2007-01-0809

An Enhancement to an Electronic Stability Control System to

Include a Rollover Control Function
Jianbo Lu, Dave Messih, Albert Salib and Dave Harmison
Ford Motor Company

Copyright 2006 SAE International

ABSTRACT driver conducts an aggressive avoidance maneuver. In

this paper, a function referred to as a Rollover Control
This paper proposes a method to enhance existing Function (RCF), designed to mitigate un-tripped rollovers,
electronic stability control systems such that a certain is studied.
level of rollover mitigation performance is achieved. Such
an enhancement is conducted through a control The simulation results studied in [2] and [3] indicate that
algorithm using only the standard ESC sensors. The an ESC system can increase a vehicles resistance to
analysis presented here reveals that a rollover mitigation un-tripped rollovers without enhancements specifically
system such as this will face a trade-off between the targeting rollover control. This may be because an
vehicles responsiveness and the control robustness due oversteer condition is one of the important contributors to
to error in the roll dynamics model and state estimation. building up large tire lateral forces which can result in un-
tripped rollovers. However, in some driving scenarios,
oversteer control may not be sufficient to counteract an
1. INTRODUCTION un-tripped rollover. If an un-tripped rollover can occur
due to lateral force buildup when the vehicle is not
Since its debut in 1995, electronic stability control (ESC) oversteering (i.e. driver intent is maintained or the vehicle
systems for automobiles have been implemented on is understeering), additional control algorithms are
various platforms [1]. Recently, the DOT announced a needed to achieve rollover control. Additionally, the co-
new proposal (Docket No. NHTSA-2006-25801, existence of traditional ESC functions and the RCF
Proposed FMVSS 126) which would require all function will likely require a new function prioritization and
automotive manufacturers to begin equipping passenger arbitration.
vehicles under 10,000 pounds Gross Vehicle Weight with
ESC systems starting with the 2009 model year and to The addition of a roll mitigation function in an ESC
have the feature available as standard equipment on all system has been actively pursued in recent years. For
vehicles by the 2012 model year. example, [4] describes an enhanced system over Driver
Stability Control systems for commercial trucks. [5]
The traditional ESC systems aim to control the yaw and proposes a stand-alone function called Anti-rollover
sideslip angle of a moving vehicle through individual Braking (ARB) when an impending rollover of a vehicle is
wheel braking and engine torque reduction such that the sensed. In [6], engineers from Bosch describe a rollover
desired path of a vehicle determined through the drivers mitigation function over its ESP system. Continental
inputs (e.g., steering input) can be maintained. That is, Teves has developed an Active Rollover Prevention
ESC systems typically help the vehicle to follow the (ARP) system. This paper provides some extension of
drivers intent such that the driver maintains control of the work developed in [7] which is similar in principle to
the vehicle regardless of the variation of the road [5], [6] and ARP but with a different control strategy. In
conditions. Since they help to keep the vehicle on the addition to ESC-based brake control, other chassis
road, ESC systems may reduce the occurrence of so- control systems have also been pursued to mitigate
called tripped rollovers, which often occur when a vehicle rollovers, see [3], [9], [10], [11], [12] and [13] for more
departs from the road surface and account for details.
approximately one third of all fatalities in single vehicle
accidents. This paper is organized as follows. A brief discussion
about vehicle roll stability is provided section 2, with a
In contrast to tripped rollover, the so-called un-tripped discussion of a roll dynamics model based on standard
rollover occurs when a vehicle stays on the road. Such a ESC sensor information in section 3. Various wheel lift
rollover is also called a friction induced rollover due to detection methods are briefly discussed in section 4.
the fact that it occurs when the tire lateral forces of the Finally, section 5 discusses the RCF control.
involved vehicle exceed certain thresholds for some
duration during aggressive steering maneuvers. For a 2. VEHICLE ROLL STABILITY
vehicle operated with loading that raises the center of
gravity, an un-tripped rollover could occur when the
The vehicular roll instability studied here is the un-tripped
rollover, where the vehicle has divergent roll motion
along the vehicles roll axis in response to steering inputs (1)
while driving on a smooth road.

Such an un-tripped rollover is typically induced by

aggressive steering inputs. More specifically, it is the then and now be expressed as:
result of steering the vehicle so aggressively that the
vehicle approaches its physical limit. Due to the
divergence of the vehicles roll motion, the vehicle can
behave unpredictably. The roll stability studied in this (2)
paper is specifically dedicated to such an un-tripped
rollover.

More rigorously, a vehicle is said to be roll stable if the

roll angle between the vehicle body and the road surface b3
is bounded by a threshold, called a rollover angle limit
,when the vehicle is driven on a smooth road surface b1
which may have an associated bank angle. Such a
b2
rollover angle limit is dependant on the vehicle loading z rr
condition, the drivers inputs and the road condition. m1 m3
zlr
zlf
The above roll stability can also be characterized using
the vertical travels of the wheel centers with respect to
the smooth road surface. That is, a vehicle is said to be m2
roll stable if it does not have sustainable two wheel lift
from the road surface when the vehicle is driven on Fig.1. The body-fixed coordinated system, the moving
smooth roads. road coordinate system and the height sensors.

A vehicle could experience an un-tripped rollover if it is

not roll stable. Physically, the un-tripped rollover can be
directly detected through sensors, for example, the laser s lf
height sensors which measure the distance of the z lf
vehicle body at the sensor mounting locations from the
road surface along the direction of the laser beam.
w lf
Considering mounting 4 laser height sensors on the 4
corners of the vehicle body shown in Fig. 1, their outputs Fig. 2. The displacements of the wheel and the
are denoted as and which measure the suspension system
relative displacements of the vehicle corners at the left-
front, right-front, left-rear and right-rear locations with The relative distance at the 4 corners can be further
respect to the road surface but defined on a body-fixed decomposed into the suspension stroke projected along
coordinate system with axes , and (called the the body vertical axis and the wheel displacement
body frame), more specifically, along the vertical between the road surface and the wheel centers which is
direction of the vehicle body. Lets define another also projected along axis. Fig. 2 shows as an
coordinate system with axes , and which is example. Notice that the suspension usually has different
geometry and the suspension strokes are not necessarily
attached to the road surface but moves and yaws with
parallel to the vehicle body axis. If the four suspension
the vehicle body (called a moving road frame). Let ts be strokes are denoted as and , the four wheel
the distance between the left and the right laser height center displacements are denoted as and
sensors, be the distance from the front to the rear
laser height sensors. Let and be the relative , then we have

roll and pitch angles between the vehicle body and the
road surface which are approximately the Euler angles
between the two coordinate systems and .
(3)
Define
where and are the projection operators that Although the aforementioned laser height sensors were
project the front and rear suspension strokes to the body instrumented in the target vehicle, their implementation
vertical axis , and and are the ratios to convert in mass production to detect the vehicle rollover
condition is generally cost prohibitive with the current
the tire displacement from the wheel location to the laser technology. In practice, we are pursuing using the
height sensor location. standard ESC sensors to detect roll instability. While [8],
[9] and [14] use an extra roll rate sensor together with the
The suspension contributions in and , which information from ESC systems to detect the roll instability,
this paper focuses on rollover detection without using a
are called suspension roll and pitch angles, can be roll rate sensor. In the following section, such a roll
expressed as in the following dynamics sensing method and its limitation in achieving
robust detection are discussed.

(4)
3. ROLL DYNAMICS MODELING

In order to use the available sensors in ESC to detect

And similarly, the wheel contributions in and , vehicular roll instability, the vehicle roll dynamics need to
be related to the sensor signals. The standard ESC
which are called the wheel departure roll and pitch sensors include a yaw rate, a lateral accelerometer, a
angles, can be expressed as steering wheel angle sensor, four wheel speed sensors,
a master cylinder pressure sensor and oftentimes a
longitudinal accelerometer. Additionally, some of the
calculated variables in ESC such as the vehicle
(5) reference velocity, the estimated brake caliper pressures,
and the individual wheel control status will also be used.

The motion sensors of longitudinal acceleration (optional

Then in ESC), lateral acceleration and yaw rate are packed in
a sensor cluster which is rigidly mounted on the body
(6) close to the vehicles center of gravity. Denote the offset
compensated and filtered longitudinal acceleration as ,
lateral acceleration as and yaw rate as . Denote
Notice that although the relationship in (2) is nonlinear,
the decomposition in (6) is linear if we work with the coordinate system defining the sensor signals as
frame (called a sensor frame). Generally speaking the
and in place of and . Also considering
sensor frame and the body frame do not coincide
that for the controlled vehicle discussed here, the due to sensor cluster mounting error and sensor element
magnitude of and are small (e.g. less than 10 alignment error within the cluster. Denote the vehicle
reference velocity calculated in ESC as . If we align
degrees), and will be
the laser height sensors with vertical direction of the
very close to the true angles of sensor frame , the similar angles to those defined in
the last section can be also defined as and ,
and . For this reason, in the sequential discussion,
.
and will be
interchangeably used with and
Notice that an accelerometer measures the total
. acceleration along the sensor measurement direction,
including the influence of gravity. The lateral acceleration
of the vehicle body at its c.g. location projected on the
Defining the time duration after a time instant t such that
lateral axis of the frame can be expressed as
the magnitude of the body-to-road roll angle is
beyond a threshold as the following
(8)
(7)
where is the distance between the origin of the ESC
Then the vehicle is roll stable if is below a motion sensor cluster and the c.g. of vehicle body.
threshold . Notice that although the above definition
In order to relate the vehicle roll dynamics with the
of the roll stability is rather kinematics in nature, the sensor signals defined in the sensor frame, the vehicles
threshold used is dynamics dependant. suspension kinematics and compliance information is
needed. In this paper, we consider an SUV with a front where is the distance between the left and right
independent suspension and a rear solid suspension. suspensions; and are the front and rear
Due to the constraints of the rigid links in the suspension,
suspension spring rates; and are the stiffness
the vehicle body can only roll and pitch along a fixed
direction, which is called the roll and pitch axes. Notice of the front and the rear anti-roll bars. and are
that the roll and pitch axes are usually different from the the front and the rear suspension damping rates.
vehicles body-fixed longitudinal and lateral axes. From
Fig. 3 for the SUV of interest, the front roll center is close and are the equivalent roll stiffness and roll
to the ground, and the rear roll center is close to the damping rate for the suspension.
vehicle floor.
Body frame
Sensor frame

a ycgs
xsw
Slr
Fyr Fzr

xwm
Srr
xm
Fig. 3. Left: front independent suspension. Right: rear
solid suspension.
Fig.4. A vehicle during a rollover event on a banked road.
3.1. Roll Sensing Based on Vehicle Body Model
From (9), the lateral tire forces defined on the sensor
Let and be the resultant forces along the lateral frame but applied to the vehicle body through its front
and rear roll centers can be computed as in the following
direction of the motion sensor but applied to the vehicle
body through the front and rear roll centers. Let the
vertical distance from the vehicle body c.g. location to the (11)
front and rear roll centers be and . Using Newton's
law in the sensor frame , we obtain the following
equations of motion and the vehicle body roll acceleration defined in the
sensor frame can be estimated as in the following

(9) (12)

where the coefficients are defined as

where and are the moment of inertia of the car

body with respect its body axes; is the sprung mass
(the mass of the car body); and are the distance
of the vehicle body c.g. to the front and rear axles with
.

The roll moment due to the suspension forces is A similar discussion could lead to the following for pitch
computed as acceleration estimation

(13)

(10) where the coefficients and can be similarly

obtained.
Considering the roll rate in (12) can be related to the which implies that the roll angle between the sensor
other roll angles as in the following frame and the moving road frame is influenced by the
wheel departure roll angle.
(14)
3.2. Roll Sensing Based on a Rear Wheel Model
where is the relative roll angle between the sensor
The aforementioned roll model of the vehicle body
frame and the moving road surface, i.e., it is the sum of demonstrates the necessity of characterizing wheel
the suspension roll angle and the wheel departure departure roll angle in determining the relative roll angle
angle . is the pitch angle between the sensor between the sensor frame and the road. Now let's
frame and the horizon, which can be approximated using compute such a wheel departure roll angle based on the
the sensor signals as wheel roll model.

The vehicle of interest has rear wheel lift earlier than

front wheel lift due to the tilted roll axis. Hence the roll
dynamics of the rear solid axle are a good early
indication of a potential vehicle rollover. Based on the
is the angle between the horizon and the road force balancing on the rear wheel set shown in Fig. 5
surface, or the road bank angle, as shown in Fig. 4. with the assumption that the vertical motion of the c.g. of
Based on (12) and (14) and taking the Laplace the wheel set is only due to its roll motion, the roll
transform, we obtain the relative roll angle between the dynamics after wheel lift has occurred are
sensor frame and the moving road frame:

(15)

where the transfer functions are defined as (17)

(16)
where denotes the roll rate of the rear wheel set,
its roll moment of inertia, its mass, the
Equation (15) shows that , which reflects the relative rolling radius of the tire, the half track of the wheel,
roll angle between the vehicle body and the road, is a is the distance from the rear roll center to the center
function of the road bank angle, the wheel departure roll
angle, the vehicle lateral acceleration and the vehicle of the rear axle, the roll moment applied from the
yaw rate. rear suspension to the wheel set which can be
approximated as , the lateral
Since in (15) is the form of a high-pass filter, when
force applied to the outside wheel from the moving road
the wheel departure roll angle is very dynamic (consisting frame, the normal force applied to the outside
of high frequency content), then (15) can be further
simplified as wheel. is the vertical force applied to the wheel set
through the rear roll center, and can be estimated from
longitudinal acceleration and the vehicle's static mass
distribution.
which indicates that the roll angle from the sensor frame
Based on equations in (17), the wheel set roll
to the moving road frame is influenced by the road bank.
acceleration can be expressed as a linear function
If the sum has only low frequency content of and but with nonlinear
(i.e., the wheel lifts slowly from the horizon), then (15) dependence on various angles such as , the wheel
can be simplified as departure roll angle and the road bank angle .

Notice that the wheel departure roll angle satisfies the

following relationship
 (18) to stay as constant. Another way to condition road bank
angle in the above computation is taking into account the
other variables. For example, during a wheel lift event,
Sensor frame Body frame is saturated at a certain level due to suspension
travel limitations, and its rate of change is close to zero.
a ycgs
xsw 3.3. Implementation Oriented Roll Approximation

Based on the discussion above, one can conclude that

the key variable used in control, namely, the relative roll
Fzr angle between the vehicle body and the road surface is
Fyr influenced by the road bank angle, the wheel departure
xwm angle, the vehicle's cornering acceleration, the vehicle's
c.g., etc.
M rollr
xm Fyrm Roof loading can significantly change the roll moment of
inertia and the sprung mass used in (9), hence
the computation of in (15) using the aforementioned
Fzrm model-based approach through sensor signals is also
influenced by roof loading.
Fig.5. The forces applied to the rear wheel set.
If in (15), the road bank angle changes slowly, while the
Using the small angle approximation, (17) can be wheel departure roll angle varies quickly, we could obtain
reduced to the approximation , where

(20)

(20) is a good approximation of if the vehicle is

driven on a smooth flat road surface without wheels lifted
(19) from the road. Notice that, if is tuned for a vehicle
without roof loading, its magnitude will be significantly
smaller than that of the true of the vehicle with roof
loading, and if it is tuned for vehicle with roof loading, its
magnitude will be significantly larger than that of the true
of the vehicle without roof loading.
Combining (18) and (19), the 4 unknowns can be solved
from the 4 equations. The 4 unknowns are Additionally, if the vehicle is experiencing slow wheel lift
and . By eliminating the other or dynamic road bank variation, computed in (20)
three unknowns, a second order differential equation for will be significantly different from the true .
can be generated. Therefore, the wheel departure
In order to implement a rollover control, we compute a
angle is a function of , lateral acceleration , roll-signal-for-control based on a weighted
and yaw rate and road bank angle . computation of and (see section 3.2)

Notice that the aforementioned computation will need to

(21)
know when the computation should begin, i.e., when
wheel lift initiates. Due to sensor offsets and low
frequency drifts, a wash-out filter is used to compute the where is a weight that is a function of the wheel lift
wheel departure roll angle in the above status and the other sensor or computed signals, and
computation. Since satisfies is another weight based on the confidence of the
computation of the wheel departure roll angle .

Figure 6 and Figure 7 display computations based on

the only variable which is not available for the sensor data in a vehicle tested on level ground. Figure 6
aforementioned model is the road bank, . Certain shows tuned for maximum-allowed loading, where
assumptions about can be made, for example, the vehicle is equipped with outriggers and roof loading
(with ESC off). It shows that during a wheel lift event,
during a rollover event, the road bank may be assumed
 is saturated around -5.5 degrees, while the true roll ineffectiveness of the brake control. On the other hand,
angle exceeds -16 degrees. overestimation of the true signal leads to unnecessary
brake control activations. Therefore, using lateral
acceleration as the key input to characterizing the vehicle
10 roll dynamics will result in making a trade-off in the
Tilde
xsm
Roll from body height sensors
control.
5

0
4. WHEEL LIFT DETECTION
degree

-5

In order to determine when the vehicle is experiencing

wheel lift such that the wheel departure roll angle can be
-10
computed based on model (15), or to prepare the brake
hydraulics for large brake pressure buildup, various
-15
wheel lift detections are conducted together with vehicle
roll determination. Three aspects of wheel lift detection
-20
2 2.5 3

Time (sec)
3.5 4 4.5 are studied here: (i) Active Wheel Lift Detection (AWLD),
(ii) Passive Wheel Lift Detection (PWLD) and (iii)
Integration of AWLD and PWLD.
Fig. 6. Comparison between and the true roll angle
based on body height sensors during a two wheel lift 4.1. Active Wheel Lift Detection
event for a vehicle with a roof loading
AWLD is used to determine if a wheel is lifted or
10 grounded by checking the wheel rotation in response to a
Tilde
Tilde
xsm
tuned for normal loading
tuned for roof loading
given brake pressure. It utilizes signals from the brake
8 xsm
Roll from body height sensors control system such as the wheel speed sensor signals,
6
the individual estimation of each of the brake caliper
pressures, etc. More specifically, it sends a small brake
4
pressure to an inside wheel, then checks the response of
that lightly braked wheel. If the vehicle lateral
degree

acceleration sensor indicates a hard cornering of the

0
vehicle on a high mu surface and the inside wheel
-2
experiences a longitudinal slip ratio larger than a
threshold in response to a relatively small brake pressure,
-4
then this inside wheel is likely to be lifted from the ground.
In order to differentiate the slip ratio of the inside wheel
generated from the low mu surface from the one of the
-6

0 2 4 6 8 10 12 14 16 18 20

Time (sec)
inside wheel generated from the lifted wheel during high
mu maneuvers, AWLD is activated when the vehicle
Fig. 7. The comparison between and the true roll achieves a certain level of cornering acceleration.
angle based on body height sensors during a slalom Implementing AWLD requires special considerations to
maneuver on a level ground. address system interactions. Notice that during driver
braking, a lightly braked but lifted wheel will likely initiate
Fig. 7 shows the computation of in a slalom on ABS to dump the brake pressure. The initial pressure
level ground for a vehicle equipped with and without roof build caused by the driver braking, followed by the ABS
load. If is tuned for a vehicle with roof loading, to action is similar to AWLD. On the other hand, in order to
maximize brake pressure build rate on the control wheel,
characterize the roll angle between vehicle body and the
AWLD brake pressure is inhibited for the wheel that is on
road surface, it overestimates for a vehicle without roof
the same hydraulic circuit as the control wheel.
loading. On the other hand, if is tuned for a vehicle
without roof loading, it significantly underestimates for Due to the reactive nature of this strategy, a lifted wheel
vehicles with roof loading. detected by AWLD suffers a time delay which can be
several hundred milliseconds. Namely, if the control
One can conclude that the ESC sensor set alone can not brake pressure command is solely based on a PL or AL
provide all the necessary information to calculate a status in AWLD algorithm, the hydraulics can not provide
relative roll angle between the vehicle body and the road sufficient pressure to control the vehicle instability. A
surface that is robust to varying load conditions and significant advantage of AWLD is the ability to reduce
varying road bank. potentially unnecessary activations when a PG or AG is
detected.
Since is used as a feedback control variable, the
underestimation of the true roll angle leads to potential
4.2. Passive Wheel Lift Detection The wheel lift status for each wheel is set to one of 5
levels. The 5 levels are: absolutely grounded (AG),
The intent of PWLD is to determine if a wheel is lifted or possibly grounded (PG), no indication (NI), possibly lifted
grounded by checking the vehicle dynamics and wheel (PL) and absolutely lifted (AL).
speed behavior without actively requesting brake
pressures. Namely, it passively monitors the wheel Fig. 8 illustrates the integration of AWLD and PWLD. A
speeds to determine if the speeds indicate a potential detailed description of the above wheel lift detection
wheel lift condition. methods can be found in [15].

One aspect of PWLD can be characterized by rolling

radius-based axis roll angle, which captures the angle
between the wheel axle and the average road surface
through the dynamic rolling radii of the left and right
wheels. Since the computation of the rolling radius is
related to the wheel speed and the linear velocity of the
vehicle at each of the four corners, such a rolling-radius
based wheel departure angle will assume abnormal
values when there are large wheel slips. This happens
when a wheel is lifted and there is torque applied to the
wheel. Therefore, if this rolling radius-based wheel
departure angle is increasing rapidly during a maneuver
with large cornering acceleration, the vehicle may have
lifted wheels. A small magnitude of this angle indicates
the wheels are likely grounded.

PWLD can also be conducted by considering the normal

loading sustained at each wheel. Theoretically, when a
wheel's normal loading decreases to zero, this indicates
that the wheel is no longer contacting the road surface. A
large magnitude of this loading indicates that the wheel is
likely grounded.

Another aspect of PWLD compares the actual road

torques applied to the wheels and the road torques which Fig. 8. The integration between AWLD and PWLD
are needed to sustain the wheels when they are
grounded. The actual road torques can be obtained
through torque balancing for each wheel using wheel 5. ROLLOVER CONTROL FUNCTION
acceleration, estimated engine torque and estimated
brake torque. If the wheel is contacting the road surface,
In this section, the RCF control structure will be
the calculated actual road torques must match or be
discussed. Fig. 9 provides a schematic overview of
larger than the torques determined from the nonlinear
RCF.
torques calculated from the normal loading and the
longitudinal slip at each wheel.

Finally, PWLD can be conducted by checking wheel Sensor/

Calculated Feedforward
longitudinal slip. If during a normal braking or throttle Signals Control
apply, the wheels on the inner curve side of the vehicle
experience increased magnitude of slip, then those
wheels are losing longitudinal road torque, which implies Feedback Control
that the wheels could be lifted during maneuvers
involving large cornering acceleration. Roll Dynamics
Front Sideslip
Angle Feedback +
Model
Notice that all the methods in PWLD are conducted Yaw Feedback
when the vehicle achieves substantial cornering
Roll Feedback
acceleration (e.g., greater than 0.5 g). Wheel Lift
Detection
4.3. Integration of AWLD and PWLD
Fig. 9. RCF Algorithm Integration
In order to capitalize on the benefits of AWLD during
steady-state driving conditions and the instantaneous
nature of PWLD, an integration of AWLD and PWLD is
required.
5.1. Mechanism to Reduce Roll Moment
Since the road bank angle influences the computation of
The front tire lateral force projected on the plane which is mainly influenced by the vehicle lateral
perpendicular to the roll axis is the main reaction force acceleration, the PD control based on will need to
causing un-tripped rollovers. Such a force is called the
effective lateral tire force (ELTF). (11) provides a simple conduct a similar trade-off between a banked road and
characterization of ELTF. Notice that a rollover is unlikely level ground.
to happen for maneuvers with a higher frequency of
steering input such as an aggressive slalom, where the Due to limited hydraulic capabilities, a leading indicator
ELTF does not have a long duration in any single vehicle of is needed to sufficiently control potential
cornering direction. However, a rollover is more likely for rollover events. Therefore another control signal used in
maneuvers with steering input of medium frequency, RCF is the model-based linear sideslip angle, , at
where the ELTF has a longer duration to accumulate and the front axle, which is the front tire lateral force times
can be sustained at critical levels. A similar situation the front tire compliance.
could occur during a maneuver with a mixed high and
low frequency steering input such as a fishhook or a J- The control based on significantly leads the
turn. Due to the tire friction circle, braking can be used to
effectively reduce the buildup of the ELTF. control. However, also has the potential to be
relatively erratic, potentially leading to a premature
5.2. Control Signals reduction in control. Therefore, a robust signal is needed
to fill in the resulting control gap between and
In order to execute RCF, a large brake pressure is
control. A yaw rate-based PD controller can
requested on the front outside wheel during potentially
unstable events. When RCF requests the maximum accomplish this, and is implemented as shown in Fig. 10.
pressure build rate, significant delays in brake pressure
buildup can occur. Therefore, if a brake pressure buildup In such a control structure including three feedback
is requested after the roll instability is underway there controllers and a feedforward controller, the phasing in a
may not be sufficient time to build an adequate control fishhook maneuver would be such that a particular
pressure to mitigate the event. The control effectiveness controller is dominant as the transitional maneuver
would be even less robust if the vehicle has large roof progresses (see Fig. 10).
load. To deal with such a brake pressure build delay, a
feedforward control is used to pre-charge the hydraulic 8 0

system. Such a feedforward control utilizes the prediction Feedforward Precharge

information based on the driver's steering and the other 7 0
Front Sideslip Angle
vehicle state information to provide a pressure build prior Yaw Rate
to the roll instability. Note that this pre-charge is 6 0

Roll Angle
designed to minimize pressure build delay, and therefore Final RCF Request
Pressure (bar)

is a relatively small pressure to establish a speed in the

5 0

brake controls pump and reduce caliper knockback. 4 0

The feedback control used in RCF is a coordination and 3 0

combination of three feedback control commands based

on three different control signals. 2 0

One of the feedback control signals used in RCF is 1 0

computed in (21). The feedback brake pressure 0

command from uses a PD feedback control 1.3 1.4 1.5 1.6 1.7 1.8 1.9 2

where the control gains and deadbands are functions of Time (s)
various sensor and computed signals. As we discussed
before, is not able to adapt to various vehicle Fig. 10. Pressure Profile of RCF Control Event
loading conditions, resulting in an unavoidable trade-off.
For example, if the PD control gains and deadbands are
tuned for a vehicle with roof loading, although the control 5.3. Arbitration and Integration
might be effective for vehicles with roof loading, it will
cause unnecessary brake activation during dynamic The ESC system gives a driver full ability to control the
maneuvers for vehicles without roof loading. Such vehicle, but with intervention when needed to help the
unnecessary braking might inhibit the responsiveness of vehicle follow the drivers intent. One of the biggest
the vehicle. Conversely, if the PD gains and deadbands differentiators between ESC and the rollover control
are tuned for a vehicle without roof loading, it will not be function is that the brake control is no longer solely in
optimized to mitigate rollover in a vehicle with roof response to driver intent.
loading.
It is possible that rollover mitigation systems may cause This compromise is driven by state estimation
the vehicle to understeer, which could lead to the techniques that can not differentiate among various road
activation of the ESC system to request understeer conditions (such as road bank) or the extremes of
control during an RCF activation, i.e., the RCF function is vehicle loading.
counteracted by the ESC understeer control. For this
reason, it is important to integrate the RCF and ESC Continuous improvement of the system performance will
functions. likely require additional hardware. A system with
additional sensors, such as the roll-rate based
On the other hand, if during an RCF activation ESC system, provides the next level of system
oversteer control is also activated, the arbitrated brake performance, where the control effectiveness and vehicle
pressure should pick the maximum between the ESC responsiveness can be simultaneously achieved. With
oversteer control pressure command and the RCF the addition of roll-rate, the roll dynamics model can be
control pressure command together with a slip control robust to varying road conditions or varying vehicle
function. loading. Additionally, some incremental benefit could be
obtained by tailoring the level of control with information
Notice that RCF also must be integrated with the ABS obtained from the roll-rate sensor. For example, while
function. While ABS aims to maintain a certain slip target the yaw rate or front sideslip angle are not sensitive to
to optimize stopping distance and steerability when in an roof loading or road bank, the same yaw rate or front
ABS event, a roll mitigation system such as RCF will sideslip angle could lead to a more unstable situation in a
likely request an alternate slip target, so as to modulate vehicle with roof load compared to a vehicle without roof
lateral forces and subsequently reduce the resulting roll load. Therefore, there could be advantages to making
moment. the controller gains and deadbands dependent on
information from the roll-rate sensor.
The RCF is targeted to reside in the brake ECU where
the ABS, TCS and ESC functions reside, such that the
integration between RCF and the existing functions can
be easily implemented. A block diagram for such a REFERENCES
configuration is shown in Fig. 11, where the lower block
depicts the brake ECU which is divided into two parts: 1. A. T. van Zanten, "Bosch ESP Systems: 5 years of
the lower portion contains the existing functions and their Experience", SAE 2000-01-1633.
priority and arbitration logic together with all the fail-safe 2. A. Ungoren and H. Peng, "Rollover propensity
and interface logic; the upper portion includes the RCF evaluation of an SUV equipped with a TRW VSC
and its priority logic. system", SAE 2001-01-0128.
3. A. Hac, "Influence of active chassis systems on
vehicle propensity to maneuver-induced rollovers",
SAE 2002-01-0967.
4. L. Palkovics, A. Semsey and E. Gerum, "Rollover
prevention system for commercial vehicles
additional sensorless function of the electronic brake
system", Vehicle System Dynamics, Vol. 32, pp. 285-
297, 1999.
5. T. J. Wielenga, "A method for reducing on-road
rollovers anti-rollover braking", SAE 1999-01-0123.
6. E. K. Liebemann, K. Meder, J. Schuh and G.
Nenninger, "Safety and Performance Enhancement:
the Bosch Electronic Stability Control (ESP)", SAE
2004-21-0060.
7. J. Lu, T. Brown and J. Meyers, "Enhanced system
for yaw stability control system to include roll stability
control function", US Patent 6654674.
8. T. Brown and D. Rhode, "Roll over stability control
Fig. 11. Function partition in a brake control ECU for an automotive vehicle", US Patent 6263261.
9. A. Hac, T. Brown and J. Martens, "Detection of
vehicle rollover", SAE 2004-01-1757.
6. CONCLUSION 10. J. Ackermann and D. Odenthal, "Robust steering
control for active rollover avoidance of vehicles with
elevated center of gravity", Proc. International
Rollover mitigation can be enhanced beyond the benefits
Conference on Advances in Vehicle Control and
provided by a traditional ESC system. However, when
Safety, Amiens, France, July 1998.
implemented with the standard ESC sensor set, this
11. Y. Lee, "Control methodology to alter automobile
study has shown that the benefits are subject to a trade-
rollover tendencies", NASA Tech Brief, vol. 24, No.
off between the effectiveness in rollover control and the
4, 2000.
responsiveness of the vehicle in roll-stable situations.
12. T. Shim and D. Toomey, "Investigation of active Email: [email protected]
steering/wheel torque control at rollover limit
maneuver", SAE 2004-01-2097. Dave Messih
13. S. Yoon, J. Jung, B. Koo and D. Kim, "Development Supervisor
of rollover prevention system using unified chassis NAE Brake Controls
control of ESP and CDC systems", SAE 2006-01- Ford Motor Company
1276. 2400 Village Rd, P.O. Box 2053
14. J. Lu and T. Brown, "Attitude sensing system of an Dearborn, MI 48124
automotive vehicle relative to the road", US Patent Email: [email protected]
6556908.
15. J. Lu, J. Meyers, M. Brewer and T. Brown, "System Albert Salib
for dynamically determining the wheel grounding and Senior Engineer
wheel lifting conditions and their applications in roll NAE Brake Controls
stability control", US Patent 6904350. Ford Motor Company
2400 Village Rd, P.O. Box 2053
Dearborn, MI 48124
CONTACTS Email: [email protected]

Jianbo Lu, Ph.D. Dave Harmison

Technical Expert Development Engineer
Active Safety Research & Advanced Engineering NAE Brake Controls
Ford Motor Company Ford Motor Company
20300 Rotunda Dr. 2400 Village Rd, P.O. Box 2053
Dearborn, MI 48124 Dearborn, MI 48124
Email: [email protected]