2008-01-0178

Analysis of Vehicle-to-Ground Impacts during a

Rollover with an Impulse-Momentum Impact Model
Nathan A. Rose
Stephen J. Fenton
Gray Beauchamp
Kineticorp, LLC

Robert W. McCoy
Ford Motor Company

Copyright © 2008 SAE International

ABSTRACT conjunction with the known initial velocity conditions, to

obtain calculated velocity changes and energy loss for
This paper explores the accuracy of a planar, impulse- each of the three vehicle-to-ground impacts. These
momentum impact model in representing the dynamics calculated values are then compared to the known
of three vehicle-to-ground impacts that occurred during a values of these parameters from the video analysis. The
SAE J2114 dolly rollover test. The impacts were accuracy of the impact model is explored and guidance
analyzed, first, using video analysis techniques to obtain is given for selecting impact model parameters that
the actual velocity conditions, accelerations, impact force maximize that accuracy.
components and the energy loss for each of the impacts.
Next, these same impacts were analyzed using the The equations of the impact model used in this paper
known initial velocity conditions and the subject impact were introduced in Reference 17 and they are given a
model. The equations of this impact model yielded fuller treatment in Reference 19. For convenience, these
calculated values for the velocity changes and energy equations are repeated in this paper and their derivation
loss for each impact. These calculated results were then is given in Appendix A. This impact model represents a
compared to the actual dynamics data from the video relatively straightforward application of the principle of
analysis of the impacts to determine the accuracy of the impulse and momentum with the exception that the
impact model results. For all three vehicle-to-ground effects of the gravity impulse are included in the
impacts considered in this study, the impact model equations. Our main purpose in using this simple, planar
results for the velocity changes and energy loss showed impact model was to determine whether such a model
excellent agreement with the video analysis results for could be used to explore the influence certain factors will
these parameters. These results suggest that it is have on rollover dynamics. For instance, could such a
reasonable to use this impact model to examine the model illuminate the factors that influence a vehicle’s
influence of various factors on rollover dynamics. deceleration rate during a rollover or those factors that
determine the number of rolls a vehicle will experience
INTRODUCTION during a rollover? These parameter influences are likely
to be more visible in a simple, planar impact model than
This paper explores the accuracy of a planar, impulse- they would be in a fully three-dimensional impact model
momentum impact model in representing the dynamics such as those used in certain accident simulation
of three vehicle-to-ground impacts that occurred during a software packages. In using the impact model of this
SAE J2114 dolly rollover test [6, 8]. The paper begins by paper to illuminate such factor influences, it would, of
introducing the equations of that impact model and then course, be beneficial to know the degree to which the
discusses video analysis of the three vehicle-to-ground results from that model will have actual relevance to
impacts that will be considered in the paper. This video real-world rollover dynamics, which are, in general,
analysis yielded the actual initial velocity conditions, highly three-dimensional. That is the issue that this
velocity changes and the energy loss for each of the paper aims at addressing.
impacts. After discussing these video analysis results,
the equations of the impact model are used, in
1
There are, of course, computer software packages with IMPACT MODEL ASSUMPTIONS
the capability of analyzing vehicle-to-ground impacts in
three-dimensions and there will clearly be vehicle-to- Before discussing the equations that will yield the
ground impacts that occur during rollovers for which velocity changes and energy loss for the impact of
such three-dimensional analysis will be essential [5, 7, Figure 1, consider the assumptions that the mathematics
21]. Nonetheless, the research reported in this paper of those equations will invoke. These assumptions
does show that, at least for the three vehicle-to-ground include the following:
impacts reported here, a simple, planar impulse-
momentum impact model can yield reasonably accurate 1. The impact will be assumed to occur entirely in a
velocity changes and energy loss for a vehicle-to-ground single plane, and thus, velocity changes along the
impact. vehicle’s longitudinal axis are neglected, as are
changes in pitch and yaw velocity.
VEHICLE-TO-GROUND IMPACT MODEL
2. The impact model equations will recognize no
Figure 1 depicts an idealized impact between a vehicle change in the position of the vehicle through the
and the ground. The impact shown here is a trailing side impact.
roof impact for a passenger’s side leading roll. This
image shows the coordinate system orientation and the 3. The impact force will be assumed to be
parameters with which the impact model equations were concentrated at a single point.
developed. The impact radius, which is the distance from
the vehicle center-of-mass to the contact point C, is 4. It has been assumed that no moment arises at the
designated with the symbol r. The impact angle, which is contact point.
the angle between the ground plane and the impact
radius, is designated with the symbol φ. The vehicle’s 5. Any effects of ground plane restitution have been
velocity vector is designated with the letter v and the neglected. In other words, the ground surface impact
vehicle’s roll velocity is designated ωr. force has been assumed to be a retarding force that
depends on relative velocity at the contact point for
its development. It is assumed that there is no
structural restitution that could potentially cause a
velocity reversal in the contact region.

The meaning of the first three of these assumptions

should be relatively clear. We suspect that the meaning
of the last two will be less clear, and thus, we refer the
reader to the extensive discussion of these concepts in
the text by Brach, titled Mechanical Impact Dynamics [2].
Relaxing these last two assumptions is possible.
However, in our judgment, this would make the modeling
reported in this paper needlessly complex.

All five of these assumptions have the potential to be

violated during any particular vehicle-to-ground impact.
The degree to which any one of them is an appropriate
assumption will depend on the specifics of the particular
vehicle-to-ground impact under consideration and on the
Figure 1 – Free-Body Diagram for a Vehicle-to- degree to which violating any of these assumptions will
Ground Impact actually degrade the accuracy of the results the impact
model yields. That issue will be addressed in the
During the depicted impact, the vehicle is subjected to “Discussion” section of this paper.
both upward and ground surface impact force
components, Fvertical and Fground, and the gravity force, IMPACT MODEL EQUATIONS
which is the vehicle’s weight. In general, Fground can act
in either the positive (left) or negative (right) direction. Once these assumptions are invoked, application of the
On the other hand, Fvertical will always act in the positive-z principle of impulse and momentum results in Equations
direction and the gravity force will always act in the (1) through (3), which yield the vehicle’s upward and
negative-z direction. ground plane center-of-mass (CoM) velocity changes

2
and the vehicle’s change in roll velocity.1 Appendix A ratio should be set at a value that incorporates that
contains a full derivation of these equations. In these snagging.
equations, vzc,i is the vertical velocity of the vehicle at
Point C immediately preceding the ground contact, kr is Within this impact model, the sign of the impulse ratio
the vehicle’s radius of gyration for the roll axis, g is the governs the direction in which the ground plane collision
gravitational constant, Δti is the duration of the impact, force acts. A positive impulse ratio produces a ground
and the letters s and c designate the sine and cosine. plane force that acts in the positive direction and a
Note that although it is assumed that the vehicle position negative impulse ratio results in a ground plane force
does not change during the impact, accounting for the that acts in the negative direction. The direction of the
effect of the gravity impulse has required inclusion of the ground surface impact force, in turn, determines whether
impact duration [9]. the vehicle will experience a positive or negative ground
plane velocity change and whether the ground surface
impact force will tend to increase or decrease the roll
⎧ k r2 ⎫
ΔV z = −(1 + e ) ⋅ v zc ,i ⋅ ⎨ 2 ⎬ velocity. Also, note that for a passenger’s side leading
2 2
(
⎩ k r + r c φ − μ ⋅ s φ ⋅ cφ ⎭ ) roll, as depicted in the image above, the ground speed
(1)
(
⎧ r 2 c 2 φ − μ ⋅ s φ ⋅ cφ ⎫
− g ⋅ Δt i ⋅ ⎨ 2
) will be negative, the roll velocity positive, and the impact
angle will always be less than 90 degrees. For a driver’s
⎬
2
(
⎩ k r + r c φ − μ ⋅ s φ ⋅ cφ ⎭
2
) side leading roll, like the crash test analyzed in this
paper, the same impact model equations apply, but the
ground speed will be positive, the roll velocity negative
(
ΔV y = μ ⋅ ΔVz + g ⋅ Δt
i
) (2) and the impact angle will always be greater than 90
degrees.

The energy loss that occurs during the vehicle-to-ground

r ⋅ (μ ⋅ sφ − cφ ) impact of Figure 1 can be written as follows:
Δω r = (ΔV z + g ⋅ Δt i ) ⋅ (3)
k r2
ΔE =
m 2
2
(
⋅ v zi − v zf2 + v yi2 − vtf2 + k r2ω r2,i − k r2ω r2, f ) (5)
Examination of Equations (1) through (3) reveals that the
initial downward velocity at the Point C (vzc,i) directly The energy loss of Equation (5) includes the energy loss
influences the velocity changes that occur during the due to vehicle deformation, ground deformation, and
impact, with the velocity changes increasing as this sliding, snagging or furrowing between the vehicle
velocity increases. This initial vertical velocity at Point C, structure and the ground.
which is given by the following equation, is related to the
vehicle’s CoM vertical velocity, its roll velocity and the ROLLOVER CRASH TEST SETUP
impact angle and radius:

v zc ,i = v z ,i − r ⋅ cφ ⋅ ω r ,i (4)

In this equation, vzi is the vehicle’s vertical velocity at its

CoM immediately preceding the impact.

Equations (1) through (3) also include the coefficient of

restitution, e, and the impulse ratio, μ. The coefficient of
restitution is the negative ratio of the post-impact to the
pre-impact vertical velocity at Point C. The impulse ratio
is the ratio of the ground surface collision impulse to the
vertical direction collision impulse. In some instances,
the impulse ratio can be thought of as a coulomb friction
value, though its application is not limited to this
interpretation [2, 3, 4, 14, 19]. In addition to the effects of Figure 1 – Rollover Crash Test Configuration
friction between the ground and the vehicle body, the
ground plane impulse may also include the effects of The rollover crash test used in this paper was run with a
forces generated by snagging between the vehicle and Ford sport utility vehicle in accordance with the SAE
the ground. When such snagging occurs, the impulse Recommended Practice J2114 (Dolly Rollover). This test
procedure involves generating a lateral roll of the test
1
vehicle by accelerating a cart, on which the vehicle sits,
The authors presented these equations in Reference 17. However, in up to the test speed, then decelerating that cart at a
that reference, Equation (2) failed to include the gravity term.
3
sufficient rate to initiate the rollover. The vehicle is accelerations at the lower A-pillar and B-pillar on both
situated on the cart perpendicular to the initial velocity sides of the vehicle. The vehicle was also instrumented
direction with an initial roll angle of 23 degrees. In the with two rotation rate sensors for each principal axis.
test under consideration in this paper, the vehicle was These were mounted on the center tunnel just rearward
situated on the cart with its driver’s side leading and the of the seats. Nine high-speed fixed cameras and one
cart and test vehicle were accelerated up to a speed of real-time panning camera recorded the test. The high-
approximately 31 mph before the cart deceleration was speed video was taken at 500 frames per second and
initiated. Figure 1 shows the test vehicle just before it the real-time video was taken at the NTSC standard
began exiting the cart. frame rate of 29.97 frames per second.

Three-dimensional dynamics data for this crash test was

obtained using video analysis that is described in detail
in a companion paper [20]. That analysis will also be
covered briefly in the next two sections.

CAMERA-MATCHING VIDEO ANALYSIS

This section describes the methodology utilized to track

the vehicle motion in the crash test video. This
methodology consisted of the following steps: 1)
preparing the test vehicle with targets that would be
tracked; 2) surveying the test facility and vehicle
geometries; 3) creating a computer environment that
included the geometries of the test facility and the test
vehicle; 4) analyzing the actual cameras and the video
images to determine characteristics, distortion and
resolution so that these could be replicated with
computer-modeled cameras; 5) accurately placing
computer-modeled cameras in a three-dimensional
computer environment and matching them to the test
video; and 6) tracking the movement of the test vehicle
by matching the location of the computer generated
vehicle targets to the targets on the test vehicle for each
frame of the video sequence. These steps are more fully
described in the following paragraphs.

1) Prior to running the crash test, the test vehicle was

marked with high-contrast yellow and black fiducial
targets. It was the motion of these targets that was
ultimately tracked in the video. In placing these
targets, the primary goal was to generate a wide
range of points on the vehicle that would be visible
and identifiable in the crash test video.

2) Once these stickers were placed, their locations

were surveyed so that they could be replicated on a
computer-generated model of the test vehicle. Also
Figure 2 – Rollover Crash Test Dynamics prior to running the crash test, a survey of the test
facility was completed. This survey provided the
The roll dynamics that occurred during this test are three-dimensional geometry of the rollover test
depicted in Figure 2. The images in this figure were facility, including the ground surface, the control
captured by a high-speed camera located downstream joints in the concrete of the test surface, the walls
of the roll. As these images show, the vehicle rolled just surrounding the test surface, and the light fixtures
past 360 degrees and then came down to rest on its located above the test surface.
wheels. During this test, the vehicle was instrumented
with sensors to measure the vehicle-fixed longitudinal, 3) Based on survey data and photographs, computer-
lateral and vertical accelerations at the center tunnel generated models of the test facility and test vehicle
between the seats and the lateral and vertical were constructed. The entire computer environment
was then oriented relative to the general roll
4
 direction. With the survey of the test vehicle aligned video. If the location and characteristics of each
to the computer model, the target stickers that were camera were set properly, then this step would yield
placed on the test vehicle, and subsequently an overlay between the video background and the
surveyed, could be transferred to identical locations computer-modeled environment. Crash test facility
on the computer model of the vehicle. These target features visible in computer model should overlay
locations were used to track the motion of the test those same features visible in the crash test video.
vehicle. Figure 3 depicts this process of camera-matching
the computer-modeled cameras to the video of the
4) The optical and geometric characteristics of all the rollover test. This step was repeated for each
cameras were documented and analyzed in order to camera position and video sequence, such that all
create computer-generated cameras that mimicked cameras and computer geometry were visually
each individual camera that captured video of the determined to be matched to their background video
rollover test. The data for these cameras came from sequence.
the survey of these cameras, analysis of the
sensors, and analysis of the technical and 6) With this determination, the computer model of the
specification drawings. Each camera was created test vehicle could then be placed in the computer-
according to its specific characteristics, since these modeled environment and positioned at each frame
characteristics differed between cameras. These to mimic the movements of the vehicle visible in
computer-modeled cameras were also located and each frame of the video sequence. The images in
oriented to be identical to the cameras surveyed at Figure 4 depict a properly matched computer model
the facility at the time the dolly rollover test was of the vehicle and facility for several frames of the
conducted. video. The mesh model of the vehicle contains
yellow dots as seen in the video image overlay.
5) Having created a computer-modeled environment These yellow dots are the surveyed points of the test
that contained the geometry of the test facility, the vehicle before it was damaged from the rollover. It
test vehicle and a series of computer-modeled can be seen in Figure 4 that these images match
cameras that replicate the actual cameras, test with all the corresponding targets that were placed
video from each camera was then designated as a on the test vehicle. In addition to the vehicle
background image for its corresponding computer- matching the video, the computer model of the
modeled camera. Each computer-modeled camera environment and background can also be seen in
could then be used to simultaneously view the this image properly oriented and positioned relative
computer model of the test facility and the crash test to the same background and environment of the
facility seen in the video.

Figure 3 – Camera-Matching Process

5
 Figure 4 – Images from Camera-Matching Video Analysis

MOTION ANALYSIS direction velocity was obtained with the following

equation:
The video analysis described in the previous section
yielded the earth-fixed x, y, and z coordinates of the xt + Δt − xt −Δt (7)
vehicle CoM and the yaw, pitch and roll angles of the vx =
vehicle at 10 millisecond intervals throughout the first 2 2Δt
seconds of the test. For the analysis presented here, the
earth-fixed coordinate system was setup with the x- Similarly, the following centered difference equation was
direction running along the initial travel direction of the used to obtain the vehicle’s vertical velocity throughout
rollover dolly and the test vehicle. The y-direction was the test:
oriented along the initial heading direction of the vehicle
and the z-axis was oriented upward. z t + Δt − z t −Δt
vz = (8)
2Δt
Once the vehicle’s positions and orientations were
obtained, they were used to calculate the velocities for In this equation, z refers to the z-coordinate of the
the vehicle during the test. For instance, the following vehicle’s CoM.
centered difference equation was used to obtain the
vehicle’s ground plane velocity throughout the test [11]:
In general, the accuracy and precision of the velocities
calculated with Equations (6), (7) and (8) will depend on
(xt +Δt − xt −Δt )2 + ( yt +Δt − yt −Δt )2 (6) the magnitude of the measurement uncertainties in the
v x− y = positional coordinates and on the time step used for the
2Δt calculation. On the one hand, if too small a time step is
used, the velocity calculations will be excessively
In this equation, x and y refer to the x and y coordinates sensitive to any measurement errors and will exhibit
of the vehicle’s CoM, the subscripts t+Δt and t-Δt excessive uncertainty. On the other hand, if too large a
reference these coordinate values from time steps time step is used, the velocity curves will suffer from
surrounding the time step of interest, and Δt is the over-smoothing error and the velocity peaks could be
duration of each time step. truncated.

Overall, it was found that the motion of the vehicle in the To explore the degree to which measurement errors
y-direction was negligible, such that the velocity of the might affect the accuracy and the precision of the
vehicle in the x-direction (along the rollover track) was velocities calculated with Equations (7) and (8), the
nearly identical in magnitude to the overall translational authors had a second analyst use the previously
speed of the vehicle given by Equation (6). This x- described camera-matching video analysis technique to
obtain a second set of motion for the vehicle in the first
6
two seconds of the test, at time increments of 30 Figure 5 contains curves for an 84% confidence interval
milliseconds. The positions and orientations of the and a 96% confidence interval. As one would expect, the
vehicle obtained with this second analysis were then uncertainty in the velocities due to potential
compared to the positions and orientations of the vehicle measurement errors decrease as the time step
obtained by the first analyst. Overall, the two sets of increases. A time step of 40 milliseconds produced an
motion data had an average difference in the x- uncertainty of 0.71 mph in the ground plane and vertical
coordinate of 0.35 inches, with a standard deviation of speeds with a confidence of 84% and 1 mph with a
0.36 inches, and an average difference in the z- confidence of 96%.
coordinate of 0.42 inches, with a standard deviation of
0.28 inches. Thus, 84% of the time, the difference Figures 6 and 7 depict the vehicle’s ground speed and
between the two analysts’ positions was less than 0.71 vertical speed for the first 2000 milliseconds of the test,
inches. At least 96% of the time, the difference between calculated with Equations (7) and (8) and with time steps
the two analysts’ positions was less than 1.0 inch. varying between 10 and 40 milliseconds.

Using differential calculus to perform an error analysis

[22], it can be shown that the uncertainty in the velocities
of Equations (7) and (8) can be estimated with the
following equations:

δx (9)
δv x =
2 ⋅ Δt

δz
δv z = (10)
2 ⋅ Δt

In these equations, δx and δz are the positional

uncertainties in the x and z coordinate directions and δvx
and δvz are the velocity uncertainties in these same Figure 6 – Vehicle Translational Speed
directions. In formulating Equations (9) and (10), it has
been assumed that the potential measurement error at
each time step is independent of those at the
surrounding time steps.

Assuming that the differences between the two analysts

reported above give us a reasonable estimate of the
uncertainty in the positional coordinates obtained with
the camera-matching technique, Figure 5 graphically
represents the uncertainty in the translational velocities
for time steps varying between 10 and 50 milliseconds.

Figure 7 – Vehicle Vertical Speed

Examination of Figure 6 reveals that, with the exception

of several small time segments, varying the time step
between 10 and 40 milliseconds made little difference to
the ground speed that was calculated with Equation (7).
Examination of Figure 7 reveals that, for most time
segments, varying the time step between 20 and 40
milliseconds made little difference to the vertical velocity
that was calculated. At a time step of 10 milliseconds,
the vertical velocity curve exhibits quite a bit of
Figure 5 – Velocity Uncertainties jumpiness that likely indicates this time step produces
excessive sensitivity to measurement errors in the z-
7
coordinate. In addition to this, the time step does have that potentially exhibits excessive noise due to
some effect on the peaks achieved by the vertical measurement errors. However, at higher time steps the
velocity curve, with the magnitude of the peaks being acceleration curves appear to be potentially subject to
diminished as the time step increases. over-smoothing error since the peak of the accelerations
are significantly influenced by the time step.
Overall, for the velocity calculations, a time step of 40
milliseconds appeared to be a reasonable compromise Any time step between 20 and 40 milliseconds yielded
between the effects of potential measurement errors and calculated peak accelerations that were significantly
over-smoothing effects. Though the peaks of the vertical lower than those exhibited by the sensor data. Visually,
velocity curve with this time step may experience some the peaks of the sensor data appear to be influenced by
truncation due to over-smoothing, this effect does not considerable noise still present in the signals, and thus,
appear excessive. it seems likely that the sensor signals in this case
overestimate the peak accelerations. However, further
Now, consider the vertical velocities of Figure 7 relative analysis would be necessary to determine the degree to
to what was physically occurring when the vehicle had which the sensor signals might be overestimating the
these velocities. This graph shows that, as it exits the accelerations and, likewise, the degree to which the
dolly and drops to the ground, the vehicle develops a video analysis data might be underestimating the
downward velocity around 3-½ mph by the time the resultant accelerations.
wheels impact the ground. As a result of the wheels
impacting the ground, the vehicle then develops an
upward velocity around 3-½ mph. The vehicle then falls
again and at the time the leading side roof rail contact
the ground the vehicle has a downward center of mass
velocity around 2-½ mph. The upward velocity change
from this first roof impact gives the vehicle an upward
velocity of 1 mph. By the time the trailing side roof
impact the ground, the vehicle has developed a
downward velocity of around 2-½ mph. The vehicle
rebounds out of this second roof contact with a vertical
velocity around 2 mph.

After calculating the vehicle’s translational velocities, the

following centered difference equation yielded the
vehicle’s ground plane acceleration throughout the test:
Figure 8 – Vehicle Resultant Accelerations
v x ,t + Δt − v x ,t −Δt (11)
ax = Using differential calculus to perform an error analysis
2Δt on Equations (11) and (12), it can be shown that the
uncertainty in the accelerations can be estimated with
Similarly, the following centered difference equation the following equations:
yielded the vehicle’s vertical acceleration throughout the
test: δx (14)
δa x =
2Δt 2
v z ,t + Δt − v z ,t −Δt (12)
az =
2Δt δz
δa z = (15)
2Δt 2
The following equation yielded the vehicle’s resultant
acceleration.
Again assuming that the differences between the two
analysts reported above give a reasonable estimate of
a res = a x2 + a z2 (13) the potential measurement errors or the uncertainty in
the positional coordinates obtained with the camera-
matching technique, Figure 9 graphically represents
Figure 8 compares the resultant accelerations calculated Equations (14) and (15) for time steps varying between
with Equations (11) through (13), with varying time 10 and 50 milliseconds. From the standpoint of the
steps, to those obtained from sensor data. The sensor uncertainty, using a time step of 40 milliseconds for
data was first filtered with a CFC 60 filter. With a time calculating the accelerations yields an uncertainty in the
step of 20 milliseconds, Equations (11) through (13) translational accelerations of approximately 0.8g, with
appear to produce a rather erratic acceleration curve
8
96% confidence, and of approximately 0.57g with 84% less than the vertical vehicle-to-ground impact force,
confidence. normalized by the vehicle weight.

Figures 10 and 11 show the vertical impact force applied

to the test vehicle during this test, calculated with
Equation (20) and utilizing the accelerations calculated
with time steps between 20 and 40 milliseconds. The
first of these graphs plots the vertical force with the
progression of time and the second plots the vertical
force with the progression of the roll. In Figure 11,
images have been included to show the orientation of
the vehicle at the peak of each of three main impulses.
These three impulses are associated with the impact
between the wheels and ground as the vehicle exits the
dolly, the driver’s side roof-to-ground impact, and the
passenger side A-pillar, roof rail and hood impact with
the ground.

Figure 9 – Acceleration Uncertainties

Having calculated the test vehicle’s CoM accelerations in

the earth-fixed coordinate system, equations can be
written relating these accelerations to the forces that
caused them. To develop these equations, again
consider Figure 1. The following equations describe the
motion of this vehicle through the impact depicted in this
figure:

macg , ground = Fground (16)

macg ,vertical = Fvertical − Fgravity (17)

Figure 10 – Vertical Impact Force

I roll α roll = Fground ⋅ r ⋅ sin φ − Fvertical ⋅ r ⋅ cos φ (18)

In these equations, m represents the vehicle’s mass, Iroll

represents its roll moment of inertia, and αroll represents
its roll acceleration. These equations of motion provide a
basis for interpreting the relationship between
accelerations recorded at the vehicle’s CoM and the
vehicle-to-ground impact forces to which the vehicle is
subjected during a rollover crash test. The first two
equations of motion can be rearranged, as follows, to
reveal this interpretation:

Fground acg , ground

= (19)
W g
Fvertical acg ,vertical Figure 11 – Vertical Impact Force
= +1 (20)
W g
The curves depicted in Figures 10 and 11 give us some
ability to judge which time step produces the most
Thus, the vehicle’s CoM acceleration along the ground accurate accelerations and forces. Physically, these
surface, in gravitational units, can be interpreted as the force curves should not drop below zero, since the
vehicle-to-ground impact force along the ground surface, vertical impact force cannot be negative. However, both
normalized by the vehicle weight. The vehicle’s vertical the 20 and 30 millisecond curves do drop significantly
CoM acceleration can be interpreted as a force that is 1g below zero over certain time intervals, and thus, they
contain physically unrealistic values. With a time step of
9
40 milliseconds, these unrealistic negative impact force significant than the discrepancies between the two
values are nearly eliminated. This gives one indication sensors themselves. Overall, the agreement between
that the forces calculated with the 40 millisecond time the video analysis and the sensor data does not appear
interval are likely more accurate than those calculated significantly affected by the time step with which the roll
with a 20 or 30 millisecond time interval. It is also likely velocity is calculated.
an indication that the sensor accelerations of Figure 8
are overestimating the peak accelerations since these
accelerations are directly related to the contact forces.
Were these sensor accelerations used to calculate
forces, they would no doubt produce peak forces well
above those calculated with the video analysis at a time
step of 40 milliseconds.

On the other hand, review of the test video appears to

show that using a 40 millisecond time step to calculate
the vertical impact force results in impact durations that
are too long. For instance, for the first wheel-to-ground
impact, the 40ms force curve indicates the impact
occurred over the time interval from 130 to 490
milliseconds. Review of the video reveals that this
impact actually occurred over the interval of time from
225 to 450 milliseconds. Thus, the 40ms force curve Figure 12 – Vehicle Roll Velocity Curves
implies an impact duration of 360ms for an impact that
actually only lasted for approximately 225 milliseconds. Using differential calculus to perform an error analysis
In terms of the impact duration, then, the 20 and 30 on Equation (21), it can be shown that the uncertainty in
millisecond force curves provide a better estimate of the the roll velocity can be estimated with the following
overall impact durations. equation:

Impact duration aside, given that the 40ms curve doesn’t δθ r

contain the physically unrealistic negative force values δωr = (22)
that the 20 and 30 ms curve do, the 40ms curve may still 2 ⋅ Δt
provide the most reasonable estimate of the peak impact
forces. If that is the case, then the first wheel-to-ground For the video analysis reported in this paper, it was
impact produced a peak vertical impact force that was found that when obtained by two separate analysts, the
approximately 335% of the vehicle weight and both the two set of motion data had an average difference in the
driver’s side and passenger’s side roof impacts roll angle of 0.58 degrees, with a standard deviation of
produced peak vertical impact forces of approximately 0.40 degrees. Thus, approximately 84% of the time, the
270% of the vehicle weight. difference between the roll angles obtained by the two
analysts was less than 1 degree. Assuming that this
Now, consider the vehicle’s roll velocity. The following average difference gives a reasonable estimate of the
difference equation will yield the vehicle’s average roll potential measurement errors or the uncertainty in the
velocity over two time steps: roll angles obtained with the camera-matching
technique, Figure 13 graphically represents Equation
(22) for time steps varying between 10 and 50
θ r ,t + Δt − θ r ,t −Δt
ωr = (21) milliseconds.
2Δt
Since the agreement between the roll velocities obtained
In Equation (21), θr is the vehicle roll angle at the with video analysis and the sensor data does not appear
specified time step. Similar equations could be written to depend significantly on the time step used to calculate
for obtaining the vehicle’s pitch and yaw velocities. the roll velocity, a 40 millisecond time step again seems
a good choice for use in the video analysis calculations.
Figure 12 compares the results from Equation (21), At this time step, the uncertainty in the calculated roll
calculated with time steps between 10 and 40 velocities will be around 18 degrees per second, with a
milliseconds, with the roll velocity obtained from the two confidence of 84%.
roll rate sensors on the vehicle. The sensor data shown
in this graph was filtered with a CFC 60 filter. In general,
the video analysis data and the sensor signals show
excellent agreement. In fact, the discrepancies between
the sensor data and the video analysis were less
10
 δΔVx = δv x2, f + δv x2,i (24)

δΔVz = δv z2, f + δv z2,i (25)

δΔω r = δω r2, f + δω r2,i (26)

When the calculations are carried out with a time step of

40ms, these equations yield uncertainties of 1.0 mph in
the ground plane and vertical CoM velocity changes and
25 degrees per second in the change in roll velocity, with
a confidence of approximately 84%.
Figure 13 – Roll Velocity Uncertainties The following equation will yield the vehicle’s energy at
any instant in time during the test:
Interestingly, the changes in roll velocity exhibited by the
roll velocity curves of Figure 12 during each impact
occurred over time intervals that were consistent with E=
m 2
(
⋅ v x − y + v z2 + k r2ω r2 + k p2ω p2 + k y2ω y2 ) (27)
what the test video showed. That being the case, the roll 2
velocity curves were used in conjunction with review of
the video to obtain the time intervals over each of the In this equation, kr, kp and ky are the vehicle’s radii of
three main impacts occurred. Those time intervals were gyration and ωr, ωp and ωy are the angular velocities.
estimated to be as follows: Figure 14 depicts the vehicle’s kinetic energy for the first
two seconds of the test.
Impact #1 – 225 to 450 ms (225ms duration)
Impact #2 – 705 to 870 ms (165ms duration)
Impact #3 – 1030 to 1200 ms (170ms duration)

Having obtained these impact durations, we can now

return to the velocity curves of Figures 6, 7 and 12 and
determine the translational and angular velocity changes
for each of the three impacts considered in this paper.
This process results in the following velocity changes for
each of these impacts:

Impact #1

ΔVx = -4.9 mph

ΔVz = 6.7 mph
Δωr = -163 deg/s Figure 14 – Vehicle Energy

Impact #2 Using differential calculus to perform an error analysis

on Equation (27), it can be shown that the uncertainty in
ΔVx = -1.1 mph the vehicle’s kinetic energy can be estimated with the
ΔVz = 3.2 mph following equation:
Δωr = 60 deg/s
v x2− y δv x2− y + v z2δv z2 + k r4ω r2δω r2 +
Impact #3 δE = m ⋅ (28)
k p4ω p2δω p2 + k y4ω y2δω y2
ΔVx = -1.5 mph
ΔVz = 4.2 mph
In this equation, δωr, δωp, and δωy are the angular
Δωr = 125 deg/s velocity uncertainties.
Using differential calculus, it can be shown that the The energy loss for each of the three main impacts that
uncertainty in these velocity changes can be estimated occurred during this test can be obtained by subtracting
with the following equations: the vehicle’s kinetic energy at the end of the impact from
11
its kinetic energy at the beginning of impact. Using Impact #3
differential calculus to determine the uncertainty in these
energy losses, the following equation is obtained: vx-y,i = 23.3 mph
vz,i = -2.4 mph
(29) ωr,i = -175 deg/s
δΔE = δE 2f + δEi2
In addition to the initial velocity conditions, the impact
In this equation, δΔE is the overall uncertainty in the model equations also call for the vehicle’s radius of
calculated energy loss, δEf is the uncertainty in the gyration about the roll axis, the impact angle and radius,
vehicle’s calculated energy loss at the end of the impact, and the coefficient of restitution and impulse ratio. To
and δEi is the uncertainty in the vehicle’s calculated estimate the radius of gyration for the vehicle in the
energy loss at the beginning of the impact. subject crash test, the vehicle weight was measured
prior to running the crash test and the radius of gyration
Given the impact time intervals reported above, the data was estimated based on the equations reported in
of Figure 14 can be used to obtain the following energy References 1 and 13.
losses for each of the three main impacts:
The impact angle and impact radius for each impact
ΔE1 = 35,775 ft-lbs ± 8,219 ft-lbs (27.0% ± 6.2%) were obtained by first selecting a vehicle position and
orientation that might best represent the “average”
ΔE2 = 8,957 ft-lbs ± 7,262 ft-lbs (9.8% ± 7.9%) vehicle position and orientation during the impact. Then
the contact point was placed in a location within the
ΔE3 = 13,749 ft-lbs ± 6,932 ft-lbs (16.0% ± 8.1%) deformation region that might best represent the point
where the resultant collision force was applied to the
The uncertainties reported with these energy losses vehicle. The impact angle and impact radius was then
have a confidence of approximately 84%. These measured.
uncertainties are rather significant relative to the best
estimate values. Clearly, considerable uncertainty has It was, of course, not apparent what vehicle position and
accrued through the course of these calculations. orientation and what contact point location would best
represent each impact. That being the case, the authors
ANALYZING THE IMPACTS took six impact angle and impact radius measurements
for each impact. These measurements were obtained
Our primary goal in the research reported in this paper from three vehicle positions per impact – the vehicle
was to determine the degree to which a simple, planar, position at first contact with the ground, the vehicle
impulse-momentum impact model would accurately yield position at the time of the maximum vertical impact
the translational and rotational velocity changes and the force, and the vehicle position at the time of separation
energy loss for the three impacts identified above. The from the ground. For each of these times, impact angle
previous section has described the actual dynamics and radius measurements were taken at the leading and
associated with each of those impacts. In this section, trailing edges of the contact region. Taking these six
the impact model introduced earlier is used to analyze measurements allowed us to establish a range of
these three impacts. possible impact angle and radius measurements.
Ultimately, it was found that taking the impact angle and
In analyzing each impact, the initial velocity conditions radius measurements at the time of the peak vertical
were obtained from the analysis of the previous section. impact force made the most physical sense and resulted
These initial velocity conditions for each impact were as in the best results from the impact model. That being the
follows: case, the authors established ranges for the impact
angle and radii by measuring to the leading and trailing
edges of the contact region at the time of this peak
Impact #1
impact force.
vx-y,i = 29.5 mph
Figures 15 through 17 are frames from the high-speed
vz,i = -3.4 mph
video of the test that visually depict one of the impact
ωr,i = -62 deg/s
radius and impact angle values for each of the impacts.
The angles and distances in these images are not to
Impact #2
scale and are simply intended to help the reader
envision how these parameters would be measured.
vx-y,i = 24.0 mph
vz,i = -2.3 mph
ωr,i = -225 deg/s

12
 Figure 15 – Impact Angle and Radius for Impact #1

Figure 18

In this figure, ri and φi are an initial estimate of the impact

radius and impact angle. The distance yc is the distance
separating a point on the ground directly below the CoM
from the point of collision force transfer. The parameter
Δyc represents a change in this distance, yc. Equations
(30) and (31) below relate the initial impact angle and
radius to a new impact angle and radius, φf and rf,
through the parameter Δyc.
Figure 16 – Impact Angle and Radius for Impact #2
r f = ri 2 + 2 ⋅ Δyc ⋅ cφi ⋅ ri + Δyc2 (30)

⎡ ri cos φi + Δy c ⎤
φ f = cos −1 ⎢ ⎥ (31)
⎢⎣ rf ⎥⎦

Thus, the location of the point of collision force transfer

could be easily iterated to obtain the best match
between the video analysis and impact model results by
simply changing the parameter Δyc. For this optimization
process, the values of the impact angle and radius
obtained with Equations (30) and (31) would be fed into
Equations (1) through (3). When iterating in this manner,
the analyst should, of course, ensure that the optimized
Figure 17 – Impact Angle and Radius for Impact #3 values of the impact angle and radius fall within their
respective ranges.
Having established a range for the impact angle and
radius for each impact, these parameters can be varied Once the contact point for each impact has been
within those ranges to obtain the best match between defined, the coefficient of restitution for that impact could
the video analysis and the impact model results. To theoretically be obtained with two different methods. The
simplify this process of obtaining the best match, a new first method would involve using the video analysis data
parameter, Δyc, was introduced into Equations (1) to measure the vertical velocity of the contact point at
through (3) to allow for easy iteration of the contact point the beginning and the end of the impact and from these
location. To see the meaning and function of this to calculate the coefficient of restitution. The second
parameter, consider Figure 18. method would involve selecting a representative vehicle

13
orientation for the impact – in this case, the vehicle results in too much uncertainty for the values to be of
orientation at the time of the peak vertical force – and any use. To see this, recall that the coefficient of
then imposing the assumption that the vehicle restitution is defined as follows:
orientation does not change during the impact and then
calculating the coefficient of restitution based on the v zc , f
initial and final vertical velocities, the initial and final roll e=− (32)
velocities, and the impact angle and impact radius. v zc ,i

To understand the difference between these two In this equation, vzc,f is the final vertical velocity at the
methods, consider that the coefficient of restitution is point of collision force transfer and vzc,i is the initial
defined as the negative ratio of the post-impact to the vertical velocity at the point of collision force transfer.
pre-impact vertical velocity at the point where the Using differential calculus, it can be shown that, for the
resultant collision force is transferred. Since the point of second method of obtaining the coefficient of restitution,
collision force transfer will lie near the perimeter of the the uncertainty in the calculated value of the coefficient
vehicle, both the vehicle’s roll velocity and roll orientation can be estimated with the following equation:
can significantly influence to the vertical velocity at Point
C at any instant in time. The vehicle’s orientation will
determine what component of the vehicle’s perimeter 1 δv z2, f + r 2 c 2φδωr2, f +
δe = ⋅ (33)
velocity contributes to the vertical velocity at C and in
which direction it contributes. The influence of these
v zc ,i (
e 2 δv z2,i + r 2 c 2φδωr2,i )
factors is displayed mathematically in Equation (4).
Equation (33) includes only the uncertainties that arise in
Now, consider that the impact model equations the coefficient of restitution due to uncertainties in the
described above assumed that the vehicle does not initial and final translational and angular vehicle
move during the impact (instantaneous impact). In velocities. It ignores uncertainties associated with the
reality, even given the relatively short impact durations impact radius and angle (though these uncertainties
associated with the impacts considered in this paper, the certainly exist). Still, when a coefficient of restitution and
vehicle traverses a relatively significant roll angle during a certainty range is calculated for Impact #1, the result is
these impacts. Specifically, in the specific test under excessive uncertainty, as follows:
consideration here, the vehicle traversed a roll angle of
approximately 31 degrees during the first impact, 31-½ e1 = 0.264 ± 0.852 (± 323%)
degrees during the second impact, and 21 degrees
during the third impact. This uncertainty range is wide enough that the value of
the coefficient of restitution could actually lie anywhere
If one uses the first method for calculating the coefficient within the physically possible range between 0.0 and
of restitution, then the vertical velocity of the contact 1.0. Given such significant uncertainty, calculated
point at the beginning and the end of the impact will be coefficients of restitution will not be useful to our
measured with the impact radius at different angles. This analysis. That being the case, in the analysis reported
difference is caused by the fact that the vehicle here, the coefficient of restitution was used as another
continues to roll throughout the impact. If one uses the optimizing parameter to obtain the best fit between the
second method for calculating the coefficient of impact model and video analysis results.
restitution, then the vertical velocity of the contact point
at the beginning and end of the impact will be calculated In addition to the coefficient of restitution, the impact
with the impact radius at a constant angle (its angle at model equations also utilize the impulse ratio as an input
the time of the peak force). Thus, the first method yields parameter. A calculated value of the impulse ratio can
a coefficient of restitution that may cause inaccuracies in be obtained for each impact with Equation (2). Using this
the impact model since it violates the instantaneous equation, the following calculated values were obtained
impact assumption of the model. The second method is for the impulse ratios associated with each of the three
more consistent with the modeling assumptions, but it impacts considered in this paper:
yields a value for the coefficient of restitution that does
not reflect the fact that the vehicle rolls throughout the μ1 = -0.421
impact. This second way of treating the coefficient of
restitution is also most consistent with the way that it has
μ2 = -0.161
typically been obtained for use with other planar,
impulse-momentum impact models [18].
μ3 = -0.189
As it turns out, this discussion is, for now, destined to
Though we have not undertaken this analysis here, one
remain a theoretical one since actually calculating the
would expect that these values may also be subject to
coefficient of restitution for the impacts of this crash test
14
considerable uncertainty. That being the case, in the Based on the results of this section and the previous
analysis reported below, the authors began by setting one, Table 1 summarizes the relevant parameter values
the impulse ratio at these calculated values but then for each impact. This table includes the 84% confidence
changed it as necessary to obtain the best fit solution for certainty ranges for a number of these parameters.
the video analysis results. As it turned out, for Impacts 1
and 3, these calculated values did yield the best-fit
solution.

Impact #1 Impact #2 Impact #3

(Driver’s Side Wheels) (Driver’s Side Roof) (Passenger’s Side Roof)
Vehicle Weight(lbs) 4494
Radius of Gyration (ft) 2.12
Initial Ground Plane Velocity (mph) 29.5 ± 0.7 24.0 ± 0.7 23.3 ± 0.7
Initial Vertical Velocity (mph) -3.4 ± 0.7 -2.3 ± 0.7 -2.4 ± 0.7
Initial Roll Velocity (deg/s) -62 ± 18 -225 ± 18 -175 ± 18
Ground Plane Velocity Change (mph) -4.9 ± 1.0 -1.1 ± 1.0 -1.5 ± 1.0
Vertical Velocity Change (mph) 6.7 ± 1 3.2 ± 1 4.2 ± 1
Change in Roll Velocity (deg/s) -163 ± 25 60 ± 25 125 ± 25
Impact Angle (deg) 88 to 109 88 to 112 68 to 117
Impact Radius (ft) 36.3 to 37.9 42.0 to 45.0 40.6 to 43.5
Impact Times (ms) 225 to 450 705 to 870 1030 to 1200
Impact Duration (ms) 225 165 170
Time @ Peak Vertical Force (ms) 330 770 1130
Impact Energy Loss (ft-lbs) 35,775 ± 8,219 8,957 ± 7,262 13,749 ± 6,932
Impact Energy Loss (%) 27.0 ± 6.2 9.8 ± 7.9 16.0 ± 8.1
TABLE 1

RESULTS results showed excellent agreement with the mean

values from the video analysis. For the first two impacts,
Equations (1) through (5) were used to calculate the best fit with the video analysis was obtained by
translational and rotational velocity changes and an placing the impact center near the geometric center of
energy loss for each of the three vehicle-to-ground the deformation regions. For the third impact, the best fit
impacts discussed in this paper. These calculations were was obtained by placing the impact center near the
optimized to obtain the best fit with the video analysis leading edge of the deformation region. For the first and
data by varying the impact angle, impact radius, third impacts, the calculated impulse ratios did yield the
coefficient of restitution and the impulse. The results of impact model solution that was the best fit to the video
these calculations are reported in Table 2. In this table, analysis data. In the case of the second impact, the
the magnitude of the over or underestimation of the optimized impulse ratio was lower in magnitude than the
actual value is listed in parenthesis. As a comparison calculated impulse ratio. This second impact was also
between Tables 1 and 2 demonstrates, the impact model the case that exhibited the greatest divergence from the
velocity changes obtained from the video analysis.

Impact #1 Impact #2 Impact #3

(Driver’s Side Wheels) (Driver’s Side Roof) (Passenger’s Side Roof)
Vertical Velocity Change (mph) 6.7 (+0.0) 4.0 (+0.8) 4.2 (+0.0)
Ground Plane Velocity Change (mph) -4.9 (+0.0) -1.0 (+0.1) -1.5 (+0.0)
Change in Roll Velocity (mph) -162 (+1) 55 (-5) 125 (+0.0)
Energy Loss (ft-lb) 35,389 (-386) 9,166 (+217) 13,220 (-529)
Energy Loss (%)
Coefficient of Restitution 0.480 0.000 0.170
Impulse Ratio -0.421 -0.125 -0.189
TABLE 2

15
DISCUSSION region. For the third impact, the best results were
obtained by placing the impact center near the leading
Impact Model Assumptions edge of the contact region.

Now, consider the impact model results of Table 2 in It is unclear to us at this point whether the fourth
relationship to the impact model assumptions. Based on assumption, that no moment arose at the point of
the results of Table 2, it can be concluded that any collision force transfer, was violated or not. Further
violation of the impact model assumptions did not research could explore the influence of a moment at the
significantly degrade the accuracy of the impact model contact point on the resulting rollover dynamics. At any
calculations. That result is encouraging and indicates rate, ignoring such a moment did not adversely affect the
that the simple planar impact model used here can offer impact results to any significant degree.
insight into rollover dynamics. That said, it still makes
sense to be specific about the degree to which certain The fifth impact model assumption, that no velocity
impact model assumptions were violated and how those reversal occurs during the impact, was satisfied for each
violations were dealt with in the modeling. It is, perhaps, of the three impacts that were analyzed in this paper.
obvious that there will be cases where violations of the This was determined by reviewing the crash test video.
impact model assumptions will be more significant that
what was observed in this test and that, in such cases, The Larger Context
the impact model used here will not be useful.
A number of studies that have sought causes of
Impacts #1 and #2 nearly satisfied the first impact model occupant injuries in rollovers have focused on crash
assumption, which was that the impact occurred entirely attributes or outcomes – the number of quarter rolls [12],
in a single plane. At the time of these impacts, the the initial vehicle translational speed [15], and the
vehicle had experienced very little pitching or yawing magnitude of roof deformation [10] or post-crash
and the vehicle velocity was still in line with its initial headroom [16] – and on how those attributes correlate to
direction. By the time of Impact #3, the vehicle had injury rates. However, crash attributes and outcomes are
developed a relatively significant forward pitch angle with not causes. In fact, they are effects that result from a
a corresponding forward pitch velocity. Thus, strictly combination of the rollover initial conditions and the
speaking, Impact #3 did violate the planar impact particular forces to which the vehicle is subjected during
assumption. Despite this, the impact model results for the rollover. It is these initial conditions and underlying
Impact #3 still showed excellent agreement with the forces that cause certain crash attributes or outcomes to
video analysis results. Thus, this violation of the planar be present. If there is a correlation between certain injury
impact assumption was not fatal to the accuracy of the types and a particular crash attribute, it is the underlying
model. forces that cause this crash attribute that could also
relate to the actual cause of that injury type. Thus, an
All three impacts violated the second assumption of the understanding of the initial conditions and underlying
impact model, which was that the vehicle experienced forces causing certain crash attributes may be significant
no change in position during the impact. As was stated to reducing injury potential in rollovers.
above, the vehicle traversed a roll angle of
approximately 31 degrees during the first impact, 31-½ In theory, an impact model such as the one used in this
degrees during the second impact, and 21 degrees paper could be helpful in understanding the underlying
during the third impact. Overcoming this discrepancy forces and physical relationships that cause certain
between the impact model assumption and the real rollover crash attributes. The impact model used in this
impacts appeared to be primarily a matter of determining paper is simple enough that parameter relationships are
a vehicle position that was the most representative of the not obscured. The equations of the impact model are
impact. In the case of these three impacts, the vehicle algebraic, and so, their solution need not be buried
position at the peak vertical force yielded the best impact within computer code. If one could have confidence that
model results. this relatively simple impact model would yield physically
realistic and accurate results, then one could
Clearly, the third assumption of the impact model, that conceivable use it to develop an understanding of why
the impact force was applied at a single point would be certain crash conditions lead to certain crash attributes.
violated for any real-world impact. Still, overcoming this
discrepancy appeared to be primarily a matter of That, ultimately, has been the goal of the research
determining a contact point location that was most reported here: to determine whether this simple, planar
representative of the point of application of the resultant impact model would yield accurate, and therefore useful,
collision force. As we have already stated, for the first results for the vehicle-to-ground impact that occur during
two impacts, the best fit was obtained with the video rollovers. The results reported above are encouraging in
analysis data by placing the point of collision force this regard. These results suggest that the impact model
transfer near the geometric center of the deformation is capable of yielding accurate results related to the
16
velocity changes and energy loss resulting from a 5. Brandse, Jeroen, “Methodology for Simulation of
vehicle-to-ground impact. That being the case, it is Rollover Cases,” SAE Technical Paper Number
reasonable to use this impact model to examine the 2006-01-0558.
influence of various factors on rollover dynamics.
6. Chou, Clifford C., McCoy, Robert W., Le, Jialiang, “A
For instance, it can be observed that the rate at which a Literature Review of Rollover Test Methodologies,”
rolling vehicle decelerates will be determined by the Int. J. Vehicle Safety, Volume 1, Nos. 1/2/3, 2005.
accumulation of the ground plane velocity changes that
occur during the rollover. Thus, any factor that influences 7. Day, Terry D., “Applications and Limitations of 3-
the ground surface velocity changes will also likely Dimensional Vehicle Rollover Simulation,” SAE
influence the deceleration rate that the vehicle Technical Paper Number 2000-01-0852.
experiences. These factors include the following: (1) the
available surface friction, (2) the ground speeds, vertical 8. “Dolly Rollover Recommended Test Procedure,”
velocities, and roll velocities experienced during the roll, Surface Vehicle Recommended Practice J2114,
(3) the orientations of the specific vehicle-to-ground Society of Automotive Engineers, October 1999.
impacts that occur during the roll, (4) the vehicle
geometry, (5) and the stiffness of the vehicle structures 9. Fonda, Albert G., “Nonconservation of Momentum
engaged during the roll. Future research could examine During Impact,” SAE Technical Paper Number
rollover test data and real-world rollovers to determine 950355.
the degree to which each of these factors might affect a
rolling vehicle’s deceleration rate over the course of an 10. Friedman, Donald, “Roof Crush Versus Occupant
entire rollover. Injury From 1988 to 1992 Nass,” SAE Technical
Paper Number 980210.
CONCLUSIONS
11. Hoffman, Joe D., “Numerical Methods for Engineers
• For all three vehicle-to-ground impacts considered in and Scientists,” Second Edition, Marcel Dekker,
this study, the impact model results for the velocity 2001.
changes and energy loss fell within the certainty
ranges obtained from the video analysis for these 12. Hughes, Raymond J., et al., “A Dynamic Test
parameters. Procedure for Evaluation of Tripped Rollover
Crashes,” SAE Technical Paper Number 2002-01-
• These results suggest that the impact model is 0693.
capable of yielding accurate results related to the
velocity changes and energy loss resulting from a 13. MacInnis, Duane D., “A Comparison of Moment of
vehicle-to-ground impact. That being the case, it is Inertia Estimation Techniques for Vehicle Dynamics
reasonable to use this impact model to examine the Simulation,” SAE Technical Paper Number 970951.
influence of various factors on rollover dynamics.
14. Marine, Micky C., “On the Concept of Inter-Vehicle
REFERENCES Friction and Its Application in Automobile Accident
Reconstruction,” SAE Technical Paper Number
1. Allen, R. Wade, “Estimation of Passenger Vehicle 2007-01-0744.
Inertial Properties and Their Effect on Stability and
Handling,” SAE Technical Paper Number 2003-01- 15. Malliaris, C., “Pivotal Characterization of Car
0966. Rollovers,” Proceedings of the 13th ESV Conference,
November 1991.
2. Brach, Raymond M., Mechanical Impact Dynamics:
Rigid Body Collisions, Revised Edition, Brach 16. Rains, Glen C., “Determination of the Significance of
Engineering, 2007. Roof Crush on Head and Neck Injury to Passenger
Vehicle Occupants in Rollover Crashes,” SAE
3. Brach, Raymond M., Brach, R. Matthew, “A Review Technical Paper Number 950655.
of Impact Models for Vehicle Collision,” SAE
Technical Paper Number 870048. 17. Rose, Nathan A., Beauchamp, Gray, Fenton,
Stephen J., “Factors Influencing Roof-to-Ground
4. Brach, Raymond M., Brach, R. Matthew, “Energy Impact Severity: Video Analysis and Analytical
Loss in Vehicle Collisions,” SAE Technical Paper Modeling,” SAE Technical Paper Number 2007-01-
Number 871993. 0726.

18. Rose, Nathan A., “Quantifying the Uncertainty in the

Coefficient of Restitution Obtained with
17
 Accelerometer Data from a Crash Test,” SAE mk r2 (ω r , f − ω r ,i ) = Py ⋅ r ⋅ sφ − Pz ⋅ r ⋅ cφ (A3)
Technical Paper Number 2007-01-0730.

19. Rose, Nathan A., “The Influence of Vehicle-to- In Equations (A1) through (A3), m is the vehicle mass, kr
ground Impact Conditions on Rollover Dynamics and is the vehicle’s radius of gyration for the roll axis, r is the
Severity,” SAE Technical Paper Number 2008-01- distance between the vehicle’s CoM and the point at
0194. which the impact force is applied (Point C), φ is the angle
between the orientation of the ground plane and the line
20. Rose, Nathan A., “A Method to Quantify Vehicle connecting the CoM to Point C, Pz and Py are the normal
Dynamics and Deformation for Vehicle Rollover (vertical) and tangential (ground plane) impulse
Tests Using Camera-Matching Video Analysis,” SAE components that result from the impact and Pg is the
Technical Paper Number 2008-01-0350. gravity impulse. Translational velocity components are
denoted with the letter v and final and initial velocities
21. Steffan, Hermann, “How to Use PC-Crash to are denoted with the subscripts f and i.
Simulate Rollover Crashes,” SAE Technical Paper
Number 2004-01-0341. The following constraint equations govern the impact
energy loss along the normal and tangential directions:
22. Taylor, John R., An Introduction to Error Analysis,
Second Edition, University Science Books, 1997. v zf − r ⋅ cφ ⋅ ω r , f
e=− (A4)
APPENDIX A v zi − r ⋅ cφ ⋅ ω r ,i

This appendix presents the derivations of Equations (1) Py

through (3) which yield the translational and rotational μ= (A5)
velocity changes for the idealized vehicle-to-ground Pz
impact shown in Figure A1.
In Equations (A4), e is the coefficient of restitution for the
impact, which is defined as the negative ratio of the post-
impact to pre-impact vertical velocities at the point of
collision force transfer (Point C). In Equation (A5), μ is
the impulse ratio, which establishes the magnitude of the
tangential impulse relative to the magnitude of the
normal impulse. This equation yields Equation (2) in the
main body of this paper when Equations (A1) and (A2)
are substituted into it.

Substituting Equation (A5) into (A3) yields the following

equation:

m ⋅ k r2 ⋅ (ω r , f − ω r ,i ) = Pz ⋅ r ⋅ (μ ⋅ sφ − cφ ) (A6)

Then, the following equation results from substituting

Equation (A1) into (A6):

Figure A1 m ⋅ k r2 ⋅ Δω r = (mΔVz + Pg )⋅ r ⋅ (μ ⋅ sφ − cφ ) (A7)

The development of these equations largely follows the
development of the planar impact equations presented in
References 3 and 4, with the exception that a gravity The gravity impulse can be rewritten with the following
equation, which can then be substituted into Equation
impulse is included.
(A7) to yield Equation (A9):
The principle of impulse and momentum dictates the
following equalities: Pg = mg ⋅ Δt i (A8)

mv zf − mv zi = Pz − Pg (A1) m ⋅ k r2 ⋅ Δω r = (mΔVz + mgΔt i ) ⋅ r ⋅ (μ ⋅ sφ − cφ ) (A9)

mv yf − mv yi = Py (A2) In Equation (A8), g is the gravitational constant and Δti is

the impact duration. Algebraic manipulation of Equation
18
(A9) yields Equation (A10), which is equivalent to
Equation (3) in the main body of this paper.

Δω r = ω r , f − ω r , i =

(ΔVz + g ⋅ Δt i ) ⋅ r ⋅ (μ ⋅ sφ2 − cφ ) (A10)

kr

Now, algebraically manipulate Equation (A4) to solve for

the final roll velocity:

ωr , f =
1
v zf +
e
(v zi − r ⋅ cφ ⋅ ωr ,i ) (A11)
r ⋅ cφ r ⋅ cφ

Now, equate Equations (A10) and (A11) through the final

roll velocity and algebraically manipulate to obtain the
following equation:

⎧ k r2 ⎫
ΔV z = −(1 + e ) ⋅ v zc ,i ⋅ ⎨ 2 ⎬
2
(
⎩ k r + r c φ − μ ⋅ s φ ⋅ cφ
2
)
⎭
(
⎧ r 2 c 2 φ − μ ⋅ s φ ⋅ cφ ⎫
− g ⋅ Δt i ⋅ ⎨ 2
) (A12)
⎬
(
⎩ k r + r c φ − μ ⋅ s φ ⋅ cφ ⎭
2 2
)
Equation (A12) is equivalent to Equation (1) in the main
body of this paper.

19