehicle types on the roads extends from the light
passenger car to the heavy commercial vehicle. Ideally, an ex-
haustive study would have consisted of collecting a large amount Hump Shapes
of vehicle data in order to perform the optimization calculations.
However, vehicles with varying specifications do not always dif- The list of commonly used speed control hump shapes includes
fer with respect to their main vibrational characteristics. sinusoidal, circular, and parabolic humps; each with two param-
For the purposes of this paper, the resonant frequencies of eters to be optimized, the base width, L, and the height, H, as
interest are those related to the bounce and pitch of the sprung shown in Fig. 3. The common use for these profiles is basically
mass on the suspensions, and the motion of the unsprung masses due to their simplicity and the resulting ease of construction. Just
over the combined tire and wheel compliance. Suspension design to examine if there may be others substantially superior to these
generally starts with the selection of a convenient body bounce simple and commonly used profiles, and being guided by Khor-
frequency and it has been known that the lower the frequency of shid and Alfares 共2004兲, it is decided to consider a number of
vibration of the car body on its suspension spring, the greater the new, but somewhat more complicated profiles.
comfort. In modern medium to large size cars, the body bounce In this study, combination hump profiles, i.e., sine-parabola,
frequency has settled to approximately 1.2 Hz 共Gillespie 1992兲. and trapezoidal profiles have been added to the list. The former is
Meanwhile, higher bounce frequencies are observed in commer- built up from ascending and descending sinusoidal parts and a
cial vehicles as their suspensions must be stiff enough to prevent parabolic top. Its design parameters are the imaginary sinusoidal
excessively large deflections. Further, studies have shown that the half cycle length L1, length of sinusoidal part L 共defining k
wheel hop 共the up and down motion of the wheel on the com- = L1 / L兲, the half length of parabolic part L2, the imaginary height
bined elasticity of the suspension and tires兲, frequency is roughly of sinusoidal half cycle H1, and the height of parabolic part H2, as
ten times greater than the body bounce frequency. Given these shown in Fig. 4. The latter consists of the horizontal length of a
guidelines, specifications of five types of vehicles were designed ramp section L1, the half length of the flat top L2, and the height
also based on the Olley criteria 共Milliken et al. 2002兲 stating that: H, as illustrated in Fig. 5.
共1兲 the front suspension should have a slightly lower ride rate than The idea of optimizing the parameters of a parabola is then
extended to consider higher order polynomials. An algorithm is
devised to handle polynomials up to 12th order. The dimensions
of the polynomial hump are the base width L, the overall height
H, and some polynomial coefficients an, as shown in Fig. 6. such that Fi共x兲 − wi* ␥ 艋 Fi* i = 1, . . . m 共8兲
Finally, as a general shape optimization problem, coordinates
of points describing the hump profile are selected as the variable The term 共wi·␥兲 introduces an element of slackness into the
parameters. Discretely defined points on the hump profile are con- problem, which otherwise imposes that the goals be rigidly met.
Downloaded from ascelibrary.org by Middle East Technical University on 10/21/14. Copyright ASCE. For personal use only; all rights reserved.
nected by straight lines. These synthesized hump profiles have The weighting vector w enables the designer to express a measure
variable parameters stored in two vectors: one for the horizontal of the relative tradeoffs between the objectives. Hard constraints
coordinates and another for the vertical coordinates of the points. can be incorporated into the design by setting a particular weight-
The number of points to be considered depends on the accuracy ing factor close to zero. GAM provides a convenient intuitive
of definition required, as shown in Fig. 7. interpretation of the design problem, which can be solved using
standard optimization procedures, like sequential quadratic pro-
gramming 共SQP兲.
Optimization
Application of Goal Attainment Method to Optimization
The techniques for multiobjective optimization are wide and var- of Speed Control Hump Profiles
ied, and the GAM first proposed by Gembicki 共1974兲 has been
We let
implemented in this study. GAM differs from the commonly used
weighted sum strategy, which converts the multiobjective prob- F1共x,Vi兲 = maximum absolute vertical acceleration at speed Vi;
lem of minimizing the vector of objectives F(x) into a scalar
problem by constructing a weighted sum of all the objectives. The
F2共x,Vi兲 = maximum absolute longitudinal acceleration
following describes how GAM has been applied to the present
problem. at speed Vi
of a point D, selected around the driver’s seat location on the
Descriptive Explanation of GAM „Gembicki 1974… plane of the vehicle body, constitute response functions to be
optimized. Here, x denotes the vector of hump dimensions and
Multiobjective optimization problems deal with the tradeoff of
Vis consist of selected vehicle longitudinal speeds.
objectives in a vector F(x) = 兵F1共x兲 , F2共x兲 , . . . , Fm共x兲其 can be de-
Here, responses F1共x , Vi兲 and F2共x , Vi兲 should be minimized
fined. F(x) may be subject to a number of constraints and bounds
below and maximized above the selected reference speed limit. In
Gi共x兲.
other words, the efficiency of the hump is assessed according to
It should be noted that because F(x) is a vector, if any of the
the slope of the response curve around the speed limit; the higher
components of F(x) are competing, there is no unique solution to
the value of the slope, the greater the efficiency of the hump will
this problem. Instead, the concept of noninferiority, also called
be.
Pareto optimality, must be used to characterize the objectives. A
One way of defining ideal response curves is the selection of
noninferior solution is one in which an improvement in one ob-
goal points. To be considered as ideal, responses should follow
jective requires a degradation of another. GAM like other multi-
goal points as closely as possible. In the present case, an ideal
objective optimization techniques is concerned with the
curve is any curve that remains between the boundaries shown on
generation and selection of noninferior solution points, a basic
Fig. 8, while following the goal points as closely as possible.
step leading to determination of the optimal solution.
Thus, the goal of the designer is to construct response curves by
GAM involves expressing a set of design goals, F* forcing them to pass through several goal points in order for them
= 兵F*1 , F*2 , . . . , Fm* 其, which is associated with a set of objectives to remain between the two design boundaries. Note that the goal
F共x兲 = 兵F1共x兲 , F2共x兲 , . . . , Fm共x兲其. The problem formulation allows points have to be selected differently for the maximum vertical
the objectives to be under- or overachieved enabling the designer acceleration and the maximum longitudinal acceleration cases as
to be relatively imprecise about initial design goals. The relative vertical acceleration levels recorded during hump crossing are
degree of under- or overachievement of the goals is controlled by substantially greater than longitudinal acceleration levels. 共The
physical explanation lies in the fact that the vehicle model expe-
riences direct vertical excitations from the road while the horizon-
tal acceleration is due to pitching caused by the lag between front
and rear suspension excitations. Obviously the latter, which is a
rather indirect effect, is less pronounced than the former.兲
In Fig. 8, F*n 共V兲 is either F*1 共V兲 or F*2 共V兲. V0 = speed value at
which the analysis begins; V1 = selected speed limit and represents
the lower limit of the interval in which the objective is mini-
mized; and V2 = upper bound of the speed interval where the vi-
brational response functions are to be maximized and is selected
as close as possible to V1 while fixing optimization goal points
Fig. 7. Synthesized hump profile F*n 共V1兲 and F*n 共V2兲 as far from each other as possible, in the
␥ = max共Ck兲 共11兲
k
Solution Technique
The solution technique is explained in what follows and is de-
picted in Fig. 10.
1 The procedure starts with the selection of the hump crossing
speed limit V1 and the speed V2;
2. Next, F*1 共V1兲, F*1 共V2兲, F*2 共V1兲, and F*2 共V2兲 must be set. In
selecting these goals for longitudinal and vertical accelera-
tions, we make use of an iterative procedure. We may start
with goals that are relatively difficult to achieve 关hence se-
lecting F*1 共V1兲 and F*2 共V1兲 as low as possible while fixing
F*1 共V2兲 and F*2 共V2兲 as high as possible兴;
3 At the onset of the optimization, we also select ␥*, which
stands for an upper limit for our objective function C and the
maximum number of iterations nmax. The attainment or non-
attainment of ␥* will determine if we should modify our
initial settings as explained below. It may happen that the
optimization routine becomes stuck close to ␥*. This is the
reason why we have to set the stopping criterion n ⬎ nmax;
and
4 Observing the output of the optimization procedure, we de-
cide whether we have to select easier goals 关setting the pair
F*1 共V1兲-F*2 共V1兲 higher and/or setting the pair F*1 共V2兲-F*2 共V2兲
lower兴 or even harder goals 关setting the pair F*1 共V1兲-F*2 共V1兲
lower and/or setting the pair F*1 共V2兲-F*2 共V2兲 higher兴. Another
possibility may be to reset the value of V2.
The above has been applied time and again in the “Case Stud-
ies.” Values for V2, F*1 共V1兲, F*1 共V2兲, F*2 共V1兲, and F*2 共V2兲 that have
been used in the next section are the result of this repetitive
Fig. 9. Iterative solution method procedure.
Case Study 1
The optimization algorithm is run for all vehicles simultaneously
for
• V1 = 18 Km/ h; V2 = 25.2 Km/ h; V3 = 36 Km/ h; V4 = 54 Km/ h;
• F*1 共V1兲 = 3 m / s2; F*1 共V2兲 = 8 m / s2; and
• F*2 共V1兲 = 1 m / s2; F*2 共V2兲 = 4 m / s2.
The optimized sine-parabola hump profile turns out to be the
most suitable hump profile for Case Study 1, displaying the best
vertical acceleration behavior. The vehicle responses over the op-
Downloaded from ascelibrary.org by Middle East Technical University on 10/21/14. Copyright ASCE. For personal use only; all rights reserved.
Case Study 2
The analysis is repeated for:
1. A typical sinusoidal hump, in current use in the United King-
dom 共4 in. in height and 12 ft in length兲;
2. Optimal sine parabola humps of unconstrained dimensions
optimized for each vehicle type separately; and
3. Sine parabola humps of dimensions comparable to the sinu-
soidal hump 共sizes corresponding to limited length in Table
2兲, and optimized for each vehicle type separately.
Vertical acceleration response are given in Fig. 11 where signifi-
cant improvement in hump performance can be observed over the
standard sinusoidal hump by using optimized sine parabola pro-
files for speed control purposes.
Case Study 3
Finally, it is desired to set the speed limit to 50 Km/ h on a resi-
Fig. 10. Optimal sine-parabola hump for Case Study 1: 共a兲 hump dential road where frequent pedestrian crossings are present. The
profile; 共b兲 enlarged scale in y-direction; 共c兲 F1 response; and 共d兲 F2 hump crossing speed limit is desired to be first about 25 Km/ h
response 共L1 = 4.000 m L2 = 2.745 m; H1 = 0.181 m; H2 = 0.316 m; and then 30 Km/ h for all kinds of vehicles that ordinarily use the
k = 2.909; cost function ␥ = 0.19兲 road. A 1-week survey showed the following distribution of ve-
hicles crossing the hump:
• Small passenger cars 共Vehicle A兲: 25%;
• Medium-size passenger cars 共Vehicle B兲: 35%;
• Large-size passenger cars 共Vehicle C兲: 20%;
• Small-size commercial vehicles 共Vehicle D兲: 15%; and
Limited length L1 = 1.999; L2 = 0.9571; H1 = 0.1762; H2 = 0.0243; k = 2.7596; Lover = 3.4 m; Hover = 0.18 m
B Optimal L1 = 4.0196 m; L2 = 2.7379 m; H1 = 0.1343 m; H2 = 0.2658 m; k = 2.8722; Lover = 8.2 m;
Hover = 0.38 m
Limited length L1 = 1.9542 m; L2 = 0.9989 m; H1 = 0.121 m; H2 = 0.0201 m; k = 2; Lover = 3.9 m; Hover = 0.14 m
E Optimal L1 = 4.5977 m; L2 = 4.6805 m; H1 = 0.1767 m; H2 = 0.5629 m; k = 4; Lover = 12 m; Hover = 0.68 m
Fig. 11. Comparison for various F1 responses over various types of humps for Case Study 2: 共a兲 Vehicle A; 共b兲 Vehicle B; 共c兲 Vehicle C; 共d兲
Vehicle D; and 共e兲 Vehicle E
• Large-size commercial vehicles 共Vehicle E兲: 5%. taken as the initial guess. Results for the trapezoidal hump are
For the case where V1 = 25 Km/ h the settings are as follows: shown in Fig. 12. The profile of the optimal synthesized hump is
• V1 = 25 Km/ h; V2 = 30 Km/ h; V3 = 40 Km/ h; V4 = 54 Km/ h; listed in Table 3, and is plotted in Fig. 13.
• F*1 共V1兲 = 4 m / s2; F*1 共V2兲 = 8 m / s2; and As the percentage of heavy commercial vehicles likely to cross
• F*2 共V1兲 = 1 m / s2; F*2 共V2兲 = 4 m / s2. the hump is only 5%, the optimization is implemented to elimi-
Optimization weights are selected according to the vehicle dis- nate these in determining the optimal hump profile. However, the
tribution: the higher the frequency of a vehicle type traveling on a effect of the hump on this category of vehicles is later assessed by
given road section, the lower the weight. Hence, the lowest plotting their acceleration response. As can be seen in Figs. 12
weights are attributed to the goals concerning the performance of and 13, the optimal hump profiles seem to offer harsh rides for
Vehicle B in this case. these vehicles as expected.
The most suitable profiles for V1 = 25 Km/ h turn out to be a The same calculations are finally performed for V1
trapezoidal and the synthesized hump profile. In synthesizing the = 30 Km/ h with the following settings:
latter, coordinates of the optimized trapezoidal hump have been • V1 = 30 Km/ h; V2 = 35 Km/ h; V3 = 50 Km/ h, V4 = 54 Km/ h;
Fig. 12. Optimal trapezoidal hump profile for Case Study 3: 共a兲 Fig. 13. Optimal synthesized hump for Case Study 3: 共a兲 hump pro-
hump profile; 共b兲 enlarged scale in y-direction; 共c兲 F1 response; and file; 共b兲 enlarged scale in y-direction; 共c兲 F1 response; and 共d兲 F2
共d兲 F2 response for trapezoidal hump 共L1 = 4.579 m, L2 = 0.922 m, response
H = 0.478 m, Lover = 11.000 m; cost function ␥ = 0.19兲
References
Notation
Fwa, T. F., and Tan, L. S. 共1992兲. “Geometric characterization of road
The following symbols are used in this paper: humps for speed-control design.” J. Transp. Eng., 118, 593–598.
a ⫽ horizontal distance from front suspension to Gembicki, F. W. 共1974兲. “Vector optimization for control with perfor-
vehicle body center of gravity; mance and parameter sensitivity indices.” Ph.D. dissertation, Case
an ⫽ coefficient for polynomial hump; Western Reserve Univ., Cleveland.