See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/245401460

On-board estimation of dynamic tyre forces from optically measured tyre

carcass deflections

Article in International journal of heavy vehicle systems IJHVS · July 2009

DOI: 10.1504/IJHVS.2009.027139

CITATIONS READS

60 2,694

1 author:

Ari Juhani Tuononen

Aalto University
51 PUBLICATIONS 1,167 CITATIONS

SEE PROFILE

All content following this page was uploaded by Ari Juhani Tuononen on 25 November 2014.

The user has requested enhancement of the downloaded file.

362 Int. J. Heavy Vehicle Systems, Vol. 16, No. 3, 2009

On-board estimation of dynamic tyre forces

from optically measured tyre carcass deflections

Ari Tuononen
Laboratory of Automotive Engineering,
Helsinki University of Technology,
P.O. Box 4300, FI-02015 TKK, Finland
E-mail: [email protected]

Abstract: Active safety systems would benefit from accurate information on

tyre forces. A special sensor module has been developed for research purposes
to study real-time tyre force estimation from tyre carcass displacements.
The forces can be measured in vertical, lateral and longitudinal directions.
The sensor measures tyre carcass movement with respect to the rim and
transmits the data wirelessly to the chassis. The raw data is analysed and tyre
force information is available on a CAN-bus. The sensor is calibrated in a tyre
test rig and proving ground measurement results are presented. The lateral and
longitudinal force estimates are accurate when compared with the vehicle
accelerations, but the vertical force is difficult to estimate during heavy braking
manoeuvres.

Keywords: active safety system; intelligent tyre; optical tyre sensor;

tyre deformation; friction estimation; advanced driver assistance; optical
position detection.

Reference to this paper should be made as follows: Tuononen, A. (2009)

‘On-board estimation of dynamic tyre forces from optically measured tyre
carcass deflections’, Int. J. Heavy Vehicle Systems, Vol. 16, No. 3, pp.362–378.

Biographical notes: Ari Tuononen received his MSc in Mechanical

Engineering from Helsinki University of Technology in 2005. He is currently
working at the same university as a researcher. His fields of interest are
vehicle stability control, tyre modelling and tyre sensors. At the moment,
his focus is on intelligent tyre and its feasibility in active safety systems.

1 Introduction

Active safety systems have become common equipment in all vehicle categories.
Estimation of individual tyre operating states could improve the performance of active
safety systems. The tyre transmits all the forces between the chassis and road, and thus
dominates the vehicle behaviour. The tyre sensors could offer several different functions:
• tyre pressure monitoring TPMS (the only one in production)
• estimation of tyre forces (vertical, lateral and longitudinal) and aligning moment
• slip ratio and slip angle

Copyright © 2009 Inderscience Enterprises Ltd.

 On-board estimation of dynamic tyre forces 363

• recognition of tyre improper and normal wearing, imbalance

• recognition of worn suspension components.
Tyre pressure monitoring is already available and it is extremely beneficial, especially
for heavy vehicles as their tyre pressure is sometimes more difficult to check than the
pressure of passenger car tyres. The low inflation pressure increases fuel consumption
(and CO2), tyre wearing and complicates vehicle behaviour in limit handling. In addition,
incorrect inflation pressure changes tyre stiffness and dynamic rolling radius, which may
mislead state estimators of active safety systems.
Vertical tyre force estimation is probably the next step after TPMS because it is
available from the acceleration sensor attached to the TPMS unit. The acceleration sensor
algorithms are discussed, for example in Morinaga et al. (2006). The piezoelectric power
source can also be implemented to estimate contact length and contact deformation,
which both correlate with vertical force (Mancosu and Di Pasquale, 2007). The first
versions of the vertical force tyre sensor may only estimate static vertical force owing to
load transfer, and not dynamic load. Dynamic wheel load estimation from vehicle sensors
is especially difficult in heavy vehicles, owing to the influence of the flexible chassis
(compared with cars) for roll stiffness distribution. Also, the payload can make load
deviation extreme.
The exact vertical force estimate together with the existing vehicle sensors provide
a good basis for estimating lateral and longitudinal tyre forces, especially for the linear
operating range of tyre. Directly measured lateral forces could allow vehicle slip angle
estimation in every condition without any tyre parameters like cornering stiffness or
friction coefficient.
Detection of broken or worn suspension parts (or tyre) can be difficult to conduct
with the use of standard vehicle sensors. In a passenger car, the driver can more easily
feel if the tyre is badly imbalanced or otherwise damaged than in heavy vehicles.
The tyre sensor can detect such a phenomenon, for example by comparing the wheel load
deviation of neighbouring wheels. This might also reveal worn damper or other potential
problems that are not otherwise easily observed.
This paper introduces an optical tyre sensor for a truck. Even if the optical
measurement devices offer some outstanding advantages, the sensor in question is not
intended to be a commercial product. Meanwhile, the optical tyre sensor module is being
developed to serve as a research platform. It offers an opportunity to study which
functions could be useful when considering product sensors.

2 Optical tyre sensor module and test vehicle

The optical tyre sensor was developed in the EC-funded Apollo-project 2002–2005 and
has been further developed in the FRICTION project (APOLLO, 2005; FRICTION,
2006; Tuononen, 2008). The sensor requires some space in the rim, and once the sensor is
installed, it is not possible to install or remove the tyre anymore. In passenger car rims,
this problem was avoided by using dividable rims. On the contrary, a dividable truck rim
was not an attractive option because of tedious assembly work and the iterative nature of
sensor development. Therefore, a special sensor module was designed to interface the rim
(Figure 1). The sensor module can be quickly removed and reinstalled without special
tools.
364 A. Tuononen

Figure 1 Sensor module installation to the test rim (see online version for colours)

1: rim; 2: flange joint welded to the rim; 3: sensor module; 5: sealing O-ring.

The tyre is installed before the sensor module and the Light Emitting Diode (LED) is
glued into the inner liner of the tyre (Figure 2). The LED is powered with wires from the
sensor module. In addition to the sensor module, a magnetic pick-up sensor is installed on
the inner edge of rim (Figure 3). The magnet is installed on the suspension to indicate the
upright position of the sensor. This enables very accurate information on the sensor
rotation angle and the data is certainly on the same time axis as the actual tyre sensor
data, owing to same signal path.

Figure 2 Cross section of optical tyre sensor

 On-board estimation of dynamic tyre forces 365

Figure 3 Optical tyre sensor components, magnet and magnetic pick-up sensor

The actual tyre sensor is a light-sensitive device called a Position-Sensitive Detector

(PSD). The PSD sensor (Hamamatsu S5991-01 Pin cushion) converts the light energy to
currents, which are measured from the corners of the sensor. The active area of the sensor
is 10 × 10 mm, the position resolution is 1.5 µm, rise time is 2 µs and photo sensitivity is
0.6 A/W (Hamamatsu, 2008). The actual displacements can be calculated as:
L (ix 2 + iy1 ) − (ix1 + i y 2 )
x= (1)
2 ix 2 + i y1 + ix1 + i y 2

L (ix 2 + i y 2 ) − (ix1 + iy1 )

y= (2)
2 ix 2 + iy1 + ix1 + i y 2

1
z= (3)
ix 2 + iy1 + ix1 + i y 2

where i indicates respective current and L is the length of the sensor active area.
Figure 4 clarifies the PSD structure and how the output current depends on position of
light spot on active area of the sensor.
However, the real-time displacement calculation was a bit different: only additions
and subtractions were executed for analogue signals. Basically, the operational amplifiers
calculated equations:
366 A. Tuononen

Vx = Gx [(ix 2 + i y1 ) − (ix1 + i y 2 )] (4)

Vy = G y [(ix 2 + iy 2 ) − (ix1 + iy1 )] (5)

Vsum = Gsum [ix 2 + i y1 + ix1 + i y 2 ] (6)

where Gx, Gy and Gsum are respective gains. The voltages are amplified and measured
with ADC. The MCU prepares the message including Cyclic Redundancy Check (CRC)
to be sent by the radio. The radio operates on 433.92 MHz. The radio receiver is located
next to the tyre (Figure 5). The data is demodulated and converted to the CAN-message
for the 1 Mbps bus. This displacement data is further analysed in a 16-bit MCU, which
calculates vertical, longitudinal and lateral tyre forces and sends the data to the 250 kbps
CAN-bus. This additional processing step was taken because otherwise tyre sensor raw
data (sampled 5100 Hz for each channel) loaded the data acquisition to the limits.
Figure 6 shows the complete signal path.

Figure 4 The PSD sensor structure

Source: Hamamatsu (2003)

Figure 5 Test vehicle Volvo FH12 with tyre sensor receiver unit next to left front wheel
(see online version for colours)
 On-board estimation of dynamic tyre forces 367

Figure 6 Signal flow from sensor to the data acquisition

The sensor and transmitter are powered by a Li-ion battery adopted from a mobile phone.
The battery is charged through a pressure proof connector in the rim. The circularly
polarised antenna and receiver unit, which can be seen in Figure 5, are installed close
to the tyre.
The optical tyre sensor output signals for approximately two rotations are shown in
Figure 7. The magnetic pick-up sensor signal is found between the contacts (magnet and
sensor position are shown in Figure 3), which are expected from the upright position of
the magnet. The intensity signal (W/m2) shows decreased intensity during contact.
The vertical displacement (inverse square relation to intensity) can be calculated as:

1
z= (7)
4095 − I PSD

where the 4095 comes from the inverse configuration of the 12-bit ADC and IPSD is
the measured intensity. The longitudinal signal shows how the contact deformation
constrains the tyre carcass in different directions on the leading and trailing edge.
This stress is released only in the exact upright and downright positions of the tyre in free
rolling, which can be detected from offset line (dash-line). The lateral signal is not
influenced noticeably in free rolling; only slight movement can be detected during
contact. This can be explained by tyre force generation from the asymmetry of the tyre or
alignment errors of the LED and sensor modules.
The influence of the slip angle can be seen in Figure 8. The generated lateral force
stretches the carcass laterally and the peak value is naturally detected during contact.
The increased vertical force, compared with Figure 7, can be seen in intensity and
longitudinal signal amplitudes.
The optical tyre sensor and receiver unit were installed into a Volvo FH12 (Figure 5
tractor with a ballast platform (without trailer). In addition to the tyre sensor and existing
ABS/ESC-sensors, the vehicle was equipped with an Inertial Measurement Unit (IMU).
The IMU measured 3 degree of freedom accelerations and rotational velocities at 100 Hz.
The IMU signals were filtered with a transfer function:
368 A. Tuononen

s 2 + 652
G(S ) = (8)
s 2 + 65s + 652

which was discretised with Tustin approximation. The natural frequency for IMU
position in the cabin was approximately 65 rad/s, which explains the filter parameters.

Figure 7 Sensor signals in free rolling (50 kph, 20 kN) (see online version for colours)

Figure 8 Sensor signal under 6° slip angle (50 kph, 30 kN) (see online version for colours)
 On-board estimation of dynamic tyre forces 369

3 Force estimation algorithms

Tyre force test rig measurements were needed to calibrate force estimation parameters.
The tyre test rig at ika RWTH (Institut für Kraftfahrwesen Aachen) satisfied all the
requirements for the force levels and for the minimum acceptable radius of the drum.
All the calibration measurements were accomplished at the ika RWTH tyre test rig
(Holtschulze, 2006; Hüsemann, 2007).
The tyre force estimate is calculated once per revolution. This means that update
frequency depends on the rotational velocity of the wheel (Figure 9). The relationship can
be simply expressed as:
v
f update = (9)
2π rdyn

where v is the vehicle velocity and rdyn is the dynamic rolling radius of the tyre. The
update rate naturally increases with speed, but on the contrary the number of samples per
rotation decreases (Figure 9). The sampling rate for the optical tyre sensor is
approximately 5100 Hz for each channel, so the number of samples for one revolution
reads:
5100 Hz
N samples = 2π rdyn . (10)
v

The parameters for upcoming force estimation models are calculated by linear least
squares fitting:
c = ( X T X ) −1 X T Y (11)
where X is the tyre sensor data, Y is the test rig data and c is the set of model parameters.

Figure 9 Update frequency and amount of samples per rotation dependence on velocity
(rdyn = 0.492 m) (see online version for colours)
370 A. Tuononen

3.1 Vertical force (wheel load)

Vertical force could be estimated from the intensity (equations (6) and (7)) or
longitudinal signal (equation (4)). The intensity signal is a natural starting point because
it is clear that the contact deformation has to be correlated with the vertical force.
However, under lateral force, the LED is shifted in a lateral direction and thus the
distance between LED and PSD is greater, even if there is no vertical movement at all.
This would disturb the vertical force estimation based on intensity signal. Furthermore,
LED intensity depends on temperature, LED alignment and supply voltage, but lateral
and longitudinal signals are independent of these.
Vertical force has influence on contact length equally. Thus, the contact length could
be indicated by the longitudinal signal amplitudes or the peak locations. Also, standard
deviation of the longitudinal signal correlates with vertical force. Nevertheless,
longitudinal signal amplitude is implemented in this study. The amplitude is calculated as
follows:
xgap = max[ x0 , xend ] − min[ x0 , xend ] (12)

where [x0, xend] includes all the x values for one complete rotation.
This is not the best formulation against noise in the signal, but the real-time capability
has to be fulfilled as well. Furthermore, the signal-to-noise ratio of the longitudinal signal
is quite good for typical wheel loads in trucks. The vertical force estimation reads:
2
Fz = xgap cz ,parabolic + xgap cz ,gain + cz , offset + [ xcz , x gain + cz , x offset ] (13)

where the parabolic term is needed due to the inverse square relation of intensity and
displacement. The x , cz,x gain and cz,x offset terms are needed to compensate the vertical
force estimate under longitudinal force, which slightly increases signal amplitude.
The calculation of x is explained in the longitudinal force estimation section.
The vertical force comparison for the test rig measurement is shown in Figure 10.

Figure 10 Tyre sensor and test rig comparison for vertical force
 On-board estimation of dynamic tyre forces 371

3.2 Lateral force

The lateral force can be calculated only from the lateral displacement signal. The peak
value of signal correlates with lateral force, but the most robust operation is achieved
from the mean value of one rotation. The lateral signal mean value for one rotation can be
calculated recursively:
yk +1 − yk
yk +1 = yk + (14)
k
where yk is reset to zero after each rotation. The lateral force reads
Fy = y c y ,gain + c y ,offset (15)

and the comparisons for test rig measurements are shown in Figure 11.

Figure 11 Lateral force estimate from the tyre sensor compared to the test rig measurements
(slip angle sweep +–6°, 50 kph, wheel load 10, 20 30 kN)

3.3 Longitudinal force

The longitudinal signal offers the possibility to estimate longitudinal force. Whereas the
amplitude of longitudinal signal indicates the vertical force, the mean value of
longitudinal signal can be implemented for longitudinal force estimation. The formula for
estimating longitudinal force is adopted from the lateral force estimation, but naturally
the longitudinal signal is exploited instead of the lateral one:
xk +1 − xk
xk +1 = xk + . (16)
k
The longitudinal force reads:
Fx = x cx ,gain + cx ,offset (17)

and the comparisons for test rig measurements are shown in Figure 12.
372 A. Tuononen

Figure 12 Longitudinal force estimate and test rig measurement for brake steps with different
wheel loads (30 kph, wheel loads 20, 30 and 40 kN)

4 Proving ground vehicle measurements

4.1 Steady-state cornering

Constant radius circle measurements were taken by very slowly accelerating the vehicle
up to the limit cornering speed. The radius of the circle was 150 m. The surface was
packed snow on an ice lake. The track was rather bumpy and resulted in great wheel load
deviations, which can be observed from the tyre sensor and vehicle acceleration sensor
signals.
The normalised lateral force for the tyre sensor (left front tyre) reads:
Fy
Fy , n = (18)
Fz

where Fy is the tyre sensor lateral force estimate and Fz is the corresponding vertical force
estimate. Lateral acceleration is measured with a standard ESP-sensor. The comparison is
shown in Figure 13, where two different measurements are depicted in the same figure
(right and left turn). It can be seen that cornering performance is clearly better for the
right turn. The tyre sensor estimate matches the acceleration signal quite well, especially
for the medium and high accelerations. There is some offset error for the left turn, where
the tyre sensor estimate is slightly greater than would be expected from the lateral
acceleration. This may come from sensor offsets, or possible toe-out or tyre asymmetry.
A stability control intervention can be observed for the right cornering (loop in
55 km/h back to 48 km/h). Figure 14 (same test as that shown in Figure 13) reveals how
the tyre sensor is able to measure short brake intervention, which also results in decreased
lateral force during braking, following the typical combined slip principle of the tyre.
 On-board estimation of dynamic tyre forces 373

The influence of lateral load transfer in steady-state cornering can be seen in

Figure 15, where the tyre sensor vertical force is plotted in time domain (same data as for
Figure 13). The uneven snow surface results in great wheel load deviations (cf. Figure 10
for even surface vertical force estimate).

Figure 13 Lateral acceleration sensor and tyre sensor lateral force estimate for constant radius
circle test (see online version for colours)

Figure 14 Tyre sensor force estimates during stability control brake intervention for left front
wheel (see online version for colours)
374 A. Tuononen

Figure 15 Tyre sensor vertical force estimate for constant radius circle test (see online version
for colours)

4.2 Lane change tests

Figure 16 shows the lateral acceleration measurement and normalised lateral force
estimate for sequential lane changes on packed snow. The measurement and estimate
match quite well during the left turn (negative values) when the load transfer is away
from the sensor tyre.

Figure 16 Lateral acceleration sensor and tyre sensor lateral force estimate for lane changes
(packed snow, 55 km/h)
 On-board estimation of dynamic tyre forces 375

Possible errors for the lateral acceleration signal are:

• roll angle introduces a gravity component to the measured body lateral acceleration
a y − g sin θ
v y =
cos θ
where ay is the measured acceleration and θ is the roll angle
• lateral acceleration sensor is not aligned with the tyre sensor, thus in transient
manoeuvre yaw acceleration has influence on lateral acceleration at the front axle
a y ,front = a y + rω

but neither of these can compensate observed bias between acceleration

measurement and tyre force estimate.
Furthermore, it has been observed that tyre sensor overestimates vertical force in some
combined slip conditions and this results in underestimation of the normalised lateral
force.
However, the data in Figure 16 should not match exactly because measured lateral
acceleration describes the lateral force requirement for a complete vehicle, but the
normalised tyre force describes the contribution of that particular tyre.
The influence of longitudinal acceleration and lane changes for vertical force can be
seen in Figure 17, which is from the same measurement as shown in Figure 16.
The longitudinal load transfer during acceleration can be observed from the decreased
vertical force. The vertical force varies between 24,000 N and 40,000 N during lane
changes.

Figure 17 Tyre sensor vertical force estimate for acceleration and lane changes
376 A. Tuononen

4.3 Braking tests

The optical tyre sensor performance to estimate longitudinal force in braking is shown in
Figure 18. The brake pressure is increased as a ramp until ABS takes control of the wheel
speed after six seconds. The contribution of the left front wheel for the vehicle
deceleration is approximated:
Fx ,tyre
ax ,tyre = g (19)
Fz ,tyre

where vertical force of the left front wheel during deceleration ax is:
gmb + max h
Fz ,tyre = (20)
2(a + b)

where m is the vehicle mass, b the is rear axle distance from centre of gravity, a is the
front axle distance from centre of gravity and h is the height of the centre of gravity.
Furthermore, pitch introduces some gravity component (Figure 19) for the measured
longitudinal acceleration and it is calculated for the global coordinate:
ax ,body − 9.81sin (ϕ )
ax = (21)
cos(ϕ )

where the pitch is simply calculated from the measured pitch rate:
t
ϕt = ∫ ϕ dt (22)
0

which is acceptable for a short period and if the pitch rate sensor offset is exactly known.

Figure 18 Longitudinal acceleration and tyre sensor estimate for ramp braking
 On-board estimation of dynamic tyre forces 377

Figure 19 Gravity component in measured longitudinal acceleration ax,body

The tyre force estimate is slightly too low when compared with the vehicle sensor
(Figure 18). The tyre sensor estimates the contribution of that particular tyre for braking,
whereas the vehicle longitudinal acceleration includes the braking forces of every tyre in
the vehicle. In other words, in the comparison, it is assumed that tyre generates braking
force in linear relation to the wheel load, which is not necessarily the case. However, the
most significant reason for deviation is probably increased tyre pressure (from increased
tyre temperature) in repetitive braking manoeuvres on proving ground. This leads to
a stiffer tyre and thus the tyre forces are slightly underestimated.
Figure 18 also shows how the tyre sensor cannot update force estimate for low
velocities. There is no new tyre force information available after 6.2 s (last marker), when
the left wheel velocity is approximately 15 km/h. The vehicle stopped so quickly after
that moment that the new rotation for the tyre sensor was never completed. In addition,
possible locking of the wheel at the end of braking would disable tyre force estimation.

5 Conclusion

The presented optical tyre sensor has proven to be a powerful tool for measuring dynamic
tyre carcass displacements. The sensor and the presented algorithms are fully real-time
capable in standard low-cost MCUs. The lateral and longitudinal force estimates seem to
be accurate in most conditions, but the vertical force estimate is not reliable in heavy
braking. Even if the tyre force estimates are available only once per revolution, transients
such as stability control brake intervention can be captured.
The accuracy of the optical tyre sensor depends mostly on the parameters, which are
highly related to the inflation pressure. However, the inflation pressure is varied due to
the temperature of the air or nitrogen in the tyre if leakage is neglected. Thus, the
measurement of tyre pressure would improve the accuracy of the optical tyre sensor
considerably and not complicate the algorithm notably.
Further research could be done with a second tyre sensor on the rear axle of the
tractor and a third one on the trailer. This would make it possible to estimate the state of
the vehicle very accurately. On the other hand, several tyre sensors on the trailer could be
used to study imbalanced tyre detection algorithms.
 378 A. Tuononen

Acknowledgements

This work was financially supported by EU-project FRICTION FP6-IST-2004-4-027006.

This support is gratefully acknowledged.

References
APOLLO (2005) Final Report, Available online at: www.vtt.fi/apollo 18.6.2008.
FRICTION (2006) webpage, http://friction.vtt.fi/ 18.6.2008.
Hamamatsu (2003) Hamamatsu PSD Selection Guide, www.hamamatsu.com
Hamamatsu (2008) Hamamatsu PSD S5991-01 Data Sheet, www.hamamatsu.com
Holtschulze, J. (2006) Analyse der Reifenverformungen für eine Identifikation des Reibwerts und
weiterer Betriebsgrößen zur Unterstützung von Fahrdynamikregelsystemen, PhD Thesis,
ika RWTH Aachen University, Aachen, April.
Hüsemann, T. (2007) Tyres in Motion at ika/fka – Tyre Investigations and Research at at ika/fka
Aachen, First FTire User Meeting, fka Aachen, 17–18 July.
Mancosu, F. and Di Pasquale, A. (2007) Adding Value to the – Next Generation Product Inside
Tire, Tire Technology Expo 2007, Cologne.
Morinaga, H., Wakao, Y.H.Y. and Kobayakawa, A. (2006) The Possibility of the Intelligent Tire,
Technology of Contact Area Information Sensing, FISITA 2006, Yokohama.
Tuononen, A.J. (2008) ‘Optical position detection to measure tyre carcass deflections’,
Vehicle System Dynamics, Vol. 46, No. 6, pp.471–481.

View publication stats