Você está na página 1de 12

IEEE/ASME TRANSACTIONS ON MECHATRONICS, VOL. 4, NO.

3, SEPTEMBER 1999 223

The Development of Vehicle Stability Control at Ford


Hongtei Eric Tseng, Behrouz Ashrafi, Member, IEEE, Dinu Madau,
Todd Allen Brown, and Darrel Recker, Member, IEEE

Abstract—This paper addresses realistic subjects encountered I. INTRODUCTION


in the challenge of achieving technology improvement in a vehicle
stability control system. They include driver intent recognition,
vehicle status measurement and estimation, control target gen-
eration, system actuation efficiency and smoothness, road bank
I N THE PAST two decades, as technology has evolved, the
use of electronic control in automotive applications has
spread from powertrain to chassis systems. Active systems
angle detection, system development and evaluation, and fault have been developed to improve vehicle control and safety.
detection. Among them, anti-lock brake systems (ABS’s), traction con-
Index Terms—Relative steering sensor, road bank estimation, trol (TC), and vehicle stability control systems (VSC’s) have
side slip observer, yaw control. already found their way into production passenger vehicles
[1], [2].
NOMENCLATURE VSC’s provide vehicle stability and handling predictability
through the interaction of multiple chassis actuators. Cur-
Distance from center of gravity (c. g.) to front rently, the external inputs to a vehicle stability controller
axle. generally consists of a steering wheel angular sensor, yaw
Distance from c. g. to rear axle. rate sensor, lateral accelerometer, and wheel speed sensors.
Wheelbase. A VSC applies vehicle yaw torque through individual wheel
Tire steer angle, positive counterclockwise braking integrated with the existing ABS/TC technology.
(CCW). These advanced technologies of interactive vehicle dynamics
Road bank or crown angle, positive as car left have been developed at Ford in the pursuit of increased safety,
side up. improved performance, and cost efficiency.
Front tire slip angle. The challenges encountered are categorized in this paper
Rear tire slip angle. as driver intent recognition, control development philosophy,
Mass of the vehicle. vehicle side slip estimation, road bank angle detection, system
Vehicle lateral velocity estimate, positive toward evaluation and development, and fault detection.
right.
Estimate of tire lateral force.
Front tire lateral force. II. DRIVER INTENT RECOGNITION
Rear tire lateral force. Since the stability control system has the ability to affect
Vehicle longitudinal velocity, positive forward. the vehicle’s attitude and motion, a function normally reserved
Yaw rate, positive CCW. for the driver, it needs to accurately interpret what the driver
Lateral acceleration, positive toward right. intends for the vehicle motion in order to provide added
Yaw rate, measured value from sensor. directional control (within physical limitations) as a driver’s
Lateral acceleration, measured value from sensor. aid. Responsiveness, consistency, and smoothness are essential
Understeer coefficient. for a driver’s confidence and comfort with the system. These
Vehicle’s moment of inertia about yaw axis. are the guiding principles for our development.
Yaw rate sensor’s drift or dc offset. A driver typically expresses directional intent through the
Lateral accelerometer’s drift or dc offset. steering wheel. The angular position of the steering wheel is
Normal ground reaction force. the first measure of driver intent.
Vehicle’s body side slip angle.
Front tire cornering stiffness. A. Steering Wheel Angle (SWA) Estimate
Rear tire corning stiffness.
Acceleration due to force of gravity. The SWA estimate is a measure of the angle of the steering
wheel with respect to the steering wheel position when the
vehicle is being driven straight. Two types of SWA sensors
exist: absolute sensors, which through the use of added hard-
Manuscript received December 21, 1998; revised May 5, 1999. Recom-
mended by Guest Editor H. Peng. ware provide an estimated SWA soon after ignition key-on,
H. E. Tseng and T. A. Brown are with Ford Motor Company, Dearborn, and relative sensors, which use software to “learn” the position
MI 48121 USA. of the steering wheel after each ignition key-on. Both absolute
B. Ashrafi, D. Madau, and D. Recker are with Visteon Automotive Systems,
Dearborn, MI 48121 USA. and relative steering sensor systems have to determine the true
Publisher Item Identifier S 1083-4435(99)07433-5. center position of the steering wheel.
1083–4435/99$10.00 © 1999 IEEE
224 IEEE/ASME TRANSACTIONS ON MECHATRONICS, VOL. 4, NO. 3, SEPTEMBER 1999

crossing. Zero degrees SWA will be approximately equal to


zero yaw rate under stable conditions. Once the SWA center
is “learned,” the SWA is estimated by

where
SWA estimate;
SWA relative to zero, the position at
ignition key-on;
estimate of the steering wheel center
position relative to zero.
After the driver turns the ignition on, there are three general
types of driving maneuvers that cover all possible stable
driving scenarios. The vehicle can be turning for a period of
Fig. 1. Relative steering wheel signal. time without crossing the true center position of the steering
wheel, the vehicle can be turning while crossing the true center
position of the steering wheel (i.e., similar to a slalom), or the
In the relative sensor system, a centering algorithm is vehicle could be driving straight. The goal is to determine the
required to estimate the SWA for each ignition key-on cycle. SWA center position of the steering wheel for each one of the
The system must first determine the “on-center” position of cases defined above.
the steering wheel which corresponds to the vehicle being 1) Turning Maneuver: If the steering wheel never crosses
driven straight in order to estimate the SWA. The VSC will not the true center position after key-on, the algorithm estimates
function until an accurate center position has been established. the SWA center position by use of the yaw rate gain equation
Therefore, the operational constraints for the algorithm are which estimates the average steer angle of the front wheels,
greater than those of an absolute sensor system. Specifically, given the variable inputs of vehicle longitudinal velocity
the algorithm must determine the center position shortly after and yaw rate
ignition on and/or shortly after the driver reaches the velocity
that enables system function, otherwise the system will be
temporarily disabled and the driver alerted.
In the absolute sensor system, a centering algorithm is used Once is calculated, the SWA can be estimated directly by a
to validate the absolute SWA and to adjust for small changes predetermined dynamic steering gear ratio which accounts for
in the center position of the steering wheel. Although this both the kinematics and compliance in a steering mechanism.
system usually knows the SWA immediately after ignition, it 2) Slalom Maneuver: By definition, a slalom maneuver
may take longer to readjust itself if the alignment of the vehicle contains an instant of directional change in vehicle yaw
has changed between ignition cycles. The centering algorithm rate that is typically accompanied by the steering wheel
would have to detect the change and relearn the absolute crossing the true center with a nonzero steering wheel angular
center. If the error is significant, system function would then velocity. Under stable conditions, the yaw rate response always
be inhibited during the adjustment period to avoid nuisance lags the steering wheel input. The lag time is related to
activation. An absolute steering sensor also costs notably more the velocity of the steering wheel as it passes through the
(on the order of four to five times the cost of a relative sensor.) true center. By monitoring the yaw rate sensor signal for
Because of the service concerns and the significant financial the directional changes in both directions and capturing the
impact of using an absolute sensor, the challenge (to improve corresponding steering wheel positions when the changes
customer value) is to use a relative sensor and develop an occur, the two captured steering wheel positions, Max_Center
algorithm that would determine quickly and accurately the and Min_Center, will be on opposite sides of the true center
angular position of the steering wheel after each ignition cycle. position (see Fig. 2). The steering wheel center estimate is
The relative SWA sensor used in our development is a stan- then given by
dard quadrature-encoded (i.e., gray code) two-channel digital
sensor which provides both relative position and direction. By
design, only one bit of the two-bit code is allowed to change
at any transition (Fig. 1). 3) Driving Straight: When a vehicle is driven straight for
At key-on, an uncentered SWA count (SWA_Uncentered) the first time after ignition key-on, the current steering wheel
variable is initialized to zero and incremented or decremented position is captured and filtered to determine the SWA center
based on the position and direction of motion of the sensor. relative to the key-on position. The algorithm further allows
All calculations are relative to the zero point of the uncentered long-term adjustment while the vehicle is being driven straight.
steering wheel position. By design, the yaw rate sensor signal The following criteria are used to determine when the
and the estimated SWA will have the same sign with minor vehicle is “driven straight.”
exceptions. These exceptions are caused by the lag between 1) Longitudinal velocity (forward or reverse) of the vehicle
vehicle yaw rate and SWA that occurred during the true center must be greater than a predefined limit, in order to guar-
TSENG et al.: DEVELOPMENT OF VEHICLE STABILITY CONTROL AT FORD 225

Fig. 2. Captured cross-zero SWA position.

antee that the yaw rate signal-to-noise ratio is adequate


for processing.
2) The vehicle must be stable.
3) The sensor must be initialized.
4) The filtered yaw rate must be less than a vehicle speed
dependent yaw rate limit.
5) The filtered yaw acceleration must be less than a vehicle
speed-dependent yaw acceleration limit.
When the previous criteria is satisfied for a predetermined
length of time, a high confidence exists that the vehicle is
“being driven straight.”
It was shown, through a variety of vehicle tests and fleet
vehicle evaluations, that the relative steering angle sensor
system performance is comparable to absolute steering angle
sensor systems and more robust to banked road disturbances.
A typical example of such vehicle test is shown in Fig. 3.
More details on the algorithms are available in U.S. Patents
5 790 966 and 5 787 375.

B. Driver-Desired Yaw Rate Fig. 3. Example of steering wheel estimation at key-on.

The basic approach to determining the driver’s directional


intent is to determine a desired yaw rate based upon a
simplified vehicle “bicycle” model with appropriate limitations
within physical constraints. The controller will then use this
information to determine when it needs to intervene and try to
control the vehicle to this target while maintaining reasonable
limits to the side slip of the vehicle. It is essential for the
target yaw rate to be responsive to the driver’s commands, to
be dynamically consistent with the base vehicle (under normal
operation), to be smooth, and to be reasonably constrained by
the physical limits of the vehicle and road surface. We take a
two-stage approach to this problem: 1) determine a yaw rate
target based upon a dynamic bicycle model appropriate for the
vehicle and 2) smoothly limit the target yaw rate based upon
the lateral acceleration of the vehicle.
The bicycle model is a dynamic model based upon the
equations of motion of a rigid body described in Fig. 4. The Fig. 4. Bicycle model.
equations of motion are

With the assumptions that the longitudinal velocity is


known via wheel speeds, the lateral tire forces can be
approximated as where is the tire slip angle and
is the tire cornering compliance. The front slip angle can be
226 IEEE/ASME TRANSACTIONS ON MECHATRONICS, VOL. 4, NO. 3, SEPTEMBER 1999

approximated as steering wheel changes. Note that the target yaw rate does
not need to be accurate at the vehicle limits of operation,
since the controller also uses vehicle side slip information in
determining the system actuation output. In other words, the
by using small angle approximations as the lateral velocity at vehicle stability controller is robust to target yaw rates that are
the tire divided by the longitudinal velocity at the tire. beyond the physical limits of the vehicle and road surface.
Likewise, the rear slip angle can be approximated as Note that nowhere in this target yaw rate determination is
the theoretical road surface coefficient directly calculated.
Although the use of the washout delta is related, it only is
A yaw rate target can then be dynamically calculated based used when the vehicle limits are exceeded by the high- de-
on the driver’s steering input and vehicle speed using Euler veloped bicycle model. Calculation of road surface coefficient
approximation of the equations of motion with a hard limit under conditions where the vehicle is not at the limits of
(in magnitude) of the yaw rate. This limit is experimentally adhesion is very difficult and simply not necessary.
determined based upon high- vehicle performance.
Note that the following vehicle parameters are required III. CONTROL DEVELOPMENT PHILOSOPHY
to tune the bicycle model: vehicle mass, c. g. location with
To provide the best customer value, the development of
respect to the front and rear axle, rotational moment of inertia
stability control systems must involve on-going subjective and
in the yaw axis, the front and rear cornering compliances,
objective evaluations, as well as persisting control refinement
and the maximum high- vehicle yaw rate. Also note that we
based on a sound system development philosophy. Target
assume that a separate function has determined the average
lateral response, system transparency, and actuation efficiency
front steer angle from the SWA. This will allow the target
and smoothness outline our development philosophy.
yaw rate determination to be decoupled from steering rack and
steering geometry changes that may take place in the vehicle
development. A. Target Lateral Response
The modification of the target yaw rate is to smoothly limit The primary objective of a stability control system is to
the target yaw rate based upon the tire/road condition using the provide vehicle stability and handling predictability. This can
lateral acceleration and longitudinal velocity of the vehicle. In be achieved by preventing excessive deviations between the
steady state, stable side slip condition intended and actual lateral response of the vehicle.
While the driver intent recognition is described in the
previous section and a target/reference yaw rate is illustrated
This provides a theoretical limit to the yaw rate that is in detail, the target lateral response is discussed here in a more
achievable at the current tire/road condition without gener- general form. The desired lateral response, in general, includes
ating excessive sideslip. Based upon a filtered version of the vehicle yaw rate, side slip angle, side slip gradient, and path
difference between the nonlimited target yaw rate, radius of curvature.
and the yaw rate limit, a washout delta is During normal driving, a driver relies on a mental model of
calculated and subtracted from the nonlimited target yaw rate the vehicle’s response to his/her input developed from previous
to get the final target yaw rate. The basic idea is to low-pass driving experience. The vehicle response of a mental model is
filter the yaw rate target that is in excess of the theoretical limit referred to as a nominal vehicle response in Fig. 5.
and then subtract the filtered excess value from the nonlimited The target response generated for stability control may be
target yaw rate. Thus, in steady-state low- conditions, the different from the nominal response, depending on vehicle dy-
final target yaw rate will approach the theoretical limit, but namics/road conditions. This difference is essential to inform
in dynamic conditions, the vehicle will still have a smooth the driver about vehicle/road variations so that the driver can
response to the driver inputs. The calculation for this washout adjust his/her command/feedback during driving.
delta is as follows: Determining the magnitude of deviations from the “nomi-
nal” vehicle response allowed by the controller is an important
if aspect in achieving the desired active vehicle handling char-
acteristics under a variety of maneuvers and disturbances.
Considerations must be given to the rejection of disturbances
else such as road bank angles, wind gusts, and sensor noises, as
well as to the actuation effort in optimizing driver, system,
and vehicle interactions (Fig. 6).
where is the low-pass filter time constant. The target response of the combined vehicle and system
Note that the final target yaw rate is then limited as was established early in the stability control development
at Ford. While a deliberate but small deviation in lateral
response was set as the most direct and effective form of driver
The use of the washout delta provides a smooth transition feedback during a stability control event, the final magnitude of
of the target yaw rate under road surface transitions, while deviation is fine tuned during the development process based
still making the target yaw rate responsive to the driver’s on the following philosophy.
TSENG et al.: DEVELOPMENT OF VEHICLE STABILITY CONTROL AT FORD 227

Fig. 5. Total vehicle dynamics control/feedback.

be minimized. They are viewed as counterproductive, as they


divert driving focus from the most critical task at hand, the
task of controlling the vehicle.
This development philosophy focuses at normal driving
because a lateral deviation of 1–2 m from the desired course
can often result in leaving the designated lane of travel. Further
compounding this is the human response delay associated
with the possible, if not probable, lack of focus on the
task of driving at the onset of stability control events (e.g.,
daydreaming, conversing, or gazing at the countryside).
This development philosophy has the added benefit of
appealing to the driving enthusiast by preserving the “fun-
Fig. 6. Driver/system/vehicle interaction. to-drive” quality for those times when the driver wishes to
explore the outer envelope of the vehicle’s handling capability.
The target vehicle response should allow the driver to The skilled driver is given sufficient vehicle feedback to
observe and respond to the driving conditions (which implies correct for oversteer or understeer before the system activates.
a deliberate and controlled driver response) and minimize the
need for the driver to react to the driving conditions (which B. System Transparency
implies a panic and uncontrolled driver reaction.) Some have contemplated that, for an active system, the
The challenge is to strike a good balance between informing actuation effort should be distinct and apparent to a driver,
drivers of the limit of road/vehicle cornering capability and so that the driver is warned of the tire/road limitations and
providing vehicle stability and similar handling characteristics, would appreciate the system’s effort during a stabilizing
regardless of whether the limit has been reached. On one event. During public road evaluations, however, we observed
hand, the target deviation (from nominal response) needs to be that design philosophy often leads to an excessive warning
small and progressive to prevent provoking a panic reaction and intrusive actuation that may panic a driver, despite the
by the driver. On the other hand, the target deviation needs to stabilizing effort.
be sufficiently large to inform the driver that the cornering A transparent and progressive stability control concept
capability of the vehicle has been approached or adverse was proposed. The design based on the transparent system
driving conditions exist. Without sufficient vehicle feedback, concept focuses on the total vehicle handling response and
the driver is deprived of critical information necessary for the optimization of driver, system, and vehicle interactions.
adapting to the driving conditions. We found that, with a transparent system, a driver can be
Other forms of feedback, such as induced deceleration, given sufficient vehicle feedback to be aware of the cornering
steering kickback, noise, and pitch or roll oscillations, should capability limit without an intrusive actuation. Additionally,
228 IEEE/ASME TRANSACTIONS ON MECHATRONICS, VOL. 4, NO. 3, SEPTEMBER 1999

this approach can enhance the driver’s perception of the IV. VEHICLE SIDE-SLIP ANGLE ESTIMATION
vehicle’s total handling characteristics. The main function of a vehicle stability system is to
Being transparent, progressive, and nonintrusive, the system limit a vehicle’s yaw dynamics and body side slip angle to
can increase its sensitivity in stability control without annoying values manageable by average drivers. While the yaw rate
or disturbing the driver. This provides significant benefits for sensor provides the necessary information on yaw dynamics
subtle maneuvers, as will be discussed in Section VI. of the vehicle, the vehicle’s body side slip angle cannot
To achieve system transparency, we need to optimize system be directly measured and is estimated using vehicle state
actuation efficiency and smoothness. variables measured from vehicle motion sensors. Often times,
the output of these sensors drift with changes in temperature, or
C. System Actuation Efficiency and Smoothness
contain a floating dc offset. In addition, a conventional lateral
Stability control systems rely on the manipulation of the accelerometer cannot distinguish between the acceleration due
brake pressure distribution and engine/drivetrain torque for to curvilinear motion of the vehicle and the acceleration
control authority. Application of brake pressure in the attempt due to gravity. Since vehicle side slip angle is one of the
to induce an understeering or oversteering moment on the controlled variables, an accurate estimation of a vehicle’s side
vehicle is highly effective, and the relative effectiveness of slip angle that is insensitive to sensor dc offset, drift, or
applying braking and traction forces at each corner of the environmental conditions, such as banked road, is necessary
vehicle has been well addressed in the literature [1]–[4]. This for robust control.
means of imposing control authority on the vehicle results in One of the first challenges is to remove the bias in sensor
an induced deceleration that could be considered as a negative measurement induced from environmental disturbances, such
attribute by the driver. This induced deceleration and the as road bank angle and temperature variation.
associated longitudinal jerk has resulted in comments for some One sensor compensation strategy may include a static and
systems as “heavy handed” or “intrusive,” indicating there is a dynamic algorithm. The static algorithm is intended for
opportunity for improving the driving experience associated situations where the vehicle is standing still, and the dynamic
with vehicles equipped with stability control systems. To algorithm compensates for sensor drift and gravity effects of
maximize system smoothness, it is important to distribute the banked roads while the vehicle is moving.
desired oversteer/understeer torque through the corners of the While the vehicle is at rest, the static compensation algo-
vehicle with the greatest yaw torque efficiency. rithm utilizes the output of the yaw rate and lateral acceleration
Actuation sensitivities and smoothness can minimize the sensors to compensate the sensor’s dc offset. While the vehicle
actual and perceived deceleration/jerk without compromising is in motion, the dynamic offset compensation algorithm of
stability. Several factors such as banked and rough roads, yaw rate sensor signal utilizes the SWA and the vehicle’s
vehicle loading and production variability, sensor errors, and yaw rate gain equation to compute a nominal yaw rate of
system actuation noises, however, work against system sensi- the vehicle. This nominal yaw rate is then compared with the
tivities. These factors push for increases in activation thresh- measured value and the difference is filtered to compute a
olds. In the development process of vehicle stability control at sensor offset. Of course, this comparison is performed only
Ford, strategies have been developed and system tuning was under certain vehicle maneuvers and conditions where the
executed to reduce the impact of such factors. Among them, model is deemed valid. After sensor offset is computed, the
vehicle side slip angle and road bank angle estimation are output of the sensor is compensated, namely,
discussed in this paper.
Also working against the stability performance and re-
finement are the actuator delays associated with physical
A similar approach is used to compensate the lateral accel-
limitations on the rate at which the brake pressure can be
eration sensor signal for dc offset, drift, and effect of gravity
built. Implementations of strategies (such as event anticipation)
on banked roads. Under certain maneuvers and conditions, the
which compensate for these time delays can lead to significant
equation of lateral motion of the vehicle is used to compute a
improvements in the performance and refinement [5].
nominal lateral acceleration and, hence, an offset, for sensor
It is noted that the driver observes more harshness (and
compensation, namely,
perceives more deceleration) if more jerky chassis pitch/roll
motions are created during the longitudinal and lateral ac-
celerations/decelerations. Therefore, while certain accelera-
tions/decelerations are necessary for stabilizing effort, careful A full description of the above algorithms can be found in
placement of the phasing between the pressure build and U.S. Patents 5 809 434 and 5 742 918.
the directional change in lateral acceleration can successfully Another challenge is to construct a robust estimator of vehi-
blend the two accelerations/decelerations together and reduce cle lateral velocity in the presence of road surface coefficient
the chassis motions and perceived harshness. uncertainty, vehicle model uncertainty, as well as sensor noise
With these in mind, we showed through development and and bias. The estimation of vehicle lateral velocity involves
tuning effort that actuation efficiency and smoothness can estimation of tire lateral forces, tire slip angles, normal ground
be accomplished to provide a transparent, progressive, and reaction forces, and coefficient of friction at the tire–ground
nonintrusive system. The developed system reduces both the interface. In order to estimate vehicle lateral velocity, the
actual and perceived deceleration. dynamic equation of lateral motion of a simplified vehicle
TSENG et al.: DEVELOPMENT OF VEHICLE STABILITY CONTROL AT FORD 229

model is used

From the theory of asymptotic observers, the vehicle’s


lateral velocity is estimated from the following relationship:

where and is an observer gain, with


Note that, when the above equation reduces
to

which represents an open-loop integration of the equation of


lateral motion of the vehicle. In the above equations,
and represent, respectively, the yaw rate and lateral
acceleration sensor signals compensated for dc offset and
sensor drift. The lateral acceleration sensor signal is also
compensated for the effect of gravity on the sensor’s output
while the vehicle is traveling on, or is in a transition to, a
banked road. The longitudinal velocity of the vehicle, , is
estimated by the anti-lock brake algorithm using individual
wheel speeds of the vehicle.
Selection of a suitable observer gain is the key to obtaining
accurate estimates of vehicle lateral velocity under various
maneuvers and driving conditions. In our strategy, the observer
gain is adaptive and varies with the coefficient of friction
available at the tire–ground interface and the behavior of
vehicle dynamics. This way, depending on the on-line vehicle
dynamics, the observer will balance the emphasis on the
measured signal and the estimated tire lateral force
approximated by

Since is known, this estimate of tire lateral force is


bounded by a maximum value defined by

Fig. 7. Slalom maneuver on ice.


where and are experimentally determined and is an
approximation of the coefficient of friction at the tire–ground velocity and body side slip angle during a slalom maneuver
interface using the magnitude of the vehicle lateral acceleration on ice.
compared with other sensor measurements. A full description of the lateral velocity observer strategy
Once vehicle lateral velocity has been estimated, vehicle can be found in U.S. Patent 5 742 919.
body side slip angle can be computed as
V. ROAD BANK ANGLE ESTIMATION
Road bank angles have a direct influence on vehicle dynam-
ics and lateral acceleration measurement. To minimize possible
The above strategy for estimating vehicle lateral velocity modeling and estimation error, it is important for a vehicle
and body side slip angle has been extensively tested and stability controller to know the road bank angle experienced
verified for a wide variety of maneuvers and road conditions, by the vehicle. Due to operating concerns on banked roads, a
including banked roads, dry asphalt, wet asphalt, snow, and controller incapable of knowing road bank angle may sacrifice
ice. Fig. 7 shows the measured and estimated vehicle lateral its control sensitivity or alternatively turn off the control
230 IEEE/ASME TRANSACTIONS ON MECHATRONICS, VOL. 4, NO. 3, SEPTEMBER 1999

system entirely under these conditions. Although this road model uncertainty. Since all three estimates contain error/bias
disturbance is neglected in the majority of academic literature induced either from vehicle transient dynamics or incorrect
discussing vehicle stability systems [6]–[9], the detection of a vehicle reference model, these estimates are not very useful
banked road has been shown to deliver considerable operating yet.
advantages for real-world applications. We then use the second and third estimates, and
For a laterally steady-state vehicle where its lateral velocity to assess the magnitude of the transient lateral dynam-
remains constant, i.e., the road bank angle experienced ics inherently contained in the model reference errors affecting
can be determined from the equation of motion for a rigid body both estimates
DFC

where represents the lateral acceleration measured by where DFC denotes a dynamic factor coefficient related to the
the sensor that includes the influence from the road bank angle transient lateral dynamics.
The DFC information is then used in the final estimation as
However, for a control system designed to activate during a multiplicative factor to reduce the bias in the estimate
vehicle dynamic/extreme lateral maneuvers, it is not appropri- assuming no transient lateral dynamics
ate to use an estimate based on a laterally steady-state vehicle.
DFC
It is also not appropriate to use an estimate collected prior
to the dynamic maneuver, as it is conceivable that the road Effectively, this function eliminates the possible bias intro-
bank angle experienced by a vehicle may change rapidly. (For duced by lateral dynamics. As the change in lateral dynamics
example, a vehicle under an accident avoidance maneuver on grows more excessive, the final bank angle estimated becomes
a highway ramp or a vehicle negotiating mountain roads may progressively conservative. This is a practical assumption that
experience large bank angle variation.) a vehicle usually does not experience excessive low- lateral
The major challenge in both the estimation of road bank dynamics change when experiencing large bank angles.
angle and the estimation of vehicle lateral velocity is to Algorithm verifications have been performed by operating
differentiate the road disturbance (road bank angle) component vehicles under a wide variety of road conditions. The evalu-
from vehicle lateral dynamics component (derivative of lateral ations showed that the bank angle estimates provided allow
velocity). The differentiation is difficult because the vehicle the controller to maintain both the desired control sensitivity
measurement is effected by both components, and and for low- extreme maneuvers and control robustness for large
both components can change rapidly. road bank angle variations.
The definite advantage of a VSC that differentiates the Figs. 8 and 9 illustrate the obtained estimation during slalom
influence of road bank angle disturbances from vehicle lateral maneuvers of a base vehicle on a banked high- surface, as
dynamics is the capability to avoid false/nuisance activation, well as a low- flat ice surface. The vehicle had its VSC
even on a highly banked road, while maintaining the desired system manually turned off to observe extreme dynamics.
control sensitivity during extreme dynamic maneuvers on low- Fig. 8 shows the bank angle estimated is fairly accurate
surfaces. during a slalom maneuver on a parabolic banked roadway.
A bank angle estimation algorithm was developed at Ford Actual bank angle experienced by the vehicle was measured
to achieve the optimization in both control sensitivity for in the test track. Despite the fact that the vehicle went
extreme maneuvers and control robustness for road bank angle up and down on the roadway and experienced bank angle
variations. Evaluation of a fleet of competitive systems showed variations during the maneuver, the algorithm was able to offer
that the Ford stability control system is best in class in handling a good estimate by differentiating the two components, and
banked road/curve disturbances. lumped in the sensor measurements, where
Using the measurements of SWA, yaw rate, and lateral
acceleration, three bank angle estimates are calculated

In other words, the algorithm detected the bank angle not


the lateral velocity derivative was the significant contributor
in and obtained the final estimate
Although the estimate assuming steady state, i.e., also
had a good estimation performance in Fig. 8, only the pro-
where denotes the transfer function relating bank angle to posed estimate, maintained its estimation performance
vehicle lateral acceleration, denotes the transfer function in a different maneuver, as shown in Fig. 9. That is, only the
relating SWA to vehicle lateral acceleration, denotes the proposed algorithm has robust performance.
transfer function relating bank angle to vehicle yaw rate, As discussed earlier, the challenge is to maintain accuracy,
denotes the transfer function relating SWA to vehicle yaw even when experiencing nonlinear tire dynamics.
rate, is the bank angle estimate assuming Fig. 9 shows the bank angle estimated during a slalom
is the bank angle estimate using lateral acceleration maneuver on an ice surface. Large tire slip angles and lat-
information assuming no model uncertainty, and is the eral velocity derivatives were experienced. We see that the
bank angle estimate using yaw rate information assuming no steady-state estimate was heavily contaminated due to
TSENG et al.: DEVELOPMENT OF VEHICLE STABILITY CONTROL AT FORD 231

Fig. 8. Estimation during slalom on parabolic bank. Fig. 9. Bank estimation during slalom on ice.

the contribution of On the other hand, our estimate


remained close to the true bank angle (which was zero in this input in steering, braking, or throttle and disturbances such as
example) despite the dynamic maneuver. road surface transitions, split driving, or road undulations
Note that the driver maneuver in Fig. 9 was actually slower can all perturb vehicle dynamics and result in undesirable
than the one in Fig. 8. This illustrates that the algorithm did deviations.
not rely on driver maneuver alone to detect the magnitude We expect a stability control system to provide significant
of vehicle lateral dynamics (which depends on both driver benefit, by preventing excessive deviation between actual
maneuver and road surfaces). and driver intended vehicle lateral response, in the presence
Also worth noting is that the proposed bank angle estimation of inappropriate driver inputs, as well as in the presence
algorithm does not use any lateral velocity estimation infor- of external disturbances. To ensure the developed system
mation, since an estimate is subjected to the influence of road functions as intended and provides significant benefit over the
bank angles and may be contaminated. base vehicle, the following is a list of example maneuvers
to evaluate vehicle stability control. Note that this heuristic
VI. DEVELOPMENT AND EVALUATION list provides maneuvers essential for vehicle development and
evaluation, but is by no means inclusive.
As previously discussed, the intervention of a VSC is
desired when excessive deviations between actual and driver-
intended lateral response exist or are developing. These de- A. Oversteer Events Provoked by Steering Input
viations can be induced by driver inputs, adverse driving Dynamic oversteer occurs in the presence of steering input
conditions, or the combination of both. Inappropriate driver with large amplitudes, quick rates, or periodic content typi-
232 IEEE/ASME TRANSACTIONS ON MECHATRONICS, VOL. 4, NO. 3, SEPTEMBER 1999

cally accompanying aggressive transitional maneuvers. These the base vehicle response. Stability control events encountered
maneuvers include single and double lane changes, panic lane in the real world are likely to be comprised of a combination
changes, j-turns, and slaloms. These maneuvers can induce of the above situations.
excessive side slip angles, lag in yaw rate response, and While the benefit of a stability control system is clear
yaw rate overshoot when the driver applies excessive steering in extreme events, it is not limited to extreme events. A
amplitudes and rates, overcorrects with countersteer, or inputs significant portion of the system value resides in more subtle
a delayed steering correction. events which are likely to be much more frequent than the
extreme events. Discussed below is an example oftentimes
B. Stability Events Provoked by Braking Input encountered during adverse weather conditions.
When the driver feathers off the brakes too slowly while
entering a turn or lightly brakes during limit cornering in a G. Nonextreme Maneuvers with Perturbation
decreasing radius or constant radius turn, the base vehicle can Consider a driver experiencing a lateral wind gust while
respond with excessive side slip angles. Sudden and aggressive traversing a straight road covered with ice. The driver corrects
application of the brakes during limit cornering can result in through steering for the initial lateral deviation induced by
an unintended increase in path radius of curvature. Braking the wind gust. When the gust quickly subsides, the driver’s
too deeply or too aggressively at the initiation of a turn can initial steering correction causes the vehicle to deviate (in
adversely affect turn-in and the ability of the base vehicle to the opposite direction of the gust-induced deviation.) The
negotiate the turn. driver then steers in the opposite direction (of initial steering
correction). If each successive steering correction is slightly
C. Stability Events Provoked by Throttle Input too delayed, and slightly too large in amplitude, the driver
An abrupt lift-off of the throttle can result in a base will soon feel uncomfortable in handling the adverse road
vehicle response with excessive slip angles, in particular, on conditions.
manual transmission vehicles or on rear-wheel-drive vehicles. Subtle maneuvers such as the above example can be used
Although traction control significantly mitigates oversteer on to evaluate a VSC because the driver steering correction effort
rear-wheel-drive vehicles and understeer on front-wheel-drive can be minimized through stability control.
vehicles in the presence of excessive throttle, stability control The above example showed a vehicle stability system can
systems have an opportunity to further improve the vehicle deliver its value not only under a rare evasive maneuver.
response in this situation. A VSC can and should provide significant benefit over a
broad spectrum of disturbance amplitudes and driver input
severities. These observations echo the development philoso-
D. Stability Events Provoked by External Perturbations
phy we proposed regarding system transparency and actuation
Road surface transitions can significantly impact the base smoothness.
vehicle response. A transition to a lower surface while
cornering may result in a significant increase in path radius
VII. MOTION SENSOR FAULT DETECTION
of curvature. A transition to a higher surface during limit
cornering can impose a large oversteering moment on the As in any active system, fault management is necessary
vehicle and can result in excessive yaw rate and side slip to the robust operation of VSC’s. In a VSC, in addition to
angle responses. The accidental placement of an outside rear the existing failsafe features for ABS and TC subsystems, a
tire on a lower shoulder of the road during limit cornering supervising level is established to monitor all signals to the
can also induce large yaw rates and vehicle slip angles. on-board system controller. Besides the standard electronic
Variations in fore–aft normal load distribution associated with diagnosis and self-test for individual sensors, each sensor is
road undulations, such as cornering over the crest of a hill, cross checked with other measurements whenever analytical
can result in undesired base vehicle response. redundancy is available.
In this paper, we focus on the fault detection of vehicle
E. System Robustness Under External Disturbances motion sensors in a VSC (i.e., yaw rate sensor and lateral
accelerometer.) These sensors provide feedback information
For vehicle maneuvers under substantial yet realistic ex-
for vehicle stability control. Due to the wide variation of
ternal disturbances, such as road bank angles, wind gusts,
vehicle dynamics under a vast operating range, such as various
bump and ditches, and sensor bias drift, the system should be
tire/road contact conditions, road bank angles, and evasive
able to reject these disturbances to avoid nuisance activation
maneuvers, the detection of an in-range fault signal within
and provide similar performance improvement under nominal
these sensors is particularly challenging.
road/vehicle conditions. Nuisance activation under these real-
Functional redundancy using the relationship among sensors
istic conditions not only distracts the driver, but also casts a
is generally used in stability control systems to detect an “in-
negative image to the customer.
range” signal fault. Because the dynamics relationship used
may not be valid for all vehicle maneuvers, the challenge is
F. Public Road Evaluation to design a proper tolerance (used in verifying the functional
In the above, several situations have been discussed in which redundancy among sensors) that avoids both false negative
the stability control system can provide significant benefit over (undetected fault) and false positive (false warning).
TSENG et al.: DEVELOPMENT OF VEHICLE STABILITY CONTROL AT FORD 233

TABLE I

Note: The subscript denotes the measured value.

Table I lists the nominal functional redundancies (or dy- is independent of an estimated vehicle lateral velocity that may
namic relationship) versus the true relationship under realistic be disturbed by sensor signal fault or sensor noise and bias.
vehicle dynamics. We see that significant deviation can occur In the third step, the algorithm maps the difference be-
through transient vehicle dynamics or road disturbances. For tween the measured and referenced signal into a “normalized
example, it is conceivable (and has been experimentally ver- residue.” Conceptually, these residues are road bank angle
ified) that all three nominal relationships are violated when estimates. For example,
a vehicle is experiencing snowplowing during a tire-limit
turning/sliding maneuver.
During the development of vehicle stability control at Ford,
the bias terms induced in various operating conditions were
analyzed and simulated. The final design tolerance are fine-
tuned throughout vehicle verifications. where the and subscripts denote the measured and
Another challenge, in addition to the robustness issue, is to referenced values, respectively.
quickly detect an in-range vehicle motion sensor signal fault. A This normalization feature allows the designer to incorpo-
vehicle motion sensor signal fault, without prompt detection, rate the influence of various road disturbances, vehicle speed,
not only may cause the controlled vehicle to deviate from and steering position into a calibrated threshold.
the driver’s intended path, but also could induce disturbances Simulated fault insertions on a vehicle under a variety of
to deter/delay fault detection. A “fault signature recognition” driving conditions showed that the algorithm delivers prompt
algorithm was developed at Ford to promote prompt detection. detection for stable driving vehicles robustly.
This approach relies on the detection of a specific failure
mode/effect. Failure mode effect and analysis (FMEA) is used
jointly to ensure various sensor failure modes (fault signatures) VIII. CONCLUSIONS
have been considered. A vehicle stability system provides stability enhancement
This “signature recognition” algorithm is composed of five and handling predictability of a vehicle. Not unlike ABS and
steps. The first step checks if the vehicle is not in an extreme TC systems, the system requires a driver’s command to assist
maneuver using sensors other than the one being inspected for the driver in achieving the desired maneuver safely under a
fault signature. This is to prevent false detection in the rare variety of driving conditions. For the system to be valuable
cases when the signals behave anomalously due to extreme to the driver/customer, it is important for the active system
maneuvers. The second step generates a reference signal to recognize the driver’s intention and assist the driver in a
from other sensors. The third step converts the difference nonintrusive manner. It is also important for the system to be
between the inspected signal and the reference signal into a cost effective.
“normalized” quantity defined as “residue.” The fourth step This paper addressed these practical concerns in vehicle
processes the normalized residue through a bandpass filter to stability systems. It described a vehicle stability system de-
capture the fault signal with certain frequency content. The veloped at Ford which utilizes a relative steering wheel sensor
fifth step further scrutinizes the signal to ensure the detected in the attempt to optimize the system performance and value.
frequency is not induced by vehicle vibrations. It showed that accurate and robust driver input recognition
To determine if a maneuver is excessive, the algorithm can be accomplished with a relative steering wheel sensor.
monitors the front-to-rear tire slip angle difference and its rate It described how vehicle status can be robustly estimated
for maneuver differentiation. This methodology is by design with an adaptive hybrid integrator/Luenberger observer. It
insensitive to road disturbances, since the front-to-rear slip also showed how road bank disturbances can be dynamically
angle difference and robustly estimated. It proposed the concept of event
anticipation, control smoothness, and system transparency
in the application of control action. It listed a variety of
maneuvers under which vehicle stability systems can provide
significant benefit over their base vehicles.
234 IEEE/ASME TRANSACTIONS ON MECHATRONICS, VOL. 4, NO. 3, SEPTEMBER 1999

IX. OUTLOOK Hongtei Eric Tseng received the B.S. degree from
National Taiwan University, Taipei, Taiwan, R.O.C.,
Rapid technology improvement in sensor/actuator technol- in 1986 and the M.S. and Ph.D. degrees from the
ogy [10], in particular, in microelectromechanical systems University of California at Berkeley, in 1991 and
(MEMS), will substantially reduce sensor/actuator cost and 1994, respectively, all in mechanical engineering.
In 1994, he joined Ford Motor Company, Dear-
allow more built-in safety features, such as more extensive born, MI, where he is currently a Technical Special-
electronic self-check and sensor hardware redundancy. The ist in the Ford Research Laboratory. His areas of
cost reduction in sensor/actuator not only will make VSC’s interest are vehicle dynamics modeling and control.
widely available, but also enable the integration of other
vehicle electronic systems. On one hand, vehicle dynamics
under vehicle stability control can be further enhanced if sys-
tems such as advanced steering [11] and suspensions [12] are Behrouz Ashrafi (S’90–M’91) received the B.S.,
integrated with vehicle stability controllers. On the other hand, M.S., and Ph.D. degrees in mechanical engineering
from Drexel University, Philadelphia, PA, in 1983,
other systems can benefit from VSC by refining their control 1985, and 1991, respectively.
algorithms with the vehicle dynamics information available. In 1992, he joined Ford Motor Company, Dear-
Furthermore, future VSC’s may be integrated with matur- born, MI. He is currently a Technical Specialist in
the Advanced Product and Manufacturing Depart-
ing technologies, such as global positioning system (GPS) ment of the Chassis Division, Visteon Automotive
guidance, lane following/lane departure warning, to provide Systems (an enterprise of Ford Motor Company),
additional safety features and further improve driver/vehicle Dearborn, MI. He is the holder of five patents
relating to vehicle controls and steering systems.
interaction. His current interests include vehicle dynamics, vehicle stability control, and
application of estimation and control theory to advanced automotive vehicles.
ACKNOWLEDGMENT
The authors would like to thank their colleagues in the Ad-
vanced Interactive Vehicle Dynamics Section, Ford Advanced Dinu Madau received the B.S. degree in electrical
engineering from Oakland University, Rochester,
Vehicle Technology, for their invaluable input during the MI, in 1986 and the M.S. degree in electrical
development process. They also appreciate the management computer controls systems engineering from Wayne
support of this program from L. M. Brown (Ford) and K. State University, Detroit, MI, in 1992.
In 1989, he joined Ford Motor Company. He
Lehmann (Continental Teves). is currently a Software Engineer in Chassis Steer-
ing and Suspension Software Development, Visteon
REFERENCES Automotive Systems (an enterprise of Ford Motor
Company), Dearborn, MI. He is the holder of two
[1] A. T. Van Zanten et al., VDC, the Vehicle Dynamics Control System of patents in fuzzy logic technology, two patents re-
Bosch, SAE 950759, 1995. lated to steering systems, and four patents pending in vehicle dynamics. His
[2] K. Koibuchi et. al., Vehicle Stability Control in Limit Cornering by Active areas of interest are vehicle dynamics and suspension controls with emphasis
Brake, SAE 960487, 1996. on algorithm development.
[3] S. Motoyama, H. Uki, K. Isoda, and H. Yuasa, “Effect of traction force
distribution control on vehicle dynamics,” in Proc. 1992 Int. Symp.
Advanced Vehicle Control (AVEC’92), Yokohama, Japan, pp. 447–451.
[4] T. Pilutti, G. Ulsoy, and D. Hrovat, “Vehicle steering intervention
through differential braking,” in Proc. American Control Conf., Seattle, Todd Allen Brown received the B.S. and M.S.
WA, June 1995, pp. 1667–1671. degrees in mechanical engineering from Michigan
[5] D. Yanakiev, J. Eyre, and I. Kanellakopoulos, “Longitudinal control State University, East Lansing, in 1989 and 1991,
of heavy vehicles with air brake acutation delays,” in Proc. American respectively.
Control Conf., Albuquerque, NM, June 1997, pp. 1613–1617. In 1991, he joined Ford Motor Company, Dear-
[6] K. H. Senger and W. Kortum, “Investigations on state observers for the born, MI, where he is currently a Technical Special-
lateral dynamics of four-wheel steered vehicles,” in Proc. 11th IAVSD ist in Vehicle Dynamics Control Systems, Ford Re-
Symp., Kingston, Ont. Canada, Aug. 1989, pp. 515–527. search and Vehicle Technology. His areas of interest
[7] L. R. Ray, “Nonlinear state and tire force estimation for advanced are vehicle dynamics and control with emphasis on
vehicle control,” IEEE Trans. Contr. Syst. Technol., vol. 3, pp. 117–124, limit handling and control system interactions.
Mar. 1995.
[8] C. S. Liu and H. Peng, “A state and parameter identification scheme for
linearly parameterized systems,” ASME J. Dynam. Syst., Meas. Contr.,
vol. 120, no. 4, pp. 524–528, Dec. 1998.
[9] J. Farrelly and P. Wellstead, “Estimation of vehicle lateral velocity,” in
Proc. 1996 IEEE Int. Conf. Control Applications, Dearborn, MI, Sept. Darrel Recker (S’85–M’88) received the B.S.E.E.
1996, pp. 552–557. degree from Iowa State University, Ames, in 1988
[10] W. Kortuem, R. M. Goodall, and J. K. Hedrick, “Mechatronics in and the M.S. and Ph.D. degrees in electrical
ground transportation—Current trends and future possibilities,” in An- engineering from the University of Illinois,
nual Reviews in Control, vol. 22. Exeter, U.K.: Elsevier, 1998, pp. Urbana–Champaign in 1991 and 1993, respectively.
133–144. In 1992, he joined Ford Motor Company, where
[11] H. Peng and J. S. Hu, “Traction/braking force distribution for optimal he has worked in vehicle controls, motor controls,
longitudinal motion during curve following,” Veh. Syst. Dynam., vol. and software development. He is the holder of four
26, no. 4, pp. 301–320, Oct. 1996. patents relating to vehicle and motor controls. He
[12] H. E. Tseng and J. K. Hedrick, “Performance of a semi-active suspension is currently a Supervisor of Software Development
with leading system information,” in Proc. 1994 Int. Mechanical Engi- for Chassis Electronics with Visteon Automotive
neering Congr. and Exposition, DSC Transportation Systems, Chicago, Systems (an enterprise of Ford Motor Company), Dearborn, MI.
IL, Nov. 1994, vol. 54, pp. 115–126. Dr. Recker is a member of the IEEE Control Systems Society.

Você também pode gostar