Model Based Estimation of Height

of Center of Gravity in Heavy

Vehicles

DELEER BARAZANJI

Master of Science Thesis

Stockholm, Sweden 2012
Model Based Estimation of Height of
Center of Gravity in Heavy Vehicles

Deleer Barazanji

MSc Thesis
Optimization and Systems theory

KTH

March 21, 2012

ii
Abstract

The center of gravity height in a vehicle affects its dynamic driving properties
but there is no accurate way of measuring the height of center of gravity
today. One example of vehicle stabilizing systems is vehicle rollover warning
and assist system which has to rely on a relatively accurate height of center of
gravity estimate in order to be implemented in vehicles efficiently and would
otherwise be considered useless. In this thesis a literature study on the height
of center of gravity in heavy vehicles in general and semitrailers in particular
is conducted at Scania CV and emphasis is towards a model relying as
little as possible on data from outside the tractor. A partly new model for
detecting the vehicle’s axle loads at different acceleration values is developed
and compared to other models, pros and cons are examined, furthermore an
estimation tool is developed for the new model in a realistically applicable
manner with regards to normal driving situations and solution limitations.
The estimation tool is tested on Scania semitrailers with different suspension
configurations and the result shows that the height of center of gravity can
be estimated as close as 4.1 (cm) from the real value for 4x2 Gen 2 Scania
tractor and 3.3 (cm) for 4x2 Gen 3 Scania tractor.

iii
iv
Acknowledgments

For the opportunity to work on this interesting subject in a motivational

and highly supportive environment I would like to thank Scania CV and my
supervisor Joseph Ah-King at REPA, all the employees at REPA, REPI,
REVC and REVM as well as the following people at Scania, Martin Sund-
berg, Jolle Ijkema and Per Elofsson for helping me in various ways in writing
this thesis. I would also like to express my gratitude to my supervisor Hu
Xiaoming at the division of Optimization and Systems theory at the Royal
Institute of Technology KTH, and Lars Drugge from the department of Ve-
hicle Dynamics at the Royal Institute of Technology KTH for providing
valuable expertise and new ideas.

v
vi
Contents

1 Introduction 7
1.1 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.1.1 Rollover . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.1.2 Static roll stability . . . . . . . . . . . . . . . . . . . . 9
1.2 Objectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10

2 Model comparison 11
2.1 Literature survey . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.2 Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.2.1 Lateral model . . . . . . . . . . . . . . . . . . . . . . . 11
2.2.2 Longitudinal model . . . . . . . . . . . . . . . . . . . . 13
2.2.3 Model complications . . . . . . . . . . . . . . . . . . . 14
2.2.3.1 Road grade . . . . . . . . . . . . . . . . . . . 14
2.2.3.2 Weight transfer . . . . . . . . . . . . . . . . 14
2.2.3.3 Acceleration . . . . . . . . . . . . . . . . . . 14
2.2.3.4 Roll angle . . . . . . . . . . . . . . . . . . . . 15
2.2.3.5 Torsional stiffness . . . . . . . . . . . . . . . 15
2.2.3.6 Rearward amplification . . . . . . . . . . . . 15
2.2.3.7 Suspension . . . . . . . . . . . . . . . . . . . 15
2.2.3.8 Load distribution . . . . . . . . . . . . . . . 16
2.2.3.9 Various other factors . . . . . . . . . . . . . 16
2.3 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16

3 Longitudinal model 17
3.1 Weight transfer . . . . . . . . . . . . . . . . . . . . . . . . . . 18
3.1.1 Tractor weight transfer . . . . . . . . . . . . . . . . . 18
3.1.2 Trailer weight transfer . . . . . . . . . . . . . . . . . . 19
3.1.3 Fifth wheel forces . . . . . . . . . . . . . . . . . . . . . 20
3.2 Parameter estimation . . . . . . . . . . . . . . . . . . . . . . 23
3.2.1 Vehicle weight estimation . . . . . . . . . . . . . . . . 23
3.2.2 Weight transfer detection and estimation . . . . . . . 23
3.2.2.1 Air suspension . . . . . . . . . . . . . . . . . 23
3.2.2.2 Suspension leaf springs . . . . . . . . . . . . 25
3.2.3 Longitudinal forces at the wheels . . . . . . . . . . . . 25

1
 CONTENTS

3.2.3.1 Tractive force . . . . . . . . . . . . . . . . . . 26

3.2.3.2 Brake force . . . . . . . . . . . . . . . . . . . 26
3.2.3.3 Retarder force . . . . . . . . . . . . . . . . . 26
3.2.4 Air resistance . . . . . . . . . . . . . . . . . . . . . . . 27
3.2.5 Road grade . . . . . . . . . . . . . . . . . . . . . . . . 28
3.2.6 Acceleration . . . . . . . . . . . . . . . . . . . . . . . . 28
3.2.7 Fifth wheel height . . . . . . . . . . . . . . . . . . . . 29
3.3 Complete model . . . . . . . . . . . . . . . . . . . . . . . . . 30

4 Results and analysis 33

4.1 Equipment . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
4.2 Test procedure . . . . . . . . . . . . . . . . . . . . . . . . . . 34
4.3 Data processing . . . . . . . . . . . . . . . . . . . . . . . . . . 34
4.3.1 Lateral and vertical acceleration . . . . . . . . . . . . 34
4.3.2 Acceleration . . . . . . . . . . . . . . . . . . . . . . . . 35
4.4 Brake calibration . . . . . . . . . . . . . . . . . . . . . . . . . 35
4.5 HCG . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
4.6 Error analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . 38

5 Conclusions 41

6 Recommendations and future work 43

A Appendix A Air spring Configuration 47

B Appendix B Vehicle data 51

B.1 Vehicle data . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51

C Appendix B Brake calibration 55

2
List of Figures

1.1 Weight transfer between the left and right wheels of a vehicle
during rolling and the effect of the HCG as well as laterally
displaced CG on mentioned weight transfer. . . . . . . . . . . 8
1.2 Figure of a truck in a turn with a laterally displaced CG. . . 9

2.1 Figure of a truck in a turn . . . . . . . . . . . . . . . . . . . . 12

2.2 Figure of a truck in a left turn. . . . . . . . . . . . . . . . . . 13
2.3 A semitrailer on a sloping road. . . . . . . . . . . . . . . . . . 14

3.1 Figure of a tractor and model related forces acting on it dur-

ing driving. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
3.2 Figure of a trailer with model forces pointed out. . . . . . . . 19
3.3 Figure of a trailer with attracting relation between trailer and
tractor. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
3.4 Figure of a trailer with repelling relation between trailer and
tractor. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
3.5 The relationship between the force on the bellow and the air
pressure & height of the air spring. . . . . . . . . . . . . . . . 24
3.6 Leaf feather on the tractor’s front axle. . . . . . . . . . . . . . 25
3.7 Different air suspension configurations in Scania’s tractors. . 25
3.8 The air resistance force and placement. . . . . . . . . . . . . . 27
3.9 Figure explaining a way of obtaining the road grade, θ =
arcsin( ∆h
∆s ). . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
3.10 Figure related to the geometrics of the suspension configura-
tion showing the vertical change in the drive axle’s position. . 30
3.11 Diagram illustrating the steps towards the solution. . . . . . . 31
3.12 The force N2 /g as a function of acceleration a for different
situations and HCG positions. . . . . . . . . . . . . . . . . . . 32

4.1 The lateral acceleration ay before and after smoothing. . . . . 34

4.2 The acceleration a before and after filtering. . . . . . . . . . . 35
4.3 Adjustable frame with blocks inside the trailer. . . . . . . . . 37
4.4 The slopes j obtained by linear regression for the different
HCG values compared to the real slopes. . . . . . . . . . . . . 38

3
 LIST OF FIGURES

4.5 The slopes j obtained by linear regression for the different

HCG values compared to the real slopes. . . . . . . . . . . . . 39
4.6 The linear regression with slope j. . . . . . . . . . . . . . . . . 40
4.7 The linear regression with slope j. . . . . . . . . . . . . . . . . 40

A.1 Gen 2 Scania tractor suspension configuration. . . . . . . . . 48

A.2 Gen 3 Scania tractor suspension configuration. . . . . . . . . 49

B.1 Conceptus with test trailer. . . . . . . . . . . . . . . . . . . . 51

B.2 Electra with test trailer. . . . . . . . . . . . . . . . . . . . . . 52
B.3 Weight blocks inside the test trailer’s container. . . . . . . . . 52

C.1 Tractor brake calibration. . . . . . . . . . . . . . . . . . . . . 55

C.2 Trailer brake calibration. . . . . . . . . . . . . . . . . . . . . . 56

4
List of notations

i=1,2,3,4 denote the tractor’s front axle, rear axle, the fifth wheel and the
trailer’s axles. x1 is the wheel base of the tractor. (m)
x2 is the distance from the tractor’s CG position to the tractor’s rear axle.
(m)
x3 is the fifth wheel to the tractor’s rear axle distance. (m)
x4 is the trailer’s CG to the trailer’s rear axle distance. (m)
x5 is the trailer’s rear axle to fifth wheel distance. (m)
y1 is the HCG of the tractor. (m)
y2 is the fifth wheel height. (m)
y3 is the HCG of the trailer. (m)
Ni is the vertical force on i. (N)
Fi is the longitudinal to road force at the front wheels of i.(N)
t is the track width of the trailer. (m)
ay is the lateral acceleration of the vehicle. ( sm2 )
a is the longitudinal acceleration of the vehicle. ( sm2 )
g is the gravitational constant. ( sm2 )
md is the mass of the tractor. (kg)
mt is the mass of the trailer. (kg)
m is the total mass of the vehicle. (kg)
r is the wheel radius. (m)
t is the track width of the trailer. (m)
s is the lateral CG displacement from the centerline through RC. (m)
Fr is the normal force on the right wheels. (N)
Fl is the normal force on the left wheels. (N)
p is the pressure applied by the brakes at the wheels. (bar)
pi is the pressure applied by the brakes at the wheels on axle i. (bar)
ps is the pressure required to start braking the wheels. (bar)
psi is the pressure required to start braking the wheels at axle i. (bar)
Tbi is the brake torque per pressure. ( Nbar
m
)
ms is the sprung weight of a vehicle. (kg)
k is the spring stiffness. (N/m)
Fb is the brake force. (N)
Te is the engine torque. (Nm)
Ft is the tractive force. (N)

5
 LIST OF FIGURES

Ut is the total transmission gain, i.e gear box and differential gear.
Fh is the retardation force. (N)
Th is the retarder torque. (Nm)
Ud is the differential gear gain.
Cd is the air resistance coefficient.
v is the vehicle speed. (m/s)

6
Chapter 1

Introduction

1.1 Background

This report is written with the purpose of estimating the HCG1 of heavy ve-
hicles, in particular a semitrailer as the one depicted in figure 2.3. Semitrail-
ers respond to driver actions in a certain way in different driving situations
and for different load distribution scenarios. Estimation of the position of
the CG is therefor fundamental in adapting vehicle control and safety sys-
tems to different load situations. One possible such application of the former
is the development of a heavy vehicle rollover system that can predict vehicle
rollover beforehand to either warn the driver or intervene by assisting the
driver in keeping control over the vehicle at an early stage to prevent full
rollover. A rollover prediction system for semitrailers is very hard to imple-
ment given only information from the sensors and control units mounted on
the tractor unit alone, today’s trailers and tractors are barely sharing any
information in-between, finding alternative ways to reduce the information
needed to be known from the trailer unit as much as possible is a must
since the criteria is to be able to couple any trailer and by only observing
the movement of the vehicle and signals and sensors available to the tractor
today be able to estimate the HCG position.

1.1.1 Rollover

Rollover of heavy vehicles is one of the most dangerous automotive vehi-

cle accident types leading to much higher fatality rates than other types of
accidents, in 2004 the fatality rate of heavy vehicle drivers in the US was
0.0482% which is about 11 times higher than the fatality rate in other oc-
cupations [1]. In the US alone there are more than 3 million heavy vehicle
1
Height of Center of Gravity.

7
 Chapter 1

Figure 1.1: Weight transfer between the left and right wheels of a vehicle
during rolling and the effect of the HCG as well as laterally displaced CG
on mentioned weight transfer.

drivers [2], this figure gives more than 1500 fatalities yearly based on 2004
year’s statistics and emphasizes the importance of a solution. Rollover of
heavy vehicles usually occur at highway exit and entrance ramps and on
other curve segments of roads where the vehicle tilts sideways due to the
CG rotating around the RC2 opposite to the turning radius direction due
to the centrifugal force opposing the lateral acceleration pointing towards
the center of the turn, see figure 2.1. The most prominent rollover inducing
factors are the HCG, track width and lateral acceleration, other parameters
are suspension and tire properties such as stiffness, dimensions and load
displacement laterally. A critical point for the rollover that denotes the be-
ginning of rollover is when weight has shifted so much to the outer wheels
that the normal force at the inner wheels is zero and they start to lift of the
road. To illustrate the relation between high center of gravity and lateral
acceleration with this critical point a graph such as figure 1.1 shows that the
higher the HCG is the less lateral acceleration is required to roll the vehicle
over, furthermore it illustrates the importance of load placement and the
dangers that moving load3 imply.
2
Roll center, a point between the vehicle’s wheels that the vehicle’s unsprung mass
sways around.
3
Load or cargo that moves and thereby alters the CG position, examples of this are
liquid cargo and live stock.

8
Introduction

Figure 1.2: Figure of a truck in a turn with a laterally displaced CG.

1.1.2 Static roll stability

A widely used ”measurement” tool for rollover is the static rollover stability
(SRS) which is a relation between the track width and the HCG derived
from the torque equilibrium equation according to figure 2.1.

t t
mt ay h = smt g + Fr − Fl (1.1)
2 2

The generally accepted rollover threshold occurs at wheel lift off meaning
that the weight that was supported by the wheels on both sides of the vehicle
is now only supported by the wheels on one side (Fl = 0, Fr = mg) when
the rolling process has reached a no turning point and thus it is referred
to as the rollover threshold, see figure 1.1. Imagine that the prior is true
and Assume further that the entire vehicle and all of its parts are rigid and
that there is no lateral displacement (s=0), with this in mind reformulating
equation (1.1) yields the following relation between the lateral acceleration
ay and the ratio of the half track width t/2 and the HCG h.

ay t
= (1.2)
g 2h

9
 Chapter 1

Equation (1.2) shows the critical lateral acceleration required to roll the ve-
hicle over, the vehicle would roll over when the left hand side of the equation
is greater than the right hand side.

1.2 Objectives

The objectives of this work are the following. A comparison of solution

models is to be conducted thereby we are to choose a solution model, de-
velop the chosen model as well as estimate the HCG with minimum prior
knowledge of the towed vehicle. The last objective implies that optimally
only information obtained from the tractor is to be considered but since this
requirement would complicate the problem at hand substantially, some in-
formation about the towed vehicle ”trailer” is assumed to be known. Some
of such information is somewhat general and apply for most trailers such
as wheel base, track width and other dimensional parameters. Since laws
and government regulations have created limits for vehicle dimensions, axle
weights as well as vehicle total weight it is only natural to assume that most
manufactured trailers are as large as possible in accordance to the laws and
regulations in each legislative area. Lastly we are to develop an estimation
tool in MATLAB.

10
Chapter 2

Model comparison

2.1 Literature survey

An extensive literature survey was made during the first phase of this work,
different models and parameter estimation methods were investigated, some
of these are mentioned in this section and one of the models is going to be
further examined, developed and chosen to be the primary way of estimating
the HCG on a test vehicle.

2.2 Models

2.2.1 Lateral model

Almost exclusively all the reports and scientific papers on the subject re-
volve around a lateral movement solution (rolling motion). A heavy vehicle
will rock and sway ”roll” around a vertical axis going through the RC, this
motion gives rise to a rolling amplitude which in turn is detected by mea-
suring a rolling angle θ (figure 2.2). The RC is different for each vehicle
combination depending on spring and tire stiffness.

One application of the lateral model revolves around finding a characteristic

frequency with which the CG rotates around the RC in a pendular motion
and the vertical ”bouncing” frequency due to the CG seen as a body of mass
supported on two springs each attached at the side of the mass.
q
1 2k
The vertical motion frequency is f 0 = 2π m and the lateral motion fre-
q
kt2
−gl
quency is f = √ 2m
, see [3] and figure 2.2. After reformulation and
2π (l2 +s2 )cosα

11
 Chapter 2

Figure 2.1: Figure of a truck in a turn

substitution of the first equation into the second one the following relation
is obtained.

4π 2 f 2 2 8π 4 f 2 cosαtan2 αf 02 t2 4π 6 f 2 f 04 t4
h +(g− )h+ cosαtan2 α−π 2 f 02 t2 = 0
cosα g g2
(2.1)

Where:
f is the roll motion frequency.
f 0 is the vertical frequency.
α is the roll angle.

The vertical frequency and roll frequency are measured with the help of
a ”shake detector” measuring the 2 frequencies f and f 0. The idea be-
hind measuring the bouncing vertical frequency is to eliminate the unknown
spring stiffness k.

Another way of applying the lateral model is to guess, estimate or measure

the spring stiffness k and only detect the rolling frequency, such work can be
read about in [4], the results indicate that since the measurements of the roll

12
Model comparison

Figure 2.2: Figure of a truck in a left turn.

angle where done on the tractor and not at the longitudinal CG position it
was attached with an error and the most prominent error factor was deemed
to be the torsional stiffness1 rendering the result based on measurements on
the tractor erroneous.
For these solutions to work more accurately, the addition of new sensors
is required. A so called tilt table test where a semitrailer is parked on a
platform able to tilt sideways can be preformed to simulate different values
of lateral acceleration and registering the related roll angle values, the ad-
vantages of the tilt table test are that one doesn’t need to drive the vehicle
in order to take measurements and that road banking is not a concern in
this case.

2.2.2 Longitudinal model

While a vehicle is accelerating or decelerating, going uphill or downhill and

any possible combinations of the former situations load is transferred in
between the vehicle’s axles, see figure 2.3. The transferred load depends
on the said acceleration or deceleration and/or road grade as well as the
1
The body of the trailer is not entirely rigid and has a twisting stiffness referred to
torsional stiffness.

13
 Chapter 2

Figure 2.3: A semitrailer on a sloping road.

HCG. Therefor if the transferred weight can be estimated then the HCG
can be estimated as well given the acceleration or deceleration and road
grade during which the weight is transferred. Very few reports and scientific
research have been done regarding longitudinal weight transfer. Patent 525
248 [5] is an example of a suggested solution using a similar model.

2.2.3 Model complications

There are parameters and factors that affect the aforementioned models
differently, the most prominent of these are discussed here.

2.2.3.1 Road grade

The road grade has to be known in order to calculate the HCG in the
longitudinal model in uphills and downhills.

2.2.3.2 Weight transfer

The weight transfer estimation is the corner pillar of the longitudinal model
and can be derived from the air suspension system’s air spring pressure
signal given knowledge about the air spring properties.

2.2.3.3 Acceleration

The acceleration whether lateral or longitudinal is a necessary factor in both

models and can be measured from the on-board tractor wheel speed signal

14
Model comparison

but would then be error borne especially while braking due to wheel slip2 .
An alternative way to measure the acceleration is to use a high frequency
GPS unit or by using an accelerometer.

2.2.3.4 Roll angle

Roll angle can be measured in a few ways, some of which are a displacement
sensor, a gyro meter measuring lateral acceleration signal that has to be
integrated to extract roll angle or a wheel speed difference signal measuring
the difference between the right and left wheels which withholds information
about turns.

2.2.3.5 Torsional stiffness

The torsional stiffness is the general name for the twisting of the trailer frame
due to its flexibility which in turn differs from one manufacturer to another
and from one type of trailer to another, the reasons for that are mainly the
difference in materials used in manufacturing, the shape and dimension of
the trailer chassis, production procedure as well as age and other factors
which makes it a whole lot harder to estimate a torsional stiffness value.

2.2.3.6 Rearward amplification

Add to the former error the known disturbance phenomenon called rearward
amplification which is a measure of the increase in lateral acceleration further
back in the trailer. It is the tendency in trailer combination vehicles to
exhibit larger swaying movements during lateral movement the further back
in the trailer one observes which in turn is a manifestation of the larger
lateral acceleration further back. The relation of lateral acceleration between
the trailer CG and the tractor CG is called ”rearward amplification” [6].

2.2.3.7 Suspension

The lateral model depends on many unknown factors such as the stiffness
of the suspension an the tires which has to be guessed, experimentally de-
rived or obtained from the manufacturer which is not an exact value since
suspension stiffness depends on many other factors such as age and usage.
2
The wheel slipping at certain times as opposed to rotating freely showing less than
real speed.

15
 Chapter 2

2.2.3.8 Load distribution

Lateral CG displacement is a consequence of uneven loading and has to be

accounted for since rolling occurs about the axis through the vertical line
between CG and RC.

2.2.3.9 Various other factors

Road banking affects the calculations in the lateral Model. Side winds have
an effect on the lateral model, pitching affects the longitudinal model mostly.

2.3 Conclusion

Due to the uncertainty in the various parameters discussed earlier in this

chapter, the model chosen to be investigated, developed further and used to
build an estimation tool is the longitudinal model. While the longitudinal
model involves less uncertain parameters it still possesses a challenge since it
involves load change detection and estimation which has always been a hard
task but with the introduction of air suspension it is possible to estimate
and new ways of load estimation using the air suspension system are being
developed as of now to give better accuracy by taking into account the air
spring’s properties.

16
Chapter 3

Longitudinal model

Figure 3.1: Figure of a tractor and model related forces acting on it during
driving.

The weight transfer between the vehicle’s axles and coupling points is differ-
ent for each acceleration value and each situation. The rear axle load is plot-
ted as a function of acceleration and this relation is different and uniquely
defined for each HCG value defined by the torque equilibrium equations
shown in this chapter.

17
 Chapter 3

3.1 Weight transfer

3.1.1 Tractor weight transfer

The Weight transfer in the tractor is investigated as a first step towards the
solution, the aim in this section is to estimate the longitudinal and vertical
position of the tractor’s CG. The weight of the empty tractor1 md,empty as
well as the normal forces of said empty tractor N1 and N2 are known. Thus
the portion of normal forces for said tractor due to a certain amount of fuel
in the fuel tank can be calculated given the longitudinal position of the fuel
tank’s CG xf :

N2,f uel x1 − mf uel gxf cos(θ) = 0 (3.1)

The same calculation can be done with respect to the normal force on the
front axle as well. Equation (3.1) is written as:

mf uel gxf cos(θ)

N2,f uel = (3.2)
x1

Which added to the empty tractor normal force gives the vehicle’s curb
weight2 :

N2 = N2,empty + N2,f uel (3.3)

The next step is to calculate the longitudinal CG position which is obtained

from the torque equilibrium equation as in figure 3.1:

N1 x1 − md gx2 cos(θ) = 0 (3.4)

Rewrite the above equation as:

N1 md gcos(θ) − N2
x2 = = x1 (3.5)
md gcos(θ) md gcos(θ)
1
Weight of tractor and driver without fuel.
2
The weight of vehicle and fuel.

18
Longitudinal model

Figure 3.2: Figure of a trailer with model forces pointed out.

The next step is to formulate the torque equilibrium equation to obtain the
vertical center of gravity of the tractor.

−N2 x1 − md ay1 + md gy1 sin(θ) + md g(x1 − x2 )cos(θ) = 0 (3.6)

Reformulating equation (3.6) gives

N2 x1 − md g(x1 − x2 )cos(θ)
y1 = (3.7)
md (gsin(θ) − a)

Insertion of x2 which is obtained from (3.5) into (3.7) as well as knowledge

about the normal force at least at one of the vehicles axles and acceleration
and road grade signals solves the equation and gives the height of CG for
the tractor.

3.1.2 Trailer weight transfer

During acceleration or retardation and on road slopes weight will transfer

between the trailer’s axles and fifth wheel in proportion to the acceleration

19
 Chapter 3

or deceleration and road grade, acceleration transfers weight from the front
to the rear axle, deceleration does the opposite [7].
The longitudinal CG position of the trailer x4 is calculated through the fol-
lowing stationary torque equilibrium equation (on a flat surface), see figure
3.2.

N3 x5 − mt gx4 = 0 (3.8)

Reformulated, equation (3.8) becomes.

N3 x5
x4 = (3.9)
mt g

mt and N3 are estimated either by weighing at least one of the tractor’s

axles and the tailer axle(s) since x3 is known or obtained from the vehicle’s
on-board control unit which calculates an estimate of the rear axle load and
of mt .

The sought-after value y3 ”HCG” can be obtained by the torque equation

from figure 3.3 as follows.

N3 x5 − F3 y2 − mt gcos(θ)x4 + mt gsin(θ)y3 + mt ay3 = 0 (3.10)

It is seen that the fifth wheel forces N3 and F3 have to be calculated in order
to estimate y3 , the next section depicts a way to estimate these forces.

3.1.3 Fifth wheel forces

The longitudinal force at the fifth wheel F3 ’s direction and size depends on
the relationship between tractor and trailer as well as roll and air drag forces
and affects the load transfer measured at the tractor’s rear axle, it therefor
has to be accounted for and estimated. If the vehicle is accelerating then the
relationship is as in figure 3.3 while braking gives different scenarios such as
if the trailer is braking only a part f or all (f=100%) of the tractor’s mass,
or if the tractor is braking a part f or all (f=100%) of the trailer’s mass
then the forces are as in figure 3.4 the following equation explains how F3 is
estimated for acceleration and braking only the tractor in this work.

F3 = mt a + FRollandDrag (3.11)

20
Longitudinal model

Figure 3.3: Figure of a trailer with attracting relation between trailer and
tractor.

Figure 3.4: Figure of a trailer with repelling relation between trailer and
tractor.

21
 Chapter 3

The following equation is valid when the trailer is braking a part of the
tractor:

F3 = f md a + FRollandDrag (3.12)

And the following equation is valid when the trailer is braking a part of itself
and the tractor is braking as well:

F3 = f mt a + FRollandDrag (3.13)

The force FRollandDrag is treated in section 3.2.4 and f is obtained given the
brake force relation between the trailer and tractor. The brake forces are
discussed in section 3.2.3.2 and the following relation is valid for f in the
situation described in equation (3.13).

mt a − F4
f= (3.14)
mt a

N3 is the vertical force at the fifth wheel and is the portion of load supported
by the tractor, it typically amounts to around one third of the trailer’s mass.
This force is in turn distributed between the tractor’s axles as follows:
Rear tractor axle:

x1 − x3
∆N2 = N3 (3.15)
x1

Front tractor axle:

x3
∆N1 = N3 (3.16)
x1

In addition the tractor axles transfer weight in-between as follows (force on

the rear tractor wheel):

md g(x1 − x2) + md ay1

N2 = (3.17)
x1

22
Longitudinal model

The forces on the rear and front tractor axles due to acceleration and decel-
eration are as follows as seen from figure 3.1.

N3 (x1 − x3 ) + y2 F3 + y1 mt a + mt g(x1 − x2 )
N2 = (3.18)
x1

N1 = N3 + mt g − N2 (3.19)

In the above equations as well as henceforth the road grade is neglected

(θ = 0) for simplicity. N3 is obtained from equation (3.18), the next step
is to use the information from the air suspension which is used on Scania
tractors to estimate the load on the rear tractor axle N2 . N2 is a given signal
and described further in other sections in this work.

3.2 Parameter estimation

3.2.1 Vehicle weight estimation

The tractor weight is assumed to be known, the trailer’s weight is estimated

from the air suspension, if air suspension is missing on the trailer then
weighing the vehicle is an alternative, there are other ways of estimating the
weight but they are outside the scope of this work.

m = md + mt (3.20)

3.2.2 Weight transfer detection and estimation

3.2.2.1 Air suspension

The on-board air suspension system is used as the primary way of estimating
the weight transfer between the vehicle’s axles, there are two main types of
air pressure suspension configurations used on Scania trucks’ rear axles, one
according to figure A.1 and 3.7 and referred to as Gen 23 consisting of two
air springs at the rear axle each situated between the chassis from above and
a connecting bar from below which in turn is mounted beneath the drive axle
and connects said air spring on one side of the drive axle to a joint allowing
it to rotate freely around the joint in 2 dimensions. The joint is connected
3
Generation 2 Scania truck.

23
 Chapter 3

Figure 3.5: The relationship between the force on the bellow and the air
pressure & height of the air spring.

to the chassis through a rigid arm and the drive axle is attached to the bar
from above almost at the middle. Gen 2 configuration therefore is not only
affected by vertical forces at the wheel but also by the longitudinal forces
such as brake forces, tractive forces and forces due to the road grade, the
reason for that is the torque in the bar between air spring and frame which
is proportional to the longitudinal forces at the wheel. These longitudinal
forces account for a part of the forces on the air spring and thus have to
be subtracted from the net force on the axle registered by the air spring in
order to obtain solely the amount of load transferred between the axles.

The other air suspension configuration is Gen 34 according to figure A.2

and 3.7 consisting of four air springs where there are two air springs on each
side of the drive axle connected through a few joints as in the figure. It
is generally accepted to assume that the air bellows in Gen 3 configuration
are not affected by the longitudinal forces and only react to vertical forces
such as the forces sought after in this experiment. The force acting on the
air spring depends on both the air pressure in the air spring and the spring
length.

4
Generation 3 Scania truck.

24
Longitudinal model

3.2.2.2 Suspension leaf springs

Figure 3.6: Leaf feather on the tractor’s front axle.

As a complement to the former mentioned weight transfer estimation the

weight transfer can be additionally measured from the leaf springs installed
on the front axle by applying Hooke’s law: F = −kx, where F is the force
on the spring, ∆x is the displacement and k is the stiffness.

3.2.3 Longitudinal forces at the wheels

(a) ”Gen 2” air suspension. (b) ”Gen 3” air suspension.

Figure 3.7: Different air suspension configurations in Scania’s tractors.

In the former section it is mentioned that the air pressure and thus the
weight estimation in the air springs is influenced by the longitudinal forces
as well. During different maneuvers different forces apply such as braking
and tractive forces, since these forces affect the axle load measurement they

25
 Chapter 3

have to be estimated in order to subtract it from the total force on the air
springs, such forces and their effect and estimation is discussed here.

3.2.3.1 Tractive force

The tractive force is the force exercised by the vehicle’s tires onto the road in
order to move the vehicle and counteract natural forces in the vehicle’s mov-
ing direction such as road grade forces, air resistance and rolling resistance.
These forces are usually lumped in the following equation [7].

Te − Je ẇ
= mv̇ + Froll + Fdrag + Fgrade (3.21)
r

The same force has to be produced from the engine and is equal to:

Te Ut
Ft = (3.22)
r

3.2.3.2 Brake force

Braking starts by applying pressure to the braking cylinders, the pressure

sent to each of the vehicle’s axles is measured, the pressure then produces a
torque that in turn gives a force stopping the vehicle. The pressure to torque
relation is normally set as 1 bar of pressure equals a torque of 5500Nm,
furthermore the pressure required to start braking is said to be somewhere
around 0.4 bar. The values 5500Nm and 0.4 bar are an approximation of
the real values which are different for each vehicle and axle depending on
several factors such as temperature and brake components wearing with age.
Thus what is done to reduce uncertainty and have a more recent and more
accurate value in this report is calibration of the brakes by performing a
rolling brake test as seen in chapter 4 and by using the following equation
for obtaining the applied brake torque per unit of brake pressure.

Fb r
Tb = (3.23)
(p − ps )

3.2.3.3 Retarder force

The Retarder is a system for retarding the vehicle by applying a force the
has the opposite effect of the tractive force. It can be simply described as

26
Longitudinal model

braking through the power train and thus only acts on the drive axle(s) of
a vehicle. The retarder is attached to the crank shaft of the vehicle between
the gearbox and the differential gear. The retarder force is calculated in a
similar way to the tractive forces.

Th Ud
Fh = (3.24)
r

3.2.4 Air resistance

Figure 3.8: The air resistance force and placement.

The air resistance on the semitrailer has to be estimated and subtracted as

well, the following is a generaly used air resistance force equation. [7] [8]

1
F = ρCd Av 2 (3.25)
2

Where:
ρ is the air density.
Cd is the air resistance coefficient.
A is the front area.

27
 Chapter 3

The values used in this report are Cd =0.73, ρ=1.225 (kg/m3 ) and the force
on the trailer is assumed to be as large as the force on the tractor, i.e.
Fa1 = Fa2 which leads to the following force Fa at the fifth wheel:

Fa = Fa1 − Fa2 (3.26)

The remaining air resistance contributions are the following equation ob-
tained by torque equilibrium as in figure 3.8 at points A and B.

h1 + h3 x1 − x3
N2a = Fa1 − (h2 + h4 ) Fa2 (3.27)
x1 x3 x5

Dimensions and parameters for the air resistance calculations are found in
Appendix B in table B.3.

3.2.5 Road grade

The road grade can be estimated in different ways, one that is used at Scania
is that a road map with the road grade along with position is known from
former logs, so by comparing the vehicle’s position at a certain time point
to Scania’s road map the road grade can be known. Since this method
requires knowledge of the route on which one drives it’s not optimal but
it’s getting better since Scania’s road map is being updated continuously
to cover more roads. Another way of estimating the road grade is through
an accelerometer’s acceleration signals (X,Y,Z) by geometrical relations or
by position signal from a GPS unit ∆h along with the distance traveled ∆s
which is illustrated in the figure below.

3.2.6 Acceleration

The acceleration can be derived from the wheel speed signal derivation, using
an accelerometer or a GPS. Acceleration in this report is obtained from an
accelerometer. Differentiation of wheel speed signal is a convenient way
of estimating the acceleration but does not work as well when braking the
wheels due to wheel slip, the differentiation can by done by a few algorithms,
the algorithms shown here are the three point central difference.

vi − vi−1
ai = (3.28)
h

28
Longitudinal model

Figure 3.9: Figure explaining a way of obtaining the road grade, θ =

arcsin( ∆h
∆s ).

And the 5 point central difference.

vi−2 − 8vi−1 + 8vi+1 − vt+2

ai = (3.29)
12h

Where:
ai : acceleration at time i. ( sm2 )
vi : speed at time i. ( ms)
h: ti − ti−1 , the adjacent measurement times t’s difference is constant, hence
the constant h.

3.2.7 Fifth wheel height

The vertical position of the fifth wheel is a variable depending on both the
distance between the axle and the frame and the wheel radius.

∆y2 = ∆yf + ∆H + r (3.30)

Where ∆H is the axle to frame distance. ∆yf is the frame to fifth wheel
coupling point.

29
 Chapter 3

Figure 3.10: Figure related to the geometrics of the suspension configuration

showing the vertical change in the drive axle’s position.

The axle to frame distance is a given signal on ELC45 equipped Scania

trucks.

The portion of the change in y2 due to tire pressure and load is assumed to
be negligible during load transfer and it is assumed that the variation in tire
radius is minimal due to assumed tire pressure regulation after reloading.
The frame to fifth wheel coupling point distance is a constant, With these
assumptions equation (3.30) can be solved and a value of y2 is estimated.

3.3 Complete model

The relationship between the axle load and the acceleration withholds infor-
mation about the HCG as seen in the sections above and this relationship
is illustrated in the following diagram along with a proposed way of imple-
mentation on a real semitrailer.
Equation (3.18) along with equation (3.11) inserted in (3.10) is reduced
to the following equation for acceleration and only braking the tractor or
5
A system measuring the distance from the axle to the frame.

30
Longitudinal model

Figure 3.11: Diagram illustrating the steps towards the solution.

applying the retarder after subtracting the air resistance force impact on the
axle load since those forces would otherwise give a nonlinear relationship
if not subtracted from equation (3.18) beforehand, so the force N2a from
equation (3.27) is subtracted from N2 in equation (3.18) before insertion.

g md g x1 x5
y3 = x4 +y2 + ( (x1 −x2 )+y1 )+y2 −((N s2 +jai ) ) (3.31)
ai mt ai mt ai x1 − x3
To get a better picture of how the forces mentioned throughout this chapter
effect the weight estimation illustrative graphs are shown below for equation
(3.31) showing the Force N2 as a function of a for known HCG values.

31
 Chapter 3

Figure 3.12: The force N2 /g as a function of acceleration a for different

situations and HCG positions.

32
Chapter 4

Results and analysis

4.1 Equipment

The tests are preformed on a Gen 3 4x2 Scania tractor as well as a Gen 2 4x2
Scania tractor along with a special trailer equipped with a changeable HCG
position, a first step is to measure dimensions and weights. A gps (Oxford
RT3040)1 is used for another acceleration signal. A laptop with Canalyzer2
installed is used to record the signals by connecting to the control unit of the
tractor as in figures 4.1a and 4.1b. Test vehicle dimensions and properties
are found in table B.1 in Appendix B

(a) Equipment and installation. (b) Equipment and installation.

1
A GPS unit with the frequency 100 (Hz).
2
A computer software tool used for signal analysis.

33
 Chapter 4

4.2 Test procedure

The tests were performed by accelerating and braking different axles of the
vehicle and different axle combinations at a time with varying brake pressure
on Scania’s test track and the tests were preformed on level ground given
the test course’s limitations and for simplicity.

4.3 Data processing

4.3.1 Lateral and vertical acceleration

Figure 4.1: The lateral acceleration ay before and after smoothing.

The lateral acceleration and the longitudinal acceleration signals are used to
eliminate road unevenness, turns and road grades in addition to a altitude
signal to detect the uphills. The lateral and vertical acceleration signals have
a frequency of 200 (Hz), they were processed by smoothing using a built in
MATLAB function for smoothing called convn which uses the principle of
moving averages, figure 4.1 shows how the signal of the lateral acceleration
was before and after smoothing.

34
Results and analysis

Figure 4.2: The acceleration a before and after filtering.

4.3.2 Acceleration

The acceleration is filtered using a low pass filter, figure 4.2 shows the signal
of the acceleration before and after filtering. The filter has a passband of
0.05, a stop band frequency of 0.2, a stop band ripple of 0.2 dB and a
passband ripple of 60 dB.

4.4 Brake calibration

(a) 4x2 tractor (Electra) & trailer A. (b) Rolling brake evaluation facility.

The rolling brake calibration test was performed on a Gen 3 4x2 tractor

35
 Chapter 4

Axle ps (bar) Tb (N m/bar)

Tractor front 0.42 4200
Tractor rear 0.42 3900
Trailer first 0.60 3800
Trailer second 0.56 3732
Trailer third 0.60 3800

Table 4.1: Brake calibration test results.

”Electra” and a coupled trailer ”Test trailer”. Before beginning the test the
semitrailer was driven around Scania’s test course and the brakes were used
frequently in order to warm them up, the vehicle 4.3a was then driven to the
test location 4.3b and the following results where obtained from the result
print outs at the test location C.1 by using the printouts to estimate Tb and
ps to be inserted in (3.23).

4.5 HCG

At each test run the HCG was changed to a different one and measured us-
ing simple center of gravity equations, see Appendix B. The relevant signals
are logged and used in the calculation of the HCG according to whether
they are suitable points or not decided by an algorithm designed to specif-
ically dispose of points that are not logged during harmonic and smooth
movements.

36
Results and analysis

Figure 4.3: Adjustable frame with blocks inside the trailer.

37
 Chapter 4

Figure 4.4: The slopes j obtained by linear regression for the different HCG
values compared to the real slopes.

Above are graphs of slopes obtained using linear regression (least squares
method), the steeper the slope is the lower the HCG is. The HCG is obtained
from (3.31) by inserting the above obtained slopes and shown in table 4.2.
In these results the slopes and thus the HCG’s are calibrated for the HCG
value 206.2 (cm).

Vehicle Electra Electra Electra Conceptus Conceptus Conceptus

Measured HCG (cm) 178.0 206.2 225.3 178.0 206.2 225.3
Estimated HCG (cm) 176.1 206.2 222.0 173.9 206.2 224.3
Difference (cm) 1.9 3.3 4.1 1

Table 4.2: The results of the HCG estimation.

4.6 Error analysis

The roll resistance is assumed to be constant in this work and the pitch
angle’s effect has not been investigated, these two factors are to be considered
error factors but the biggest error in the line fitting arises from the error
in the axle load estimation and the acceleration signal processing. The axle
load signal is attached to an error of 300 (kg) according to Scania CV, this
figure can become bigger when subtracting the disturbance forces since these
forces are attached with errors as well. The Number and placement of the

38
Results and analysis

Figure 4.5: The slopes j obtained by linear regression for the different HCG
values compared to the real slopes.

Error factor y1 y2 Te Th r
Error margin ±10cm ±10cm ±100Nm ±100Nm ∓10cm
∆y3 ±7.26cm ±35.42cm ∓10.88cm ∓8.50cm ±49.34cm
Error factor x3 x5 mt md
Error margin ±10cm ±10cm ±100kg ±100kg
∆y3 ±3.7cm ±1.5cm ±1.1cm ∓0.6cm

Table 4.3: Error analysis.

sampled points decide the accuracy in the results. To illustrate this the
following two examples are shown.
The slope in fgure 4.6 is j=372 giving a HCG value of 377 (cm) wich is way
off the real value.
The slope in figure 4.7 is j=899 giving a HCG value of 206.2 (cm). Thus
this confirms that the wider the acceleration span is the more accurate the
solution is. To fully understand the errors one should have knowledge of the
distribution function of the above plots 4.6 and 4.7 should such a distribution
function exist.

39
 Chapter 4

Figure 4.6: The linear regression with slope j.

Figure 4.7: The linear regression with slope j.

40
Chapter 5

Conclusions

The proposed and applied way of estimating the HCG in this work proved to
be a very efficient one leading to quite accurate and satisfactory results and
the HCG was estimated within 4.1 (cm) for a 4x2 Gen 2 Scania tractor and
3.3 (cm) for a 4x2 Gen 3 Scania tractor. The solution proposed and used
in this work can be applied to non articulated vehicles and other vehicle
combination by altering and re-modifying a few model components. The
results indicate that the 4x2 Gen 3 Scania tractor’s axle load calculation
could be slightly affected by the longitudinal forces which was assumed to
be not true in this work. The former conclusion is based upon the fact that
after calibration the axle load differs substantially from the true one.

41
 Chapter 5

42
Chapter 6

Recommendations and
future work

Finding a more suitable acceleration estimation. Adding braking sensors to

detect the brake force and by that having a broader span of points across
the acceleration spectrum which theoretically leads to a more accurate slope
value and thus leading to a better HCG estimate. Improve the axle load
calculation by taking into account the air springs length which is already
under development. Make trailer dimensions available to tractor unit, for
this work the wheel base is the only required trailer dimension parameter.

More accurate road angle estimation. A more accurate selective road seg-
ment calculation leading to a better algorithm for the time points at which
one runs the estimation algorithm. The roll resistance is assumed to be con-
stant during the entire test and is thus left untreated while in reality this is
not true, so finding a way of estimating the roll resistance and subtracting
its effect from the axle load should lead to better results.

43
 Chapter 6

44
Bibliography

[1] “Truck driver safety and health,” NIOSH Science Blog, 2007. Re-
trieved January 03, 2012 from http://www.truckinfo.net/trucking/
stats.htm.

[2] “Trucking statistics,” TruckInfo. Retrieved January 03, 2012 from

http://www.cdc.gov/niosh/blog/nsb111907_truck.html.

[3] Y. Watanabe, “Patent us 2010/0198492 a1,” 2010. Center of-Gravity

Detection System, Lateral Rollover Limit Velocity Estimation System,
And Cargo Weight Estimation System.

[4] J. Ah-King, Center of Gravity Height Estimation of Heavy Vehicles.

2007.

[5] N. Vågstedt, K. Linqvist, and N. Pettersson, “Patent 525 248,” 2005.

Arrangemang och förfarande för att uppskatta tyngdpunktshöjden hos
ett fordon.

[6] C. Winkler, P. Fancher, Z. Bareket, S. Bogard, G. Johnson, S. Karami-

has, and S. Mink, “Heavy vehicle size and weight, test procedures
for minimum safety performance,” tech. rep., The University of Michi-
gan, Transportation Research Institute, 2901 Baxter Road, Ann Arbor,
Michigan, 1992.

[7] T. D. Gillespie, Fundamentals of Vehicle Dynamics. 1992.

[8] E. P., “Air resistance parameter calculations,” 2012.

45
 Appendix

46
Appendix A

Appendix A Air spring

Configuration

47
 Appendix A

48
Appendix A Air spring Configuration

49

Figure A.2: Gen 3 Scania tractor suspension configuration.

 Appendix A

50
Appendix B

Appendix B Vehicle data

B.1 Vehicle data

Figure B.1: Conceptus with test trailer.

51
 Appendix B

Figure B.2: Electra with test trailer.

(a) Front block group. (b) Rear block group.

Figure B.3: Weight blocks inside the test trailer’s container.

52
Appendix B Vehicle data

Conceptus 4x2 Gen2 tractor Electra 4x2 Gen 3 tractor Test trailer
y1 (cm) 125 125
y2 (cm) 115 115
x1 (cm) 370 355
x3 (cm) 63 51
x5 (cm) 780
N s1 /g (kg) 5700 6100
N s2 /g (kg) 1980 2170
N s3 /g (kg) 10000
N s4 /g (kg) 18440
r (cm) 50.0 52.5 47.5

Table B.1: Test vehicle’s dimensions and weights.

Weight (kg) HCG (cm)

Frame 5670 120
Axles 2100 47.5
Container 4670 268
Blocks 16000 189.5-289.5
Total 28440 178-234

Table B.2: Table of test trailer properties.

h1 h2 h3 h4
200 (cm) 100 (cm) 40 (cm) 140 (cm)

Table B.3: Table of air resistance related dimensions.

53
 Appendix B

54
Appendix C

Appendix B Brake
calibration

(a) Front tractor axle. (b) Rear tractor axle.

Figure C.1: Tractor brake calibration.

55
 Appendix C

(a) Front trailer axle. (b) Middle trailer axle. (c) Rear trailer axle.

Figure C.2: Trailer brake calibration.

56