Volume 9, Issue 2

2014 ARC--CSI DATA DVD

C

2014 Vol.9 Iss.2

ollision C The International Compendium for Crash Research

Volume 9 Issue 2

FALL 2014

Collision: The International Compendium for Crash Research

Using Inequality Contraints in the Probability Analysis of

Motor Vehicle Accidents

Volume 9, Issue 2 - FALL 2014

Ignition Switch Data in CDR Reports

“Low Velocity” Testing collisionmagazine.com

CRASH SITE EVIDENCE

A 1968 Mustang GT, driven by Lt. Frank Bullitt, is chasing a 1968 Dodge Charger R/T, driven by a man known only as “The Killer,” down Taylor Street on Russian Hill in San Francisco.

We have the STIFFNESS data you need for your calculations!!

Now includes a FORCE BALANCE calculation module!!

4N6XPRT StifCalcs® puts the NHTSA Crash Test database on YOUR 32 or 64 bit MS-Windows computer, automatically calculates A-B-G values Front, Side, and Rear - and Test Specific Crush Factors based on the NHTSA test data, allows the creation of “CLASS” vehicles and determine the “CLASS” stiffness values, PLUS conduct FORCE BALANCE calculations when needed, all of this with no need for you to get on the internet, and all for the LOW price of $650.00*!

Can you do this yourself “for free”?? Only if YOUR TIME = $0.00!!

For more information, program order form, and sample printouts please visit our web site at http://www.4n6xprt.com.

4N6XPRT StifCalcs® is provided by 4N6XPRT Systems

For a full reconstruction of this 46 second chase sequence, visit www.edccorp.com/bullitt

8387 University Avenue - La M esa, CA 91942 - USA Phone: 800-266-9778 / (619) 464-3478 Fax: (619) 464-2206

Individual Vehicle Stiffness Data

HVE directly answers the following questions using Newton’s laws of motion:

Questions

Answers

Speed on Taylor St.

30 – 37 mph (both vehicles)

Driver Inputs

Mustang: 2-3 lb brake pedal force, 200 deg left steer onto Filbert, then 25-65 %WOT Charger: 3-5 lb brake pedal force, 175 deg left steer onto Filbert, then 70-75 %WOT

Maximum Undercarriage Impact Force

Mustang: Fx= -1730 lb, Fy= -13 lb, Fz= -2120 lb Charger: Fx= -8610 lb, Fy= -120 lb, Fz= -11100 lb

Maximum Suspension Force

Mustang: 4330 lb @ L/F; 3270 lb @ R/R Charger: 4810 lb @ R/F; 3530 lb @ L/R

Hubcaps Lost

Mustang: N/A (American Racing Mags) Charger: Five

Other software may claim to answer these questions. Have you closely evaluated their claims and the basis for their results?

So, you want Stiffness data for a vehicle, QUICKLY & CONVENIENTLY, but you don’t want to buy a program? Order your data on a vehicle-by-vehicle basis from us at only $40.00 per test!

Year: _________ Make: _____________________ Model:____________________________________ Impact on (circle one): Front - Side - Rear Impact speed (circle one): High - Low YOUR CONTACT INFORMATION Name: ________________________________________ Company: _____________________________________ Address: ______________________________________ City/State/Zip: _________________________________ Phone: _______________________________________ Fax: __________________________________________ Email: ________________________________________ Visa/MC/AmEx: ________________________________ Exp.:____/____ Case Reference: ___________________

To learn more about HVE and all of the available tools for crash investigators, call 503-644-4500 or visit www.edccorp.com. Our professional sales staff will help you get started using HVE today!

Send Order to:

4N6XPRT Systems Fax: 619-464-2206

KNOWLEDGE = POWER Engineering Dynamics Corporation • 8625 SW Cascade Blvd., Suite 200 Beaverton, OR 97008 USA • Phone (503) 644-4500 • www.edccorp.com

Ph: 800-266-9778 8387 University Avenue / La Mesa, CA 91942 / USA E-mail - 4n6@4n6xprt.com

Are you ready for REALITY REALITY? All-in-One 64-bit Crash and Crime Solution for advanced professionals 0 15 30

Realize your full potential Reality, by ARAS 360 Technologies, is the most advanced all-in-one software application in the forensic industry.

Why REALITY REALITY? n

A complete 3D drawing system allowing users to generate precision diagrams of crash, crime or any incident scene with stunning details and realism.

n

Physics based 3D animation system allows auto-synchronization of up to 10 events, animation of any model in the Reality system complete with rotating and turning wheels, sub animation of human models, and complete control of lighting.

n

Point Cloud management system With features not found in any other application with its true 64 bit application, users can download point clouds up to 2 billion and view the cloud at maximum density. Reality supports all point cloud scanners including Z+F, Lecia, FARO, and Reigl. Only Reality allows users to instantly generate models from objects in the scan and move them to other positions and other scenes.

n

Advanced crush features

Only Reality has a fully validated method of analyzing scanned vehicles to determine crush volume and calculate energy loss and EBV.

s Combine high density point cloud scenes – with Reality advanced animation to create a realistic driver’s view.

n

Complete and validated 3D simulation system Only Reality allows simulation of unlimited moving vehicles and objects, even in a point cloud scene. Add mesh based collision modelling, suspension effects, and 3D terrain following and you have the most modern simulation system ever introduced to the forensic crash industry.

n

s Side view of camera

placement in high density cloud

s Download the cloud of post impact positions

Direct compatibility with all measurement systems Our powerful graphics engine allows you to generate realistic scenes under any light conditions. Why not generate professional looking scenes?

n

Direct integration with Google imagery including Google terrain data so users can choose to generate digital terrain models of any crash or crime scene from total station data, google terrain data or point cloud data.

Call for a demo today: 877 814 2360 www.aras360.com

Contents

Fall 2014

Volume 9 Issue 2 8

inside 4

Letter From the Editor

6

NAPARS: Letter From the President

features 8

Using Inequality Constraints in the Probability Analysis of Individual Motor Vehicle Accidents by: Scott Kimbrough

26

Wheel Slip and Its Effect on Reported Vehicle Speed by: David M. Hallman

42

How Long Can You Stay Awake? by: Clinton Marquardt

46

Low Velocity Piston Isolator Testing of a Ford Crown Victoria by: Lissette Ruberte, Susan Lantz, Robert Thompson, Michael Chan, Bobby Clemence, Greg Dycus, Mitchell Burton, Brian Chan, Terry Wong, and Billy Cox

60

Vehicle Dynamics and Resultant Occupant Accelerations Caused by Vehicle Wheel Separation by: C. Brian Moody, Orion P. Keifer, Bradley C. Reckamp, and Wes Richardson

80

CDR Report Data from Vehicles Subject to the GM Ignition Switch Recall with the “Epsilon” ACM by: W. R. Rusty Haight

106

Determination of Motorcycle PreCollision Speed - Part 1 & 2 by: Oren Masory, Wade Bartlett, and Bill Wright

www.collisionpublishing.com

26

80

60

Collision Magazine - Volume 9 Issue 2 3

Letter from the editor In the past, the discussion of “peer review” as a concept and as a process as it's used for crash related papers has played out on the pages of Collision and elsewhere. Not long ago, one organization felt it necessary to circulate a memo at a meeting defending their peer review process in response to our then stated position and indictment of that process. More recently, we’ve seen another obvious example of how that process failed. We stand by the previously discussed belief that “peer review” is demonstrably used to stifle dissent or opposing views or, perhaps worse, advance an agenda no matter the result.

At the same time, we decided to take on a different approach to how we represent the presentation of articles in Collision. Beginning with this issue, we will start indicating whether an article is “peer reviewed” or simply “editorially reviewed.” While we still believe it's our job to publish whatever we get (within reason...) and let the end readers, not a cabal bent on publishing like minded material, be the final judge of a document’s value. Below is the distinction we plan to illustrate between the two categories: Peer Reviewed There would be a banner on the first page of the article that would indicate the paper had been reviewed by our review committee and approved as published. We have in place a more formal review process than we’ve had in the past and will mark, on the first page, whether or not that article has been through that process.

Editorially Reviewed There would be a banner on the first page of the article that would have been reviewed only by editorial staff. These articles are reviewed but not by way of a more “formal process” including, for example, they might be rejected without changes which would be the case with those which have been more formally “peer reviewed.”

So, what's the difference?

4 Collision Magazine - Volume 9 Issue 2

As we envision it, there are submissions (other than those which are clearly already opinion content such as “My Turn At The Wheel”) which should be published because we believe it’s our job to distribute information but don’t necessarily rise to the need for more formal review.

For example, a review of how a computer program might be used or how an agency/company does this or that would not necessarily rise to the level of requiring a “peer” review. Similarly, the transcript of a mock hearing, a discussion of legal concepts by a lawyer - are content not requiring any sort of peer review; they can be “editorially reviewed.”

We believe there are things that - and frankly this is our ONLY purpose in contemplating this path - might benefit some if they could say the Collision article they read or relied on was “peer reviewed” - a representation generally made with no understanding of the underlying process or lack thereof. For that reason, we are going to start sending out select articles to qualified reviewers from within the community to get the kind of reviews that might make them better and then identify them as “peer reviewed” starting with this issue of Collision.

Do you want to participate? Would you like to be a reviewer?

If you are interested in being on our review board, drop me a note, let me know how to send you an article in the future. Reviewers might get one, maybe two papers to review as the next issue approaches and have a chance then to participate in this process.

Managing Editor sbaker@collisionpublishing.com

www.collisionpublishing.com

BACK ISSUES

Collision Publishing Inc PO Box 892885 Temecula, CA 92589 Toll Free: 800-280-7940 Email: editor@collisionpublishing.com Web: www.collisionmagazine.com ISSN: 1934-8681

If you would like purchase back issues of Collision Magazine, please visit www.collisionmagazine.com or call us toll free 800-280-7940.

ADVERTISER INDEX

COLLISION STAFF Scott Baker Sean Haight Tonya Baker Carla McLeod

Owner, Managing Editor Senior Editor Advertising Manager Subscription Services

All rights reserved, © 2014 Collision Publishing Inc. The opinions and conclusions expressed in this publication and attached data disk and throughout the articles attributed to specific authors are the opinions and conclusions of the authors noted and not necessarily the editorial staff or anyone else for that matter.

While some articles have been reviewed for content, the accuracy of reprinted models or equations cannot be fully guaranteed. It is the responsibility of the reader to apply critical thinking to an individual review of the content and make their own personal judgements as to its value to them, individually. At the end of the day, facts belong to everybody, any other opinions to us. The distinction is yours to draw...otherwise, the opinions expressed herein are not necessarily those of any employer, not necessarily ours and probably not necessary. Dis-

senting opinions, discussion or conclusions which may express an adverse position to those expressed herein by specifically cited authors can be addressed in writing by sending an email to the editor at editor@collisionpublishing.com or sending a “regular mail” letter to us at PO Box 892885, Temecula, CA 92589. By

sending correspondence to either our email or snail mail addresses listed in this publication you are agreeing that: (1) we are by definition, “the intended recipient” (2) all information in the email is ours to do with as we see fit and make such financial profit, political mileage, or good joke as it lends itself to and, (3) this overrides any disclaimer or statement of confidentiality that may be included on your original message. The entire physical universe, including this publication and attached data disk, may one day collapse back into an infinitesimally small space. Should another universe subsequently re-emerge, the existence of this publication and data disk in that universe cannot be guaranteed.

SUBMIT AN ARTICLE If you would like to submit an article to be published in Collision please email us your article for review or call us toll free 800-280-7940. Email: sbaker@collisionpublishing.com

ADVERTISING RATES & INFO If you would like to advertise your products or services in Collision Magazine, rates, availability and other information are available online at: www.collisionmagazine.com

4N6XPRT Systems 4N6XPRT Systems ARAS 360 ARC Network ARC-CSI CDR User's Summit Collision Publishing Collision Safety Institute Crash Data Group Engineering Dynamics FARO FARO Laser Technology Inc MapScenes MapScenes Vericom Computers Visual Statement Visual Statement

PAGE 118 Inside Back 1 44 59 23 116 105 45 Inside Front 24-25 54 40 2 55-58 41 78-79 Back Cover

Included with this issue of Collision

is the 2014 ARC-CSI Crash Conference data DVD. The ARC-CSI DVD includes all the photos, video, instrumented crash testing data, scaled diagrams, speaker PowerPoints, and any other supporting data compiled in a easy to review, menu-driven data DVD.

Next ARC-CSI Crash Conference June 1-4 2015 www.arccsi.com

www.collisionpublishing.com

Collision Magazine - Volume 9 Issue 2 5

Premier Industry Partner Letter from the President

Continual research, experimentation, and exchange of ideas through discussion and debate are paths of development of a professional body of knowledge about all levels of technical accident reconstruction.

Substantial and purposeful academic study is a pre-requisite for acquiring, understanding and adding to the body of knowledge of professional accident reconstruction.

Maintenance of the highest standards of ethics and integrity is imperative to the improvement of professional accident reconstruction.

To inform and educate the public on accident investigation and reconstruction issues.

It is hard to believe that nine years have passed since NAPARS began an Industry Partnership with a brand new publication - Collision Magazine. Congratulations to Scott Baker and the team at Collision Publishing for developing their original idea for a cutting edge crash publication into the quality product we see today.

2015 begins a new chapter for NAPARS. By now the end of year elections are complete and a new Board of Directors has been chosen to lead us for the next two years. Congratulations to both the old and new leadership and thank you for your support for the principles that guide NAPARS:

NAPARS would also like to publicly thank our outgoing Treasurer Chip Johnston for his many years of service not only to NAPARS but the crash reconstruction community. Chip, may you enjoy your retirement as much as we have enjoyed your support, loyalty, and friendship.

As always, if you have any questions or suggestions for NAPARS, you can email us at President@NAPARS.org.

Chuck Veppert, President NAPARS

The National Association of Professional Accident Reconstruction Specialists is a non-profit organization whose members have joined together to share the challenge of dealing with the complex problems of accident reconstruction and to upgrade and ultimately professionalize the accident reconstruction field.

6 Collision Magazine - Volume 9 Issue 2

www.collisionpublishing.com

www.napars.org

Industry Partners IAARS

is one of the oldest associations dedicated to expanding the knowledge and training of automobile crash Reconstructionists. IAARS was started in 1980 as a resource for members to use when the inevitable questions arose while trying to reconstruct a crash. We have members from several different countries as well as many different states across the U.S. The annual seminars not only provide instruction on specific topics by experts in the field, they provide face-to-face discussions on how other people approach reconstructing a crash. The old saying "two heads are better than one" is definitely true in accident reconstruction.

Membership in IAARS has other advantages, including a discount (from IAARS) on a subscription to Collision Magazine; access to electronic delivery of the quarterly publication of the Society of Accident Reconstructionists, and electronic delivery of bi-monthly editions of Accident Reconstruction Journal. Members also receive discounts at our training seminars.

We have annual conferences with training opportunities across the country. IAARS was the first association to due actual crash testing in 1982 with T-Bone and Head-on collisions and we continue to try and have a dynamic full scale vehicle crash or testing at each of our seminars. As a member you have full access to the setup and test with all of its data. As a member you have We have three levels of membership. Our highest level, Felaccess to a vast archive of videos and data from many different low, includes a peer review of a reconstruction project. This and sometimes unique tests. Some of these tests include vehicle differs from many other associations in that with this process for membership it can help the member as peer review is a spe- to vehicle, truck to vehicle, motorcycle to vehicle and vehicle to pedestrian crashes, semi rollovers, tip-over testing and brake cific area of judicial consideration when admitting an expert opinion in court. This review, along with training and experi- efficiency testing. IAARS sometimes partners with other associaence, go a long way if you happen to come across a stickler for tions when providing seminars or conferences. We partnered a judge. We offer opportunities for a member to publish work with WREX2000 at Texas A&M in the year 2000 and we are working with other associations for a large seminar in Florida or research in the reconstruction field and an opportunity to in 2016. IAARS will be partnering with SCARS for this years present materials at our annual conferences. conference in Charleston South Carolina in July of 2015.

Our association has "old-timers" that have worked thousands IAARS enjoys working with all of the associations and organizaof accidents. If you have a question, chances are they have tions that are involved in helping get answers to how and why encountered a similar situation in the past and can give some insight on how they worked it out. We also have members very collisions occur and how to make our roads and highways safer. familiar with the ins and outs of new technology (Crash Data Hope to see you at one of the conferences.

Retrieval, Photogrammetry) and are willing to help with any Sincerely, Fred Rice, President, IAARS www.iaars.com questions or concerns you may have.

The ARC Network

was founded in 1998 and since has grown to be the largest and most respected organization for Accident Reconstruction and Traffic Investigation.

The ARC Network is more than just an organization it is a research and communications Internet portal for everything accident reconstruction. The ARC Network provides research tools, Expert directory, event calendar, industry news feeds, education directlory and a Member's Only area that gives access to various online databases and information.

The ARC Network offers various membership levels to fit your needs and budget. Membership starts at just $89.00 per year.

However, that is where the value of the ARC Network starts.

With your membership you not only get access to everything in the ARC Network, you also receive a, $50 discount to the annual ARC-CSI Crash Conference, a $50 discount to the CDR

User's Summit and a free copy of Greg Russell's Excel Crush Stiffness spreadsheet program. Extend your savings and sign up as a Police Agency member or Platinum member and receive a complimentary subscription to Collision Magazine!

In addition, as a member you have the choice of being listed in the most-visited Expert Directory on the Internet. Our Expert directory boasts a #1 ranking in Google searches (a feat worth much more than the cost of membership)!

Scott Baker President The ARC Network (800) 280-7940 www.accidentreconstruction.com arc@accidentreconstruction.com

For information on becoming a Collision Industry Partner send an email - editor@collisionpublishing.com www.collisionpublishing.com

Collision Magazine - Volume 9 Issue 2 7

r Pee

ed

ew evi

R

Using Inequality Constraints in the Probability Analysis of Individual Motor Vehicle Accidents Scott Kimbrough MRA Forensic Sciences

A

bstract Accident analysis often involves using derivations to convert raw input information into more useful and comprehensible information. For example, analyzing a motor vehicle accident often involves converting raw inputs such as the lengths and directions of tire marks into more immediately meaningful outputs such as the speeds of vehicles at impact. Newton’s laws are often used in such derivations. Abstractly, this process can be thought of as involving three entities, namely: 1) an input space containing information about an accident; 2) a set of derivation procedures; and 3) an output space containing the results of the derivations. Each derivation procedure “maps” points in its input space to corresponding points in its output space; thusly, the terms “derivation procedure” and “mapping” will be used interchangeably.

8 Collision Magazine - Volume 9 Issue 2

www.collisionpublishing.com

Often some of the most important information about an accident can take the form of inequality constraints that place boundaries on things like magnitudes and when and where events happened. Such inequality constraints can apply to input variables, output variables, or both. When the constraints apply to the input variables, they can be applied directly to limit the inputs. However, when the constraints apply to output variables, the derivation must usually be carried out before the constraints can be applied. When inequality constraints are applied to the outputs, they separate the output space into feasible regions and infeasible regions, where the feasible regions of the output space contain those outputs that satisfy the constraints. At the same time, separating the output space into feasible and infeasible regions also implicitly separates the input space into corresponding feasible and infeasible regions, where the feasible regions of the input space contain those inputs that map to feasible outputs. www.collisionpublishing.com

Collision Magazine - Volume 9 Issue 2 9

This paper describes a simple method for determining the feasible regions of the input space corresponding to the feasible regions of the output space. The proposed technique begins by using Monte Carlo Simulation to create what will be called the “mapping matrix”. As the Monte Carlo Simulation goes through its derivation loops, at the end of each loop, a row of data is constructed that contains the particular sampled inputs used in that loop together with the resulting derived outputs; this row of data is then appended as an additional row to a matrix whose existing rows contain the inputs and the corresponding outputs of the previous loops. At the conclusion of the Monte Carlo Simulation, this matrix is called the mapping matrix. Because each set of inputs is stored in the same row of the mapping matrix as the resulting outputs, the mapping matrix can be used to identify those regions in the input space that map to specified regions in the output space; this is a valuable capability, because inequality constraints on the output variables can be used to determine those regions of the output space that are consistent with aspects of the information surrounding an accident. Then, using the mapping matrix, corresponding regions of the input space, which contain the only combinations of inputs that could have led to the accident, can be determined.

Although the method to be described is simple, it opens the door to many useful forms of analysis. The mere act of identifying the feasible regions of the input space can provide significant information about the inputs themselves.

Moreover, once the regions of feasible inputs have been identified, what-if studies can be conducted to determine how the probability of the accident occurring would have been affected by hypothetical changes in the inputs. Additionally, inequalities can be used to define different scenarios and other events so that the relative likelihoods of the different scenarios or other events can be determined.

Also, because the described method builds upon a foundation of Monte Carlo Simulation it provides all the attendant benefits of probability based sensitivity analysis.

I

ntroduction Accident reconstruction can be viewed as the process of converting the raw evidence and information about an accident into more immediately meaningful forms of information. For instance, an accident reconstructionist might convert raw measurements of tire marks and vehicle crush into information about the initial speeds of the vehicles involved. This process of converting raw evidence to other forms of information often utilizes derivation procedures we will refer to as mappings, which can range from simple equations to detailed simulation programs. The properties of mapping will be further discussed below.

10

Collision Magazine - Volume 9 Issue 2

Frequently, an accident reconstructionist uses Newton’s equations of motion to determine the time-distance aspects of an accident. The inputs to the equations of motion typically include positions and velocities together with assigned parameters such as drag factors and reaction times.

Executing the attendant equations of motion produces time-distance trajectories based upon the assigned inputs.

However, there is no assurance that the chosen inputs are uniquely correct. For this reason, it is better to study a range of inputs and to obtain a range of outputs, for example by using Monte Carlo Methods. The application of Monte Carlo Methods in accident reconstruction has been the subject of many papers 1,2,4,5,6,7,8,9,10. One pioneering paper in particular 1 highlighted the utility of applying inequality constraints on outputs with the purpose of identifying the corresponding feasible regions of the input space, but this paper did not cover some important conceptual and operational details required for implementing the proposed methods and properly interpreting the results.

In this paper, a framework is developed that enables a systematic probabilistic analysis of individual accidents. It will be shown that the information surrounding accidents often includes valuable information that can be cast in the form of inequality constraints on the derived outputs. For example, it might be known that a certain vehicle was traveling within a specified range of speeds when it impacted an object; this is the kind of information that can be cast as a pair of inequality constraints, with one specifying a lower bound and the other specifying an upper bound.

The framework that is developed extracts the maximum information from such constraints and offers powerful capabilities for understanding what led to an accident and for quantifying how various factors contributed to the probability of the accident occurring. Several case studies will be presented to demonstrate these claims.

The paper is tutorial in nature and includes content to help and encourage readers to create their own computer programs to implement the techniques presented. In practice, many accidents to be analyzed will be discovered to possess unique features that necessitate the authorship of custom computer programs for their analysis. However, this will not be a major obstacle for accomplished programmers, because the required set of subroutines is relatively small and many of these subroutines are native to widely used computer languages and programming environments such as spread sheets 5,8.

A

Framework for Probability Analysis of an Individual Accident The information set associated with an accident will typically contain such things as physical evidence,

www.collisionpublishing.com

testimonial evidence, circumstantial evidence, and experience-based knowledge. Various mapping can be applied to this information. For example, consider the mapping defined by the equation

where: v is the speed at the start of a skid left by a vehicle that stopped. g is the gravitational acceleration. µ is the drag factor. x is the length of the skid This mapping is often used in accident reconstruction to determine how fast a car was traveling when the car first started to skid. The inputs the mapping operates on are the length of the skid mark, x, and the estimated drag-factor, µ. As it is typically used, this mapping operates on x and µ and generates a value for v. Often the length of the skid mark is based upon a field measurement and the estimated drag factor is based upon experience and taking into account the type of roadway surface and the type of vehicle involved.

The information about an accident frequently includes conditions which can be formulated as inequality constraints on derived outputs. In the example just presented, it might be known that the initial speed of the vehicle was close to the pace speed on a particular stretch of freeway.

This information can be cast as two inequality constraints on v; one constraint specifying that v must be greater than some lower bound, and the other constraint specifying that v must be less than some upper bound. Taken together, these two constraints define a feasible region in v space, which in turn defines a corresponding feasible region in the input space roperties of Mappings: In this section we will discuss some important properties of mappings. To begin with, mappings have input spaces and output spaces, the dimensions of which are generally the number of input variables and the number of output variables respectively. The set of permissible inputs is called the domain of the mapping and the set of all outputs obtained by mapping the entire domain into the output space is called the range of the mapping.

Frequently, the dimension of the input space and the dimension of the output space are not the same. For example, revisiting the mapping introduced above, namely

v = (2gµx)1/2

0 < µ < µmax and 0 < x , And the range is the 1-dimensional space of v bounded by the constraint

v = (2gµx) 1/2

P

The domain of this mapping is the 2-dimensional space of µ and x, bounded by the constraints

v > 0.

So, this mapping is from a subset of 2-dimensional space (denoted as R2) to a subset of 1-dimensional space (denoted as R1).

Many times it is desired to go backwards through a mapping and determine what values of inputs generated a specific value of the output, in other words, we are seeking to invert the mapping. Returning to our example mapping, v = (2gµx)1/2 attempting to invert it produces the equation µx = v2/2g So, for each value of v we cannot determine unique values of µ and x; all we can determine is that the corresponding values of µ and x lie on a curve defined by µx = v2/2g 0 < µ < µmax and 0 < x A mathematician would say that the mapping is not invertible because it cannot identify a unique point in the input space for each unique point in the output space.

However, even though the mapping is not invertible, it does tell us something about the inputs corresponding to any particular value of the output. In this case the mapping was simple enough that we could algebraically determine a relationship between the inputs corresponding to any specific value of output. However, in cases involving complex mappings, finding a formula for the sought-after relationships can be very difficult, if not impossible. So, in the general case it is usually better to use numerical methods to obtain the desired relationships. That is the subject of the next section.

M

apping Matrix A mapping matrix is a matrix of data where each row of the matrix consists of a set of inputs together with the derived outputs corresponding to those inputs. In this section a simple mapping matrix will be constructed to illustrate the process. Returning to our example mapping,

v = (2gµx) 1/2 www.collisionpublishing.com

Collision Magazine - Volume 9 Issue 2 11

we will make a simple mapping matrix for this. We start by considering only three values of µ and three values of x. Let {µ} = {.2, .3, .4} and {x} = {20, 25, 30} Where the notation { } means a sequence of values. Then the mapping matrix becomes µ 0.2 0.2 0.2 0.3 0.3 0.3 0.4 0.4 0.4

x 20 25 30 20 25 30 20 25 30

v 16.05 17.94 19.66 19.66 21.98 24.07 22.70 25.38 27.80

Figure 1: Typical "pdf" plot

The headings of the columns have been kept separate from the matrix so the matrix can be operated on mathematically. However it is imperative that the headings and the ordering of the columns be known and be kept consistent.

The example mapping matrix is small because the input sequences were coarsely divided. One can imagine that if the ranges of µ and x were more finely divided and their sequences had more elements, then the size of the mapping matrix would grow and the data in it would more accurately approximate the input-output properties of the actual mapping. In practice, in order to achieve workable accuracy, the completed mapping matrix might have millions of rows.

P

robability Density Functions and Mapping Matrices If the inputs to the mapping are sampled from probability density functions (hereafter referred to as "pdf ’s") and a sufficient number of repetitive derivations are made, where at the beginning of each repetition a set of inputs is compiled by sampling from the assigned input pdf ’s, then the resulting mapping matrix will contain valuable information. The process just described is commonly referred to as Monte Carlo Simulation 16,17, which is simply a numerical method for determining the pdf ’s of the output variables of a mapping, given the pdf ’s of the input variables of that mapping.

As a refresher, the pdf of a random variable can be thought of as a finely spaced histogram that indicates the likelihood that the variable of interest lays in any particular interval of its range. A plot of a typical pdf is shown below 17.

12

Collision Magazine - Volume 9 Issue 2

On the left vertical axis, the values of the pdf have been normalized (i.e., scaled) so that the integral of the entire pdf is 1.0. This is done so that the area under the pdf curve between any two values of the random variable of interest is the probability that a sample of that random variable will fall between those two values. Also appearing on the plot is a curve representing the so called “probability function”, which is the integral of the pdf. Its value appears on the right vertical axis. The probability function indicates the probability that the variable of interest lies to the left of any particular value. The probability function is useful for determining the area under the curve between any two values. For example, in reference to the plot above, the probability that the random variable is between 62 mph and 65 mph is the value of the probability function at 65 mph minus the value of the probability function at 62 mph, or 0.55 - 0.08 = 0.47 = 47%.

There are two dominant types of pdf ’s used to model inputs, namely a uniform pdf and a Gaussian (a.k.a., normal or bellshaped) pdf. The plot of a uniform pdf looks like a horizontal line between the lower and upper limits of the variable of interest. The plot of a Gaussian pdf looks like the profile of a bell. One typically assigns a uniform pdf to a variable about which little prior information is known, but for which upper and lower bounds can be set. One typically assigns a Gaussian pdf to a variable whose values are known to be likely centered around a particular value, which is the mean value. The spread of a Gaussian variable can be expressed in terms of a deviation, where 68% of the samples of the random variable will fall within plus or minus one deviation of the mean.

Note: with a true Gaussian pdf the range of the variable is theoretically infinite in each direction from the mean, even though the probability of distant values is exceedingly small.

www.collisionpublishing.com

However, since Monte Carlo simulations can involve million of iterations, it is possible that some distant values could be sampled, which can sometimes create numerical problems if the sampled value has an illogical sign or is close to zero.

Therefore, the author’s bellshape subroutine has been programmed to discard any sampled values which are over 2.5 times the deviation away from the mean, which resulting in discarding only about 1% of the samples.

Numerically, a pdf can be thought of as a 2-row matrix, with one row being the midpoints of small intervals on the horizontal axis that cover the range of the random variable of interest, and the other row holding the corresponding values of the density at those midpoints. This construct is similar to the plotting data for a histogram. A histogram subroutine generally needs two inputs, the first being the interval size for the bins on the horizontal axis and the second being a sequence of values to be sorted out and placed in the appropriate bins. A histogram subroutine simply tallies the number of values of the variable of interest that fall within the bounds of each and every bin of the horizontal axis. To prepare for plotting, a histogram subroutine creates two vectors, one containing the individual bin counts (i.e., the density), and the other one containing the midpoints of the corresponding bins. Once the vector of bin counts is normalized by dividing each bin count by the total of all bin counts times the width of the bins, then the two vectors become a numerical representation of a pdf. Logically, the smaller the width of the bins (and equivalently the greater the number of bins) and the larger the number of values to be sorted into the bins, the more accurately the resulting histogram plot data will approximates the probability density function of interest.

It is useful to keep in mind the essential equivalence between a finely-spaced normalized histogram and a pdf. A thorough understanding of this equivalence will aid in the understanding of many aspects of the probability analysis presented in this paper.

Once the inputs to the mapping have been assigned pdf ’s, each successive row of the mapping matrix is generated by assembling a set of inputs obtained by sampling each input from its respective pdf and operating on this assembled set of inputs to obtain a corresponding set of derived outputs.

Once the inputs have been sampled and the derivations performed, then a row is added to the mapping matrix that consists of the assembled inputs together with the corresponding derived outputs. Henceforth, the resulting mapping matrix will be referred to as a pdf mapping matrix.

Random number generators that are shaped to particular pdf ’s are used for sampling the pdf ’s. Most advanced programming languages provide such subroutines 8. In addi-

tion, there are many texts that describe numerical methods for sampling from an arbitrary pdf 14,16,17.

If the rows of the mapping matrix are generated repetitively and sequentially a sufficient number of times by assembling a set of inputs sampled from their respective pdf ’s, then the data in the columns of the resulting pdf mapping matrix can be converted to normalized histograms that closely approximate the pdf ’s of the respective variables. Since the pdf ’s of the inputs were already known, the fact that their columns can be transformed to approximate pdf ’s is unimpressive (but a good check on the performance of the computer program). However the columns representing the derived output variables can be transformed to obtain the previously unknown pdf ’s of the outputs.

In practice it is not difficult to determine when a sufficient number of loops have been executed. One begins by performing the prospective analysis at some initial number of loops. Then, the analysis process is repeated with a greater number of loops. If the answers from the two analyses are very similar, enough loops probably have been used, otherwise the number of loops should be increased until the answers from subsequent analysis converge. There are also analytic equations that can be used to estimate the appropriate number of loops 8.

R

egion Inversion Using a pdf Mapping Matrix Region inversion means using a pdf mapping matrix to determine the regions of the input space that mapped to specified regions of the output space.

This process of region inversion is accomplished by applying simple sorting and selection operations on the rows of the pdf mapping matrix. The process begins by identifying a region of interest in the output space by applying one or more inequality constraints to the columns of the pdf mapping matrix containing the outputs to be constrained, and then selecting out those rows that contain output values satisfying the inequality constraints. Then, the columns of the reduced pdf mapping matrix can be processed (via a finely-spaced normalized histogram subroutine) to obtain numerical representations of the pdf ’s of the feasible outputs and inputs, which incorporate the information provided by the constraints.

Example 1: In this section an example problem is presented. The goal of the example is to evaluate the driver’s reaction time. This example will illustrate the valuable information that can be obtained by identifying the feasible region of the input space. In this example, little is initially known about the driver’s reaction time, but once the feasible region of reaction times is identified by applying inequality constraints

www.collisionpublishing.com

Collision Magazine - Volume 9 Issue 2 13

Figure 2: Scene diagram on the impact speed and performing region inversion, much more becomes known. A relatively simple problem has been chosen for the example so that it can be analyzed both by region inversion and by direct application

by Monte Carlo Simulation. The example is first solved by region inversion and then by direct application of Monte Carlo Simulation, after which the solutions are compared.

Suppose a SUV stalls in the roadway. Further suppose that an oncoming truck crests a hill and comes upon the stalled SUV in its lane. The truck tries to stop, but unfortunately hits the back of the stalled SUV. The diagram below illustrates the situation.

The equations of motion to be used for mapping are: v(t) = v0 + µg(t-τ)(t>τ) ; notice the Boolean term (t>τ) x(t) = d – v0t - .5µg(t – τ)2(t>τ) ; notice the Boolean term (t>τ) both valid for t < tstop, where tstop = τ + v0/µg ti2 - 2titstop + (τ2 - 2d/µg) = 0 ; equation to determine the time of impact vi = v(ti ) = v0 + µg(ti-τ)(ti>τ) ; equation to determine the speed at impact both valid for ti < tstop where: d is the detection distance at which the oncoming driver detects the stalled vehicle. v0 is the initial speed of the oncoming vehicle. µ is the drag factor for the stopping rate of the oncoming vehicle. g is the acceleration of gravity.

tstop is the time required for the oncoming truck to come to a complete stop (assuming it hit nothing). ti is the time of impact if impact occurs.

In this example, initial pdf ’s were assigned to each of the input variables required to calculate the impact speed of the truck. The input variables were v0, d, µ, and τ, and the output variables were ti and vi. At the beginning of each derivation cycle, each input variable was sampled from its respective pdf. Once the pdf mapping matrix was obtained, a pair of inequality constraints on the magnitude of the impact velocity vi were applied to determine which rows of the pdf mapping matrix contained feasible values of vi satisfying both inequality constraints. At that point a reduce matrix consisting of only those rows of the pdf mapping matrix containing feasible values of vi was created.

The information set regarding the subject accident contained the following: 1. Based upon the speed limit, witness testimony, and the driver’s habits, the truck was believed to be traveling about 65 mph when it crested the hill. Accordingly, the pdf of the initial speed of the truck, v0, was modeled as bellshaped with a mean value of 65 mph and a deviation of 2 mph. The assigned deviation was based upon knowledge of how tightly the speed of a vehicle is commonly held. Recall that 68% of samples will fall within plus or minus 1 deviation of mean.

2. Based upon measurements, the crest of the hill was about 450 feet from where the impact occurred.

Accordingly, the pdf of the detection distance, d, was modeled as bellshaped with a mean value of 450 feet and a deviation of 15 feet. The assigned deviation is based upon the inherent accuracy of the measurement that was taken.

τ is the reaction time of the driver of the oncoming vehicle.

14

Collision Magazine - Volume 9 Issue 2

www.collisionpublishing.com

3. Based upon published test data, the drag factor for such a truck would be expected to be about -0.65.

Accordingly, the pdf of the drag factor, µ, was modeled as bellshaped with a mean of -0.65 and a deviation of 0.03. The assigned deviation was based upon the normal variance of drag factors.

4. Since little is known about the reaction time of the driver of the truck, its pdf was modeled as uniformly distributed over the range of 0.5 to 4 seconds. Once the region inversion is performed this pdf will be reshaped and contain much more information.

5. Based upon crush analysis it was estimated that truck was traveling about 18 mph to 22 mph at impact. This will be used to formulate two inequality constraints on the output variable vi.

The basic steps required to create the desired pdf mapping matrix are outlined below.

Step 1: Sample v0, d, µ, τ Step 2: Calculate tstop = τ - v0/µg ; recall that µ is negative.

Step 3: Calculate xstop = τv0 - 0.5µg(tstop – τ)2 Step 4: Check is ( xstop < d ) ? ; asking the question “will the vehicle stop before impact occurs?”

If so, then set vi = 0 , ti = 10 ; in theory ti is infinite, but any relatively large number will suffice to mark this instance Go to Step 8.

Step 5: Calculate tmin = d/v0 Step 6: Check is ( τ < tmin ) ? ; asking the question “does the driver react before impact occurs?”

If not, then set vi = v0 , ti = tmin ; this because the vehicle will impact at full speed.

Go to Step 8.

Step 7: Solve the quadratic equation ti2 - 2ti tstop + (τ2 - 2d/µg) = 0 which, via the quadratic formula has the solution ti = tstop – ((tstop2) - (τ2 - 2d/µg))1/2 Step 8: Store

in proper order as row of pdf mapping table Step 9: Have sufficient iterations been performed?

If not, then go to Step 1 Step 10: Determine those rows of the pdf mapping table that contain vi values in the range of 18 mph to 22 mph.

Identify these rows as the feasible rows. Create a reduced matrix consisting of only the feasible rows of the pdf mapping matrix; this reduced matrix will be referred to as the feasible pdf mapping matrix.

Step 11: Create pdf ’s using the values in the columns of the feasible pdf mapping matrix using a subroutine that creates a finely-spaced normalized histogram.

Results for Example 1: Recall that v0 , d, µ, τ were sampled from their respective assigned pdf ’s. Therefore, if things have gone properly, the pdf ’s created from the data in the columns of the full matrix should closely resemble the pdf ’s that were assigned to the respective variables. After 1 million iterations (which created a pdf mapping matrix with 1 million rows), the program produced these plots for v0 , d, µ, and τ. (see figure 3).

It can be verified that these plots match the assigned input pdf ’s. After producing the check plots shown above, the program then produced pdf plots for both vi and ti , using the full matrix. Recall, these variables were the derived output variables. At this point the inequality constraints on vi had not yet been applied yet and all the rows of the full matrix were used.

At first, these last two plots might appear strange. For example, the density function for vi has a spike at 0 mph and its probability function starts out at about 0.55 instead of 0; the reason for this is that many times in the derivations, based upon the particular set of input variables sampled for that derivation, the vehicle would stop short of collision, about 55% of the time which is why the probability for vi starts at 0.55. As another example, the density function for ti has a spike at 10 seconds and its probability function jumps from 0.45 to 1.0 at that value; the reason for this is because in step 8 we arbitrarily used a value of 10 seconds to signify that no collision would occur. Theoretically, if the vehicle stops before collision, then the time until impact is essentially infinity, but it is sufficient to just use a large enough number that it is separated from any legitimate impact time.

v0 , d, µ, τ, ti , vi www.collisionpublishing.com

Collision Magazine - Volume 9 Issue 2 15

Figure 3: Plots for v0, d, µ, and t

Figure 4: "Pdf" plots for both vi and ti

16

Collision Magazine - Volume 9 Issue 2

www.collisionpublishing.com

After producing the check plots, the program then sorted through the rows of the full pdf mapping matrix to find those rows in which the calculated value of vi was within the constrained range of 18 mph to 22 mph. Retaining only these rows, the program created the feasible pdf mapping matrix. The program then created numerical representations of the “feasible” pdfs by operating on the columns of the feasible pdf mapping matrix. The plots shown below display the feasible pdf ’s so obtained. (see figure 5) Obtaining the last plot shown, for τ (i.e., tau), was the stated goal of the example. Recall that in the beginning there was no solid information concerning the driver’s reaction time and it was assigned a uniform pdf between 0.5 and 4 seconds. However, after applying the inequality constraints, the resulting feasible pdf for the driver’s reaction time is far more informative. It indicates, among other things: 1) the probability that his reaction time lies in the interval of 2.25 to 3.0 seconds is about 80%; 2) the mean value of his reaction time is about 2.6 seconds; 3) the most likely reaction time is about 2.6 seconds; and 4) the spread in reaction time is about 1.8 seconds.

Validation of Results for Example 1: By design, the example problem was simple enough that the pdf of the driver’s reaction time could be obtained directly from a Monte Carlo Simulation, wherein the reaction time, τ, is now an output and the impact velocity vi is now an input. The pertinent equations are: δx = -(v02 – vi2)/2gµ and τ = (d – δx)/v0 where δx is the deceleration distance over which the oncoming vehicle braked. The sampled input variables in the equations above are v0 , vi , µ, and d. In the general case, this approach also involves an inequality constraint on an output, namely that the derived reaction time, τ, must be greater than 0 (or more precisely the quickest time possible) to be feasible. However, this constraint will automatically be satisfied over the range of parameters involved.

The following pdf ’s were assigned to the input variables: v0 - bellshaped distribution with a mean of 65 mph and a deviation of 2 mph. vi - uniform distribution ranging from 18 mph to 22 mph. µ - bellshaped distribution with a mean of -0.65 and a deviation of 0.03. d - bellshaped distribution with a mean of 450 ft and a variation of 15 ft.

“

After applying the inequality constraints, the resulting feasible pdf for the driver’s reaction time is far more informative.

”

The goal was to obtain a pdf for τ from this direct approach and to compare it to the pdf for τ obtained in the example above. This was done and the two plots below compare the pdf of τ obtained from the direct method to the pdf of τ obtained from the region inversion method. (see figure 6) As one can see, the two plots are essentially identical. One might question the value in using region inversion if the desired information can be obtained directly. The answer is that many times there will not be an easily implemented direct way.

P

erforming What-If Analysis: This section shows how to conduct a what-if study concerning the driver’s reaction time. The study begins with the feasible pdf mapping matrix obtained from the original example problem above, which is so because it is the columns of the feasible pdf mapping matrix that contain the combinations of inputs that are feasible and could have led to the accident. Accordingly, the whatif answers the question, given the feasible regions of inputs (i.e., the probabilistic equivalent of the conditions that led to the accident), what output changes would occur if some aspect of the inputs was changed.

In the original example above, the feasible pdf of the driver’s reaction time was found to have a mean value near 2.6 seconds, which is a little slow compared to norms in similar situations involving immediately apparent hazards.

This section determines what would have happen in a

www.collisionpublishing.com

Collision Magazine - Volume 9 Issue 2 17

Figure 5: Numerical representations of the “feasible” “pdfs” by operating on the columns of the feasible pdf mapping matrix

18

Collision Magazine - Volume 9 Issue 2

www.collisionpublishing.com

Figure 6: Comparison of the “pdf ” of τ obtained from the direct method to the “pdf ” of τ obtained from the region inversion method probabilistic sense if the driver’s average reaction time was 0.25 seconds faster.

Before embarking on a solution, an important distinction needs to be made. In the example problem above we were able to sample the input variables (namely v0 , d, µ, and τ) individually and independently from their respective pdf ’s.

This was possible because the input variables were logically independent of each other. For instance, it was logical to assume that the initial speed, v0 , was not dependent upon or affected by the distance that the stalled vehicle could be seen, d, or the friction factor, µ, or the driver’s reaction time, τ. And likewise, none of the other inputs were dependent on any others. Using samples from these independent pdf ’s as inputs, we were able to use Monte Carlo Simulation to calculate the outputs. After that, by identifying and keeping only the outputs consistent with the constraints of the accident, we obtained the smaller feasible pdf mapping matrix. However, something else has occurred; whereas the columns of the full pdf map-

ping matrix associated with these input variables contained independent and uncorrelated values (since the values were sampled from independent pdf ’s), the same columns of the reduced feasible pdf mapping matrix now contain values which are typically correlated to the values in the other columns. Therefore the columns (and the pdf ’s they represent) are no longer independent, and taken all together, the columns constitute what is called a multidimensional joint pdf. Here is what is important about that: if correlations exist between two or more input variables, then once one or more of the input variables is sampled, then any subsequent input variables must be sampled from what are called conditional pdf ’s, which can be difficult to obtain.

Fortunately, there is a way to avoid this difficulty; it depends on the recognition that, taken together, the input values in each row of the feasible pdf mapping matrix represent a single multidimensional sample from the multidimensional joint pdf. Therefore, taken together, the input values in any row of the feasible pdf mapping matrix can be used as a sample from the multidimensional joint pdf.

This approach was employed in this example what-if study.

At at the start of each calculation loop of the attendant Monte Carlo Simulation, the inputs were all sampled simultaneously using the set of inputs found in the rows of the feasible pdf mapping matrix as they were sequentially processed. In other words, instead of individually sampling the inputs from separate pdf ’s, as in the first example, in this case, all of the inputs are sampled simultaneously according to the values found in individual rows of the feasible pdf mapping matrix, as the rows are sequentially processed.

Having obtained the feasible pdf mapping matrix , the example what-if began by subtracting 0.25 seconds from each value in the column of the feasible pdf mapping matrix corresponding to the reaction time. Then, another Monte Carlo Simulation was conducted where at the start of every calculation loop, the next row of the feasible pdf mapping matrix was selected and its input values were treated as a single sample from the multidimensional joint pdf of the inputs. All of the inputs in each calculation loop are simultaneously assigned the corresponding values of the selected row.

Because the driver was now reacting on average 0.25 seconds faster than before, logically there will be fewer calculated collisions, because there will be more instances in which the oncoming vehicle will be able to stop before impact. Indeed, the number of times it is calculated that the oncoming truck would have been able to stop before a collision under the new conditions determines the probability that the accident would have been avoided.

www.collisionpublishing.com

Collision Magazine - Volume 9 Issue 2 19

The results that follow were obtained by performing 1,000,000 loops, which generated a full pdf mapping matrix with 1,000,000 rows. The inequality constraints on impact velocity were then applied and afterwards the feasible pdf mapping matrix had only about 25,000 rows.

Then, after shifting the driver’s reaction times in the appropriate column of feasible pdf mapping matrix by 0.25 seconds, and running the loop calculations again, but using the input sections of the rows of the feasible pdf mapping matrix as input samples, it was discovered that about 85% of the time the oncoming vehicle would now be able to stop before impact. Thusly, if the driver had reacted on average 0.25 seconds quicker than he did, the probability of the accident occurring could have been reduced by 85%.

In addition, if an impact still did occur it would be at a lower velocity. The plots below compare the impact velocity pdf from the original problem (above) to the impact velocity pdf from the what-if problem in which the driver reacted 0.25 seconds faster (below). According to these plots, the average impact speed has fallen from about 20 mph to about 4 mph.

So, if the driver reacted 0.25 seconds faster, not only would the probability of the collision occurring have been cut by about 85%, but the pdf of the impact velocity would have been shifted lower.

Example 2: This second example is more realistic and shows how the proposed technique can be applied in more complex situations. Consider an incident in which a straight chassis truck changes lanes without the driver looking carefully and impacts a car that was next to it. Suppose the witnesses stated that the car was traveling slightly faster than the truck and was back near the tailgate of the truck when the truck started to move over. Furthermore, suppose the witnesses stated that the driver of the car started to brake once he realized what was happening, but in spite of that, the front of his car still advanced to the vicinity of the front of the cargo box of the truck before starting to fall back.

Unfortunately, before the car could fall completely back, it was struck by the tailgate of the truck. The diagram below shows the basic situation. (see figure 8) The information provided by the witnesses involves the position and speed of the car relative to the truck at different times. Therefore, to apply this information we must derive the relative positions of the vehicles as a function of time. It turns out that the following 3 equations and 1 definition can be used to derive the longitudinal motion of the two vehicles relative to each other. tstop = tr + (vc + δvc )/µg tlessor is the smaller of t and tstop xc(t) = x0 + (vc + δvc ) tlessor + [0.5(µg)(tlessor – tr )2 (t > tr )] ; valid for 0< t<ti Notice the Boolean operator (t > tr ) which is either 1 if t > tr or 0 otherwise. xt(t) = tv0; where, tstop is the time required for the car to come to a complete stop, if it does so. xc(t) is the longitudinal position of the front of the car as a function of time t. xt(t) is the longitudinal position of the tailgate of the truck as a function of time t.

Figure 7: Comparison of impact velocities 20

Collision Magazine - Volume 9 Issue 2

x0 is the initial position of the front of the car relative to the tailgate of the truck.

www.collisionpublishing.com

Figure 8: Scene diagram v0 is the initial velocity of the truck.

4 feet > absolute value of max ( xc(t) – xt(t) – 25 feet)

δvc is the initial difference in speed between the car and the truck. µ is the spatial average deceleration rate of the car once it begins to brake. tr is the reaction time of the driver of the car. ti is the time that impact occurs.

By subtracting the second equation from the first, we can obtain an equation for the position of the front of the car relative to the tailgate of the truck, namely xc(t) - xt(t) = x0 + (vc + δvc ) tlessor + [0.5(µg)(tlessor – tr )2 (tlessor > tr )] - tv0 ; valid for 0< t<ti Since the witness statements describe how far beyond the tailgate of the truck the car advanced before braking and then falling back, we will need to determine the maximum value of xc(t) – xt(t) over the time interval 0< t < ti. This can be achieved by evaluating the equation above over a sequence of closely spaced times spanning the interval 0< t < ti, and then searching for the maximum difference of xc(t) – xt(t) over that sequence of times. This is an example of a mapping that is a combination of an equation and a numerical operation. The witnesses also described how the car was very near the back of the truck when the impact occurred. This mapping is just the evaluation of xc(t) – xt(t) at t = ti. This is an example of a mapping that is a simple equation.

Witness statements were used to formulate inequality constraints on the outputs xc(t) and xt(t). The first constraint follows from the testimony that the car advanced up to near the front of the box before falling back. Accordingly, the maximum value of xc(t) – xt(t) is within an estimated 4 feet of the length of the box on the truck, which is 25 feet, or

The second constraint follows from the testimony that the impact occurred somewhere near the tailgate of the truck, 0< xc(ti ) – xt(ti ) < 4 feet.

The third constraint follows from the reality that the driver of the car had to react before the impact occurred, or else he could not have gained on the truck and then fallen back. tr < ti.

The pdf ’s assigned to the input variables were: v0 - bellshaped with mean of 40 mph and deviation of 1.5 mph. δvc - bellshaped with mean of 10 mph and deviation of 1 mph. x0 - uniform from 0 feet to 5 feet. ti - bellshaped with mean of 5 seconds and deviation of 1 second. tr - uniform from 1 second to 5 seconds. µ - bellshape with mean of -0.65 and deviation of 0.05.

The plots shown below compare the initially assumed pdf for the driver’s reaction time with the refined pdf derived from the values in the column of the feasible pdf mapping matrix corresponding to the driver’s reaction time. (see figure 9) As expected, after identifying the feasible regions of the input space, a much more informative pdf has been obtained. As examples of what has been learned, it is now known that there is a 50% probability that the driver responded in less than 1.3 seconds and that there is a 90% probability that the driver reacted in less than 1.6 seconds.

www.collisionpublishing.com

Collision Magazine - Volume 9 Issue 2 21

Figure 9: "Pdf" for the driver’s reaction time

C

onclusions: Frequently, some of the most valuable information related to an accident can be cast in the form of inequality constraints on variables that must be derived (mapped) from more basic forms of evidence. The mappings can range from simple equations to complex computations. Once appropriate mappings were formulated, Monte Carlo Simulation can be used to obtain a pdf mapping matrix. From there, the applicable inequality constraints can be used to select out those rows of this matrix that contain feasible values consistent with the applied constraints (and even more importantly, with the information used to formulate the constraints). The product of the selection process is a feasible pdf mapping matrix containing only the feasible rows of the pdf mapping matrix.

The columns of the feasible pdf mapping matrix associated with input variables contain samples from the joint pdf of the input variables, and the columns associated with output variables contain samples from the joint pdf of the output variables. Compared to the original pdf ’s found in the entire pdf mapping matrix, the resulting multidimensional joint pdf ’s found in the feasible pdf mapping matrix contain more information since they incorporate the information conveyed by the constraints. It was shown that the multidimensional joint pdf of the inputs is a natural starting point for what-if studies that seek to calculate the effects of hypothetical changes on the probabilities associated with the accident, since this joint pdf is the probabilistic equivalent of the initial conditions that led to the accident.

The author has used these methods extensively and has found that it usually takes custom programming to solve real life problems. However, the number of subroutines required to write custom programs is relatively small and the necessary subroutines can be found in many program22

Collision Magazine - Volume 9 Issue 2

ming languages and environments 8 or can written with reasonable effort. As shown in the examples, a major task is formulating the appropriate mapping equations to convert the information available regarding an accident to the outputs which meaningful inequality constraints can be applied to. However, these mappings can often be derived manually from basic equations of dynamics.

In order to use Monte Carlo Simulation it is vitally important to understand the differences between marginal, conditional, and independent pdf ’s 11,16,17. This is not a difficult task, but it is essential for the proper formulation Monte Carlo Simulations and proper interpretation of the results.

Acknowledgements: The author thanks the reviewers for the many useful suggestions made.

R

eferences: 1. Wood, D. and O’Riordain, S., “Monte Carlo Simulation Methods Applied to Accident Reconstruction and Avoidance Analysis”, SAE 940720.

2. Kost, G. and Werner, S., “Use of Monte Carlo Simulation Techniques in Accident Reconstruction”, SAE 940719.

3. Brach, R., “Uncertainty in Accident Reconstruction Calculations”, SAE 940722 4. Daily J., “Monte Carlo Techniques for Correlated Variables in Crash Reconstruction”, SAE 2009-010104.

5. Ball, Danaher, and Ziernicki, “Considerations for Applying and Interpreting Monte Carlo Simulation Analyses in Accident Reconstruction”, SAE 2007-010741

www.collisionpublishing.com

6. Kimbrough, S., “Probability Analysis of Motor Vehicle Accidents”, ASME IMEC2002-32957 7. Kimbrough, S., “Determining the Relative Likelihoods of Competing Scenarios of Events Leading to an Accident”, SAE 2004-01-122 8. Bartlett, W., “Conducting Monte Carlo Analysis with Spreadsheet Programs”, SAE 2003-01-0487 9. Moser, A., et. al., “Application of the Monte Carlo Methods for Stability Analysis within the Accident Reconstruction Software PC-CRASH”, SAE 200301-0488

12. Baker, J. and Fricke, L., Traffic Accident Investigation Manual, Northwestern University Traffic Institute, ISBN 0-912642-06-8 13. Limpert, R., Motor Vehicle Accident Reconstruction and Cause Analysis, Lexis Publishing, 1999, ISBN 0-327-04974-X 14. Press et al., Numerical Recipes, Cambridge Press, 1986, ISBN 0 521 30811 9 15. Olsen, P., Forensics Aspects of Driver Perception and Response, Lawyers and Judges Publishing Company, 1996, ISBN 0-913875-22-8

10. Kimbrough, S., “Monte Carlo Analysis of Safe Following Distances under Different Road Conditions”, ASME IMEC2003-42865

16. Rubinstein and Kroese, Simulation and the Monte Carlo Method, Wiley 2008, ISBN978-0-470-177945

11. Papoulis, A., Probability, Random Variables, and Stochastic Processes, McGraw Hill, 1965, ISBN 07048448-1

17. Shonkwiler and Mendivil, Explorations in Monte Carlo Methods, Springer 2009, ISBN 978-0-38787836-2

r Pee

ed

R

ew evi

heel Slip David M. Hallman, M.S., P.E.

A

bstract On ice, on wet roads, in the air ... in the right conditions, wheels can slip and when they do there is the potential for a reported speed (whether on the speedometer or in EDR related data) to be different than the actual vehicle speed at the time the information was recorded. This article will look at the sources of reported vehicle speed such as vehicle speed sensors and wheel speed sensors and some of the conditions which might affect the reported speed. Based on prior testing performed coupled with testing done at the 2014 ARC-CSI Crash Conference, this article will explore the causes and implications of wheel slip and some of the steps a reconstructionist might take to account for the phenomenon where appropriate.

Also discussed here are recently added (by the manufacturer) data recording and storage features specific to some Toyota vehicles which were discovered during vehicle testing.

I

ntroduction Crash investigators have many different sources of information regarding what occurred in the moments leading up to and immediately following a crash. One of those sources is the retrievable data stored in the airbag control module (ACM) in a steadily increasing number of vehicles on the road. Crash data has gained wide acceptance among experts, attorneys and the court system 9 as being an accurate representation of certain aspects of vehicle performance and operator behavior in the moments leading up to a crash as well as containing crash and postcrash information. It is important to understand that the stored ACM data reflects a particular sensor output and interpretation of that output at the time it was recorded, but may not accurately represent the behavior of the physical vehicle in all cases.

26 Collision Magazine - Volume 9 Issue 2

As defined by the Code of Federal Regulations (CFR) Title 49, Part 563, vehicle speed in the stored crash data is required to be the indicated vehicle speed. Depending on the year, make and model of the vehicle, indicated vehicle speed could be supplied by a speed sensor in the transmission or wheel speed sensor(s). It could be reported from those sensors to the ACM via the engine controller, instrument panel controller, body controller, or a variety of other modules on the vehicle. If a wheel or wheels are spinning, slipping or locked prior to, during, or immediately after a crash the speed information may not be directly indicative of actual vehicle speed. The Data Limitations text from most Bosch CDR reports states “SDM (ACM) Recorded Vehicle Speed accuracy can be affected by various factors, including but not limited to the following…wheel lockup and wheel slip" (or words to that effect).

An example of crash data collected where ACM reported speed was suspect is shown in Figure 1. In the case of this GM non-deployment data, vehicle indicated speed is shown between 103 miles per hour and 106 miles per hour in the four seconds leading up to algorithm enable (AE) and 20 miles per hour in the final second with an increase in throttle percentage from 25% to 49%. How is this possible? More importantly, how could this be a nondeployment event with an indicated speed decrease of 83 miles per hour reported from 2 seconds to 1 second?

This data was collected from a run-off-road crash which occurred on a two lane road in rural North Dakota. The vehicle operator first crossed over the oncoming travel lane (which was empty at the time) and drove off of the pavement. After leaving the roadway and driving down into the far side ditch it struck a dirt “ramp” which was put in place to allow a farmer to drive equipment into his field without having to drive down through the steep ditch

www.collisionpublishing.com

along the roadway. This ramp was oriented perpendicular to the paved roadway and the parallel ditch. When the vehicle struck the ramp, the unbelted driver slid forward in the seat resulting in the increased throttle percentage. At the same time, the front wheel rotation slowed dramatically but the vehicle did not, resulting in a very low reported speed with an actual vehicle speed that was significantly higher. After striking the dirt ramp, the vehicle became airborne for approximately 250 feet and then impacted the ground and rolled multiple times end over end, ejecting and killing the unbelted driver.

This article will begin by discussing some reasons why wheels can slip and how wheel slip is controlled by different vehicle systems depending on the type of slip which is occurring. Three different systems are included in modern vehicles for limiting and controlling wheel slip under various conditions. Those systems are antilock braking systems (ABS), traction control systems (TCS) and the recently mandated 4,5 electronic stability control (ESC). These systems, how they operate and federal regulations governing them will also be discussed.

Data which was collected from several different vehicles to examine the effects of wheel slip on reported vehicle speed will be examined.

During testing on a 2013 Toyota Rav4, an interesting new feature specific to some Toyota vehicles was discovered. The 2013 - 2015 Toyota Rav4, the 2014 - 2015 Toyota Highlander and 2014 - 2015 Highlander Hybrid all record data in various modules when

www.collisionpublishing.com

Collision Magazine - Volume 9 Issue 2 27

Figure 1: GM non-deployment crash data certain triggers are met. This data is referred to by Toyota as vehicle history data. The recorded data is not reported as crash data in the ACM, but is collected and stored in various locations depending on the trigger which leads to the data storage. This data is only retrievable using the Toyota TechStream scan tool. Information regarding this new stored data will also be discussed.

W

hy do wheels slip?

Wheels can slip for many reasons, sometimes intentionally, but most times unintentionally.

In the case of a car built for drag racing, the wheels are intentionally spun immediately prior to a race to heat the tires, soften the tires and increase the friction coefficient between the tire and track surface. A higher friction coefficient between the two means more power can be delivered by the tire to the road surface with minimal slippage. The car can then accelerate faster without spinning away available power through slipping the tires during take-off.

All control and acceleration forces must be reacted by the tire-to-road contact patch in order for the car to accelerate and remain controllable by the driver. Once the force that can be reacted by that tire-to-road contact surface is exceeded, the tire will slip. In passenger cars and light trucks, tires can slip due snow, ice, or other friction reducing material on the road surface. Reduced tread on the surface of the tire due to wear can also reduce a tires ability to react load across its contact with the road surface.

Excessive water on the road can cause a phenomena know as hydroplaning where the tire is actually lifted off of the road surface by a film of water. This can cause significantly

28

Collision Magazine - Volume 9 Issue 2

reduced tire-to-road friction and can lead to wheel slip and loss of vehicle control.

In order to properly explain tire-patch-to-road reaction forces, a basic definition of velocity and acceleration is required. Vehicle velocity is a vector quantity and is defined by both a speed and travel direction. Acceleration is defined as a change in velocity over time. This applies not only when accelerating forward (positive), but also applies when a vehicle is accelerating sideways (cornering), or accelerating rearward (slowing, negative). There is finite amount of force which can be reacted by the tire across its contact patch. When a vehicle is traveling in a straight line, most of that available force can be used for either positive or negative acceleration. When cornering is added, some of the total force which was originally available for acceleration (positive or negative) is used to corner the vehicle which leaves less force available for acceleration. This means that under the combination of heavy acceleration (or braking) and cornering, the tire is more likely to slip laterally.

The friction circle (Figure 2) shows a graphical representation of how acceleration can cause tires to slip. When the combination of the forces applied across the tire patch is within the circle (approximately 1.0 G or “gravity” for dry pavement, less for icy or wet conditions), the tire will not slip. If the applied force level moves outside of the circle, the tire begins to slip. One “G” is equal to the acceleration of earth’s gravity (32.2 ft. /sec2 or 9.81 m/sec2), which in the case of tire to road contact is measured in a direction across the tread face instead of the typical vertical direction. A tire can react 100% of its maximum force by turning, braking, accelerating or any combination of

www.collisionpublishing.com

concerns related to controllably stopping a vehicle, wheel lockup and vehicle directional control. The brakes on the vehicle can only slow the rotation of the wheels. It is the tire’s ability to react load across the road surface that stops the vehicle (back to the friction circle). When brakes without ABS are applied with enough force to lock the wheels, the vehicle slides because traction between the tires and the road surface has been lost. While the wheels are sliding, steering control is also lost.

Figure 2: Friction Circle those maneuvers. If the tire is at the maximum force when braking and the driver attempts to add steering input the tire will slip and the vehicle will continue along its original trajectory. If the tire is at its maximum force when turning and a driver attempts to add braking or acceleration, the vehicle will lose turning capacity and slide.

ABS operates by measuring the wheel speeds and electronically modulating brake pressure to the individual wheel brakes to prevent the wheels from locking. Early systems monitored the front wheels individually and the rear differential speed to control the rears together. Current systems monitor each wheel’s speed for better control. As ABS is a supplement to the basic braking system, if an ABS system malfunctions, normal braking will not be affected, the ABS will simply not function and the wheels will be able to lock.

B

asic Brakes, ABS Passenger vehicle brakes have become much more sophisticated over the last 30 years. Passenger vehicles and light trucks were originally equipped with drum brakes on both axles. These drum brakes were directly hydraulically connected to the brake master cylinder which was directly activated by the brake pedal in the vehicle. A vacuum powered brake booster mounted between the brake pedal and the master cylinder was added to this simple system which assisted the operator in applying the brake pedal and allowed the brakes to be applied with more force. Four wheel drum brakes on vehicles were later replaced by a combination system with disc brakes on the front axle and drum brakes on the rear axle.

The disc/drum brake combination evolved to four wheel disc brakes on passenger vehicles and eventually most light trucks followed as well. Medium and heavy trucks and busses can vary from this, but in most cases disc brakes are normally used on steer axles and drum brakes are used in other locations including trailer axles. Many newer passenger vehicles also have TCS, the newly mandated stability control and some collision mitigation systems capable of applying the brakes for the driver in the event of an impending crash.

The first anti-lock braking systems (ABS) were implemented after the introduction of disc brakes. ABS addresses two

Figure 3: Wheel speed sensor Vehicle ABS is integrated into the conventional braking system. A computer controlled pump and actuator unit, installed between the brake master cylinder and the brake calipers, is used to control and regulate brake system hydraulic pressure to the individual wheels. ABS is capable of modulating the brake pressure at a given wheel many times per second and controls each wheel brake individually. Modern ABS Systems use a speed sensor and toothed wheel at each wheel to monitor individual wheels speeds (Figure 3). The individual speed sensors are monitored by a dedicated electronic control module (ECU). These speed sensors are also the heart of the ABS system. Without them, the system would not be able to function.

The wheel speed sensors are a hall-effect type sensor. Halleffect sensors rely on a permanent magnet and a toothed wheel to generate a signal. The toothed wheel is mounted directly to the rotating portion of the wheel hub assembly

www.collisionpublishing.com

Collision Magazine - Volume 9 Issue 2 29

located at each wheel on the vehicle and the “sensor” portion is mounted to the non-rotating portion of the hub assembly. This allows the toothed wheel to rotate at the same speed as the vehicle’s wheel. Each time a tooth on the toothed wheel passes the sensor magnet, a signal pulse is generated. By counting the pulses and knowing how many teeth are on the wheel and the tire’s rolling radius, wheel speed can be calculated and interpreted by the ABS controller. The signal produced by the wheel speed sensor pick-up is an analog signal that is converted into a digital signal for interpretation by the ABS controller.

Wheel speed sensor operation (Figure 4) is similar to the magnetic pick-up in an ignition distributor. When the teeth of the toothed wheel pass the iron core, the magnetic lines of force cut through the coil windings causing a voltage to be induced into the coil. When the tooth is centered on the iron core the magnetic field does not move and zero volts are induced in the coil. As the tooth moves away from the iron core the magnetic field expands, resulting in a negative voltage. As the rotation of the sensor rotor increases, the voltage and the frequency of this signal increase, indicating to the controller a higher wheel speed.

tor potential wheel slip. ABS is not necessarily active during all stops, only those where the brakes would normally lockup. During ABS braking, the controller calculates the slip ratio of each wheel and controls the brake actuator to provide appropriate braking pressure to each wheel’s brake caliper to keep it at an “impending slip.” Impending slip is the point at which the maximum amount of force can be reacted across the tire to road contact without a loss of control or total slippage. The pressure delivered to the brakes will be less on slippery pavement than it would on dry pavement. As a result, braking distance may increase but directional control can be maintained.

According to one NHTSA study,1 when introduced, antilock brake systems had almost no effect on fatal crash involvements. Runoff-road crashes increased significantly after ABS was originally implemented in passenger vehicles. The increase in those types of crashes was offset by reductions in collisions with pedestrians and collisions with other vehicles on wet roads. A second generation of NHTSA analyses around the year 2000 showed a smaller increase in run-off-road crashes after drivers presumably became better educated in the proper use of ABS. Also according to a NHTSA study 1, ABS is effective in nonfatal crashes, reducing the overall crash-involvement rate by 6 percent in passenger cars and by 8 percent in light pickup trucks, SUVs and vans.

When ABS was initially implemented into passenger vehicles, and for many years following, there was no federal requirement for it. With the recent federal mandate for electronic stability control (covered later in this article) on passenger vehicles which requires an ABS system to operate, ABS is now mandated on passenger vehicles as well.

The newly required ABS must comply with Federal Motor Vehicle Safety Standard (FMVSS) 135.

D Figure 4: Hall Effect Operation The wheel speed sensors are constantly monitoring individual wheels speeds at all times while the wheel is rotating.

The ABS controller calculates the speed of each wheel from the supplied wheel speed sensor data and the changes in individual wheel speeds over time. These speeds are compared to each other and reported ground speed to moni30

Collision Magazine - Volume 9 Issue 2

ifferentials With limited exceptions (mostly racing applications), every car has a differential (Figure 5). In order for a car to travel smoothly around a corner, the outer wheels must travel further than the inner wheels and the typical differential allows that to occur. The inboard and outboard wheels of the car must travel along different radius arcs while turning a corner. The further away from the center point of the arc, the larger the distance travelled. The inner wheel travels a smaller arc and the outer travels the larger one. This means the outer wheel needs to rotate more times than the inner one in order to go around a corner. If the outer and inner wheels were connected together with a solid axle, one of them could not turn more times than the other and the driven end of

www.collisionpublishing.com

Figure 5: Typical open differential the vehicle would not corner smoothly. Differentials allow the two wheels on the driven axle to turn at different rates and typically supply some amount of gear reduction as well.

As well as allowing vehicles to turn corners smoothly, the differential acts as a torque-splitter which splits the input torque to two output shafts which can turn at different rates. On a two-wheel-drive car there is only a differential on the driven axle. On four-wheel-drive or all-wheel-drive vehicles both the front and rear axles have differentials because they are both driven. In the case of all-wheel-drive vehicles, there is often a third differential located in the driveline from the front to the rear to allow the front and rear axles to spin at different rates.

There are several different types of differentials in use on vehicles today. The most basic and common type is the open differential. Open differentials have very few internal components as seen in Figure 5. The pinion gear is solidly mounted to the pinion shaft and is driven by the output shaft from the transmission. The pinion, in turn,

drives the ring gear and the combination of the two is what sets the gear reduction for a given differential. Attached to the ring gear is the cage, which contains two captive pinion gears which are meshed with the two output pinion gears. Each output pinion is connected to one axle which drives one wheel. The captive pinions are free to rotate as required. As the input pinion spins and drives the ring gear, the ring gear rotates and the captive cage and the two captive pinions rotate along with it. When travelling in a straight line, both drive pinions rotate at the same rate, so the captive pinions don’t spin and the turning of the ring gear is translated directly to both drive pinions. When the vehicle starts to turn a corner, the outboard wheel must rotate more times and more quickly than the inboard wheel and the captive pinions allow the two drive pinions to spin at slightly different speeds while still transmitting torque to them both.

One downside of the open differential is it cannot compensate for one drive pinion being held solidly compared to the other. This explains why a two-wheel-drive vehicle can easily get stuck when one wheel has less friction with

www.collisionpublishing.com

Collision Magazine - Volume 9 Issue 2 31

the ground than the other. When this occurs, all of the power input gets redirected to the drive pinion with the least resistance. This means all of the power is directed to the wheel which is slipping. This can also be a problem in a four-wheel-drive vehicle with two open differentials. If both wheels on one side or opposing wheels on opposite axles are slipping, all of the power available gets directed to the slipping wheels and the vehicle is stuck.

The concerns with the open differential can be avoided by utilizing a limited slip differential. The limited slip differential does exactly what the name says, limits slip within the differential through several different methods. This can be done mechanically through the use of springs, clutches or a viscous material or it can also be done electronically.

There’s not a lot of difference between the design of an open differential and a limited-slip differential. Limited slip still has the same components as an open differential with a variety of extra elements depending on the type.

As an example, some limited slip differentials use spring pressure plates placed in the cage between the two drive pinions. This spring arrangement pushes the drive pinions outwards into a set of clutch packs. The backside of the drive pinions also have friction material on them which presses against clutch plates built into the cage. The clutch is always engaged by the springs and both output pinions attempt to spin at the same speed as the ring gear and cage as if the car were moving in a straight line. When a car with a limited-slip differential begins to corner, there is enough applied force to compress the springs and cause the drive pinions to slip against the clutch material allowing them to turn at different speeds again. The spring pack stiffness and the friction of the clutch pack determine the amount of force needed to slip the clutch and allow smooth cornering.

T

raction Control (TCS), ESC, Federal Regulations The maximum possible torque which can be transmitted from the tire to the road surface is determined by the coefficient of friction generated between the road and the tires (friction circle). If applied force exceeds that level, the wheels can spin relative to the road. If a vehicle is unable to gain traction for some reason, one wheel will spin while the other remains stopped. This is due to the open differential that most vehicles utilize causing the wheel with the least traction to receive the most power.

The purpose of the TCS is to prevent wheel spin from occurring during acceleration. There are currently no federal requirements for traction control on passenger vehicles.

slippage by activating the individual wheel brake on the slipping wheel. In order for the system to operate properly, it must be able to monitor the speed of the individual wheels on the vehicle and compare those speeds to one another as well as to vehicle speed. TCS monitors individual wheel speeds using the same wheel speed sensors utilized by the ABS system. If a wheel is slipping, TCS can activate the ABS and individually apply the brakes to the slipping wheel, reduce engine power by closing the throttle or do both. Applying the brakes to the slipping wheel will cause power to transfer to the non-slipping wheel. Conditions for TCS activation may include slippery road surfaces but heavy acceleration or acceleration while cornering can also cause system operation. Once activated, the TCS can control drive wheel speed as necessary to bring the vehicle under driver control. It is intended to improve vehicle stability when starting, accelerating or turning.

Multiple control modules on the vehicle must work together for TCS to operate properly. The TCS monitor signals from the four wheel speed sensors to determine the speed of each wheel and determine vehicle speed. When wheel slippage occurs the ABS system can be commanded to apply hydraulic pressure to the brakes at the slipping drive wheel and the engine control module (ECM) can reduce engine power by closing the throttle plate. At the same time, the slip indicator light is turned on to notify the driver of TCS operation. The TCS module and anti-lock brake module may be separate modules, but may also be integrated together as they rely on the wheel speed sensors and the hydraulic modulator and both require the ability to individually slow each wheel to maintain vehicle control. The ECM must also be in the control loop so engine power can be reduced as necessary to reduce wheel spin.

Electronic Stability Control (ESC) systems are different from TCS systems and are now required in all passenger cars and light trucks (49 CFR, part 571 & 585). ESC is a computerized control system designed to control vehicle over-steer (and possibly understeer) by applying the brakes to individual wheels and reducing engine power. This system builds on the existing ABS and TCS technology. ESC was introduced on passenger vehicles prior to this federal mandate in the mid 1990’s by Mercedes. This technology is now required on passenger vehicles in an attempt to prevent loss-of-control vehicle crashes. A NHTSA study 3 has indicated that ESC could potentially reduce passenger car crashes 35 – 40% and SUV crashes by up to 67%.

The same study indicates that single vehicle fatality crashes could be reduced by 56%. All passenger vehicles manufactured after 9/1/2011 must have an ESC system 7.

TCS cannot increase traction between a slippery road surface and the tire contact patch. Rather it controls wheel 32

Collision Magazine - Volume 9 Issue 2

www.collisionpublishing.com

The basic federal requirements for the ESC system are as follows: • •

• • • • • • • • •

The system must have the means to apply braking to all four wheels individually through the use of a control algorithm.

The system must be operational at all times except: • When it is shut off.

• When the vehicle speed is less than 9.3 mph.

• When the vehicle is in reverse.

The system must remain active even if TCS or ABS systems activate.

There must be an ESC malfunction / operation telltale.

There may be an optional ESC Off switch and telltale.

• Specific symbols required (FMVSS 101).

The system must pass a specific dynamic test which is detailed in FMVSS 126.

The system must pass specific functional tests for lateral stability and responsiveness.

Manufacturers may also: • Control throttle, Reduce fuel to engine • Implement “roll stability control” The system must meet over-steer correction requirements.

There is currently no understeer requirement.

This rule also requires FMVSS 135 compliant ABS on passenger vehicles.

The six main components and their typical locations within an ESC system are shown in Figure 6: wheel speed sensors,

control module, steering angle sensor, yaw rate sensor, accelerometer, and the hydraulic modulator. The hydraulic modulator is the same one used in the ABS and TCS. ESC adds the yaw sensor, an accelerometer, and a steering angle sensor to the standard ABS and TCS. During a swerve or sudden turn, the control module monitors the difference between the intended path of the vehicle measured by the steering angle sensor and the actual vehicle path monitored by the yaw rate sensor and sends a signal to the hydraulic unit, commanding it to apply the brakes to the appropriate wheel to correct the difference. Wheel braking causes the vehicle to rotate to correct the difference between the monitored angles. The engine control module can also reduce engine power (similar to TCS operation) by sending a signal to the throttle actuator commanding it to close the throttle or by shutting down individual fuel injectors.

In an over-steer situation, the rear tires lose lateral stability and the vehicle can fish tail or “drift” causing it to spin out. This can occur during radical steer inputs or evasive maneuvers. When the ESC system detects this yaw rotation, braking is applied to the outside front wheel and the throttle can be adjusted if necessary. This correction is also accompanied by a warning light to let the driver know the system is active. In an understeer situation, lateral tire traction can’t prevent lateral motion (friction circle) and the vehicle slides outboard on a curve. This can occur when speed is too high or the driver applies too much steer input and the tires plow causing the vehicle to slide outboard of the intended path. When ESC detects this yaw rotation

Figure 6: Typical ESC components www.collisionpublishing.com

Collision Magazine - Volume 9 Issue 2 33

braking is applied to the inside rear wheel and the throttle may be adjusted. A warning light would also be active in this case.

D

ata collection and analysis Multiple years, makes and models of vehicles were tested and evaluated for this study. The first vehicle tested was a 2004 Toyota Highlander. The Highlander was a front wheel drive model with a TCS and ABS. This vehicle was not equipped with ESC. Data was recorded during testing using the Toyota TechStream Lite scan tool. Multiple data sets were recorded with each front wheel spinning individually and both front wheels spinning as well. In all cases, the TCS did activate and eventually controlled the wheel spin in approximately 3 to 4 seconds.

Figure 7 shows an example of the data collected with both front wheels spinning from a stop. The accelerator is applied to the floor, but as shown, once the wheels begin to slip, the ECU reduces the throttle opening in an attempt to stop the wheels from slipping. At the second data point,

vehicle speed follows the slipping front wheel speeds up near 15 kilometers per hour (KPH) while the rear wheels are moving at approximately 5 KPH. At this point, the rear wheels are a better indicator of actual vehicle speed as they are unpowered and along for the ride. At the third data point, the indicated vehicle speed almost exactly splits the difference between the left front and right front wheels and is about 3 KPH higher than the indicated rear wheel speed. Only the left front wheel appears to be slipping at this point. The TCS is successfully controlling the wheel slippage.

Another vehicle tested for this study was a 2009 Honda Ridgeline. The Ridgeline was a four wheel drive model equipped with ABS, TCS and ESC. Data was recorded using the Auto Enginuity Scan tool equipped with the enhanced Honda / Acura package. Multiple data sets were recorded. In all cases, the TCS did activate and controlled the wheel spin very quickly.

Figure 7: 2004 Toyota Highlander wheel slip 34

Collision Magazine - Volume 9 Issue 2

www.collisionpublishing.com

Figure 8 shows an example data set of heavy accelerator application from a stop, executing an ABS stop and a second heavy accelerator application. As shown during both throttle applications, engine RPM lags accelerator application indicating the TCS is limiting engine power to control wheel slip. During both acceleration events, all four wheel speeds track very closely together indicating the TCS is controlling well. The only indicated divergence in wheel speeds is during the ABS event where the two left side wheels slide to a stop more quickly than the right side wheels. Otherwise this vehicle controls individual wheels speeds well and quickly.

Another vehicle tested for this study was a 2013 Toyota Rav4. This vehicle was also a four wheel drive vehicle with ABS, TCS and ESC. Data was collected and recorded during testing using the Toyota TechStream Lite scan tool.

Multiple data sets were recorded with all wheels spinning.

During this testing, it was discovered that some Toyota vehicles are storing what Toyota refers to as “Vehicle Control History” when certain trigger events occur. The stored

data is extensive as are the triggers which cause it to be stored.

According to the Toyota Information System (TIS), the 2013 - 2015 Rav4 and the 2014 - 2015 Highlander and the 2014 – 2015 Highlander Hybrid currently store this history data. It is considered likely by the author that additional vehicles will be added in the future. The data is recorded when triggered by specific vehicle behavior. Some other manufacturers store freeze frame data when a diagnostic trouble code (DTC) is set. Freeze frame data typically gives you one “frame” of data which occurred at the moment the DTC was set. These Toyota vehicles also store freeze frame data for non-DTC events but the history data is much more extensive. This history data is different from other manufacturers in that it does not require a DTC to be set in order to record the data.

These history recordings are stored in different storage areas depending on the trigger group and up to 140 items can be stored. The oldest data can be overwritten by future events when current space is exceeded for a particular

Figure 8: 2009 Honda Ridgeline wheel slip www.collisionpublishing.com

Collision Magazine - Volume 9 Issue 2 35

trigger group. The sampling rate, number of data points, points recorded before and after the trigger and recording time all vary by trigger item. The sampling rate varies from 2 Hz to 6.67 Hz; number of data points varies from 10 to 35 before and / or after the trigger and recording time varies from 5 to 7.5 seconds before and / or after the trigger.

Basic operation of the data recording system is shown in Figure 9. The trigger is detected and both pre-trigger and post trigger event data is stored. The trigger which led to the data being stored is identified by a code which can be interpreted using the on-line Toyota Information System (TIS). There are currently 26 different triggers listed and some examples are unexpected accelerator position compared to vehicle speed, simultaneous brake and accelerator application, ESC operation, TCS operation, ABS operation, sudden vehicle turns, vehicle pre-collision system operation, high accelerator angle during cruise operation and activation of the lane departure system.

There are 86 different record types listed in TIS and depending on trigger type, data stored can include vehicle speed, accelerator application, throttle opening, brake

pressure, steering angle, engine load, cruise control set speed, individual wheel speeds, turn signal position and many other data points. If the vehicle is equipped with a factory navigation system, event time information may also be stored.

Vehicle history data is not cleared by the battery being disconnected and cannot be cleared using the Toyota scan tool. This data can only be accessed using the Toyota TechStream scan tool. A “lite” version of this tool is available for purchase by the aftermarket community and was utilized for this testing. This data should absolutely be accessed in the event of a crash investigation involving a Toyota vehicle. Remember, this information is not permanently stored like deployment information stored in the ACM. Vehicle control history can easily be overwritten by subsequent triggers which may occur during normal operation. In the event of a crash investigation involving a 2013 - 2015 Toyota Rav4 or 2014 - 2015 Highlander or Highlander Hybrid, the vehicle should not be driven (or even started and run) until the vehicle control history can be downloaded by a Toyota dealer or someone who has the proper equipment. This history is not accessible using the Bosch CDR Tool as it is not crash data and is not regulated by 49CFR563.

An example of the Toyota history data is shown in Figure 10. This data was recorded on a day when the roads were snow covered and stored as the result of TCS activation. Only a portion of the data is shown. The trigger in this case occurred when the TCS first activated at 1186.2 seconds as indicated on the horizontal axis of the diagram. The first half of the data was recorded when the vehicle was stationary and is not shown.

Figure 9: Toyota vehicle control history operation 36

Collision Magazine - Volume 9 Issue 2

www.collisionpublishing.com

Through the center of the graph, this data set shows that the indicated vehicle speed follows the individual front wheel speeds. It also indicates a divergence of the actual indicated vehicle speed from the individual rear wheel speeds by up to 8 KPH. Lastly, it also shows the ECU taking control of the throttle away from the vehicle operator. At approximately 1189.2 seconds, the driver pushes the accelerator pedal

Figure 10: 2013 Toyota Rav4 traction control event data further down but the throttle plate moves the opposite direction, closing instead of opening further, effectively limiting power until wheel slip is brought under control.

Once the individual wheel speeds begin to converge, the throttle is slowly opened allowing engine power to increase and the vehicle to speed up under control. As indicated, this vehicle took about 1.3 seconds to get the wheel slip under control.

Immediately prior to crash day at the 2014 ARC-CSI crash conference, data for this study was collected from one of the Ford Crown Victorias made available for use by the crash team. This data was PCM data collected using the Bosch CDR system. An event was created, the key was shut off and then a PCM download was performed with the Bosch system. The data collected was limited to indicated speed, accelerator position, throttle position and engine RPM. In order to compare wheel slippage between data sets, a baseline was created by accelerating on dry pavement prior to accelerating with each individual rear wheel and both rear wheels on gravel. Figure 11 shows the data collected when both rear wheels were spinning

on gravel. Note the relatively rapid increase in indicated vehicle speed and engine RPM. Also note the difference between the applied accelerator and the throttle opening indicating the likely presence of a TCS. This Crown Victoria did not have a direct connection from the accelerator pedal to the throttle plate, it utilized a throttle-by-wire system with a solenoid driven throttle plate.

To develop a meaningful analysis of the Crown Victoria PCM data, speed traces from the collected data were graphed together as seen in Figure 12. The traces labeled as “RR Spin” and “Both Spin” in the graph were both from data sets collected at the same throttle opening (100%) as the trace labeled “Pavement.” This indicates that during the same level of applied throttle and acceleration, the indicated vehicle speed was approximately 5 miles per hour (mph) higher than the vehicle was actually moving at the time the data was recorded due to wheel spin. Indicated vehicle speed is higher than vehicle speed and would be recorded as such in the event of a crash.

www.collisionpublishing.com

Collision Magazine - Volume 9 Issue 2 37

Figure 11: Crown Victoria EDR wheel spin data

Figure 12: Crown Victoria data comparison 38

Collision Magazine - Volume 9 Issue 2

www.collisionpublishing.com

C

onclusions Crash investigators have access to many sources of information to reconstruct what occurred in the moments leading up to and immediately after a crash. One of those tools is the ACM stored crash data which is available for download in an ever increasing number of vehicles. It is important to remember, however, the stored data only reflects a particular sensor output at the time it was recorded but may not accurately represent the actual behavior of the vehicle. Vehicle speed in the stored crash data is the indicated vehicle speed provided to the ACM from another module. Depending on the year, make and model of the vehicle, indicated vehicle speed could be supplied to the reporting module by a speed sensor in the transmission or wheel speed sensor(s). If a wheel or wheels are spinning or slipping during the crash sequence, the indicated and reported speed information may not be indicative of actual vehicle speed.

This article discussed some reasons why wheels slip and how wheel slip is controlled by different vehicle systems.

Three different systems were discussed including antilock braking systems (ABS), TCS systems and electronic stability control (ESC). Federal regulations governing the applicable systems were also summarized. All of these systems can affect how a vehicle immediately prior to and during a crash. All of these systems must be considered when performing a crash reconstruction.

Data which was collected from several different vehicles to examine the effects of wheel slip on reported vehicle speed was presented and discussed. An interesting new data storage feature specific to some Toyota vehicles was summarized and discussed. As of the submission of this article, only the 2013 - 2015 Toyota Rav4 and the 2014 - 2015 Toyota Highlander and Highlander Hybrid reportedly store Vehicle Control History data, but more are expected in the future. The recorded data is not reported as crash data in the ACM and is therefore not governed by Part 563. This data is only retrievable using the Toyota TechStream scan tool and must be accessed in the event of a crash involving one of these vehicles.

So what does all of this information mean to the Reconstructionist? A complete analysis of the data collected for this study indicates that when the wheels slip, actual vehicle speed can be different from the indicated vehicle speed. According to the requirements detailed in CFR Title 49, Part 563, indicated vehicle speed is reported to and stored by the ACM in a crash. In the research here, reported speed for vehicles with spinning wheels was expectedly higher than actual vehicle speed. In some cases where the individual driven wheel speeds were different,

“

If a wheel or wheels are spinning or slipping during the crash sequence, the indicated and reported speed information may not be indicative of actual vehicle speed.

”

indicated speed spilt the difference between the two spinning wheels. In those cases, indicated speed was still not precisely indicative of the actual vehicle speed. If wheel slip is a possibility based on crash scene evidence or road conditions at the time of the crash, the Reconstructionist should not make the assumption that the EDR speed data accurately indicates the actual vehicle speed at the time the data was recorded.

R 2.

3.

4.

5.

6.

7.

8.

9.

eferences 1. National Highway Traffic Safety Administration (August, 2009) “The Long-Term Effect of ABS in Passenger Cars and LTVs”. Report number: DOT HS 811 182.

National Highway Traffic Safety Administration (April, 2003) “Sampling Issues in Rear-End Pre-Crash Data Collection”. Report number: DOT HS 809 541.

National Highway Traffic Safety Administration (September, 2004) “Preliminary Results Analyzing the Effectiveness of Electronic Stability Control (ESC) Systems”. Report number: DOT HS 809 790.

49CFR571 – Federal Motor Vehicle Safety Standards 49CFR585 – Federal Motor Vehicle Safety Standards 49CFR563 – Federal Motor Vehicle Safety Standards FMVSS No. 126 Electronic Stability Control Systems Toyota Information System subscription service, accessed on-line on multiple occasions.

People v. Luis Miguel Noriega, California Appellate Court, B236018, 3/18/2013

www.collisionpublishing.com

Collision Magazine - Volume 9 Issue 2 39

CRASH INVESTIGATION LASER MAPPING + Best alternative to total stations or a clipboard and wheel + Achieve professional-grade accuracy at a fraction of the cost + Save time and collect field measurements quicker, easier and safer TruPulse® 200X Laser Rangefinder

LaserSoft® Measure App

• Achieves +/- 4 cm accuracy without a prism • Calculates distances, heights and 2-D slopes • Has an in-scope display that adjusts to variable lighting conditions

• Runs on your smart phone • Captures, stores and sends raw laser measurements with scope shots • Works well for for revisiting the scene at a later date

Only $9.99 • Search “LaserSoft Measure”

QuickMap® 3D Kit • Saves money without sacrificing accuracy • Exports CAD-compatible data in convenient formats • Deploys quickly and packs up nicely

MapStar® TruAngle® • Helps collect more points faster from a single vantage point • Measures to a prism pole and/or directly to landmarks

1.877.696.2584 www.lasertech.com/collision www.facebook.com/LaserTechnologyInc

L ria

o

t Edi

ed

ew i v e

R

How long can you stay awake?

Clinton Marquardt Human Fatigue Specialist

T

he longest well-documented period without sleep is 264 hours and 12 minutes…about 11 days1. Seventeen year old Randy Gardner broke the previous record of 260 hours for his highschool science fair project. This was back in 1965 and there hasn’t been anything like it since.

mental and physical tests remained remarkably intact.

In fact, William Dement, the grandfather of sleep medicine and originator of the term “REM sleep”, tells a story of how Randy repeatedly beat him on a baseball game at a penny arcade in the middle of the night.1

What was amazing about Randy’s Guinness Book of What was not so amazing, was that the longer Randy World Record breaking feat, stayed awake, the harder it was to keep him from was that his ability to falling asleep. Randy was fatigued. perform well “Fatigue” is used to describe on many many different human states. Here are six uses of the term:

42 Collision Magazine - Volume 9 Issue 2

www.collisionpublishing.com

At 264 hours and 12 minutes of wakefulness, it would be hard to argue that Randy’s hand-eye coordination 2. Physical Fatigue = a human physical state re- would have been betsulting from body movements like exercise. ter than someone with 3. Lethargic Fatigue = a human psychological a BAC of 0.05%. If Randy’s performance was worse and physiological condition of lethargy that than this, should he be trusted to drive a car? If not, can result from a number of illnesses such as when would you tell Randy to stop driving and get some sleep? depression or the common cold.

4. Receptor Fatigue = a biological response to The threshold where almost all human performance over stimulation of one of the senses; for ex- measures decline rapidly is 22 hours. This is because ample, the olfactory receptors will fatigue af- it becomes impossible to maintain your awake state ter prolonged exposure to a smell, this results (a.k.a. state instability) at 22 hours of continuous in a temporary inability to distinguish that wakefulness. Micro-sleeps, short periods of sleep last-4 ing three to four seconds, begin at this threshold specific smell. providing that countermeasures such as strategic use 5. Task Fatigue = a point of diminishing returns, of caffeine and alertness sustaining medications have where the effort required to stay on task is not been ingested or you don’t have Randy’s stimulagreater than the effort required to move for- tion team sitting beside you in the car. ward on the task.

For accident reconstructionists, the 17 and 22 hour 6. Sleep related fatigue = a human biological thresholds can serve as indicators for investigating fatigue. At 17 hours of wakefulness, fatigue should be state of sleepiness. considered a potential contributing factor and other Randy was experiencing sleep-related fatigue. There sources of fatigue should be investigated to determine is a range in sleep-related fatigue levels. At the lowest the likelihood that fatigue may have played a role in fatigue level you would feel wide-awake you would the accident. At 22 hours of wakefulness, a full invesfind it hard to fall asleep within ten minutes.2 At the tigation of fatigue should be considered. Fatigue will other end of the range, the extreme fatigue level, like most likely be a contributing factor.

Randy, you would find it really difficult to stay awake eferences and, without someone forcing you to stay active and 2 1. Dement, W. (1972). Some must watch aroused, you would fall asleep within five minutes. while some must sleep. San Francisco: W.H. Randy’s team of friends and researchers kept him Freeman and Company. stimulated and awake. Without your own stimulation 2. For more information on measuring sleepiness team, just when should you cut off the wakefulness see: Carskadon, M., Dement, W., Mitler, M., and think about getting some sleep?

Roth, T., Westbrook, P., & Keenan, S. (1986).

Guidelines for the multiple sleep latency test Back in 1997, a re(MSLT): A standard measure of sleepiness. Sleep, search team from 9(4), 519-524.

Australia compared 3. Dawson, D., & Reid, K. (1997). Fatigue, alcohol cognitive psychoand performance impairment. Nature, 388, 235. motor performance 4. Beaumont, M., Batejat, D., Pierard, C., Coste, after long periods O., Doireau, P., Van Beers, P., Chauffard, F., of wakefulness to Chassard, D., Enslen, M., Denis, J., & Lagarde, performance after D. (2001). Slow release caffeine and prolonged drinking alcohol.3 What they found was that if you (64-h) continuous wakefulness: Effects on vigistay awake for 17 hours, your hand-eye coordination lance and cognitive performance. Journal of Sleep would be about the same as if you drank enough alcoResearch. 10(4), 265-276. hol to bring your blood alcohol concentration (BAC) level to 0.05%.

1. Mental Fatigue = a human psychological state resulting from spending extended or intense periods of time on a task like studying for an exam.

R

www.collisionpublishing.com

Collision Magazine - Volume 9 Issue 2 43

r Pee

ed

ew evi

R

Low Velocity Piston Isolator Testing of a Ford Crown Victoria

Lissette Ruberte Robert Thompson Bobby Clemence Mitchell Burton Terry Wong

46 Collision Magazine - Volume 9 Issue 2

Susan Lantz Michael Chan Greg Dycus Brian Chan Billy Cox

www.collisionpublishing.com

I

ntroduction

Accident reconstructionists are often asked to determine the change in velocity (delta-V) sustained by passenger cars with minimal visible damage to the bumper assembly beyond scratches and scuff marks to the plastic bumper cover. In vehicles equipped with energy-absorbing piston isolators, evidence marks on the exposed piston tube in conjunction with isolator compression data can be used to estimate the change in velocity experienced by the vehicle. However, bumper specific data relating isolator compression to changes in velocity of the vehicle are often unavailable.

Trends in isolator performance specific to vehicle manufacturers have previously been reported by Siegmund et al. (1994) based on over 600 vehicle-to-vehicle and vehicle-tobarrier collisions. In that study, several of the vehicles tested were within the 1980 to 1982 model year, when US Federal Motor Vehicle Safety Standards required that passenger cars sustain 5-mph longitudinal barrier and 3-mph pendulum impacts without sustaining significant structural damage. In 1982, the standard was lowered to 2.5-mph longitudinal barrier impacts and 1.5-mph corner pendulum impacts (La Heist and Ephraim, 1987). Of interest is whether the dynamic behavior of the piston isolators in the 2007 Ford Crown Victoria compares with the isolator data reported in the earlier work by Siegmund et al. (1994).

A series of low-speed, full-width, rear-into-barrier impacts were conducted with a 2007 Ford Crown Victoria. The tests were conducted on a single day as part of the ARC-CSI Boot Camp Experience. Two instrumented females were seated in the driver seat, one for three tests, the other for five tests. Neither occupant reported any pain or symptom of injury at the three-day posttesting follow-up examination. This article focuses on the dynamic behavior of the Ford Crown Victoria piston isolators.

M

ethods

Eight full-width vehicle-into-barrier tests were conducted with a 2007 Ford Crown Victoria (VIN: 2FAFP72V47X147799, weight: 1853 kg).

The vehicle was originally part of a police fleet, then a taxi cab fleet.

The rear bumper assembly of the 2007 Ford Crown Victoria consisted of a pair of energyabsorbing piston isolators with an exposed piston tube length of approximately 40 mm.

The bumper assembly was inspected prior to testing and showed no evidence of prior structural damage or repairs.

The vehicle was subjected to sequential barrier impacts with increasing closing speed to achieve changes in velocity (delta-V) of 1.0 to 2.6 m/s (1.7 to 5.8 mph). The repeatability of Ford (Mercury) piston isolators in low-speed impacts has previously been questioned (King et al., 1993). In order to verify the repeatability of the tests results, two series of sequential barrier impacts were conducted with the same vehicle.

The first sequence consisted of three impacts with a closing speed range of 0.56 to 1.38 m/s (1.3 to 3.1 mph). The second sequence consisted of five impacts with a closing speed range of 0.46 to 1.77 m/s (1.0 to 4.0 mph). These changes in velocity were selected to develop a detailed profile of isolator compression versus velocity change. At this severity level, the vehicle was expected to withstand multiple impacts without sustaining structural damage. The barrier consisted of a stack of ten concrete barrier blocks. The impact surface of the barrier was covered with a 13-mm thick sheet of plywood.

Each successive impact occurred within 2 to 5 minutes. The vehicle’s transmission was put in neutral, and the vehicle was accelerated to the desired closing speed by manually pushing the vehicle while monitoring a digital speed readout (VBOX). The vehicle was released less than two feet before impact and struck the stationary rigid barrier with the full bumper width. After each impact, the rear bumper was inspected and photographed. As shown in Figure 1, a plastic strap was added to each piston at the compression point, and post-impact displacement was measured to the nearest 0.8 mm (1/32 inch).

www.collisionpublishing.com

Collision Magazine - Volume 9 Issue 2 47

ata Collection The Ford Crown Victoria was fitted with a triaxial block of ICS 3031-050 (50g) accelerometers screwed to the front center tunnel forward of the front seat near the center of gravity of the vehicle to measure acceleration in the X-, Y-, and Z-directions. Vehicle acceleration data was collected at 10,000 Hz. All axis systems were in accordance with SAE J211-1 recommended practice with the X-, Y-, and Z-axes directed forward, to the right, and downward, respectively. Data collection was initiated by a pressure-sensitive tape switch mounted on the vehicle’s rear bumper. A time trap triggered by pressure-sensitive tape switches recorded the vehicle’s velocity.

The time trap was placed such that the rear wheels would make contact with the tape switches approximately 0.3 m before barrier contact.

D

ata Analysis In all tests, the vehicle was aligned to contact the barrier with the full bumper width. Thus, only the longitudinal (X-axis) acceleration was analyzed. To determine vehicle peak acceleration and delta-V, the data was filtered with channel frequency classes 60 and 180, respectively (SAE J211-1). The crash pulse was determined from the filtered vehicle acceleration (CFC 60) and defined as the time lapse between the last time the acceleration passed through zero at the beginning of the trace to the first time the acceleration passed through zero at the end of the trace. This was done to eliminate a time delay in the acceleration trace due to the plywood not being flush with the barrier. The time delay at the beginning of the acceleration trace averaged 31 ± 3 ms. To calculate deltaV, the vehicle acceleration was integrated within the crash pulse time. Restitution (ε) was calculated based on the closing speed and the change in velocity via the equation

D

of conservation of momentum, assuming the barrier was infinitely massive. esults A summary of the vehicle kinematics, including displacement values for the right and left piston isolators, are reported in Table 1. The isolators exhibited a linear increase in isolator compression for the changes in velocity tested, as shown in Figure 2. No isolator displacement was noted for changes in velocity of 1.0 m/s or less. This change in velocity was associated with a vehicle acceleration of less than 1.7 g. For the first sequence of impacts, there was no significant difference between the right and left piston-isolator compression. During the second sequence of impacts with changes in velocity of up to 2.6 m/s, differences between the right and left piston-isolator compression varied up to 10 mm. These differences between the right and left sides were likely due to one side of the bumper striking the barrier slightly before the other.

The isolators compressed up to approximately 25 mm and did not bottom out. The coefficient of restitution ranged from 0.46 to 0.82, and it tended to decrease as the change in velocity increased, as shown in Figure 3. The duration of the crash pulse ranged from 117 to 154 ms. The vehicle xacceleration trace consistently displayed a high frequency variation between 100 to 110 ms, except during the two tests with the smallest closing speed (Test A and Test D).

The peak vehicle acceleration recorded was 7.6 g for an impact speed of 1.77 m/s.

R

iscussion Although only a 2007 Ford Crown Victoria vehicle was tested, the Mitchell Collision Estimating & Reference Guide indicates that the same rear piston isolators and reinforcement bar were used in Ford Crown Victoria and Mercury Grand Marquis vehicles from 1998

D

Figure 1: Photographs of the exposed piston tube length prior to impact (left) and after an impact of 1.34 m/s (right).

48

Collision Magazine - Volume 9 Issue 2

www.collisionpublishing.com

to 2011. Thus, it is likely that results from this study can be used as reference when analyzing Ford/Mercury vehicles within the above listed year range.

Overall, the piston isolators displayed good repeatability, as shown by the clustering of the data points in Figure 2. This repeatability confirmed that the bumper components remained undamaged after each test. Although the vehicle acceleration trace consistently displayed a high frequency variation between 100 to 110 ms, there was no evidence that maximum compression or bottoming out had occurred. The same amount of compression was never measured on both isolators. Additionally, no plateau was observed in the curve relating isolator compression to velocity change. Piston isolators have been reported to have a maximum compression distance that is less than the exposed piston tube length and within 36 to 80 mm (King et al., 1993; Siegmund et al., 1994). An OEM rear piston isolator manufactured in 2008 and used rear piston disassembled from a 1997 Ford Crown Victoria were placed under a hydraulic press and compressed until the maximum stroke length was reach. Although, there were visual similarities between the older and newer piston isolators (Figure 4), the older model had an exposed piston tube length of approximately 35 mm while the newer model had an exposed piston tube length of 40 mm (Figure 5).

Both piston isolators reached a maximum stroke length that was approximately equal to their exposed piston tube length (Figure 6). During testing of the exemplar 2008 OEM piston under a hydraulic press, an unknown fluid was observed to exit the cylinder tube. No fluid leakage was observed in the used prior model year range piston isolator. Further testing is required to determine if the results from this study are applicable to Ford Crown Victoria/ Mercury Grand Marquis vehicles models prior to 1998.

Test

Closing Speed [m/s]

delta-V [m/s]

A B C D E F G H

0.56 0.95 1.38 0.46 0.89 1.36 1.77 1.77

1.02 1.42 2.04 0.76 1.35 2.09 2.59 2.59

In Siegmund (1994) the authors compared restitution to delta-V and reported that the coefficient of restitution initially decreases with increasing velocity and that they were most effective at absorbing energy before they began to bottom out. In our study, restitution decreased with increasing changes in velocity. The largest restitution (e = 0.82) was observed in the first impact test. It was likely that the Ford isolators were initially seized, and as a result, a higher speed change was required for the isolators to become functional.

In our study, the crash pulse ranged from 117 to 154 ms and was shortest for changes in velocity between 1.4 and 2.1 m/s. The longest crash pulse duration was recorded for the smallest closing speed tested. Siegmund et al. (1994) reported that the duration of the collision decreased with increasing velocity change, and the collision durations were less than 200 ms for barrier impacts with changes in velocity greater than 2 m/s. Similar findings reported by other investigators during low to moderate speed, fullwidth, rear-into-barrier impacts are summarized in Table 2.

When using isolator travel to determine changes in velocity, certain factors should be taken into consideration. As discussed by Siegmund et al. (1996), “The vehicle-to-vehicle data, when compared to the barrier data, indicates that direct use of barrier data will not provide an accurate estimate of a vehicle’s speed change in all cases.” When comparing piston measurements of the barrier impacts to vehicle-to-vehicle impacts in millimeters travelled per speed change, the barrier impacts were consistently less than the vehicle-to-vehicle impacts. A variety of real world factors can influence the amount of isolator travel and, thus, speed change, including variability in bumper construction, mass of the vehicles, vehicle stiffness, alignment of the bumpers on impact, and restitution. Isolator displacement is a valid

Isolator Compression Right [mm] Left [mm] 0.0 0.0 4.8 6.4 12.7 14.3 0.0 0.0 9.5 3.2 14.3 11.1 15.88 23.81 25.4 15.88

Vehicle apeak [g] 1.8 (91 ms) 2.8 (56 ms) 4.6 (63 ms) 1.4 (92 ms) 3.1 (67 ms) 5.5 (56 ms) 7.6 (61 ms) 6.4 (76 ms)

Crash Pulse Restitution [ms] 152 117 126 154 123 129 134 130

0.82 0.50 0.47 0.64 0.53 0.54 0.46 0.46

Table 1: Isolator compression and vehicle kinematics for each test www.collisionpublishing.com

Collision Magazine - Volume 9 Issue 2 49

Figure 2: Comparison of 2007 Ford Crown Victoria rear isolator compression with results reported by Siegmund et al. (1994) for Ford/Mercury vehicles

Figure 3: Restitution v. velocity change for a 2007 Ford Crown Victoria, rear impact into barrier test 50 Collision Magazine - Volume 9 Issue 2

www.collisionpublishing.com

Figure 4: Photograph of the Ford Crown Victoria/Mercury Grand Marquis rear piston isolator for the 1998 to 2011 MY (left) and 1992 to 1997 MY (right).

Figure 5: Photograph comparing the exposed piston tube length an OEM 2008 Ford Crown Victoria/ Mercury Grand Marquis rear piston isolator (left) and a used 1997 Ford Crown Victoria (right).

Figure 6: Photograph showing bottomed out isolator pistons of an OEM 2008 Ford Crown Victoria/ Mercury Grand Marquis rear piston isolator (left) and a used 1997 Ford Crown Victoria (right). www.collisionpublishing.com

Collision Magazine - Volume 9 Issue 2 51

scientific method for determining delta-V if the above factors are taken into account (Bailey et al., 1995; Cipriani et al., 2002; King et al., 1993). onclusions Examination of the piston isolators in a 2007 Ford Crown Victoria revealed shorter exposed piston tube length and higher stiffness than previously reported by Siegmund et al. (1994) for Ford/Mercury vehicles. The results of the current study demonstrated that advances in bumper design have resulted in changes in the performance of the Ford piston isolators. The dynamic behavior of the 2007 Ford Crown Victoria piston-isolators was shown to be repeatable and the rear bumper assembly was capable of withstanding multiple low-speed impacts without structural damage. Relying on the prior testing by Siegmund et al. (1994) to estimate the delta-V for a late model vehicle, based on isolator compression alone, may slightly underestimate the delta-V of the vehicle. The subject Ford Crown Victoria impact speed was not increased to the point where the isolator pistons were bottomed out.

A series of similar tests, with two or more similar vehicles, wherein the delta-V is incrementally increased to the point where the isolators are fully compressed may be beneficial.

3.

eferences 1. Bailey, M.N., Wong, B.C., and Lawrence, J.M. (1995) Data and methods for estimating the severity of minor impacts. SAE Paper Number 950352, Society of Automotive Engineers, Warrendale, PA. 2. Cipriani, A.L., Bayan, F.P., Woodhouse, M.L., Cornetto, A.D., Dalton, A.P., Tanner, C.B., Timbario, T.A., and Deyerl, E.S. (2002) Low Speed Collinear

8.

C

R

4.

5.

6.

7.

9.

Impact Severity: A Comparison Between Full Scale Testing and Analytical Prediction Tools With Restitution Analysis. SAE Paper Number 2002-01-0540, Society of Automotive Engineers, Warrendale, PA. Goodwin, V., Martin, D., Sackett, R., Schaefer, G., Olson, D., and Tencer, A.F. (1999) Vehicle and Occupant Response in Low Speed Car to Barrier Override Impacts. SAE Paper Number 1999-01-0442, Society of Automotive Engineers, Warrendale, PA. King, D.J., Siegmund, G.P., and Bailey, M.N. (1993) Automobile Bumper Behavior in Low-Speed Impacts.

SAE Paper Number 930211, Society of Automotive Engineers, Warrendale, PA. Linder, A., Avery, M., Krafft, M., and Kullgren, A. (2003) Change of velocity and pulse characteristics in rear impacts: real world and vehicle tests data. Accident Reconstruction Newsletter. 10(1).

Siegmund, G. (2005) Variability in Vehicle and Dummy Responses in Rear-End Collisions. Traffic Injury Prevention. 6(3): 267-277.

Siegmund, G.P., Bailey, M.N., and King, D.J. (1994) Characteristics of Specific Automobile Bumpers in Low-Velocity Impacts. SAE Paper Number 940916, Society of Automotive Engineers, Warrendale, PA. Siegmund, G.P., King, D.J., and Montgomery, D.T. (1996) Using Barrier Impact Data to Determine Speed Change in Aligned, Low-Speed Vehicle-to-Vehicle Collisions. SAE Paper Number 960887, Society of Automotive Engineers, Warrendale, PA. West, D.H., Gough, J.P., and Harper, G.T.K. (1993) Low Speed Rear-End Collision Testing Using Human Subjects. Accident Reconstruction Journal. 5(3): 2226.

Impact Velocity (m/s)

Isolator delta-V Crash Coefficient Compression (m/s) Pulse (ms) of restitution (mm)

OW3500 (1999)

5.1

3.0

68

OW3539 (1999) OW3594 (1999) OW3660 (1999) OW3718 (1999) OW9999 (1998) OW3737 OW3749 OW3763 OW3760 CR98001 CR98002 CR98006 1981 Ford Granada

6.9 9.8 8.3 11.1 5.1 14.6 10.3 9.9 12.8 6.7 6.7 6.7 0.6 1.4 1.6

3.7 3.6 4.8 5.0 2.8 4.8 5.1 4.8 4.8 4.6 4.0 4.1

80 84 69 65 92 103 82 89 74 130 72 129 125 172 107

Vehicle

52 Collision Magazine - Volume 9 Issue 2

www.collisionpublishing.com

Source/ Comments Linder et al. (2003) foam core bumper

2-2 1-13 9-15

West et al. (1993)

2.0 1.9 2.1 2.6 3.3 3.6 4.3 1984 Volvo 760 1.0 1.8 2.0 2.6 3.2 1990 Ford Escort 1.3 2.1 1989 Chevrolet Sprint 1.1 2.0 1979 Cadillac Biarritz 1.3 1.9 1997 Chevrolet Cavalier 2.3 2.3 1.7 1.7 2.8 1994 Toyota Corolla LE 2.2 2.1 1998 Toyota Corolla VE 2.2 1.7 1.7 2.8 2.8 3.4 2.2 1996 Honda Civic Si 2.3 2.2 1995 Dodge Neon 2.2 1992 Jeep Cherokee Sport 2.2 2002 Dodge Caravan SE 2.2 2002 Ford F150 XLT 2.2 1999 Ford Explorer Limited 2.3 1999 Mazda Protegé SE 2.2 2003 Volkswagen Jetta 1.8T 2.2 1.6 1.6 2.7 2.3 1999 Volvo S80 2.9 2.1 2001 Subaru Forester S 2.2 2001 Hyundai Accent GS 2.2 2.2 2000 Chevrolet Malibu LS 2.2 2000 Honda Accord V6 2.2 1998 Ford Windstar 6L 2.2 2001 Ford Focus ZX3 2.2 1998 Chevrolet 1500 Silverado 2.2 1990 GMC Sierra S15 Sonoma 2.2

1.9 3.0 1.6 3.0 1.6 2.3 3.5 3.5 2.7 2.7 4.0 3.2 3.1 3.3 2.6 2.6 4.1 4.2 4.9 3.2 3.5 3.2 2.8 2.8 2.8 2.9 3.3 3.5 3.5 2.7 2.7 4.3 3.2 3.1 3.2 3.3 3.4 3.2 3.4 3.4 2.8 2.8 2.8

128 160 114 113 110 115 198 151 134 125 117 126 70 90 100 90 140 110 105 98 113 102 121 98 117 121 117 117 133 106 164 117 117 133 121 137 145 129 121 137 137 148 141 125 125 109 125 109 129 117 188 152 94 82 82

5-28 0-33 19-32 40-47 49-57 44-54 57-57 1-11 5-18 23-25 32-35 31-50 2-13 32-44* 25-32 38-44 22-60* 48-60

0.5 0.46 0.46 0.47 0.21 0.21 0.52 0.54 0.62 0.62 0.44 0.43 0.45 0.48 0.53 0.55 0.48 0.5 0.46 0.47 0.48 0.47 0.28 0.29 0.31 0.34 0.45 0.58 0.62 0.65 0.66 0.59 0.44 0.47 0.46 0.51 0.53 0.47 0.52 0.54 0.27 0.27 0.27

Goodwin et al. (1999)

Siegmund et al. (2005)

*Isolator maximum stroke

Table 2: delta-V, Crash Pulse Duration, Isolator Compression, and Coefficient of Restitution Reported by Other Investigators

www.collisionpublishing.com

Collision Magazine - Volume 9 Issue 2 53

ACCIDENT RECONSTRUCTION AND FORENSIC MAPPING SOLUTIONS DESKTOP MAPPING & ANIMATION SOFTWARE | EVIDENCE COLLECTION SOFTWARE & HARDWARE | TRAINING

WHEN YOU’RE LOOKING FOR THE BEST ACCIDENT RECONSTRUCTION SOFTWARE, YOU NEED TO LOOK AT

EVIDENCE RECORDER

FORENSIC CAD

Evidence collection software built specifically for the professional reconstructionist.

Law enforcement and accident investigators use MapScenes Forensic CAD.

Using a Windows Mobile or tablet device, Evidence Recorder allows you to experience the power, precision and productivity gains associated with being able to visually verify your ongoing field evidence collection – on-the-spot, right in the palm of your hand. From there, it can seamlessly integrate with the MapScenes Forensic CAD desktop software.

MapScenes Forensic CAD is the most precise desktop forensic mapping software available in the world. This powerful and feature-rich application provides law enforcement professionals and accident reconstruction specialists, like you, with confidence that the evidence they present is precise, compelling, and irrefutable.

LET’S GET BACK TO BASICS How quickly and easily can you deliver a simple 2D scene for court?

Let’s face it, most of the drawings prepared for case files or courtroom presentations are simple, straightforward 2D diagrams.

The most critical and stressful time spent is at the scene. The reconstructionist needs the best scene evidence collection software and the best diagramming software. That’s it.

So how does it work? The process begins with Evidence Recorder (EVR).

With the MapScenes system, your diagramming begins in the field.

When using EVR to take measurements, you’re actually building the framework for the diagram in real-time. The user interface is crafted to work very specifically with your workflow to ensure a brisk pace when taking measurements via total station or GNSS receiver.

The data transfer between Evidence Recorder and Forensic CAD is unmatched. The diagram you can see in Evidence Recorder appears in MapScenes.

The evidence (actual measurements made by the instrument) is available and verifiable. The drafting commands are powerful and flexible. Your final deliverable is both credible and reliable.

When you do end up working on a scene that requires more visual appeal, MapScenes still has you covered.

With add-on modules, you can bring your diagrams to life with animation and simulations, as well as the ability to view the point cloud data, take measurements, and create your diagram right within the point cloud.

WHAT IS THE

SYSTEM?

EVIDENCE COLLECTION SOFTWARE & HARDWARE

DESKTOP MAPPING SOFTWARE

WORLD-CLASS, CERTIFIED TRAINING

Reduce errors, save time, and eliminate frustration.

When accuracy and precision are non-negotiable.

Training by professionals, for professionals.

Evidence Recorder is supported with more total stations, GNSS receivers, data collectors, and PC tablets than ever.

Not only is Forensic CAD the best choice for forensic mapping, but it is also the best-priced desktop software package in this industry.

Combine with your existing hardware, or choose from one of our hardware bundles. Evidence Recorder is included on all hardware configurations and offers the absolute best in value.

Starting at: $995 per user.

Training is essential in all facets of law enforcement. MapScenes acknowledges its responsibility to the industry by ensuring that all trainers and educational courses are kept up to date through a base educational standard, preserving quality and integrity.

At any time, you can add more functionality with our add-on modules–Animation module and Point Cloud module–for a truly custom approach to your exact requirements.

We have set an industry benchmark by developing a comprehensive framework that provides investigators with the knowledge and skills required to use the MapScenes system easily, efficiently, and expertly.

TO LEARN MORE, OR TO OBTAIN A FREE TRIAL VERSION, CALL US OR VISIT OUR WEBSITE: 1.800.668.3312

www.mapscenes.com

® MapScenes is registered in the U.S. Patent and Trademark Office by MicroSurvey Software Inc. MicroSurvey is part of Hexagon.

ed

R

ew evi

Vehicle Dynamics and Resultant Occupant Accelerations Caused by Vehicle Wheel Separation

r Pee

C. Brian Moody P.E., Bradley C. Reckamp P.E.,

A

Orion P. Keifer P.E., Wes Richardson

bstract Passenger car wheels are typically attached to the suspension hubs with lug nuts threaded onto wheel studs, or with lug bolts threaded into mating female threads in the hubs. These threaded connections can, for various reasons, fail. Such a failure often causes the wheel and tire to separate from the vehicle while in motion. This separation invariably results in the vehicle falling onto its suspension / brake components, and in many cases, an impact between the ejected tire and the vehicle. Accelerations imposed on the vehicle’s occupants during such incidents have been blamed for injuries and have been the subject of personal injury litigation. In order to better understand the accelerations and forces experienced by vehicle occupants during wheel separation from a moving vehicle, testing was performed involving a 1988 Nissan Maxima and further testing of a 1998 Ford Escort sedan. The series of Nissan tests included a static test set, in which one of the Nissan’s rear wheels was removed and the hub was lifted and dropped from various heights, as well as a dynamic test set, in which one of the rear wheels was caused to eject while in motion. The series of Ford tests included a dynamic test set in which a wheel was caused to eject while the Ford was traveling at various speeds, as well as a second dynamic test set in which the Ford was driven over speed

60 Collision Magazine - Volume 9 Issue 2

bumps and parking lot wheel stops at various speeds. Acceleration data experienced by the vehicle was recorded during all tests. This data was then compared to various activities with vertical acceleration and existing design standards for amusement park rides.

I

ntroduction A wheel may detach from a passenger vehicle in motion due to a variety reasons. Such accidents occasionally result in loss of control of the vehicle leading to a collision, for which standard collision reconstruction techniques are applicable. Invariably, however, the wheel ejection also results in an impact between the suspension, or brake components, and the pavement and possibly between the ejected wheel and the vehicle as the wheel departs. The forces from these impacts have been the focus of many personal injury claims over the years. In a common analytical approach, a reconstructionist uses the Impact-Momentum equation, that is the change in momentum (mass times change in velocity) is equal to the impulse (force times time over which the force is applied). Generally, a drop in height was assumed in order to calculate the impact velocity/change in momentum and an assumed impact time duration of the impact is used to calculate the force on the vehicle. In practice, the calculated force can

www.collisionpublishing.com

www.collisionpublishing.com

Collision Magazine - Volume 9 Issue 2 61

“

Two sets of dynamic wheel ejection tests were performed, the first in 1996 and the second in 2012.

”

vary by orders of magnitude, primarily depending upon the assumed time duration of the impact. In order to better understand and quantify the forces experienced during such events, two sets of tests were conducted in which a vehicle was instrumented and purposely caused to eject a wheel while in motion. Data from accelerometers were collected and analyzed. Data were also collected on static drops of a vehicle without a wheel and a vehicle when it was driven over speed bumps and wheel stops.

C

oordinate System A vehicle is assigned an SAE standard coordinate system 1 of positive X being the forward direction, positive Y being toward the right side of the vehicle, and positive Z pointing straight down toward the earth as the car is at rest on its tires. Using this convention, all accelerations can be trigonometrically divided into their X, Y, and Z components, Gx, Gy, and Gz respectively. Some lateral accelerations (Gy) are to be expected, but in small intensities compared to accelerations in the X and Z directions (Gx and Gz).

T

esting Two sets of dynamic wheel ejection tests were performed, the first in 1996 and the second in 2012.

In tests conducted in 1996, a 1988 Nissan Maxima was used to perform several tests to measure accelerations on the vehicle and at the head of a front right passenger during the event of a rear wheel separating from the vehicle. Later, in 2012, a 1998 Ford Escort was used to perform several tests to measure accelerations experienced by the vehicle and the interface between the operator and the seat during a single wheel separation from the vehicle while in motion. Additionally, the vehicle was driven over wheel stops and speed bumps, for comparison.

62 Collision Magazine - Volume 9 Issue 2

1996 Tests Initial experiments conducted in 1996 were designed to replicate an actual specific ejection of a left rear wheel and were published previously 2. A summary is included here as a background to the more recent testing.

A 1988 Nissan Maxima, occupied by a driver and a front right passenger test subject, was instrumented and tested.

A tri-axial piezoelectric accelerometer array (VibraMetrics Model 1136) was located on the floor to the rear of the passenger seatback (Photograph 1), and another was fixed to the left temple area of a custom head harness worn by a human test subject (female, age 39, 66 kg, 162 cm) seated in the front right passenger seat (Photograph 2). Two arrays of eight resettable Impact-O-Graph shock indicators (2, 4, 6, 8, 10, 15, 25, and 30 g’s) were used to detect peak accelerations, one located on the floor just to the rear of the passenger seat back (Photograph 1) and the other in the trunk above the left rear tire. Impact-O-Graph shock indicators feature a resettable ball bearing and spring arrangement that allows them to simply indicate whether or not a peak acceleration level was exceeded. Simple low pass resistor/capacitor filters, with -3 dB breakpoint of 100 Hz, were used between the VibraMetrics accelerometers and the personal computer (PC) data acquisition system. National Instruments VirtualBench Logger TM software was used.

Three stationary tests were performed by removing the left rear wheel and resting the brake rotor on the pavement.

The wheel hub was then raised to the test elevation change and maintained at that height with a simple lever using a jack stand as a fulcrum. The lever was then twisted out of the way allowing the vehicle to drop back to the ground through test elevation changes of approximately 15 cm, 30 cm, and 61 cm (Photograph 3).

Two dynamic tests were performed using the same instrumentation arrangement as described. In each dynamic trial, the Nissan was positioned at one end of the test track and the lug nuts were removed from the left rear wheel.

The test driver then accelerated and turned the steering wheel quickly, causing the wheel to eject and roll in the original direction in which the Nissan was traveling. The test speed was not measured or recorded but was estimated to be approximately 24 kph.

1996 Test Results Table 1 is a summary of the peak Z-axis (vertical) accelerations (-Gz) at the floor and at the test subject’s head as recorded by the VibraMetrics accelerometers. The attenua-

www.collisionpublishing.com

Photograph 1: Accelerometer array located below the passenger seat back.

Photograph 2: Tri-axial accelerometer attached on a head band (model in photograph was not the test subject).

2012 Tests The 2012 tests were designed to further examine the resultant accelerations arising from a wheel ejection from a vehicle in motion. A 1998 Ford Escort sedan was used as the test vehicle. The vehicle was fitted with four aftermarket rims and new Achilles ATR Sport 205/40RZ17 84W XL tires to replicate a specific actual incident.

Photograph 3: Static drop test apparatus. tion of the seat and body of the test subject was calculated and presented.

The acceleration in X-axis (Gx) for the two dynamic runs at the passenger’s head was less than 1.1g’s for test 1 and 0.8 g’s for test two.

Stationary Run #1 #2 #3 Dynamic Run #1 #2

Acceleration data was collected using four Crossbow Model CXL25LP3 tri-axial 25g accelerometers and one Crossbow Model CXL25LP3 tri-axial 2g accelerometer. The 25g accelerometers were mounted as follows: one on the floorboard beneath each of the front seats’ leading edges, one on the center tunnel forward of the rear seat, and one on a rectangular nylon board, along with the 2g accelerometer, that was placed on the driver’s seat and on which the driver sat during the tests. The dimensions of the nylon board were 20.3 cm wide by 15.2 cm long by 0.9 cm thick. The two accelerometers were mounted side by side on the middle of the board and the board was placed in the middle of the seat such that the accelerometers were between the board and the seat, with the long dimension of the board in the Y (transverse) direction. The accelerometer data

Peak Gzfloor -0.97 -1.75 -1.53

Duration 0.08 0.11 0.23

Peak Gzhead -0.53 -0.53 -0.85

Duration 0.19 0.20 0.24

Attenuation (Percent) 45% 70% 44%

-3.28 -2.38

0.12 0.13

-2.22 -1.29

0.15 0.20

32% 46%

Table 1: Vertical (Gz) Acceleration Peaks www.collisionpublishing.com

Collision Magazine - Volume 9 Issue 2 63

Photograph 4: The data collection system. were processed via a National Instruments 9205 analog input module data collection system in a CompactDAQ chassis with National Instruments LabView on a notebook computer (Photograph 4). Data was collected at 1400 Hz and at 16 bit resolution.

Four video cameras were used to document the wheel separation tests. One was anchored to an exterior superstructure on the vehicle to view the subject wheel and tire during testing. A second was anchored to the interior of the passenger side window to view the driver during testing (Photograph 5). A third was set to the side of the test area with a fixed wide view to observe the vehicle from the side during testing. A fourth, with a narrower view, was set up to manually pan with the vehicle during testing.

Two subjects were used as drivers and passengers, as summarized in Table 2. Subject 1 has training and experience as an amateur racecar driver. He has participated as a driver in other vehicle performance and crash testing. Subject 2 has participated as a driver in other vehicle performance and crash testing including driver for the second dynamic test in the 1996 wheel ejection testing.

For each of the wheel separation tests (tests 1 through 9) the vehicle was driven to the start point, jacked up, and the subject wheel was removed. Double-sided foam tape was applied between the wheel and brake hat surface and the

Subject 1 2

Gender Male Male

Age 34 60

Height 178 cm 188 cm

Photograph 5: A video camera was anchored to an exterior superstructure to view the subject wheel and another video camera was mounted on the passenger side window for a view of the driver. wheel was reinstalled and lowered to the ground. The lug nuts were then removed from the subject wheel. The vehicle was accelerated in a straight line to achieve the nominal test speed, as registered on the speedometer, at which time the driver jerked the steering wheel slightly to cause the subject wheel to separate from the vehicle.

Vehicle speeds were calculated by integrating the recorded Gx accelerations for tests 1 through 9. Tests 1 through 4 were ejections of the left front tire at nominal speeds of 16.1, 32.2, 48.3 and 64.4 kilometers per hour. Tests 5, 6, and 7 were left rear tire ejections. On test 5, at the beginning of the run, the tire was abruptly ejected, well below the 32.2 kilometers per hour target speed. On test 6, the wheel came off the studs; however, the brake drum continued to roll on the inside of the wheel rim and complete ejection was not obtained. Test 7 took significant steering input to eject the wheel. It should be noted that a factory alloy rim with a size 185/65R14 tire was used for Tests 6 and 7 due to the inability to reach the nominal speed (the tire ejected prematurely without driver input) with the aftermarket tire and rim. All other tests were conducted with the aftermarket tires and rims. Test 8 and 9 were right front

Weight 84 kg 106 kg

Tests Driver 1-7, 11&12 8-10

Table 2: 2012 Test subject data 64

Collision Magazine - Volume 9 Issue 2

www.collisionpublishing.com

Passenger 12

tire ejections at nominal speeds of 32.2 and 24.1 kilometers per hour respectively.

For the left front wheel ejection tests, the bolt and nut that attaches the lower ball joint assembly to the lower control arm contacted the ground and dragged on the pavement following the wheel ejection (Photograph 6). The nut and bolt were photographically documented and replaced after each test (Photograph 7). The fender was also photographically documented and was replaced with a salvaged fender following trials that resulted in significant damage (Photograph 8).

For the wheel stop and speed bump tests, the accelerometers and data collection remained the same; however, the only video recorded was of the driver, with the video camera attached to the passenger side window.

The speed bump test was performed in the parking lot of a strip mall that had three speed bumps evenly spaced at approximately 45.7 meters. The vehicle was driven back and forth over all three speed bumps at various speeds so that each speed bump was driven over 8 times for a total of 24 speed bump tests ranging in speed from 8.4 kph to 34.6 kph.

2012 Test Results Fifteen channels of accelerometer data were collected and processed. The data was transformed from the accelerometer axis to the axis of the vehicle and the acceleration of gravity was subtracted from the z-axis data. The raw data was filtered using a 4-pole Butterworth filter with a cutoff of 100Hz and 5Hz. The 100Hz filter was chosen based

The wheel stop was located in a parking lot with pavement on both sides such that the vehicle could be driven over the wheel stop. The contour of the wheel stop is shown on Photograph 9. All tests were conducted from the tapered side of the wheel stop, that is, the vehicle approached from the right side as shown in Photograph 9. The vehicle was positioned perpendicular to the wheel stop at an appropriate distance in order to approach the stop at the respective nominal test speed. The test vehicle was then accelerated to the nominal speed and the left wheels were driven over the wheel stop. In each trial both wheels of the left side were driven over the wheel stop.

Photograph 7: Lower ball joint nut and bolt.

Note the damage due to abrasion on the asphalt.

Photograph 6: Gouge caused by the lower ball joint nut and bolt riding on the ground.

Photograph 8: Damage to the left front fender, rear of the front wheel well due to the impact with the ejected wheel

www.collisionpublishing.com

Collision Magazine - Volume 9 Issue 2 65

on the previous testing and that it approximates the CFC60 recommendations for collision simulation input 1. The 5 Hz filter was chosen due to its use in amusement park ride assessment 7, and so that accelerations imposed upon the driver of the tests could be compared to allowable accelerations for common amusement park rides.

T

ire Ejection Tests In review of the data, the most significant forces on the vehicle were in the vertical direction (Gz).

Therefore to summarize the results, and for the purposes of this paper, the peak values of the vertical channel of the accelerometer on the driver’s seat surface are presented.

Figure 1: Vertical acceleration (Gz) floorboard under the driver’s seat for Wheel separation test 5.

100 Hz filtered data.

Figures 1 through 3 show Gz data collected during wheel separation test 5 which exhibited the largest accelerations experienced of all wheel separation tests. Figure 1 shows data collected by the accelerometer attached to the floorboard under the driver’s seat and filtered at 100 Hz. Figures 2 and 3 show data filtered at 5 Hz and 100 Hz respectively of the 2g accelerometer which was attached to the nylon board in the driver’s seat.

Table 3 lists the maximum Gz and minimum Gz accelerations measured at the driver’s seat for each of the tests. Table 4 lists the maximum Gz and minimum Gz accelerations measured at the floorboard under the driver’s seat for the same tests.

As measured at the driver’s seat interface, test 1 had the highest accelerations for an ejected front wheel. When the wheel was ejected, the advancing fender, specifically the rear of the wheel well, came into contact with the wheel. Since the ejected wheel was still rolling, the interaction of the wheel with the vehicle body damaged the fender, as shown in photograph 8, and caused it to be lifted up causing a negative (upward) acceleration spike. Figures 4 through 6 in the appendix show the test 1 100 Hz filtered data collected from the accelerometer located on the floorboard under the forward portion of the driver’s seat.

Figure 10 in the appendix shows the 100 Hz filtered Gz data at the surface of the 66

Collision Magazine - Volume 9 Issue 2

Figure 2: Vertical acceleration (Gz) driver’s seat surface for Wheel separation test 5. 5 Hz filtered data.

Figure 3: Vertical acceleration (Gz) driver’s seat surface for Wheel separation test 5. 100 Hz filtered data. www.collisionpublishing.com

Test #

Desc

Min unfilt

Max unfilt

Min 100Hz Max 100Hz

Min 5Hz

Max 5Hz

1 2 3 4 5

29.2 kph LF -1.49 0.93 -1.47 0.93 -0.89 0.84 28.1 kph LF -0.44 0.55 -0.36 0.48 -0.18 0.24 53.0 kph LF -1.05 0.65 -1.02 0.68 -0.65 0.33 71.2 kph LF -1.16 0.78 -0.96 0.76 -0.55 0.55 24.4 kph LR -2.06 1.42 -1.88 1.37 -0.96 1.07 42.5 kph LR 6 -0.55 0.4 -0.55 0.38 -0.21 0.25 partial depart* 7 49.3 kph LR -0.92 0.63 -0.64 0.6 -0.3 0.3 8 43.8 kph RF -0.38 0.56 -0.37 0.53 -0.15 0.28 9 32.6 kph RF -0.84 0.56 -0.81 0.56 -0.17 0.25 Mean -0.99 0.72 -0.90 0.70 -0.45 0.46 * During test 6 the left rear wheel departed from the hub, but remained with the car with the brake drum riding on the inside of the rim.

Table 3: Maximum and minimum vertical Accelerations (Gz) of driver’s seat surface with tire ejection.

Test #

Desc

Min unfilt

Max unfilt

Min 100Hz Max 100Hz

Min 5Hz

Max 5Hz

1 2 3 4 5

29.2 kph LF -3.68 3.03 -1.21 2.35 -0.80 0.54 28.1 kph LF -2.14 2.14 -0.82 1.12 -0.10 0.25 53.0 kph LF -3.27 3.07 -1.42 0.94 -0.51 0.32 71.2 kph LF -5.03 4.47 -0.47 0.44 -0.47 0.44 24.4 kph LR -2.40 2.03 -1.62 1.24 -0.66 0.49 42.5 kph LR 6 -0.67 0.79 -0.46 0.47 -0.11 0.17 partial depart* 7 49.3 kph LR -2.71 1.91 -0.69 0.59 -0.18 0.20 8 43.8 kph RF -2.12 2.35 -1.32 0.78 -0.21 0.24 9 32.6 kph RF -2.86 2.39 -2.41 1.83 -0.17 0.28 Mean -2.76 2.46 -1.16 1.08 -0.36 0.33 * During test 6 the left rear wheel departed from the hub, but remained with the car with the brake drum riding on the inside of the rim.

Table 4: Maximum and minimum vertical Accelerations (Gz) of the floorboard under the driver’s seat with tire ejection.

driver’s seat for comparison with Figure 9 in the appendix.

Figures 7 through 9 in the appendix show the 5 Hz filtered data for the three axes measured by the accelerometer at the seat interface. This was chosen as closely replicating the testing for amusement park ride acceleration.

Photograph 7 shows the lower ball joint nut and bolt following one of the front left wheel ejection trials.

The maximum range of accelerations, at the driver’s seat interface, for a front tire ejection occurred during test 1 with an acceleration range (maximum minus minimum) of 2.42 g unfiltered, 1.73 g filtered at 5 Hz, and 2.40 g filtered at 100 Hz. For rear tire ejections, the maximum range of accelerations occurred during test 5 with an acceleration range of 3.48 g unfiltered, 2.03 g filtered at 5 Hz, and 3.25 g filtered at 100 Hz. The average range of

www.collisionpublishing.com

Collision Magazine - Volume 9 Issue 2 67

accelerations measured for all tire ejection tests was 1.71g unfiltered, 0.91 g filtered at 5 Hz, and 1.59 g filtered at 100 Hz. Figures 6 and 10 in the appendix show vertical (z-axis) accelerations measured by accelerometers located at the floor below the driver’s seat, and at the driver’s seat surface respectively, both filtered at 100 Hz. A comparison shows the attenuation effects of the driver’s seat for test 1. A similar comparison of figure 13 and figure 17 in the appendix shows the attenuation effect of the driver’s seat for test 5.

W

heel Stop Tests Minimum and maximum vertical accelerations measured at the driver’s seat surface are presented in Table 5. These represent accelerations imposed on the driver of the vehicle. A comparison of figure 20 and figure 24 in the appendix shows the attenuation effect of the driver’s seat for test 11e. The maximum range of accelerations occurred during test 11e conducted at a speed of 8.2 kph. The acceleration range was 2.67 g unfiltered, 1.69 g filtered at 5 Hz, and 2.64 g filtered at 100 Hz. On average, the range of accelerations measured for all wheel stop tests was 1.84 g unfiltered, 1.23 g filtered at 5 Hz, and 1.83 g filtered at 100 Hz. Table 6 lists the maximum Gz and minimum Gz accelerations measured at the floorboard under the driver’s seat for the wheel stop tests.

S

peed Bump Tests Three speed bumps at a commercial shopping center were used for the tests. The speed bumps used were substantially similar. They measured approximately 5.7 cm tall and 66.0 cm wide. Photograph 10 shows one of the speed bumps used. Table 7 is a summary of the vertical acceleration data measured at the driver’s

Photograph 9: Contour of the wheel stop

68

Collision Magazine - Volume 9 Issue 2

seat surface. A comparison of figures 27 and 31 in the appendix shows the attenuation effect of the driver’s seat for test 12v. The maximum range of unfiltered and 100 Hz filtered accelerations occurred during test 12v, which was conducted at a speed of 29.0 kph. The acceleration range was 2.18 g unfiltered, 0.76 g filtered at 5 Hz, and 2.08 g filtered at 100 Hz. The maximum range of 5 Hz filtered acceleration occurred during test 12d at 12.4 kph. The acceleration range for test 12d was 1.05 g unfiltered, 0.89 g filtered at 5 Hz, and 1.05 filtered at 100 Hz. On average, the range of accelerations measured for all speed bump tests was 1.24 g unfiltered, 0.68 g filtered at 5 Hz, and 1.23 g filtered at 100 Hz. Minimum and maximum vertical accelerations measured at the floorboard under the driver’s seat are presented in Table 8 for the speed bump tests.

L

iterature Review Curb Impacts A search of the literature showed no previous reported wheel ejection testing; however, curb impacts with human occupant testing are documented and are believed to be the most similar event to a wheel ejection, since they cause a vertical excursion to a vehicle.

Ogan, Alcorn and Scott 3 performed seven tests, running a 1994 Chevrolet Astro van into a six inch barrier curb at an approach angle of 60 degrees at speeds of approximately 5.1 to 24.5 kph. They also performed eight tests, running a 1991 Geo Prizm at 30 to 90 degrees into 15.2 cm mountable curb at speeds of 24.1 to 43.5 kph. The driver was fitted with a two axis accelerometer, using an adjustable nylon strap at the base of the driver’s neck in the front of his body. Thirteen of the tests were with the accelerometers oriented in the longitudinal (Gx) and lateral (Gy) directions. The accelerometers were oriented in the longitudinal (Gx) and vertical (Gz) direction on two tests to the mountable curb, one at a 60 degree approach angle and at 43.5

Photograph 10: One of the three speed bumps used in the speed bump tests.

www.collisionpublishing.com

Test #

Speed (kph)

Min unfilt

Max unfilt Min 100Hz Max 100Hz

Min 5Hz

Max 5Hz

10a 10b 10c 10d 10e 11a 11b 11c 11d 11e 11f 11g 11h

5.6 10.1 3.5 3.5 3.7 4.7 7.1 4.3 10.0 8.2 6.4 11.7 6.8 Mean

-1.2 -1.15 -1.24 -1.47 -0.74 -0.71 -0.79 -0.99 -1.02 -1.98 -0.81 -1.05 -1.37 -1.12

0.7 0.96 0.8 0.79 0.34 0.36 0.76 0.41 1.04 0.69 0.41 1.25 0.88 0.72

-0.79 -0.59 -0.72 -0.85 -0.3 -0.53 -0.57 -0.52 -0.74 -1.11 -0.55 -0.49 -0.94 -0.67

0.66 0.74 0.67 0.74 0.18 0.31 0.59 0.35 0.71 0.58 0.38 0.83 0.61 0.57

-1.19 -1.16 -1.23 -1.48 -0.73 -0.69 -0.77 -0.98 -1.01 -1.95 -0.8 -1.03 -1.36 -1.11

0.7 0.96 0.81 0.78 0.34 0.36 0.76 0.41 1.03 0.69 0.41 1.24 0.87 0.72

Table 5: Wheel stop maximum and minimum vertical accelerations (Gz) of driver’s seat surface.

Test #

Speed (kph)

Min unfilt

Max unfilt Min 100Hz Max 100Hz

Min 5Hz

Max 5Hz

10a 10b 10c 10d 10e 11a 11b 11c 11d 11e 11f 11g 11h

5.6 10.1 3.5 3.5 3.7 4.7 7.1 4.3 10.0 8.2 6.4 11.7 6.8 Mean

-4.41 -1.47 -1.47 -3.32 -2.04 -1.68 -5.20 -1.38 -2.54 -6.74 -1.41 -2.83 -4.13 -2.97

0.74 2.32 0.94 1.29 0.74 1.31 0.97 0.51 1.95 2.72 0.77 2.38 1.39 1.39

-0.59 -0.47 -0.44 -0.64 -0.30 -0.42 -0.50 -0.30 -0.56 -1.23 -0.46 -0.36 -0.71 -0.54

0.45 0.56 0.40 0.43 0.15 0.26 0.55 0.27 0.61 0.64 0.34 0.57 0.56 0.45

-2.72 -1.32 -1.00 -2.68 -1.85 -1.14 -2.86 -1.35 -2.10 -6.52 -1.40 -1.40 -2.49 -2.22

0.64 2.17 0.76 1.09 0.59 0.36 0.93 0.50 1.56 2.26 0.64 1.88 1.25 1.13

Table 6: Wheel stop maximum and minimum vertical accelerations (Gz) of the floorboard under the driver’s seat. kph and one at a 45 degree approach angle and at 43.5 kph. The kinematic responses of the drivers were minimal, with no secondary contact with any interior components of the vehicle and were not “outside the confines of normal driving experiences”. The highest acceleration measured was 3.8 g combined Gx/Gy in the 6.9 kph 60 degree test into the barrier curb, where the wheels did not mount the curb.

Hicks, Field, Bowler, Ptucha and King 4 conducted curb impact tests, where the acceleration was measured on the tunnel, near the center of gravity. They tested a 1977 Chevrolet Malibu, at approach angles of 90 and 60 degrees and speeds between 20 and 53 kph. They also tested a1982 Toyota Tercel at an approach angle of 90 degrees at speeds of 32 and 42 kph. They reported the highest -Gz peak vertical acceleration of -20 g’s (data collected at 400 Hz and

www.collisionpublishing.com

Collision Magazine - Volume 9 Issue 2 69

Test #

Speed (kph) Min unfilt

Max unfilt

Min 100Hz Max 100Hz Min 5Hz

Max 5Hz

12a 12b 12c 12d 12e 12f 12g 12h 12i 12j 12k 12l 12m 12n 12o 12p 12q 12r 12s 12t 12u 12v 12w 12x

8.4 7.4 8.4 12.4 11.7 11.4 12.9 15.4 15.4 20.9 21.4 20.4 21.9 24.8 24.1 22.7 34.4 28.5 20.0 12.6 16.3 29.0 34.6 32.8 Mean

0.49 0.38 0.48 0.6 0.54 0.47 0.57 0.6 0.64 0.57 0.54 0.54 0.66 0.66 0.45 0.63 0.73 0.67 0.57 0.53 0.64 0.66 0.64 0.72 0.58

-0.3 -0.27 -0.36 -0.45 -0.33 -0.4 -0.36 -0.48 -0.49 -0.7 -0.62 -0.52 -0.97 -0.9 -0.8 -0.65 -0.69 -0.67 -1.34 -0.37 -0.42 -1.43 -0.92 -1.05 -0.65

0.43 0.38 0.45 0.51 0.44 0.38 0.29 0.33 0.51 0.38 0.32 0.3 0.27 0.33 0.27 0.33 0.33 0.35 0.25 0.31 0.49 0.3 0.28 0.32 0.36

-0.3 -0.27 -0.36 -0.45 -0.33 -0.4 -0.36 -0.49 -0.49 -0.7 -0.63 -0.53 -0.97 -0.9 -0.8 -0.64 -0.7 -0.66 -1.4 -0.36 -0.42 -1.52 -0.93 -1.06 -0.65

0.48 0.38 0.48 0.6 0.54 0.47 0.53 0.61 0.64 0.57 0.54 0.54 0.66 0.65 0.46 0.63 0.72 0.66 0.58 0.53 0.64 0.65 0.64 0.71 0.58

-0.27 -0.24 -0.31 -0.38 -0.29 -0.29 -0.25 -0.23 -0.3 -0.39 -0.31 -0.24 -0.39 -0.35 -0.34 -0.34 -0.34 -0.23 -0.41 -0.2 -0.29 -0.46 -0.41 -0.48 -0.32

Table 7: Speed bump maximum and minimum vertical accelerations (Gz) of driver’s seat surface. filtered to 40 Hz) for the two Toyota tests, but noted that these peaks are consistent with saturation of the analog to digital card used in the collection of data and the actual peak could have been larger.

Meyer, Haden, Herbst, Hock and Forrest 5 reported testing four vehicles, each at approach angles of 30, 60 and 90 degrees into a15.2 cm high curb. At least five speeds up to 65 kph were tested for each of the three approach angles.

A tri-axial accelerometer was attached near the center of gravity of the vehicles. The peak vertical acceleration -Gz measured for the 1981 Datson 210 was -12.34 g’s, for the 1987 Jeep Grand Wagoneer was -15.8 g’s, for the 1985 Nissan Pickup was -15.1 g’s and for the 1991 Subaru Legacy was -28.5 g’s. It was reported that none of the drivers reported the vertical accelerations as being significant and 70

Collision Magazine - Volume 9 Issue 2

film analysis confirmed the vertical motion was relatively minor (3 - 5 cm).

Bayan, Cipriani, Cornetto and Timbaro 6 reported 46 curb impact tests of a 1996 Dodge Ram 1500 Pickup at speeds of up to 24.1 kph with approach angles of 90 and 45 degrees. The vehicle was instrumented with a Vericom VC3000 with a tri-axial accelerometer. The data was recorded at 100 Hz. A 4th order phaseless Butterworth filter was used to remove noise and vibration; however the frequency of the filter was not listed. The highest vertical -Gz acceleration was approximately -4.0 g’s during the 90 degree impacts at approximately 24.1 kph. While it is clear that the vehicle contained passengers, no information on the passengers was presented. Presumably, the testing was conducted without injury.

www.collisionpublishing.com

Test #

Speed (kph) Min unfilt

Max unfilt

Min 100Hz Max 100Hz Min 5Hz

Max 5Hz

12a 12b 12c 12d 12e 12f 12g 12h 12i 12j 12k 12l 12m 12n 12o 12p 12q 12r 12s 12t 12u 12v 12w 12x

8.4 7.4 8.4 12.4 11.7 11.4 12.9 15.4 15.4 20.9 21.4 20.4 21.9 24.8 24.1 22.7 34.4 28.5 20.0 12.6 16.3 29.0 34.6 32.8 Mean

0.51 0.41 0.52 0.57 0.53 0.42 0.49 0.82 0.63 0.89 1.06 0.91 1.17 0.92 0.84 0.94 0.99 1.02 1.20 0.47 0.59 1.90 0.91 1.11 0.83

-0.31 -0.29 -0.36 -0.38 -0.37 -0.43 -0.39 -0.56 -0.63 -0.69 -0.58 -0.68 -0.90 -0.97 -0.73 -0.64 -1.05 -1.15 -1.03 -0.43 -0.47 -1.45 -1.00 -1.23 -0.70

0.29 0.33 0.42 0.37 0.33 0.31 0.25 0.28 0.34 0.25 0.24 0.22 0.22 0.22 0.25 0.25 0.21 0.23 0.24 0.17 0.35 0.23 0.19 0.19 0.27

-0.32 -0.31 -0.37 -0.40 -0.40 -0.47 -0.43 -0.58 -0.65 -0.71 -0.65 -0.69 -1.04 -1.03 -0.78 -0.68 -1.08 -1.18 -1.08 -0.47 -0.52 -1.72 -1.08 -1.27 -0.75

0.49 0.41 0.49 0.53 0.50 0.40 0.47 0.80 0.62 0.86 1.00 0.89 1.03 0.86 0.81 0.91 0.96 0.99 1.16 0.47 0.58 1.72 0.87 0.99 0.78

-0.26 -0.27 -0.32 -0.31 -0.24 -0.24 -0.20 -0.15 -0.25 -0.27 -0.20 -0.20 -0.27 -0.26 -0.28 -0.26 -0.24 -0.17 -0.44 -0.17 -0.26 -0.24 -0.21 -0.26 -0.25

Table 8: Speed bump maximum and minimum vertical accelerations (Gz) of the floorboard under the driver’s seat. striction. It should be noted that the data presented in this paper excludes gravity and the limits cited above would therefore be -5.0 and -1.0 g’s.

Biomechanical Literature Design Standards The ASTM International publishes the Standard Practice for Design of Amusement Rides and Devices (ASTM F 2291) 7. Included in the standard are design acceleration limits for acceleration pulses of over 0.2 seconds. The limits are based on the data being process through a 5 Hz 4-pole, single pass filter and are plotted as acceleration verses time duration. Acceleration in the downward positive z direction (Gz) of up to 1.5 g’s is permitted for up to 4 seconds and up to 1.1 g’s is permitted without time restriction. Acceleration in the upward (-Gz) of up to -6.0 g’s (including gravity) is permitted for up to 1 second and -2.0 g’s (including gravity) or less is permitted without time re-

Human Activity Research Szabo and Welcher 8 noted that SAE Class 60 filter (lowpass filter with cutoff frequency of approximately 100 Hz) as likely adequate for low speed human subject body segment acceleration processing. Further, they presented vertical acceleration data for riding in a boat, horseback riding, and riding a roller coaster as comparative activities for lumbar spine injury potential. Data for a Ski-Nautique ski boat traveling across a fresh water lake at approximately 48 kph showed -Gz peaks at over 6 g’s. Data for a horse with a western style saddle while galloping was presented, with

www.collisionpublishing.com

Collision Magazine - Volume 9 Issue 2 71

-Gz peaks at approximately -12 g’s. Peak -Gz accelerations at approximately -7 g’s were recorded on a double looping roller coaster.

Ng, Bussone, Duma, and Kress 9 instrumented 18 subjects, categorized as large male (4 subjects), mid. male (5 subjects), mid. female (9 subjects) and small female (4 subjects) with accelerometers on their back approximating the level of the T1 and L5 vertebrae. Data was taken at 2000 Hz; however, no processing was noted. Seven activities were recorded, including sitting in a chair, sitting quickly in a chair, walking, running, jumping jacks, vertical leap, and jumping off of a 20 cm step. It was found that in general that the acceleration measured by the upper thoracic spine (T1) accelerometer was less than that of the lower lumbar spine (L5) accelerometer, which illustrates the force attenuation of the spine. The highest peak lumbar accelerations in the –Gz direction were recorded in the maximum vertical leap event. For the large males the average was -11.27 g’s (range -6.03 to -21.47 g’s); for the mid. males, the average was -11.45 g’s (range -8.58 to -14.56 g’s); for the mid. females the average was -8.78 g’s (range -5.60 to -13.28 g’s); and for the small females the average was -6.16 g’s (range -3.77 to -9.91 g’s).

Manoogian, Funk, Cormier, Bain, Guzman and Bonugli [10] also performed testing of volunteers to vertical loading scenarios. Eighteen subjects were tested in four activities, a chair plop, jump from a platform 30, 60, and 90 cm high, seated drop from 5, 7.5, and 10 cm and a chair tip activity where the center of the chair back was dropped at heights of 5, 7.5, and 10 cm. The accelerations were transformed to ground coordinates and filtered to CFC180. The seated drop produced vertical forces in the spine and was more controlled for more consistent data. The average lumbar acceleration for the 10 cm seated drop in the -Gz direction was -7.3 ±3.1 g’s. The highest vertical accelerations -Gz were from the 90 cm platform jump, the average of which was -13.2 ±5.3 g’s. As noted in the report, “this study has reported the data that corresponds to events which do not produce lasting pain or injury.”

Also noted in this testing, is that there is considerable attenuation of the peak acceleration between the chassis of the vehicle (floor) and the head of the passenger seat occupant. The vertical acceleration peaks occurred when the left quarter panel contacted and rose up over the ejected tire. These peaks were not attenuated by the vehicle’s suspension. It was estimated that when the tire was fully ejected, the left rear of the vehicle fell approximately 61.0 cm, therefore, the peak acceleration at the wheel was assumed to be comparable to the 61.0 cm static drop test. The lack of suspension attenuation on the peaks in the dynamic tests can be clearly seen since the chassis acceleration peaks of the two dynamic tests are much higher that of the 61.0 cm static test.

2012 Tests The two test drivers participated in multiple tests. All of the wheel ejection tests in 2012 were performed in a single day. A later day was used for all of the wheel stop and speed bump tests. Subjectively, both drivers described the accelerations as “mild” and unlikely to result in injury.

The average negative peak 100 Hz filtered vertical acceleration for the wheel ejections, wheel stops test, and speed bump tests at the driver’s seat was -0.90 g’s, -1.11 g’s, and -0.65 g’s respectively. Similarly for the 5 Hz filtered data, the average negative peak accelerations were -0.45 g, -0.67 g, and -0.32 g respectively. The average negative peak 100 Hz filtered for the wheel ejections, wheel stops test, and speed bump tests at the floorboard under the driver’s seat

D

iscussion 1996 Tests The initial assumption was that the peak forces on the vehicle chassis were from the ground impact of the brake components, therefore static tests of increasing height were performed as a precursor to the dynamic tests. During the dynamic tests, however, it was found that the rear wheel separated, remained upright and was impacted by the lower rear corner of the wheel well.

The rear quarter panel then rolled up on the wheel, rapidly lifting the vehicle body, producing distinct but subtle quarter panel damage as shown in Photograph 11.

72

Collision Magazine - Volume 9 Issue 2

Photograph 11. Damage to the left rear wheel well quarter panel which occurred due to the impact and subsequent riding over the detached wheel.

www.collisionpublishing.com

was -0.97 g, -1.13 g, and -0.78g respectively. This indicates that driving over a wheel stop produces peak accelerations grossly similar and slightly higher than a wheel ejection event. The speed bumps produced significantly lower vertical accelerations.

While the Standard Practice for Design of Amusement Rides and Devices (ASTM F 2291) 7 does not address short duration (less that 0.2 second) acceleration pulses, it does give a known benchmark. The filtering at 5 Hz removes the very short duration pulses that exist in amusement park rides. None of the recorded peak accelerations recorded during these tests, including the unfiltered data, exceeded the standard’s limits.

The curb tests cited above give acceleration data for various points on the respective test vehicles and subjects’ bodies.

Nonetheless, the accelerations measured in these tests are significantly below the peak values above, and therefore are well within tolerable levels. This is to be expected since the curbs used in general were standard six inches high curbs.

These are higher than the wheel stop testing reported here.

Further, the highest accelerations were reported were at 90 degree impacts where both wheels of an axle would climb the curb simultaneously.

The human activity research Ng, et al, and Manoogian, et al, measured acceleration at the test subject’s upper lumbar spine and upper thoracic spine. They noted significant acceleration attenuation of the spine in the measured accelerations of the upper to lower spine measurements. The tests presented here measured the acceleration at the seat / test subject interface and therefore the measured acceleration would be expected to be approximated and perhaps slightly higher than the accelerations at the upper lumbar spine.

The highest -Gz peak measured in the wheel ejection tests was from test 5. The acceleration at the seat interface was -2.06 g’s unfiltered and -1.88 g’s filtered at 100 Hz. These values are well below those reported in the cited studies.

In these studies, the wheel ejection event corresponds to events which do not produce lasting pain or injury.

S

ummary/Conclusions Driving over a wheel stop produces peak accelerations similar to, but slightly higher than, a wheel ejection event. The speed bumps produced significantly lower vertical accelerations than a wheel ejection event.

Wheel ejection events produce vertical accelerations of the vehicle well below those caused in testing of curb impacts with human occupants.

The accelerations measured at the driver’s seat were well below accelerations produced in the cited testing of human activities.

The measured accelerations in the testing of wheel ejections were well within the limits established by ASTM for amusement park rides. The wheel ejection events reported in this study, correspond to events that do not produce lasting pain or injury.

R

eferences 1. Surface Vehicle Recommended Practice “Instrumentation for Impact Test – Part 1 – Electronic Instrumentation”, SAE J211/1, SAE International, Warrendale, PA, 2007

2. Layson P. D., “Occupant Dynamics in Stationary and Moving Vehicles Subjected to Vertical Impacts: An Experimental Study,” Eleventh National Conference on Undergraduate Research Proceedings, 1997 3. Ogan, Jerry S., Acorn, Thomas L., and Scott, John D., “Acceleration Levels and Occupant Kinematics Associated with Vehicle-to-Curb Impacts”, Accident Reconstruction Journal, May/June 1995, pp. 54-56.

4. Hicks, B., Field, D., Bowler, J., Ptucha, S., and King, D., “Speed Change from Curb Impacts”, Accident Investigation Quarterly, Summer 1997, pp. 20-26.

5. Meyer, S. E., Haden, J., Herbst, B., Hock, D., and Forrest, S., “Curb Impacts – A Continuing Study in Energy Loss and Occupant Kinematics”, SAE 2002010-557, SAE International, Warrendale, PA, 2002.

6. Bayan, F. P., Cipriani, A. L., Cornetto, A. D., and Timbario, T. A., “Vehicle Speed Changes and Accelerations Associated with Curb Impacts and a Comparison to HVE SIMON”, SAE 2005-01-1175, SAE International, Warrendale, PA, 2005.

7. “Standard Practice for Design of Amusement Rides and Devices”, ASTM 2291-06a, ASTM International, West Conshohocken, Pa, 2006.

8. Szabo, T. J. and Welcher, J. B., “Human Subject Responses to Various Acceleration Fields”, Low Speed Collision TOPTEC, Society of Automotive Engineers, Vancouver BC, August 1996.

9. Ng, T. P., Bussone, W. R., Duma, S. M. and Kress, T. A., “Thoracic and Lumbar Spine Acceleration in Everyday Activities”, Biomedical Sciences Instrumentation, Vol 42, pp. 410-5, 2006.

www.collisionpublishing.com

Collision Magazine - Volume 9 Issue 2 73

10. Manoogian, S., Funk, J., Cormier, J., Bain, C., Guzman, H., and Bonugli, E., “Evaluation of Thoracic and Lumbar Accelerations of Volunteers in Vertical and Horizontal Loading Scenarios”, SAE 2010-010146, SAE International, Warrendale, PA, 2010.

Appendices (figures 4-31):

Figure 4: Axial acceleration (Gx) floorboard under the driver’s seat for Wheel separation test 1. 100 Hz filtered data.

Figure 5: Lateral acceleration (Gy) floorboard under the driver’s seat for Wheel separation test 1.100 Hz filtered data.

Figure 6: Vertical acceleration (Gz) floorboard under the driver’s seat for Wheel separation test 1. 100 Hz filtered data.

Figure 7: Axial acceleration (Gx) driver’s seat surface for Wheel separation test 1.

5 Hz filtered data.

Figure 8: Lateral acceleration (Gy) driver’s seat surface for Wheel separation test 1.

5 Hz filtered data.

Figure 9: Vertical acceleration (Gz) driver’s seat surface for Wheel separation test 1.

5 Hz filtered data.

74 Collision Magazine - Volume 9 Issue 2

www.collisionpublishing.com

Figure 10: Vertical acceleration (Gz) driver’s seat surface for Wheel separation test 1. 100 Hz filtered data.

Figure 11: Axial acceleration (Gx) floorboard under the driver’s seat for Wheel separation test 5. 100 Hz filtered data.

Figure 12: Lateral acceleration (Gy) floorboard under the driver’s seat for Wheel separation test 5. 100 Hz filtered data.

Figure 13: Vertical acceleration (Gz) floorboard under the driver’s seat for Wheel separation test 5. 100 Hz filtered data.

Figure 14: Axial acceleration (Gx) driver’s seat surface for Wheel separation test 5.

5 Hz filtered data.

Figure 15: Lateral acceleration (Gy) driver’s seat surface for Wheel separation test 5.

5 Hz filtered data.

Figure 16: Vertical acceleration (Gz) driver’s seat surface for Wheel separation test 5.

5 Hz filtered data.

Figure 17: Vertical acceleration (Gz) driver’s seat surface for Wheel separation test 5.

100 Hz filtered data.

www.collisionpublishing.com

Collision Magazine - Volume 9 Issue 2 75

Figure 18: Axial acceleration (Gx) floorboard under the driver’s seat for Wheel stop test 11e.

100 Hz filtered data.

Figure 19: Lateral acceleration (Gy) floorboard under the driver’s seat for Wheel stop test 11e.

100 Hz filtered data.

Figure 20: Vertical acceleration (Gz) floorboard under the driver’s seat for Wheel stop test 11e.

100 Hz filtered data.

Figure 21: Axial acceleration (Gx) driver’s seat surface for Wheel stop test 11e.

5 Hz filtered data.

Figure 22: Lateral acceleration (Gy) driver’s seat surface for Wheel stop test 11e.

5 Hz filtered data.

Figure 23: Vertical acceleration (Gz) driver’s seat surface for Wheel stop test 11e.

5 Hz filtered data.

Figure 24: Vertical acceleration (Gz) driver’s seat surface for Wheel separation test 11e.

100 Hz filtered data.

Figure 25: Axial acceleration (Gx) floorboard under the driver’s seat for Speed Bump test 12v.

100 Hz filtered data.

76

Collision Magazine - Volume 9 Issue 2

www.collisionpublishing.com

Figure 26: Lateral acceleration (Gy) floorboard under the driver’s seat for Speed Bump test 12v.

100 Hz filtered data.

Figure 27: Vertical acceleration (Gz) floorboard under the driver’s seat for Speed Bump test 12v.

100 Hz filtered data.

Figure 28: Axial acceleration (Gx) driver’s seat surface for Speed Bump test 12v.

5 Hz filtered data.

Figure 29: Lateral acceleration (Gy) driver’s seat surface for Speed Bump test 12v.

5 Hz filtered data.

Figure 30: Vertical acceleration (Gz) driver’s seat surface for Speed Bump test 12v.

5 Hz filtered data.

Figure 31: Vertical acceleration (Gz) driver’s seat surface for Speed Bump test 12v.

100 Hz filtered data.

www.collisionpublishing.com

Collision Magazine - Volume 9 Issue 2 77

ed

ew evi

CDR Report Data from Vehicles Subject to the GM Ignition Switch Recall with the “Epsilon” ACM

r Pee

R

W. R. Rusty Haight Collision Safety Institute

80

Collision Magazine - Volume 9 Issue 2

www.collisionpublishing.com

I

ntroduction Crash data retrieved from a vehicle which is identified as part of the “GM ignition switch recall” may contain data parameters which may be associated with the type of failure detailed in the recall.1 The focus of this article is on identifying those parameters, understanding their impact on the remaining data elements which may be retrieved and potentially eliminating data which is not associated with the ignition switch failure.

This article will explore how the ignition switch function is related to the data stored in a specific airbag control module found in those vehicles associated with the recall and what the “normal” versus “abnormal” data states are for some of the data elements retrieved from the identified vehicles in the recall group. This article is not intended to rehash all the underlying issues with the ignition switch as described in the recall. For the purposes of this narrative, one should assume that the description of the ignition switch “issues” are discussed sufficiently in the recall and elsewhere including the Valukas report 2 and press releases 3 and hearing information from the Energy and Commerce Committee of the US House of Representatives, 113th Congress. The focus here is on the interpretation and use of data from a subset of the vehicles (more specifically, data from the airbag control module type found in those vehicles) associated with the recall in a crash reconstruction context.

In the preparation of this article, the author has relied on publically available information such as the “document binder” materials from the Energy and Commerce Committee 4 and resources/references listed in those documents, crash tests conducted at the 2014 ARC-CSI Crash Conference,5 the NHTSA Recall documents, and a review of more than 200 CDR data files from a variety of vehicles/modules.

S

cope Six vehicle models (some are sisters/ clones) are identified in the NHTSA Recall. They are:

•

2005-10 Chevrolet Cobalt

• 2006-11 Chevrolet HHR • 2007-10 Pontiac G5 • 2006-10 Pontiac Solstice • 2007-10 Saturn Sky • 2003-07 Saturn Ion Of these, the Saturn Ion was equipped with a different Airbag Control Module (ACM) than that found in the remaining vehicle models. The Ion ACM data set does not include all the same data elements found in the ACM installed in the remaining vehicles and an analysis of data from an Ion’s ACM should be handled differently than that recovered from the remaining vehicles/modules identified in the recall. By way of further identification in this respect, we might turn to two sources of information regarding those vehicles and the ACM type(s).

First, the Bosch CDR Tool Help file 6 can be used to identify the cable which would be used if one were to image data from an ACM found in an Ion or one of the other listed vehicles when connected “direct-to-module.” The direct-to-module cable for the Ion is identified in the Help File as cable number 02003004. The direct-to-module cable number called for in the Help File for the remaining listed vehicles is the 02003321 cable. The Help File cable number can be thought of as generally identifying a “family” of ACMs which may be found in a variety of vehicles.

www.collisionpublishing.com

Collision Magazine - Volume 9 Issue 2 81

Here, for example, the Cobalt, HHR, G5, Solstice and Sky all use the same type of ACM; however, the calibration or setup of the ACM for a specific vehicle model within that group may not be the same as that found in another model. The ACM accessible using the 0200321 cable is found listed for other GM vehicles which, notably, are not part of the ignition switch recall. One should also recognize that the “ignition switch recall” is not a recall of the ACM identified as having been installed in any of those vehicles.

In short, as will be discussed in more detail later, the issue here does not have to do with the ACM but rather the separate ignition switch found in the identified vehicles.

Secondly, we might consider the “Data Limitations Version Identifier Code” found at the end of the Data Limitations text of Bosch CDR Tool reports. For the identified Cobalt, HHR, G5, Solstice and Sky vehicles, that code is 01016_SDMEps_r0066. Generally, the module type found in these vehicles can be referred to as the “Epsilon” (where “SDM” has long been understood to stand for Sensing Diagnostic Module, the GM-specific name for the airbag control module). By comparison, the module found in the Ion is identified in the Data Limitations text as the 01014_SDMDW_r0046; generally referred to as the “DW.” The Epsilon module was made by the supplier Siemens AG which was acquired by another supplier - Continental AG - in 2007. The “DW” was built by Delphi, a different supplier 7 altogether.

There are specific elements and processes discussed in several references 4 which apply to the Epsilon module but do not apply to the DW and which were examined, at least in part, at the 2014 ARC-CSI Crash Conference in testing. For this reason, the scope of this article is limited to the crash data found in a “CDR Report” for data retrieved from an Epsilon 8 module which may be found in a vehicle subject to the recall.

T

he “Switch Issue” To more fully understand the data which might be retrieved from vehicles subject to the recall 8 it may be useful to understand some of the otherwise unique attributes of these vehicles as they relate to ignition, power and bus communication and the corresponding behavior of the ACM. Notably, many of these characteristics are shared by other GM vehicles and module types as well as other modules used by and vehicles made by other manufacturers. Moreover, it is important to keep in mind throughout this discussion that the Epsilon ACM is not the subject of the recall.

82

Collision Magazine - Volume 9 Issue 2

In the Bosch CDR Tool Help File, the first reference we see to a vehicle/module which would call for the cable number 02003321 for direct-to-module imaging is the 2004 Malibu (Maxx); a vehicle not included in the underlying recall.1 CDR Tool reports for the 04 Malibu indicate it has an Epsilon module (although some of the data parameters we find in later incarnations aren’t present in reports for the 04 Malibu). It would appear that the 04 Malibu is the first time the Epsilon module was used in an accessible GM vehicle. That is also the first time we see a reference to “Note 6” in the CDR Tool Help File.6 “Note 6” (specifically for GM vehicles) reads: “A key is not required to image these vehicles through the DLC, so you can leave ignition off or on during imaging.” CDR Tool users have come to understand that while the ACM remains installed in the vehicle and there is sufficient invehicle power, one need not necessarily turn the ignition key to the “run” position to establish communication with and image a module by way of the communication bus access at/through the Diagnostic Link Connector (DLC or OBD II port).9 Prior to the module introduced in the 2004 Malibu, to access an ACM’s Event Data Recorder (EDR) function for data retrieval through the DLC, a given module had to be powered up 10 by turning the ignition key to the “run” position. Starting with the 2004 Malibu and continuing on through most recent model GM vehicles, the EDR function in ACMs may typically be accessed without the ignition key. To permit this, these newer ACMs have what might be generally compared to a PC’s “wake up on LAN” feature. When the ACM has available power from the vehicle (meaning the battery is connected and there’s sufficient voltage) and the communication bus is intact such that messages sent on the GMLAN 11 can be received by the ACM, the ACM will respond to a message broadcast from the CDR Tool software to “wake up” that ACM before retrieving an EDR record(s).

Figure 1 graphically illustrates the concept that, with the ignition in the “off/lock” position, if there is power from the vehicle to the module, the ACM’s EDR function can be accessed by way of the GMLAN allowing data to be imaged using the Bosch CDR Tool. For what might be described as “Non-Note 6” GM vehicles, the ignition would have to be in the “run” position to image the module through the DLC. This might be confusing to some familiar with a CDR Tool report from this type of GM vehicle where, in the Data Limitations, one finds this entry:

www.collisionpublishing.com

Note 6: A key is not required to image these vehicles through the DLC, so you can leave ignition off or on during imaging.

ACC

Off/Lock Accessory Run Start (“Crank”)

(Bosch CDR Help File)

+ Battery

SDM

+

+

EDR function available when there's sufficient in-car power available.

Fuse Figure 1: With the ignition in the “off/lock” position, if there is power from the vehicle to the module, the ACM’s EDR function can be accessed by way of the GMLAN allowing data to be imaged using the Bosch CDR Tool.

“The ignition cycle counter relies upon the transitions through OFF->RUN->CRANK power moding messages, on the GMLAN communication bus, to increment the counter. Applying and removing of battery power to the module will not increment the ignition counter.”6 What this means is that when imaging is done after a crash event either “direct-to-module” where the module would be powered up without the benefit of the power-moding message from the Body Processor Module (“BPM” also referred to elsewhere and in most GM documentation as the Body Control Module or BCM) or where the imaging is through the DLC but the ignition had not been cycled prior to the start of the imaging process, the CDR Tool Report parameters “Ignition Cycles At Investigation” and “Ignition Cycles At Event” would be the same (Figure 10).

To be clear, the passage above from the Data Limitations text for this type of module specifically addresses the method by which the ignition cycle counter is increased (or incremented “up”) for the data parameter “Ignition Cycles At Investigation.” The Data Limitations entry does not mention anything about the ignition switch relative to the actual data imaging and, of course, the Data Limitations text wouldn’t normally be seen until after imaging was completed (whether or not there were event(s) recovered) so such mention would be redundant at that point.

However, this suggests that there might “something else” going on in the vehicle/module when the actual ignition is cycled from “off” through “run” to “crank (start)” which is

different than when the ACM’s EDR function is accessed by way of the GMLAN with a message from the CDR Tool software while the ignition switch is not in the “run” position which would not, of course, include a “cycling of the ignition (switch).”

That “something” seems to be identified in one of the documents submitted to the US House Energy and Commerce Committee found in the “Document Binder”4 as “Document 21.” In the “Binder Index,” it is referred to as the “Continental Automotive Systems US, Inc. Field Event Analysis Report GMHEC000003129 3142” (document) and herein as the “Continental Report” 12 from 2008. Here is where the previous mention of the Epsilon module supplier(s) makes further reading clear. In another document in the Committee binder we find a report from 2005 which will be referred to as the “Siemens Report.”13 To minimize confusion, one should recall that Siemens was the original supplier of the Epsilon module and Siemens was acquired by Continental around 2008. For that reason, the report formats and content are nearly identical but for the masthead.

In the “Continental Report,” imaging using the Bosch CDR Tool isn’t separately discussed or even seemingly anticipated. Instead, that report addresses the various messages from the BPM based on the cycling of the actual ignition switch and a response to that action by the BPM in terms of broadcast messages on the GMLAN. Figure 2 graphically illustrates what the Continental Report tells us about the process such that, while the ACM would be

www.collisionpublishing.com

Collision Magazine - Volume 9 Issue 2 83

W hen the ignition is transitioned from " lock" through " accessory to " run" the BP M sends a power mode message on the G MLAN. W hen the ACM detects that power mode message, it initializ es it’ s crash sensing functions.

Signal

BP M

+

Signal

+ Battery

SDM

+

+ Fuse Block

Figure 2: The process to initialize crash sensing functions drawing power from the vehicle’s electrical system, to initialize crash sensing functions - as opposed to the EDR function alone - the ACM (SDM) requires a specific Ignition Logic Level (ILL) message from the BPM on the GMLAN indicating the ignition has passed through the “run” position.

The Continental Report also indicates that when the ignition is cycled out of “run” through “accessory,” the BPM broadcasts a message which initiates a shutdown process at the ACM first disabling the Electronic Crash Sensors (ECS) (Figure 3) and then disabling the ACM’s deployment algorithms (front and side) all within about 60-150ms. For some time after that, as the energy reserve in the ACM draws down and the capacitors begin to discharge,14 the ACM’s “X” and “Y” accelerometers remain active such that acceleration (and later delta-V) might be captured then recorded particularly since the EDR function remains similarly active. Both the Continental and the Siemens reports suggest this draw down time could approach 4.5sec from the initiation of the signal; however, no specific time is offered.

For example, in the Continental report, one finds a discussion of the behavior of the SDM (ACM) where, in testing, Continental injected various signals and created a “loss of battery” at various times relative to EDR “Event Enable” or what is often referred to in the “CDR vernacular” as “algorithm enable (AE).” Here, again, overlapping termi84

Collision Magazine - Volume 9 Issue 2

nology may lead to some confusion. Where the seasoned CDR Tool user might be familiar with the term “Algorithm Enable (AE)” in the context of identifying a point in time where the “wake up” or “trigger” threshold for a given GM ACM has been met and the deployment decision making process has begun, we find in the Continental and Siemens Reports a description of a point where “algorithms (are) enabled” (or disabled). In that context, in those reports, the use is not the same. “Algorithms enabled” in these reports should not be confused with the CDR Tool use of the term “algorithm enable.” In fact, in the Continental report, the author(s) use the term “Event Enable” to indicate the point where the “wake up” or “trigger” threshold (“AE” in the CDR Tool vocabulary) for the ACM’s decision making process has been reached rather than the CDR Tool user’s more typical understanding of “algorithm enable,” a convention, for clarity, which is adopted for the remainder of this article.

In contrast, in the Siemens and Continental Reports, the terms “algorithm disable flag” and “disable the deployment algorithm” are used to indicate that the ACM is set to a condition where, by way of analogy, we might find that the ACM “believes” it is supposed to power down as a function of the signal from the BPM indicating the vehicle has been affirmatively shut off by the driver. Again, by way of analogy, it would be as if the AMC is “thinking” that “the car’s in the garage, the driver’s turned off the car so we can shut everything down now;” a behavior described in

www.collisionpublishing.com

W hen the ignition switch is transitioned out of the " run" position through " accessory," the BP M sends a power mode message on the G MLAN. W hen the ACM detects that message, it begins to shut down the deployment algorithm capability but the EDR functions remain active.

Signal

BP M

+

Signal

+ Battery

SDM

+

+ Fuse Block

Figure 3: When the ignition is cycled out of “run” through “accessory,” the BPM broadcasts a message which initiates a shutdown process at the ACM first disabling the Electronic Crash Sensors (ECS) and then disabling the ACM’s deployment algorithms the Valukas 2 report as “...a deliberate and sensible decision made to prevent passengers from being injured by airbags in parked cars. ...” This general logic is actually common to other ACMs used by other manufacturers and is a reasoned response to the use case of the ignition affirmatively being switched by the driver from the “run” position through “accessory” to “off.”

down” process, unless “Event Enable” has already occurred, the ACM will not command the deployment of supplemental restraints (airbags/safety belt pretensioners) because the deployment algorithms have been disabled. In the anticipated use case of the shutdown process initiated by the driver affirmatively turning the key from the “run” position, understandably this wouldn’t “be a problem.”

The problem, as it relates to the ignition switch identified by way of “the recall,” is that the switch may, in the right conditions, unexpectedly move out of the “run” position essentially creating a BPM message condition which the ACM is programmed to accept and act on in a specific way but which is occurring in an otherwise unanticipated scenario. In short, the idea that the ignition key being affirmatively turned to “off” by a driver would reasonably begin the process of the ACM power down is anticipated if not essentially an industry standard. However, the unanticipated condition occurring in the case of the vehicles equipped with the recalled switch is that the switch failure leads to an anticipated message being sent at an inappropriate time.

Where the key moves out of the “run” position and the vehicle’s not “parked in the driveway” but instead headed down the road “toward a crash,” the result is the effect noted in the recall: crashes without supplemental restraint system deployments where those deployments might otherwise be anticipated. Where the switch moves from the “run” position as the vehicle is moving down the road and the shutdown process has begun, the deployment algorithms cannot be re-enabled for at least another 2.5sec assuming the driver recognizes the problem and turns the key back to the “run” position to start that process.

In summary, this means that about 150ms after the ACM receives the message from the PBM initiating the “shut

In the Continental Report, there is a time line offered as part of the an analysis of a path toward disabling deployment algorithms as it relates to the “SDM Power Mode” state. The Continental Report also lays out a time line with

www.collisionpublishing.com

Collision Magazine - Volume 9 Issue 2 85

respect to the deployment algorithms being re-enabled.

Here, Figure 4 brings both of those occurrences together.

Read from left-to-right, Figure 4 is a time line which starts out suggesting the vehicle’s been started and the ignition is in the “run” position. At that point, the ACM Power Mode state is reported as “RUN.” Assuming, by way of example, the vehicle is being driven down the road for some time more than 5 seconds, what is identified later as the “pre-crash” data is being captured in the ACM in a circular buffer in the ACM. At some random point in time, the ignition switch is moved from the “run” position to/through the “accessory” position. At about the same time, subject to latency on the GMLAN, the BPM broadcasts a power mode message indicating the power down as a function of the ignition position. At that point, systems without reserve energy such as the Powertrain Control Module (PCM), Transmission Control Module (TCM) and Automatic Occupant Sensing (AOS) power down and most meaningful messages from those systems (whether or not on the GMLAN) are effectively lost. The ACM sees this message and “... immediately initiates power down sequence by shutting off the Electronic Crash Sensor (ECS) and dropping its

“

... as is so often the case when one reviews a CDR Tool Report, one cannot work in a vacuum: the whole report should be considered carefully ...

”

86 Collision Magazine - Volume 9 Issue 2

internal phase to a lower phase (but not completely off).

In this phase, the X and Y sensors are still active, as is the EDR. ...”15 In this “lower phase,” the ACM’s deployment algorithms are disabled and, after about 150ms, a deployment will not/cannot be commanded; however, since the X and Y sensors and EDR function are still active for some time, the ACM can detect and may even record a crash pulse should there be an event.

The time line (as seen in Figure 4) is such that if the ignition switch were to be returned to the “run” position, the BMP would broadcast a power mode message which, when received by the ACM, would result in the ECS being re-initialized after about 600ms and, after an additional 1900ms (or more), the deployment algorithms would be re-enabled.

As noted, were a crash to occur when the ACM was in a powered down state, certain data parameters may be recorded while the EDR is still active and the crash reconstructionist with at least this rudimentary understanding of the process should be in a better position to evaluate the CDR Tool Report elements as described later in this narrative.

D

river Experience To a lay driver, if the ignition were to be moved from the “run” position, the driver’s experience, their perception, would most likely be seen as if the vehicle were “stalling:” the dash malfunction indicator lamps would illuminate and there would be a loss of vehicle power like the engine had been shut off. Of course, essentially that would be the case or at least be consistent with observation. In fact, many of the customer complaints received by GM and complaints received by NHTSA prior to the recall were just that: of a vehicle stalling (also referred to as a “moving stall”) without apparent warning. The Valukas 2 report discuses the idea that one of the reasons the ignition switch issue wasn’t discovered earlier was that driver reports/complaints (chiefly those other than those involved with crashes) were treated as “stalling” issues which were almost impossible to replicate from a dealer/service perspective lacking some clue(s) that the problem would be part of a component used/built in much the same way for years without problem.

While the perception of “stalling” may be useful in evaluating parameters found in the CDR Tool Report data it demonstrably didn’t make a potential defect investigation any easier whether that was being done by GM or the National Highway Traffic Safety Administration (NHTSA).

www.collisionpublishing.com

I gnition Switch P osition Run

G MLAN message from Body P rocessor Module is updated to “Run”

60 0 ms

At the SDM, the “V ehicle P ower Mode Status” SDM P ower Mode changes to “Run” and Run “Run/Crank I gnition Switch Logic Level” shows “Active”

“T on”

I gnition switch changed from ACC to Run I gnition Switch P osition Run

I gnition Switch P osition Accessory

SDM P ower Mode “Accessory” Deployment Algorithm disabled

Any T ime Electronic Crash Sensors (ECS) are re-initializ ed and sending data SDM Deployment Algorithm Enabled

1 9 0 0 + ms

“Vehicle Power Mode Status”

I gnition switch changed from Run to Accessory I gnition Switch P osition Accessory

SDM P ower Mode Run At the SDM, the “V ehicle P ower G MLAN message Mode Status” from Body P rocessor changes from Module is updated “Run” to “Accessory” from “Run” to and “Run/Crank “Accessory” and I gnition Switch the SDM starts Logic Level” shows shutting down “I nactive”

Approx . 1 5 0 ms

These data parameters; however, are observed in another similar module which is used in GM vehicles which are not part of the ignition switch recall but which were manufactured during the same time frame. In what is identified (as discussed previously) as the “SDMC” supplied by either/both Delphi and Autoliv, the “Vehicle Power Mode Status” and “Run/Crank Ignition Switch Logic Level” parameters are reported as in Figure 6. No CDR Tool Reports from SDMC ACMs have been discovered where the “Vehicle Power Mode Status” or “Run/Crank Ignition Switch Logic Level” were anything other than “Run” and “Active” respectively.

For the Epsilon ACM, the recorded element “Vehicle Power Mode Status” may be observed in CDR Tool Reports as “Run,” (Figure 5) “Accessory” (Figure 7) or “Off” (Figure 8). This reported status is a reflection of that status broadcast by the BPM on the GMLAN and considerations including message latency and whether or not the

T ime (relative to SDM)

“T off”

The discussion that follows address some of the data parameters found in the CDR Tool Reports for the Epsilon ACM and how they might be evaluated by the reconstructionist. For example, the Epsilon ACM was one of the first to include data elements associated with the “Vehicle Power Mode Status” and “Run/Crank Ignition Switch Logic Level” as seen in Figure 5. These two data parameters are obviously useful in evaluating what the “observed” ignition state was at the ACM at “Event Enable” (as it is referred to in the Continental Report) while in Figure 5 reported as part of the “System Status at AE (Algorithm Enable).” Notably these are not parameters found as part of the data elements reported in the DW module found in the Saturn Ion.

Any T ime

The author has attempted to compare the data from those reports to the information found in the Continental and Siemens Reports and identify commonalities, trends or similarities that might be useful to the crash analyst in a scenario where they are trying to evaluate, against the background of a crash, the data retrieved from the ACM in one of the recalled vehicles.

EDR Function Active (assuming power available from battery or capacitors)

D

ata Elements from the Epsilon CDR Report As part of this analysis, more than 200 CDR Tool Reports for/from a variety of GM vehicles were reviewed for similarities or commonalities and a series of patterns emerged. Some of these CDR Tool Reports came from vehicles used in testing conducted by the Collision Safety Institute at the 2014 ARC-CSI Crash Conference and elsewhere.

Figure 4: Time line with respect to the deployment algorithms being reenabled

www.collisionpublishing.com

Collision Magazine - Volume 9 Issue 2 87

Figure 5: Data elements associated with the “Vehicle Power Mode Status” and “Run/Crank Ignition Switch Logic Level”

Figure 6: “Vehicle Power Mode Status” and “Run/Crank Ignition Switch Logic Level”parameters from the “SDMC”

Figure 7: “Vehicle Power Mode Status” observed in CDR Tool Reports as “Accessory”

Figure 8: “Vehicle Power Mode Status” observed in CDR Tool Reports as “Off”

88 Collision Magazine - Volume 9 Issue 2

www.collisionpublishing.com

event was completely recorded may effect the reliability of this particular parameter as reported.

“Run/Crank Ignition Switch Logic Level” The parameter “Run/Crank Ignition Switch Logic Level” as observed in Epsilon ACM reports is reported as either “Active” (Figure 5) or “Inactive” (Figures 7 and 8). Associated with the “Vehicle Power Mode Status,” this reported status is also a reflection of that broadcast by the BPM on the GMLAN and considerations including message latency and whether or not the event was completely recorded may effect the reliability of this particular parameter as reported. An examination of CDR Tool Reports together with the time line offered in the Continental Report, this message appears to be the first of these two elements received or captured by the ACM when the shut down process begins.

Referring back to the time line described in Figure 4, in normal operation, the “Vehicle Power Mode Status” should be recorded as “Run” and the “Run/Crank Ignition Switch Logic Level” as “Active.” When the BPM gets the signal from the ignition switch that the position has changed out of “Run,” the “Run/Crank Ignition Switch Logic Level” will change to “Inactive.”

When the ignition switch position moves out of the “run” position, the next position would be “accessory” (Figures 1 and 3). In most of the field examples reviewed for this article, the reported status, when it wasn’t “Run” was “Accessory” but for a small group of the reports reviewed.

In those few examples, three commonalities emerge: (1) where the “Run/Crank Ignition Switch Logic Level” is observed as “Inactive,” (2) the “Vehicle Power Mode Status” is reported as “Off” (Figure 8) and (3) the vehicles associated in these reports were all manual transmission vehicles. Another data parameter is identified or associated with manual transmission vehicles as discussed later in this article.

“Events Recovered” In the File Information data table (Figure 9) is found on the first page of all Bosch CDR Tool reports. For reports based on data from the Epsilon module, the last parameter indicates the “Events Recovered” - the name(s) of or type(s) of event(s) retrieved from the imaged module using the CDR Tool. For this module, the normal event names would be “Non-Deployment,” “Deployment,” and “Deployment #2.” Described in the Data Limitations16 text for this module, the ACM would record and the report would reflect up to two recorded events (either “Deployment” and “Non-Deployment” or “Deployment” and “Deploy-

Figure 9: File Information data table is found on the first page of all Bosch CDR Tool reports ment #2”) or a single event limited to a “Non-Deployment” or a “Deployment.”

Where the recalled vehicle ignition switch has moved from “Run” and an event occurs when the deployment algorithms have been disabled, the “Events Recovered” will always be reported as a “Non-Deployment.” Referring back to the time line (Figure 4) unless Event Enable has occurred before the deployment algorithms are disabled, a deployment cannot be commanded and, for that reason, regardless of the crash severity, when an event is recorded, it will be identified as a “Non-Deployment” (event).

At this juncture, where one were to see a “Non-Deployment” recorded after an event and find the “Vehicle Power Mode Status” showing “Accessory” (or “Off”) and “Run/ Crank Ignition Switch Logic Level” showing “Inactive,” the reasonable conclusion might be that, for whatever reason, the ignition switch had moved from the “Run” position prior to the event (for this ACM in the recall vehicles).

While this would often be the case, there are other parameters identified in the review of CDR Tool Reports and the Continental Report which should also be reviewed to more fully evaluate the scenario.

While the “Events Recovered,” “Vehicle Power Mode Status” and “Run/Crank Ignition Switch Logic Level” parameters are clearly meaningful in the evaluation of these scenarios, as is so often the case when one reviews a CDR Tool Report, one cannot work in a vacuum: the whole report should be considered carefully. There are often other recorded parameters found in a CDR Tool Report which should be considered in an overall, thorough evaluation of the retrieved data and its application to a situationally complete analysis17,18,19.

Such is the case here. The three parameters discussed thus far against the background of the recall and the module behavior time line (Figure 4) make a strong case for an ignition switch failure prior to a crash event; however, there remain other parameters in a CDR Tool Report translated

www.collisionpublishing.com

Collision Magazine - Volume 9 Issue 2 89

(reported) with the latest version of the CDR Tool software which should also be considered.

may not reflect the actual state captured at the time of the event.

“Event Recording Complete”

“Event Recording Complete” (where the translated value is either “yes” or “no”) is found in the data set for the Epsilon module in the “System Status At (Event Name)” data table in the CDR Tool Report (Figure 10)6.

As is the case in virtually any CDR Tool Report, the parameter “Event Recording Complete” (also seen as “Complete Event Recorded” for some systems) is a very useful to data element indicating whether or not there had been a power interruption during the crash event which then lead to a power depletion during the recording process. After the event, where a depletion occurs during that recording process, a power depletion may preclude the complete recording of one or more of the data elements an EDR function is designed to record (conceptually to include here the “Vehicle Power Mode Status” and “Run/Crank Ignition Switch Logic Level” parameters). As such, the retrieved raw data for one or more of those unrecorded elements

“Automatic Passenger SIR Suppression System Validity Status at AE” Also found in the “System Status At (Event Name)” data table in the CDR Tool Report (Figure 10)6, are the parameters “Automatic Passenger SIR Suppression System Validity Status at AE” and the associated parameter “Automatic Passenger SIR Suppression System Status at AE.” These describe conditions associated with the (right front) passenger Automatic Occupant Sensing (AOS) system; a

Figure 10: “System Status At Non-Deployment” data table in the CDR Tool Report 90 Collision Magazine - Volume 9 Issue 2

www.collisionpublishing.com

system designed to detect an occupant in the (right front) passenger’s seat and enable or disable the passenger airbag.

Not all of the vehicles identified in the recall have an AOS (a system that was introduced into this line of GM vehicles in about 2005). Where the system is not present on a given GM vehicle, the Bosch CDR Tool Report will not include these parameters (compare Figures 10 and 11).

Where those parameters are present, the reported values, particularly the validity status, is significant in terms of evaluating the actual power state in a given vehicle at Event Enable.

The AOS doesn’t have an energy reserve so, when the power is “cut” (where the key is moved from “run” as previously discussed), the power to the various vehicle systems is shut off. At that point, since the AOS is wired directly to the ACM, there is a near instantaneous indication that the status of the AOS is “invalid.” Given that the AOS is wired directly to the ACM, there is virtually no latency in that reporting compared to the potential latency anticipated with respect to the BPM broadcasting on the GMLAN.

This observation is consistent with the CDR Tool Reports reviewed during the research for this Article as well as the information contained in the Continental Report. Where Continental injected a “loss of battery” to the system(s) during their research at various times relative to an Event Enable signal, they found that “... the Suppression System behavior provides further evidence that the vehicle experienced a Loss of Battery prior to impact. ...”.15 In one field example (which may be complicated by other conditions not yet fully evaluated in that case) the timing of the “Automatic Passenger SIR Suppression System Validity Status” signal versus the reported values for the “Vehicle Power Mode Status” and “Run/Crank Ignition Switch Logic Level” parameters further supports this observation (Figures 12 and 13). Here (Figure 12) we see that the “Vehicle Power Mode Status” is “Run” but the “Run/ Crank Ignition Switch Logic Level” is “Inactive.” Keeping in mind these are based on messages broadcast by the BPM on the GMLAN, they might be contemplated together in light of the “Automatic Passenger SIR Suppression System Validity Status” showing “Invalid” at Event Enable.

Figure 11: “System Status At Non-Deployment” data table in the CDR Tool Report

Figure 12: “Vehicle Power Mode Status” is “Run” but the “Run/Crank Ignition Switch Logic Level” is “Inactive”

Figure 13: “System Status At Non-Deployment” data table in the CDR Tool Report www.collisionpublishing.com

Collision Magazine - Volume 9 Issue 2 91

“Vehicle Event Data (Pre Crash) Associated With This Event” Found in the “System Status At (Event Name)” data table in the CDR Tool Report (Figure 10)6, this is a “yes” or “no” parameter indicating whether or not the Event Data page of the CDR Tool Report is associated with an identified event. The Epsilon ACM will only record one set of “pre-crash” data even if there were two individual events recorded. The analyst would look to the translated “yes” or “no” value for this parameter found in the “System Status at (Event Name)” data table to evaluate whether or not the “pre-crash” data is associated with that named event.

This parameter is useful toward a further evaluation of the recorded pre-crash data elements by way of determining whether or not they’re associated (related) to the recorded “Non-Deployment” in the case used case anticipated in this article.

“Vehicle Speed” Found in the “Pre-Crash Data” of the Epsilon CDR Tool Report, “Vehicle Speed” refers to the Transmission Control Module’s (TCM) reported speed of the vehicle not necessarily the speed of the center of mass of the vehicle.

This calculated value relies on power up sensor(s) sending a message to the TCM on the GMLAN which, at the TCM, is used to calculate a “Vehicle Speed” (also seen in other systems as “Speed, Vehicle Indicated”20).

When those sensors are powered down because the vehicle is powered off such as when the ignition moved from the “run” position through “accessory,” there is no signal to process and the indicated speed is most frequently reported as “zero (0)” (Figure 14). Other pre-crash parameters such as “Engine Speed” (engine RPM) may be reported similarly or (as in Figure 14) may be marked as “invalid.”

An opportunity presents itself here to review the retrieved data in another format unique to GM data sets: the Engineering Translation Report (ETR). In the ETR for the same data set from which Figure 14 was developed, a review of the ETR indicates that there were values captured pre-crash for the “Engine Speed” and, like “Vehicle Speed”

near the crash (nearer to time “-1" as a frame of reference), the “Engine Speed” drops to “zero.” Because the “Engine Speed” has an “Data Invalid” flag set, in the CDR Tool Report translation, in place of numerical values, we find the word “invalid.” Notice here (Figure 15) also that the “Vehicle Speed” values correspond to what we see in the CDR Tool Report AND those values are associated with a “Data Valid” flag.

In particular for certain Cobalts which are subject to the recall, end users of the Bosch CDR Tool have found data sets where the “pre-crash” data parameter “Vehicle Speed” has been marked “Invalid” rather than showing numerical values greater than 0 leading to zeros close to time “-1.”

An example of this is found in Figure 16.

For the moment, let’s call the vehicle from which the data set used to make Figures 14 and 15 vehicle “A” and the vehicle from which the data set used to make Figures 16 and 17 vehicle “B.” Note the differences between Figures 14 and 16 as it relates not only to “Vehicle Speed” but also “Engine Speed” (for these examples). Then consider the content of Figure 15 against that observed in Figure 17. In Figure 15 (and correspondingly in Figure 14 where Figures 14 and 15 are from the same imaged data set) we find the “Vehicle Speed” as a number greater than 0 then “dropping” to 0 for the last samples leading to time “-1” and we see the “Engine Speed” marked as “Invalid.” In contrast, in Figures 16 and 17 (again these two Figures from the same imaged data set) we find “Vehicle Speed” values and the “final” zero BUT we observe the validity status as “Invalid.”

Both vehicles from which these figures are developed were in crashes and were both identified as having experienced a power mode change from “run” prior to Event Enable while the car was in motion. The difference between these two Chevrolet Cobalts is that vehicle “A” has an automatic transmission and vehicle “B” has a manual transmission.

Where the Cobalt has a manual transmission, there is a crash and crash data is recovered, the CDR Tool Report shows the “Vehicle Speed” as “Invalid” without respect to the ignition switch observation/reported state.

Figure 14: Indicated speed is reported as “zero (0)” 92 Collision Magazine - Volume 9 Issue 2

www.collisionpublishing.com

Figure 15: Engineering Translation Report (ETR) showing “Vehicle Speed” values that correspond to what we see in the CDR Tool Report AND those values are associated with a “Data Valid” flag

Figure 16: “Pre-crash” data parameter “Vehicle Speed” has been marked “Invalid” rather than showing numerical values greater than 0 leading to zeros close to time “-1”

Figure 17: Engineering Translation Report (ETR) showing "Engine Speed" and “Vehicle Speed” values

www.collisionpublishing.com

Collision Magazine - Volume 9 Issue 2 93

T

esting to Verify Observations In addition to the review of the previously described documents (i.e.: the “Continental Report”) and other CDR Tool Reports for vehicles with the Epsilon ACM, two 2005 Cobalts were purchased by Collision Publishing specifically for the purpose of conducting a series of crash tests to examine, in a controlled setting, the data retrieved when the ignition switch moves from the “run” position. A third Cobalt was obtained by the Major Crash Assistance Team (MCAT) - Lake County (IL) together with Dynamic Safety, LLC for use in a crash test as part of MCAT member in-service training and where the same sort of scenario could be created the controlled environment of a crash test.

In the interest of full disclosure, all three of the Cobalts purchased for this testing were obtained “on the street” from private parties - not through a dealer, not through GM, but from private parties specifically for the test series described. The two Cobalts used at the 2014 ARC-CSI Crash Conference were driven in various conditions before the testing described here including one which was driven without incident from Southern California to Las Vegas for the 2014 ARC-CSI Crash Conference. One of the Crash Conference cars was sold to the author’s team with the recall notice the seller had only recently received from GM (Figure 18) although the owner had not taken the car in for the recall repair(s). At no time prior to the affirmative switch position changes which are part of the testing described here were the ignition switches observed to move from “run,” “fail” or otherwise appear to malfunction on any of the cars.

The two Crash Conference Cobalts were both 2005 model year vehicles, both had automatic transmissions, one was a four door the other a two door model. As part of the crash data developed at the 2014 ARC-CSI Crash Conference, there are photos, videos, data files .CDRx files, and drawings of these two vehicles and the crash tests they were used in included with this issue of Collision on the DVD. The data from the third 2005 Cobalt (used by MCAT) is part of the in-service training which will be described more fully in a later issue of Collision.

The previous sections of this article dealt with the interface between the ACM and the other components of the car as it relates to the CDR Tool Report data and then with the data parameters one should consider when examining a CDR Tool Report for a vehicle covered under the recall after it has been involved in a crash. To examine these areas together, in a controlled setting, Collision Publishing obtained the two 2005 Cobalts previously described and included them in the 2014 ARC-CSI Crash Conference

94

Collision Magazine - Volume 9 Issue 2

Figure 18: Example GM recall notice designing certain aspects of the Conference test series to highlight the data parameters previously described. The same can be said, of course, for the MCAT Cobalt.

The data from the two Crash Conference Cobalts may be identified by VIN and are referenced by the color of the car to match the photos and videos on the DVD accompanying this issue of Collision. The first of the vehicles (the “white Cobalt”) can be identified in the data set by the last six digits of the VIN “605718.” The second (the “blue Cobalt”) can be identified by the last six digits of the VIN “785998.” The “white Cobalt” was used in a barrier test (identified in the test series as “Test 6”) and the “blue Cobalt” was used first as part of a “low delta-V” test series and then in a barrier test (identified in the test series as “Test 7”).

To create a condition where the ignition switch moved from “run,” the ARC-CSI Crash Crew took two approaches. In one of the “low delta-V” tests, the human subject in that test was instructed to turn the ignition from “run” through “accessory” before impact. In the remaining tests in that series, the ignition was left in the “run” position.

For the two barrier tests, inasmuch as there was no human subject onboard, the switch was controlled remotely using a switch built by RSR Engineering which, on command about 2-3 seconds pre-impact, turned the switch from the “run” position to the “accessory” position.

In preparation for the barrier tests, the Crash Crew observed that for these two automatic transmission Cobalts, it was impossible to turn the key even forcibly by hand from “run” to “lock” while the car was in gear. It could be turned to the “accessory” position but, unless the car was parked, out of gear, the key wouldn’t turn to “lock.”

www.collisionpublishing.com

The third Cobalt (referred to as the “MCAT Cobalt”) was used in a car-to-motorcycle crash test where the author was the driver of that Cobalt in the test and the ignition was manually moved from the “run” position pre-crash. Similarly, with this automatic transmission Cobalt, the ignition wouldn’t cycle from “run” to “lock” while the car was in gear but would turn to “accessory.”

T

he “Low delta-V Series” In total, during the 2014 ARC-CSI Crash Conference, there were 7 “low delta-V” tests conducted where the blue Cobalt was one of the test vehicles (for purposes of that series, “low delta-V” was defined as a

test where the delta-V of one of the vehicles was less than 10mph (16km/h)).

Table 1 is a summary of the crash tests involving the two Crash Conference Cobalts and the MCAT Cobalt. In Table 1, the first tests (identified by test letter) are from the “low delta-V” test series at the 2014 ARC-CSI Crash Conference: tests “A” through “I” (leaving out tests “B” and “C” which did not involve the Cobalt). Prior to test “A,” the blue Cobalt ACM had no recorded events (“Event(s) Recovered” = “None”). The first two, tests “A” and “D,” were conducted such that the delta-V captured by Crash Crew instrumentation was, in both cases only about 4mph

Blue Cobalt for Tests A-I Test ID:

A

D

E

F

G

H

I

White Cobalt test 6

Blue Cobalt test 7

"MCAT" Cobalt

(New) "Event Recovered"

NONE

NONE

ND

NONE

NONE

ND

ND

ND

ND

ND

“Maximum SDM Recorded Velocity Change (MPH)”

5.84mph (9.39 km/h)

7.79mph (12.53 km/h)

8.44mph (-13.58 km/h)

-33.1mph (-53.26 km/h)

-32.47mph (-52.25 km/h)

-5.88mph (-9.46 km/h)

“SDM Longitudinal Axis Recorded Velocity Change (MPH)” (max obs)

6.1mph (9.8 km/h)

-8.13mph (-13.08 km/h)

-8.81mph (-14.17 km/h)

-34.5mph (-55.52 km/h)

-33.8mph (-54.39 km/h)

-6.1mph (-9.81)

-8mph (-12.9 km/h)

-8.4mph (-13.5 km/h)

-34.01mph (-54.7 km/h)

-33.6mph (-54.07 km/h)

-7.8mph (-12.5 km/h)

Crash Crew control instrument delta-V

4mph (6.43 km/h)

4mph (6.43 km/h)

7.1mph (11.4 km/h)

6.6mph 5.1mph (10.6 (8.2 km/h) km/h)

“Vehicle Power Mode Status”

Run

Accessory

Run

Accessory

Accessory

Accessory

“Run/Crank Ignition Switch Logic Level”

Active

Inactive

Active

Inactive

Inactive

Inactive

“Automatic Passenger SIR Suppression System Validity Status at AE”

Valid

Invalid

Valid

Invalid

Invalid

NOT LISTED

“Automatic Passenger SIR Suppression System Status at AE”

Not Suppressed

Not Suppressed

Not Suppressed

Suppressed

Suppressed

NOT LISTED

SIR Warning Lamp status / On/Off time (sec)

Off / 5090

Off / 5730

Off / 6970

Off / 7990

Off / 9720

Off / 655200

"Vehicle Speed" (-5 to -1)

"0mph"

9, 10, 0, 0, 0

9, 10, 11, 11, 12

21, 27, 0, 0, 0

25, 30, 0, 0, 0

25, 30, 38, 43, 0

Vehicle speed at impact (control instrumentation)

0mph (N/A not in motion)

12mph (19.3 km/h)

12mph (19.3 km/h)

31mph (49.88 km/h)

30.2mph (48.6 km/h)

44mph (71 km/h)

Ignition switch moved from "run" before impact (approx sec)

n/a

approx 2.753.25sec

n/a

approx 2.53.0sec

approx 2.02.5sec

approx 0.751.25sec

Table 1: Test Data Compilation and Comparison www.collisionpublishing.com

Collision Magazine - Volume 9 Issue 2 95

(6.43km/h). In the Data Limitations text6 for the Epsilon module, the end user is informed that “The minimum SDM Recorded Vehicle Velocity Change, that is needed to record a Non Deployment Event, is five MPH.” Notably, this passage doesn’t call for a specified/required polarity or axis, simply a magnitude. For this reason, for tests “A” and “D,” no new event was recorded because the Cobalt didn’t experience a delta-V greater than 5mph (8km/h).

At test “E,” the Cobalt was hit from behind by another car and the control instrumentation showed a +7.1mph (11.4km/h) delta-V and a Non-Deployment event was recorded. Relative to the Data Limitations text, this is consistent with (a) the requirement of a delta-V with a magnitude greater than 5mph but (b) no specified polarity. In this case, the polarity for the X delta-V was positive: rearto-front on the vehicle. For some older GM vehicles, the ACM would only record a negative delta-V. This ACM is not subject to that polarity “restriction.”

What’s also missing is a specified axis relative to qualifying recording magnitude/threshold. A review of other field CDR data involving this ACM and previous crash tests at the ARC-CSI Crash Conference has demonstrated that this ACM type will record data for either the X or Y axis where the delta-V is greater than 5mph as well in a scenario where the combination of the X and Y axis delta-Vs is greater than 5mph. This is an illustration of where, in a Data Limitations text - particularly for a GM set - what it doesn’t say is usually just as important as what it does say.

Going back to Table 1 and examining the data for Test “E,” one would note that the ACM reported two seemingly contradictory values for delta-V. The parameter “Maximum SDM Recorded Velocity Change (MPH)” (from the “System Status at [Event Name]” data table) shows the translated value of 5.84mph but, in the “SDM Longitudinal Axis Recorded Velocity Change (MPH)” we find the maximum value (actually an average value for a 10ms period) is 6.1mph. The observation that these two CDR Tool Report values don’t match each other “better” is not part of this particular study since it’s not identified as related to the ignition switch data but is one which has been observed with reports for this module type before and is found in other tests data in the remainder of Table 1.

While the tests described in Table 1 are all almost exclusively acting on the test vehicle’s X axis it might be tempting to compare an average against the control instrumentation; however, “Maximum SDM Recorded Velocity Change (MPH)” is a calculated value; the combination of what the ACM recorded for the X and Y axis in a given crash inasmuch as the “Maximum SDM Recorded Veloc-

96

Collision Magazine - Volume 9 Issue 2

ity Change (MPH)” is described in the Data Limitations text as 6 : “... Maximum Recorded Vehicle Velocity Change is the maximum square root value of the sum of the squares for the vehicle’s combined “X” and “Y” axis change in velocity. If a CDR Printout user were to calculate resultant velocity change using X and Y axis time history data, the calculated value may be different than the Maximum SDM Recorded Velocity Change parameter value displayed in the CDR report. This is due to the rounding that occurs within the SDM while calculating the Maximum SDM Recorded Velocity Change value. ...”

Using the data from the tests described in Table 1, that approach doesn’t really make a difference in the data disparity and doesn’t account for differences between the two “CDR Printout” reported values. All of the tests in Table 1 are acting almost exclusively on the X axis (relative to the test vehicle). In “low delta-V” tests “E,” “H,” and “I” where there to have been a lateral component greater than 0 that would seemingly make a larger difference as part of a combined value because the X component is relatively small itself; however, all of the Y axis data (for those tests) is “zero” in the “time history data (tables).” With that in mind, one is left to contemplate what data is being used to make the calculation for “Maximum SDM Recorded Velocity Change (MPH)” because, given the data thus far available, it is demonstrably not based on the maximum time line X and Y delta-V values (taken at the same point in time).

Another tempting approach might be to consider comparing the just the X axis time series data from the CDR Tool Report with that from the control instrumentation’s observation on the X axis thereby eliminating the effect of the combination of the X and Y values. As can be seen in Table 1, that doesn’t create a better fit.

One may contemplate the effect of varying sample rates, an offset/latency with respect to Event Enable, smoothing, averaging (of the time history data) and even potential conversions between metric and imperial but none seem to provide a clear, consistent trend. If, for example, the “Maximum SDM Recorded Velocity Change (MPH)” were to be considered the actual maximum single value (for those tests where we only see X axis data greater than zero), and the time history data displayed for 10ms data points are averages, the “maximum” would necessarily be greater than any of the average X axis values reported; however that’s not the case in any of the test data sets shown in Table 1.

www.collisionpublishing.com

As discussed later in this narrative, the time history data found in the CDR Reports is generally more consistent with the delta-V calculated from control instrument acceleration separate and apart from the “Maximum SDM Recorded Velocity Change (MPH)” reported value.

The blue Cobalt was also used in Tests “F” and “G” and we see, for those tests in Table 1, the control instrumentation showed delta-Vs of 6.6 and 5.1mph, respectively. While the qualified for recording based on magnitude alone, these events failed to overwrite the data from Test “E” because, in the Epsilon Data Limitations6 text, we find this passage with respect to a Non-Deployment where one: “...can be overwritten by an event that has a greater SDM recorded vehicle velocity change. This event will be cleared by the SDM, after approximately 250 ignition cycles.”

In the case of Tests “F” and “G” in this series, neither had a delta-V greater than that observed in Test “E” and 250 ignition cycles had not yet passed. So, where the delta-V in Test “E” was 7.1mph, the delta-Vs in Tests “F” and “G” were less than that and they were less than 250 ignition cycles after test “E” and neither were recorded.

Editor's Note

Almost every day there seems to be something in the news about “the GM Ignition Switch Recall.” Whether reported accurately or not, “the cat’s out of the bag:” there has been an issue identified with the ignition switch in certain GM vehicles where, unexpectedly, the car is effectively shut down while in motion because the ignition switch moves out the “run” position. But this article isn’t about the “why” of the recall and isn’t offered to discuss the details of the recall itself, but while necessarily addressing some functional details of the ignition switch, it is focused instead on the impact the switch “issue” has on CDR Tool Reports. What should you be looking for in a CDR Report that might be related to the ignition switch issue?

Part of the research done for the article includes crash tests with two Cobalts which Collision Publishing purchased specifically for this purpose.

As described in the article, they were purchased essentially at random from private sellers, driven to Las Vegas and then used in testing not for the performance of the ignition switch but to get data related specifically with the CDR report parameters described in the accompanying article.

Two things here stand out. First, these Cobalts were driven and used in a number of settings prior to testing including one which was driven from Southern California to Las Vegas to be part of the 2014 ARC-CSI Crash Conference without experiencing anything unusual associated with the ignition. Second, to create the scenario where the ignition moved out of the “run” position for this testing, a special remote control system was developed which would turn the key in the ignition switch from the “run” to “accessory” position, remotely, as the test vehicle moved into a designed crash.

Two of the tests conducted at the 2014 ARC-CSI Crash Conference were barrier tests using the RSR Phantom Drive System and the RSR Ignition Switch Remote Control Device. A third, low delta-V test was conducted where the driver manually turned the key to the “accessory” position. In that regard, then, the entire ARC-CSI Crash Conference Crash Crew contributed to the content of this article, the research and successful conduct of the crash tests as they have done every year and without their inestimable contributions, this research would not be as complete as it is.

As part of the crash data developed at the 2014 ARC-CSI Crash Conference and found on the disk included with this issue of Collision, there are photos, videos, data files and drawings of these two Cobalts and the crash tests they were used in. This data is also available from Collision Publishing when one orders a back issue/second copy of this issue of Collision.

www.collisionpublishing.com

Collision Magazine - Volume 9 Issue 2 97

For the next test, Test “H,” the planned delta-V would exceed that observed in the previous tests and, anticipating the event would overwrite the data from Test “E,” the test vehicle occupant was instructed to turn the ignition toward “off” prior to impact. The subject Cobalt had, as noted, an automatic transmission so the anticipated and later observed position the occupant could turn the key to while the vehicle was in motion, was limited to “accessory.”

On approach, at about 2.5 seconds before impact, the occupant successfully turned the ignition to “accessory” as seen in in-car video and as is reflected in the data seen in Table 1. The CDR report reflects, as seen in Table 1, the “Vehicle Power Mode Status” indicated “accessory” and the “Run/Crank Ignition Switch Logic Level” showed “Inactive.” The translated values for the parameters “Automatic Passenger SIR Suppression System Validity Status at AE” and “Automatic Passenger SIR Suppression System Status at AE” are also noted in Table 1 as “Invalid” and “ Not Suppressed,” respectively.

For these two tests, the planned impact speed was about 30mph (48km/h). The vehicles were driven into the barrier by way of the RSR Engineering remote control system which, as a single unit, controls the vehicle’s accelerator, brakes and steering. For these tests, the plan was to move the ignition from the “run” position about 2 seconds preimpact. To accomplish this, a second system, the RSR Engineering remote control system, was designed to control move the ignition switch was installed on each vehicle (Figure 19).

As anticipated, in Test “H,” the last few samples of vehicle speed are recorded as “zero (0)” and the engine speed marked “invalid.” If one were to review the actual CDR file (found on the DVD associated with this issue of Collision), it would be evident that there were values for engine speed recorded but they were marked “invalid;” an observation consistent with previous field observations.

The final test planned for the “low delta-V” series was Test “I” which anticipated a greater delta-V than Test “H” and in which the occupant was instructed to leave the ignition in the “run” position. That having been done, as seen in Table 1, there was a new event recorded (the delta-V was greater than that in the previous test as planned), the “Vehicle Power Mode Status” reflects “Run” and the “Run/ Crank Ignition Switch Logic Level” shows “Active.” For this last test, the parameter “Vehicle Speed” shows appropriate speeds for the approximately 5 seconds pre-impact.

The reader is offered the opportunity to reexamine, first hand, the underlying CDR files as well as the control instrumentation data (and video, photos and diagrams) which accompany this issue of Collision for the “low delta-V” test series; a feature not found in other published works.

T

he Barrier Tests After the “low delta-V” test series at the 2014 ARCCSI Crash Conference, the “blue” and “white” Cobalts were used in one barrier test each. The tests were planned as frontal impacts, slightly offset (to the vehicle’s right meaning the main contact would be to the left front).

98 Collision Magazine - Volume 9 Issue 2

Figure 19: Remote control system to control move the ignition switch was installed on each vehicle The first of the two tests conducted involved the “white Cobalt” where it struck the barrier at about 31mph (50km/h) and the ignition switch was observed to be moved from the “run” position approximately 2.5 3.0sec pre-impact.

In the second test, where the “blue Cobalt” was used, the impact speed was approximately 30mph (49km/h) and the ignition was observed moved from the “run” position at approximately 2.0 2.5sec pre-impact as seen in Table 1.

www.collisionpublishing.com

In both cases, the CDR Report for that test indicated that the “Event(s) Recovered” were “Non-Deployments.” Setting aside, for the moment, that we know that an airbag deployment isn’t a function of a specific delta-V (after all, the delta-V isn’t known until the crash pulse is over) and leaving aside the previous discussion found in this narrative that, when the ignition is found to have moved from the “run” position with sufficient time pre-impact the deployment algorithms are disabled and, therefore, the only event that will be recorded is a Non-Deployment, one might consider the possibility that, in these tests, we might have actually expected a Non-Deployment but it was “masked” as a function of the ignition switch movement.

To examine this possibility, we might compare the acceleration recorded in these two barrier tests with the acceleration from a similar barrier test. A 56km/h (35mph) New Car Assessment Program (NCAP) Frontal Impact Barrier Test 21 was selected as the closest representative publically available barrier test with which comparisons might be made (referred to hereafter as “the NCAP test”). The NCAP test resulted in airbag deployments and a delta-V of a little more than about -56km/h (-35mph).

The comparison that might be best made between the NCAP test and the ARC-CSI Crash Conference tests would be the acceleration and change in acceleration observed in the first several milliseconds after impact. In that regard, one might look to the time between Event Enable

and deployment command which, we find for the driver and passenger airbags as recorded and part of the CDR report for the NCAP test, was 14ms for the first stages and 30ms for the second stages.

In Figure 20, we see the acceleration plot for the crash pulses for the blue and white Cobalts as well as the acceleration trace for the NCAP test Cobalt. Generally, from about “time zero” through the 14ms and on to the 30ms point (with some noise observed the white Cobalt test), the trend for the change in acceleration is comparable. Leading to the 14ms point, all three of the tests reflect acceleration in excess of -10g. Reasonably, one might conclude that the 30mph barrier tests conducted at the ARC-CSI Crash Conference should have resulted in deployments where, as noted, Non-Deployment events were recorded as would be the case where the ignition was switched from the “run” position prior to impact.

At this point, another associated data element which might be examined is the delta-V as a time history plot from the ARC-CSI crash test vehicles control instrumentation versus the reported delta-V time history graph. Figures 21 and 22 reflect that comparison. In Figure 21 we see the comparison from the white Cobalt test (2014 ARC-CSI Test 6) and Figure 22 the blue Cobalt (2014 ARC-CSI Test 7) crash test delta-V time histories from the respective CDR Tool Reports and the Crash Crew instrumentation.

Figure 20: Acceleration plot for the crash pulses for the blue and white Cobalts as well as the acceleration trace for the NCAP test Cobalt www.collisionpublishing.com

Collision Magazine - Volume 9 Issue 2 99

Figure 21: Crash test delta-V time histories from the respective CDR Tool Reports and the Crash Crew instrumentation for the White Cobalt

Figure 22: Crash test delta-V time histories from the respective CDR Tool Reports and the Crash Crew instrumentation for the White Cobalt While there was some discussion previously of the quantitative differences between the CDR reports delta-V data elements and then that captured by the Crash Crew control instrumentation, the slope overall values seen in these compare favorably.

100 Collision Magazine - Volume 9 Issue 2

The MCAT test is a separate issue in this respect. In that test, the subject Cobalt was driven into the rear of a motorcycle (with a dummy figure on the motorcycle) and the total Crash Crew instrumentation delta-V observed for the crash was -7.8mph (-12.5km/h). Perhaps more importantly, the peak g’s in the first 30ms never exceeded

www.collisionpublishing.com

-5.4g. Inasmuch as this is a dissimilar test as those shown in Figure 20, that data wasn’t included there graphically. It is the author’s belief that this test would have resulted in a Non-Deployment whether or not the ignition was in the “run” position at impact; however, there are other notable parameter data elements which will be discussed later in this narrative.

The remaining examination of the data retrieved as a function of the ARC-CSI Crash Conference tests and the MCAT test should then reasonably be focused on whether or not the other previously discussed parameters were correctly captured and what that parameter indicates. For those tests where a CDR retrieved EDR data set was reported for that specific event, those parameters and that discussion follows.

“Event Recording Complete” For all of the tests being discussed here, the CDR report indicates that the events were recorded completely indicating no power interruption during the test which lead to a power depletion during the recording function post-crash.

Given this observation, one may conclude any information which may have been captured in volatile memory by the ACM pre-crash was successfully/completely recorded by the ACM’s EDR function post-impact.

“Vehicle Power Mode Status” and “Run/Crank Ignition Switch Logic Level” In the two ARC-CSI barrier tests as well as the “low deltaV” test series and for the MCAT test, the state of these two parameters is reported consistent with previously discussed logic. The position of the ignition in the “accessory” position is observed as changing in the in-car videos as well remaining in “accessory” after the test.

In the tests discussed here, the ignition was moved from “run” as close to impact as approximately 0.75sec to as much as about 3.25sec before impact. Not tested in this series and potentially the subject of later testing would be a comparison with results in a scenario where the ignition moved from the “run” position at a time greater than 3.25sec pre-impact. As previously discussed, the ACM would have at least capacitor power available for some period after the ignition were to have been turned from “run” during which a Non-Deployment event could be recorded if there were a crash.

In the Continental Report15 where one finds the previously described “battery power loss” testing, for the tested ACM, Continental reported that where the power loss occurred at 4.5seconds before the injected impact signal (their “time zero”) there was no data recorded. While this might rea-

sonably suggest an upper limit to the amount of time a module might still be capable of detecting then recording an event of about 4.5 seconds, one should remember that this will be dependent on a number of factors not the least of which would be the actual timing of the “power off” relative to impact. The vehicle’s electrical state and the condition/capabilities of the ACM’s capacitors should also be contemplated in this respect.

Nonetheless, given the data found in the Continental Report and the testing related in this narrative, one may conclude that if the ignition were to move from “run” at about more than about 3.25 to 4.5 seconds before an impact, instead of a “Non-Deployment” being recorded, there may be no events recorded at all.

“Automatic Passenger SIR Suppression System Validity Status at AE” and “Automatic Passenger SIR Suppression System Status at AE” Previously discussed in this narrative, the AOS does not have internal power and so as soon as power is shut down (when the key would be moved from “run”) the directly connected AOS shuts down and the signal to the ACM is interpreted as “Invalid.” In all of the tests described here, the validity status was reported as anticipated.

For “low delta-V” tests “E,” “H,” and “I,” the validity status is reported as would be anticipated for the associated “Vehicle Power Mode Status” and “Run/Crank Ignition Switch Logic Level” state. One should notice in the summary found in Table 1 that once the validity status changes, that change is not persistent into another ignition cycle/test meaning that as the “Vehicle Power Mode Status” changes from “run” to “accessory” to “run” across three separate tests, the validity status is reported as “valid” and then “invalid” as anticipated rather than, for example, changing to “invalid” and staying that way until there was some other intervention such as the use of a scan tool. The same can be said for the two 2014 ARC-CSI barrier tests.

A separate observation is made for the parameter “Automatic Passenger SIR Suppression System Status at AE” for the barrier tests versus the “low delta-V” tests. In the “low delta-V” tests in which a Non-Deployment was anticipated whether or not the ignition was in the “run” position, the “Automatic Passenger SIR Suppression System Status at AE” is reported as “Not Suppressed” (emphasis added).

Perhaps one way of thinking of this is there was no need to “suppress” the deployment, there wasn’t going to be one given the crash severity, so the indication is, correctly, “not suppressed.”

www.collisionpublishing.com

Collision Magazine - Volume 9 Issue 2 101

However, in the two barrier tests from the 2014 ARCCSI Crash Conference, that status shows “Suppressed.”

As previously discussed, one would anticipate that these would have been deployment events given the acceleration observed compared to the NCAP test. The “Suppressed” status doesn’t appear to be directly associated with the ignition position because, as seen in the “low delta-V” test series, that status didn’t change (while event recording was complete as previously noted) while the validity status did change depending on ignition power mode status.

It appears, from the data available here, that this parameter is more reflective of there having been an event otherwise sufficient to deploy an airbag; however the last known status for the AOS was insufficient to enable the passenger airbag for deployment because there was no occupant detected in the right front/passenger seat.

One might wonder, at this point: why no mention thus far of these parameters from the MCAT testing CDR report?

Earlier in this article under the parameter heading “Automatic Passenger SIR Suppression System Validity Status at AE” and the associated “Automatic Passenger SIR Suppression System Status at AE” parameter was the discussion of these parameters relative to Figures 10 and 11. The relevant differences noted in Figures 10 and 11 with respect to the listed data parameters (the descriptions found in the left hand columns) is that, in Figure 10, these parameters are listed but in Figure 11 they are not. Both figures come from Epsilon module ACM CDR files (as previously noted).

The parameters are absent in Figure 11 (which is from the MCAT testing CDR file/report) because where the vehicle is identified as not having an AOS installed, those parameters are omitted from the CDR tool report. One should remember here that a CDR Tool Report is like a form or template. The parameter descriptions found in tables of a CDR Tool Report are part of a specific report form associated first with a module type then with specific subset(s) of that module type and/or vehicle configuration (for some OEs that is associated with a specific VIN or VIN range) and then with the types(s) of event(s) recovered.

In the case of the 2014 ARC-CSI Cobalts, those vehicles had an installed AOS. In the case of the MCAT Cobalt, it did not. What we might refer to by way of example as Epsilon “report form A” was used for the ARC-CSI Cobalts but “report form B” was used for the MCAT Cobalt when the CDR Tool created the final CDR reports.

This is a good opportunity to remind the reader that a CDR Tool Report is basically unique to a module/ACM

102 Collision Magazine - Volume 9 Issue 2

type and the parameter names/descriptors in that report are present or absent not as a function of whether or not there was a power interruption during the crash or power depletion during the recording process but rather as a function of the CDR Tool software’s identification of the module, its capabilities and/or calibration, or that of the vehicle and then for the event(s) recovered. There may be some instances, for some modules where other factors may influence the CDR Tool Report format including using a VIN other than that from the vehicle in which the module was originally installed (i.e.: “VIN spoofing”). Many times; however, differences or specifics of a report or pair of reports can only be identified by way of a thorough reading of the complete report(s) in the latest version of the CDR Tool software.

“Vehicle Event Data (Pre Crash) Associated With This Event” In all of the CDR reports for the various crash tests discussed here, the retrieved “pre-crash” data was associated with the recorded Non-Deployment event.

“Vehicle speed” None of the vehicles used in the testing here were equipped with a manual transmission, therefore, none exhibited the “invalid” vehicle speed feature. When the ignition was in the “run” position (and the vehicle moving), the reported speed compared favorably with the speed observed using the control instruments.

For each test where the ignition was moved to “accessory,” the reported vehicle speed after that point was, as anticipated, reported as “zero” and not marked “invalid.” Other pre-crash parameters such as engine speed were observed to both be displayed with “zeros” closest to Event Enable or to have been marked “Invalid” even though data associated with those elements would be found using the Engineering Translation Report. Although a working theory here is that whether or not Engine Speed was marked “invalid” might have to do with the time between the point when the ignition is moved from “run” prior to Event Enable, there are insufficient data points in this study and none out further than about 3.25 seconds before impact which might fully support that hypothesis.

For the three tests in this series where the ignition switch was moved from “run” and there were vehicle speeds recorded other than “zero,” those speeds were consistent with observations made in each test and would be, by themselves, reliable. The “zeros” were, of course not necessarily “unreliable” but were a function of the power lost to the

www.collisionpublishing.com

source components as the ignition was moved from the “run” position.

switch may or may not have moved from the “run” position prior to a crash.

onclusions There are no shortcuts, cheat sheets of “check box” forms that will readily serve the end user well as it relates to the CDR Technology. Just as each crash is unique, each CDR Tool Report is unique and each should be individually reviewed in their entirety - from the File Information data table to the hexadecimal data (even if only in the ETR) - to “do the job right.” An admonishment to simply “read the Data Limitations text” is of little real service to the end user who is then left unprepared to competently read the full report content in a fully informed manner.

Given the information regarding the system’s interdependent functions contained in this article, one should better understand how the various data elements may be recorded, how they are related to one another and how to use them together in an analysis of a crash where the ignition switch may have moved from the “run” position and eliminate those where that’s not the case.

C

One cannot take an individual data parameter and assume it will stand on its own independent of the remainder of the report in a given analysis. The many data parameters in a report are often interdependent to the point of making an analysis based on a single data element incomplete perhaps to the point of illustrating an analyst’s ignorance if not incompetence.

CDR Tool reports are unique to a module type and then essentially filtered for a variety of reasons before the end user sees the finished product: the report. In some cases, the CDR Tool software’s programming to identify characteristics of a specific vehicle may rely on information including the appropriate VIN for the vehicle/module being imaged. In short, “VIN spoofing” outside of a research setting where the end product is otherwise predictable is inappropriate and may result in otherwise unanticipated data translation(s).

In general, this article should reinforce that just because one or perhaps a couple elements in a given CDR Tool report is marked “invalid” or have otherwise “unrealistic” or “inconsistent data” (i.e.: zero for vehicle speed close to impact), that doesn’t mean the “whole report” is invalid and should be cast aside. Just as it is true that one head light on a vehicle may go out doesn’t mean the other head light or the tail lights will go out because each of the lamps on a vehicle are independent of one another, the data elements are individually recorded and have separate origin(s) in a vehicle although a collection of data elements may have an interpretation based dependency during a later analysis.

Specific to the Epsilon ACM installed in the GM vehicles subject to the recall, this article sets out the characteristics associated with various data elements which may help identify whether or not the data set making up the CDR Tool Report is related to a situation where the ignition

E

ndnotes and References 1. For purposes of this narrative, “GM ignition switch recall,” “ignition switch recall,” “the recall,” “ignition switch failure,” “NHTSA Recall” or words to that effect will be used here to refer, primarily, to NHTSA Recall Campaign Number: 14V047000 of February 10, 2014. One may find other associated recalls (i.e.: NHTSA Recall Campaign Number: 14E021000 for aftermarket replacement ignition switches) but this article deals more narrowly with specified vehicles identified as part of Recall Campaign 14V047000.

2. Valukas, Anton R., “Report to Board of Directors of General Motors Company Regarding Ignition Switch Recalls,” May 29, 2014 3. U. S. House of Representatives, 113th Congress, Energy and Commerce Committee, “Committee Report Details NHTSA Failures in GM Ignition Switch Recall,” web September 16, 2014, at http://energycommerce.house.gov/press release/committee report details nhtsa failures gm ignition switch recall 4. U. S. House of Representatives, 113th Congress, Energy and Commerce Committee, “The GM Ignition Switch Recall: Why Did It Take So Long?,” web April 2, 2014, at http://energycommerce.house.gov/hearing/“ gm ignition switch recall why did it take so long” 5. The data, photos and videos of those crash tests are found on the accompanying disk with this issue of Collision or can be ordered separately from Collision Publishing.

6. At the time of this writing, Bosch CDR Tool software version 14.1 was the most current version of the Tool software. Help file and Data Limitations references are made relying on software release 14.1. In this version, the revised number for the Data Limitations text for the Epsilon module is identified as Revision 006.

www.collisionpublishing.com

Collision Magazine - Volume 9 Issue 2 103

7. “Delphi Mechatronics” is identified as the supplier for the original ignition switch which is the subject of the underlying recall. The airbag and seatbelt assets of Delphi North America have since been acquired by another supplier: Autoliv 2.

8. From this point forward in the article, references to “the vehicles” or “subject vehicles” or words to that effect refer to the identified year models of the Cobalt, HHR, G5, Solstice and Sky vehicles listed in the recall to the exclusion of the Ion unless otherwise noted.

9. At one time, for example in now out-of-date CDR Tool version 3.4, “Note 6” in the Help File read: “A key is not required to download these vehicles, leave ignition off.”

10. For purposes of this discussion, we will set aside the technique known as “back powering” or, essentially, applying power to/powering up a module through its fuse for DLC access data retrieval.

11. GMLAN is the GM-specific name for the Controller Area Network or “CAN” communications bus architecture in these vehicles. The 04 Malibu was the first CDR Tool accessible vehicle to utilize the new GMLAN (CAN) bus architecture in the US which eventually replaced the GM legacy protocol such as the “Class2" serial bus. It is then the 2004 and newer GMLAN vehicles which have this “wakeup” capability.

12. The document is identified in the Congressional Committee document set specifically as “HHRG 113 IF02 20140401 SD030.pdf ” and referred to here as the “Continental report.”

13. The document is identified in the Congressional Committee document set as “HHRG 113 IF02 20140401 SD026.pdf ” and referred to here as the “Siemens report.”

14. That time is unspecified but may depend on the condition of the electrical system in the vehicle as well as the capacitors in the ACM. 15. From the “Continental Report” identified in the Congressional Committee document set specifically as “HHRG 113 IF02 20140401 SD030.pdf ” at page 12 of 14.

16. From the Epsilon ACM Data Limitations, using CDR Version 14.1: “There are two types of recorded crash events. The first is the Non Deployment Event. A Non Deployment Event records data but does not deploy 104 Collision Magazine - Volume 9 Issue 2

the air bag(s). The minimum SDM Recorded Vehicle Velocity Change, that is needed to record a Non Deployment Event, is five MPH. A Non Deployment Event may contain Pre Crash and Crash data. The SDM can store up to one Non Deployment Event.

This event can be overwritten by an event that has a greater SDM recorded vehicle velocity change. This event will be cleared by the SDM, after approximately 250 ignition cycles. This event can be overwritten by a second Deployment Event, referred to as Deployment Event #2, if the Non Deployment Event is not locked.

The data in the Non Deployment Event file will be locked, if the Non Deployment Event occurred within five seconds of a Deployment Event. A locked Non Deployment Event cannot be overwritten or cleared by the SDM. The second type of SDM recorded crash event is the Deployment Event. It also may contain Pre Crash and Crash data. The SDM can store up to two different Deployment Events. If a second Deployment Event occurs any time after the Deployment Event, the Deployment Event #2 will overwrite any non locked Non Deployment Event. Deployment Events cannot be overwritten or cleared by the SDM6.”

17. Haight, W. R., “Using CDR System Data in Crash Reconstruction or What Does the Term "Complete Reconstruction" Really Mean?”, Collision Magazine, Volume 6, Issue 1, 2011.

18. “Situationally complete analysis” is a term first adopted as part of the Collision Safety Institute’s CDR Data Analyst Course to replace the misused and often misunderstood term “complete reconstruction” as applied to the use of CDR Tool Reports in crash analysis. A situationally complete analysis is one which takes into account the available factors in a unique crash being examined where those factors may not be present from one crash under investigation to the next.

19. “Use of Event Data Recorders in Light Duty Vehicle Crash Reconstruction”; Chapter 19 of the Traffic Crash Reconstruction Manual, Northwestern University, Center for Public Safety, 2010, Fricke, L., et al (Chapter by Green, T., Haight, W. R., DiTallo, M.).

20. See also “Speed, Vehicle Indicated” at 49CFR563.5.

21. Report Number CAL-05-11, New Car Assessment Program (NCAP) Frontal Impact Barrier Test for a 2005 Chevrolet Cobalt, NHTSA Number M500105 conducted March 3, 2005.

www.collisionpublishing.com

www.collisionpublishing.com

Collision Magazine - Volume 9 Issue 2 105

r Pee

ed

ew evi

R

Determination of Motorcycle Pre-Collision Speed Oren Masory

Department of Ocean and Mechanical Engineering, Florida Atlantic University

Wade Bartlett

Mechanical Forensics Engineering Services

Bill Wright

Institute of Police Technology and Management

Part I : Physical Models

A

bstract The purpose of this paper is to evaluate different formulas, based on principles of mechanics, which are used to estimate motorcycle’s precollision speed in motorcycle-to-car accidents. These formulas use measurable physical parameters such as motorcycle’ wheelbase and car crush depth which were obtained from experiments performed by the authors as well as published experimental data. The results of these evaluations indicate that the motorcycle’s precollision speed cannot be estimated with high accuracy using these formulas. Therefore, during accident investigation, additional information is needed for accurate reconstruction.

I

ntroduction According to the National Highway Traffic Safety Administration (NHTSA), in 1969, there were 2.3 million motorcycles registered in the United States. This number increased dramatically during the next 40 years and currently there are over seven million motorcycles registered in the United States. Obviously, as the number of motorcycles on the road increased, so did the accident rate. In 1969 69,500 motorcycle

106 Collision Magazine - Volume 9 Issue 2

accidents were reported, in which 1,855 persons were killed. In 2004, 4,008 motorcyclists were killed in the US, a 116% increase from 1969. The death rate per 100 million vehicle miles traveled for a motorcyclist is over 35 times higher than of an automobile occupant 1.

One of the main and unequivocal reasons motorcyclists are killed in crashes is because the motorcycle itself provides virtually no protection in a crash. For example, approximately 80 percent of reported motorcycle crashes result in injury or death while a comparable figure for automobiles is about 20 percent.

The earliest and the most frequently cited testing was published by Severy in 1970 2, which included one test at 20mph, one test at 40mph, and 5 tests at 30mph.

That paper included a chart showing a linear relationship between wheelbase reduction and the motorcycle’s pre-collision speed with high correlation factor of 0.975.

The Severy data is not relevant to today's motorcycles: chassis design, wheels, materials, engine mounting, and front end suspension styles have changed enough that his data is not applicable for today motorcycles.

www.collisionpublishing.com

In 1976, Professor Harry Hurt3 of the University of Southern California conducted a study on 3,600 motorcycle accidents in the Los Angeles area. The results of this study have become the basis for much motorcycle rider training and research. The study found that in 65 percent of the multi-vehicle cases the automobile violated the motorcycle right of way. A similar survey conducted by the Philadelphia Police Department showed

that such a violation is the cause in only 45 percent of all motorcycle accidents in the city of Philadelphia. Hurt’s study also found that in 64 percent of single-vehicle motorcycle accidents, the operator was responsible for his accident. The majority of accidents occurred on a clear, dry day during daylight hours and typically, the motorcycle operator had less than six months riding experience and no formal training.

www.collisionpublishing.com

Collision Magazine - Volume 9 Issue 2 107

In McNally 8 the principles of linear and angular momentum conservation were investigated. It was concluded that “Use of the sensitivity analysis will allow the reconstructionist to determine if the techniques should be applied to the given analysis or be abandoned in favor of other methods of speed analysis”.

In Adamson 9, the validity of 10 different relationships, used to determine the motorcycle’s pre-collision speed strikes a passenger car, were evaluated. In contrary to expectations, the correlations between speed and crush were always stronger than any relations involving energy. When calculating speed from motorcycle’s crush, it has been commonly assumed that that the type of wheel, cast or spiked, was an important consideration. This analysis showed that while wheel type does affect the motorcycle-to-car crush ratio, the total crush was not significantly affected.

The equations that most accurately predicted the reported impact speed of the test motorcycles were the Eubanksform equations, with the modifications suggested by this analysis that account for the differences between the door/ fender impact areas and the pillar/axle areas. The door/ fender collision speeds were found to be predicted with a 95% confidence range of plus or minus 20% of the nominally calculated value, while the pillar/axle speeds were predicted within a range of plus or minus 28%.

From an accident reconstruction point of view there is a limited amount of data, compared with vehicle collision data, which makes accurate modeling very difficult. Collision tests involving cars moving at high enough speeds to affect the motorcycle’s trajectory are very rare. In these cases it appears to reduce the total measured crush, such that using the relationships developed here will under predict speeds, typically by 15%. The additional complication of a moving target vehicle also increases the data spread significantly. The limited car-bumper impact testing appeared to be very similar in crush characteristics to the pillar/axle impacts, but the data set was too small for significant analysis.

This paper provides analysis of test results of 13 different tests in which the motorcycles were crashed into the side of a car (See Appendix – Table A.2). The impact velocity ranged from 40-58[mph] and the impact point on the car varied. The motorcycle’s wheelbase change, width and the maximum indentation and the movement of the vehicle were recorded. In addition, data from previous 56 tests were used to check the analysis results (all data are provided in the Appendix).

108 Collision Magazine - Volume 9 Issue 2

M

odeling And Data Analysis Model I – Wheelbase Reduction

In this model the pre-collision speed of the motorcycle is correlated with its wheelbase reduction.

This model is the earliest model, proposed by Severy 2.

The wheelbase reduction data of the previous 56 tests were graphed versus the pre-collision speed and are shown in Figure 1. As shown the data is scattered and does not fit a straight line since a correlation factor of 0.27 is very poor.

Model II – “Total” Crush With this method the motorcycle speed is correlated with the “Total Crush” which is the sum of the motorcycle’s wheelbase reduction and the maximum indentation of the vehicle’s crush. The correlation, if exists, eliminates the need to know the stiffness of the motorcycle structure as well as the local stiffness of the vehicle at the impact location. However, two different cases are being considered: 1) The motorcycle strikes vehicle at a “soft” location such as a door or a fender; and 2) A “hard” point, such as a pillar or within 3 inches of an axle, is being struck. For the first case the following linear relationship with a correlation factor of 0.8 was obtained (see Figure 2): V = 1.43(L+C)+10.4

(1)

where: V – Motorcycle pre impact speed (mph) L – Motorcycle wheelbase reduction (inches) C - Maximum indentation in the vehicle’s crush zone (inches) Similarly, for impacts at “hard” location the following relationship with a correlation factor of 0.76 was obtained (see Figure 3): V = 1.59 (L+C)+14.72

(2)

Using these models, the pre-collision speed of the motorcycle was estimated and then compared to the experimental data. For further analysis the data was divided according to the location of the impact as follows: 1) Fenders and door; 2) Pillar ±3 inches; and 3) Axle ±3 inches. The reason behind this classification is the difference in the stiffness at these locations. The estimation errors are illustrated in Figures 4 through 6 for different impact location on the car. As shown, in cases where the impact location is “hard” (pillar or axle) the largest estimation errors are approximately 30% while when the impact location is “soft” the error reach 40%. In both cases, the results are not acceptable.

www.collisionpublishing.com

Figure 1: Results of model I

Figure 2: Motorcycle speed estimate “Soft location”

Figure 3: Motorcycle speed estimate “Hard location”

Figure 4: Motorcycle speed estimation error where impact at the axle

Figure 5: Motorcycle speed estimation error where impact at the pillar.

Figure 6: Motorcycle speed estimation error where impact at the door.

www.collisionpublishing.com

Collision Magazine - Volume 9 Issue 2 109

As shown the only “good” correlation is in the case where the impact location is around the axle, probably because the stiffness at this point is very high for all vehicles.

Model III – Conservation Of Linear Momentum The Law of Conservation of Momentum dictates that the total momentum just prior to the collision is the same as immediately after the collision. In most motorcycle collisions this basic formula must be expanded to include both motorcycle and rider post-impact velocities, since the motorcycle and rider seldom stay together following the collision. In these tests the impacted vehicle was in rest thus the relationship is given by: (Mm+Mr )V = McVc+MmVm+MrVr

(3)

Mm - Motorcycle’s mass Mr - Rider’s mass Mc - Vehicle’s mass Vc - Vehicle’s post impact speed Vm - Motorcycle’s post impact speed Vr - Rider’s post impact speed For our test conditions Eq. 3 can be simplified since there was no rider on the bike: (4)

To estimate post impact speed of the motorcycle conservation of energy relationship is used: Vm = √ 2µm gl

(5)

l - Motorcycle’s skid marks length µm - Coefficient of friction between the motorcycle and the skidding surface The same way the post impact of the vehicle can be calculated:

L - Vehicle’s skid marks length

110 Collision Magazine - Volume 9 Issue 2

(7)

The estimation errors of the motorcycle’s pre-collision speed using the principle of linear momentum conservation are shown in Figure 7, indicating poor results. In some cases the estimation errors exceeding 100%. It is interesting to note that all errors are negative which means that the model overestimated the actual motorcycle’s speed.

Model IV – Conservation Of Angular Momentum

In side collisions the impulse applied to the vehicle results in the struck end of the vehicle “sliding,” while the opposite end acts as a pivot point. When this type of vehicular motion occurs, it is possible to calculate the torque acting on the vehicle by (assuming uniform distribution of the vehicle weight):

τtire = (1/2) Mc gWb µc

(8)

where:

τtire - Torque caused by the tires sliding sideway The value of the torque calculated above can then be used in the following formula, which calculates the rotational velocity of the vehicle.

g - Gravitational acceleration

where:

V = (√ 2g /Mm ) (Mc √ µc L + Mm√ µm l )

Wb - Vehicle’s wheelbase

where:

Vc = √ 2µc gL

Substituting the above results in Eq. 3 the pre-collision speed of the motorcycle can be determined by:

The McNally Rotational Momentum Method 8 utilizes vehicle information (weights, weight distributions, impact location, vehicle rotation), and scene information (post impact speed and direction of both vehicles). The first step in the rotational analysis is to determine the total amount of torque acting through the tires/roadway interaction to slow the angular velocity of the vehicle following the collision.

where:

MmV = McVc+MmVm

µc - Coefficient of friction between the vehicle and the skidding surface

(6)

ω = √ (2τtireθ ) / (I+Mcd 2)

(9)

where:

ω - Angular velocity of the vehicle θ - Vehicle’s angle of rotation I - Yaw moment of inertia of the vehicle d - Distance of the farthest axle from the impact point to the center of gravity of the vehicle

www.collisionpublishing.com

Figure 7: Estimation errors of the pre-collision motorcycle speed using linear momentum conservation.

At this point the change in the motorcycle speed, due to the impact, can be determined: ∆Vm = (ω (I+Mc d 2)) / (SMm )

(10)

where S is the length of moment arm determined by measuring the perpendicular distance from the principal direction of force to the center of mass of the vehicle.

If the motorcycle’s direction of travel at impact and its post-collision speed are known, then impact speed of the motorcycle can be determined. Figure 8 show the estimation errors of the motorcycle’s pre-collision speed determined by this model. As in the previous model, the errors are large and reaching 80%. Similarly, in most cases the pre-collision speed is over estimated.

C

ONCLUSIONS The paper describes 5 different methods, based on the principles of mechanics, by which motorcycle’s pre-collision speed with the side of a vehicle, can be estimated. None of these methods produced satisfactory results (pre-collision speed bounded by 10% error). More-

over, from accident reconstruction point of view, the use of the conservation momentum relationships is very limited since in many cases the investigation is taken place sometime after the accident. As a result, required measurements, such as skid marks, are not available anymore. Thus, the “Total Crush” method is the most useful, in spite of the large error it produces, since at the best both vehicles are still available for inspection and the motorcycle’s wheelbase reduction and the vehicle’s crush can be still measured.

R

EFERENCES

1. http://www.edgarsnyder.com/motorcycleaccidents/statistics.html

2. Severy, D., Brink, H., and Blaisdell, D., "Motorcycle Collision Experiments," SAE Technical Paper 700897, 1970.

3. Motorcycle Accident Cause Factors and Identification of Countermeasures, Volume 1: Technical Report, Hurt, H.H., Ouellet, J.V. and Thom, D.R., Traffic

www.collisionpublishing.com

Collision Magazine - Volume 9 Issue 2 111

Figure 8: Estimation errors of the pre-collision motorcycle speed using angular momentum conservation.

Safety Center, University of Southern California, Los Angeles 4. Fricke, Lynn B., and Riley, Warner W., Reconstruction of Motorcycle Traffic Accidents, Topic 874 of the Traffic Accident Investigation Manual, Northwestern University Traffic Institute, 1990 5. Brown, John F., and Obenski, Kenneth S., Forensic Engineering Reconstruction of Accidents, Charles C. Thomas, 1990 6. Obenski, Kenneth S., Motorcycle Accident Reconstruction: Understanding Motorcycles, Lawyers & Judges Publishing Co., 1994

112 Collision Magazine - Volume 9 Issue 2

7. Niederer, Peter F., Some Aspects of Motorcycle-Vehicle Collision Reconstruction, SAE Paper 900750, Society of Automotive Engineers, 1990 8. Bruce F. McNally, Wade Bartlett, 20th Annual Special Problems in Traffic Crash Reconstruction at the Institute of Police Technology and Management, University of North Florida, Jacksonville, Florida, April 15-19, 2002 9. Adamson, Kelley S.; Alexander, Peter; Robinson, Ed L.; Burkhead, Claude I.; McMannus, John; Anderson, Gregory C.; Aronberg, Ralph; Kinney, J. Rolley; Sallmann, David W.; Johnson, Gary M.; Seventeen Motorcycle Crash Tests Held at WREX2000, WREX2000 Conference Results, 2002

www.collisionpublishing.com

Part II : Self Learning Models

A

BSTRACT In this paper a self learning method, based on Abductive Networks, is being used for the development of a model that related motorcycle’s precollision speed to post-collision measurable data. The best results were obtained where the motorcycle wheel base reduction and the maximum crush of the vehicle were used as input the model. This result is, to some degree, similar to the results obtained in part 1 of this paper where these two parameters were assumed to be linearly related to the motorcycle pre-collision speed. The results obtained with tis methods indicate that the motorcycle’s pre-collision speed cannot be estimated with high accuracy. Therefore, during accident investigation additional information is needed for highly accurate reconstructions.

were not satisfactory in part due to the fact that very few crash tests are performed on motorcycles.

I

This self learning method, which will be described in the following section, is called “Abductive Networks. This method was used successfully in many application one of which was to determine collision force and crush energy in vehicle collisions 10.

NTRODUCTION The increasing number of motorcycle accidents is a major concern to policy makers. The understanding the causes of this accidents by accident reconstruction tools, will allow law makers to address this issue by regulating design features, safety requirements, road design modifications etc.1.

Different models, based both vehicle damage and on onsite measurements, were reported in the literature 2-9. Some are based on physical principles like conservation of linear and/or angular momentum, and some based on correlation between measurable data. In both cases the results

The method described in this paper falls into the second category where a correlation between measurable data and the motorcycle’s pre-collision speed is being searched. Previously, a specific model was assumed and then the measured data was correlated to it. Thus, the produced models were not necessarily the best ones and mostly simple as linear polynomials. In this case, a self learning method is being used which automatically will find the best model that will fit the data. As a result higher order polynomial might be found. In any case it should be emphasize that since the model are “just” correlation to measured data they do not shed any light on the physics of the crash.

A

IM – ABDUCTIVE NETWORKS AIM 11 is a powerful supervised inductive learning tool for automatically synthesizing network models from a database of input and output values.

The model emerging from the AIM synthesis process is a robust and compact transformation implemented as a layered abductive network of feed-forward functional elements as shown in Figure 1.

Figure 1: Example of Abductive Network www.collisionpublishing.com

Collision Magazine - Volume 9 Issue 2 113

All the functional and connection elements are learned from the input data. Currently AIM has seven types of elements. The algebraic form of each element is a polynomial where Wn are the coefficients determined by AIM and Xn are the input variables (Table 1 shows sample elements).

All terms in an element’s equation may not appear in a node since AIM will throw out or carve terms which do not contribute significantly to the solution.

Single

W0+W1* X1+W2* X12+W3* X13

Double

W0+W1* X1+W2* X2+W3* X12+W4* X22+ W5* X1* X2+W6* X13+W7* X23

5. Evaluate the performance of the model using the rest of the above data set (the portion that was not used for the train the network.

6. Once the network (model) performs to satisfaction an explicit model can be derived and implemented.

T

Table 1: Example of elements definition The eligible inputs for each layer and the network synthesis strategy are defined in a set of rules and heuristics which are an inherent part of the synthesis algorithm.

AIM automatically determines the best network structure, element types, coefficients and connectivity by minimizing a modeling criterion which attempts to select as accurate a network as possible without over fitting the data.

The modeling criterion used within AIM is the Predicted Squared Error (PSE). The PSE is a heuristic measure of the expected network squared error for independent data which is not included in the training database. The PSE is given by: PSE = FSE +KP

4. Train the Abductive Network using a portion of the above database.

he Data Sets And Data Allocation Two data sets with experimental data were available to the authors:

1. Data Set I: contains data from 47 crush tests that were performed previously by different investigators.

2. Data Set II: contains data from 13 crush tests that were performed by the authors.

Both data sets include the following information:

(1)

where FSE is the fitting squared error of the model to the training data and KP is a complexity penalty term determined in AIM by the equation: KP = CPM* (2K / N)* sp

2

(2)

a) Pre-collision motorcycle’s speed (VM). b) Motorcycle’s wheelbase reduction due to the impact (WR). c) Vehicle’s maximum crush (see Figure 2) (DC). d) The impact location with the car, “Hard” or “Soft” locations. “Hard” location is considered as a location 3” from a pillar or axle and a “Soft” location is any other place on the car (e.g. doors or fenders) (H or S). e) The weight of the motorcycle (WM).

where K, N and sp2 are determined by the database of examples used to synthesize the network and CPM, the Complexity Penalty Multiplier, is a variable the user can select.

The default value of CPM is 1; a lower value decreases the complexity penalty collision and results in a more complex network and inversely for a higher value.

1. To create a model using AIM one has to follow these steps: 2. Decide what are the inputs and the output to the model.

3. Create a database which includes sets of inputs and the corresponding outputs from the process being modeled.

Figure 2: Vehicle’s maximum car crush.

114 Collision Magazine - Volume 9 Issue 2

www.collisionpublishing.com

As mentioned before, the first step is to “train” the system by providing a set of inputs and their corresponding outputs to the system. The inputs are post-collision measurable quantities such as motorcycle’s wheelbase reduction.

The output, in the particular case, is only the pre-collision motorcycle’s speed. The training data set is selected randomly from both data sets. Once a model is established (the network converged), the rest of the data is used to evaluate the prediction achieved by the model.

M

odeling Model #1

This model relates to the model proposed in 2 where the motorcycle’s pre-collision speed was related only to its wheelbase reduction. Thus, the only input to this model is the wheelbase reduction of the motorcycle (WR) and the only output is the speed. The model has been trained using measurements from 24 different crash tests from Set I. Upon convergence, the model was used to predict the motorcycle speed of the other 23 crash tests, from the same set, which were not used for training. The average error of the predicted speed was 8.45% and the maximum error was 23.8%. At this point the model was used to predict the motorcycle’s speed using data set II. In this case the average prediction error was 17.8%.

(∆DC ) – Error in the measurement of the vehicle’s maximum crush As an example assume that (∆WR ) = (∆DC ) = 0.25”. It will result in: (∆VM ) = 0.6125+0.019 (∆DC )

(5)

Eq. (5) indicates that the error in the prediction of the motorcycle’s speed is linear with the value of the vehicle’s maximum crush measurement, meaning that during the accident reconstruction process this value has to be measure with high accuracy.

The model was used to predict the motorcycle’s pre-collision speed of the other 23 cases in data set #1. The errors in the predictions of the motorcycle’s speed had average of 4.64% and the maximum of 11.47%.

The same model was used to predict the motorcycle’s preimpact speed of the cases provided in data set II. It resulted in an average error of 12.8% and maximum error of 22%.

The histogram of these prediction errors are shown in Figure 4.

Model #2 This model relates to the model proposed in 2 where the motorcycle’s pre-collision speed (VM ) was related to its wheelbase reduction (WR ) and the maximum crush of the car (DC ). Again, 24 different randomly selected tests were used to train the network. Once the model was obtained, upon convergence, an explicit formula can be extracted: VM=13.48+2.05(WR )+0.40(DC )+0.038(DC )2

(3)

The capability to obtain an explicit relationship is very advantageous compared to other self-learning methods (e.g. Neural Networks). For example, the sensitivity of the speed prediction to the errors in the measurements of the wheelbase reduction and the vehicle’s maximum crush, can be determine:

Figure 3: Histogram of the errors in prediction of the motorcycle speed using data set I

(∆VM )=(∂VM / ∂WR )(∆WR )+(∂VM / ∂DC )(∆DC ) (∆VM )=2.05(∆WR )+0.40(∆DC )+0.076(DC )(∆DC ) (4) where: (∆VM ) – Error in motorcycle’s speed (∆WR ) – Error in the measurement of the motorcycle’s wheelbase reduction

Figure 4: Histogram of the errors in the prediction of the motorcycle’s pre-collision speed using model #2 and data set II

www.collisionpublishing.com

Collision Magazine - Volume 9 Issue 2 115

Other Model Few other models, which differ from each other by: 1) The set of inputs; 2) Inputs configuration; and 3) Training sets, were tried (see Table 2). Observing the results given in Table 2, one can reach the following conclusions: 1. In cases where the training data and the verification data are from the same data set, the models are better (smaller errors)(tests 1, 2, 5, 6 and 7).

2. Some set of input produces better results (within the same set). Compare the results of models 1 and 2 and 5 and 6.

3. That the mixed data set produced relatively inferior results is likely that the tests were probably controlled differently.

Data Set 1 I

Inputs

Error [%] Average Max. 4.64 11.47

2 I

(WR ), (DC ), (H) or (S) (WR ), (WM )

6.27

21.74

3 I & II

(WR ),(DC )

5.61

20

4 I & II

(WR ),(WM )

7.49

17.42

5 II

(WR ),(DC )

59.78

93

6 II

(WR ),(DC )

48

102

7 I

(WR ),(DC ),(WM )

6.94

17.92

onclusions A simple, self learning models for a motorcycle’s pre-collision speed estimation were presented. The models are not based on principles of mechanics but on the correlation of experimental data to the motorcycle’s pre-collision speed. The models were obtained automatically using the AIM system and some are reasonably accurate to be used in accident reconstruction. The models obtained by this method are explicit in contrast to non-explicit model that can be obtained by other learning procedures such an Artificial Neural Networks. As such it provides better understanding to the effect of each measurement on the model and the sensitivity of the model to each of the inputs. This self learning should be used with great caution since the quality of predicted result is highly depends on the training data set. Thus, in order to obtain good results the training data set should be large

116 Collision Magazine - Volume 9 Issue 2

R

eferences 1. National Highway Transportation Safety Administration. http://www.edgarsnyder. com/motorcycle-accidents/statistics.html

2. Severy D. Brink, H. and Blaisdell, D., "Motorcycle Collision Experiments," SAE Technical Paper 700897, 1970.

3. Motorcycle Accident Cause Factors and Identification of Countermeasures, Volume 1: Technical Report, Hurt, H.H., Ouellet, J.V. and Thom, D.R., Traffic Safety Center, University of Southern California, Los Angeles.

4. Fricke, Lynn B., and Riley, Warner W., Reconstruction of Motorcycle Traffic Accidents, Topic 874 of the Traffic Accident Investigation Manual, Northwestern University Traffic Institute, 1990.

5. Brown, John F., and Obenski, Kenneth S., Forensic Engineering Reconstruction of Accidents, Charles C. Thomas, 1990.

6. Obenski, Kenneth S., Motorcycle Accident Reconstruction: Understanding Motorcycles, Lawyers & Judges Publishing Co., 1994 7. Niederer, Peter F., Some Aspects of Motorcycle-Vehicle Collision Reconstruction, SAE Paper 900750, Society of Automotive Engineers, 1990.

Table 2: Results for other models

C

and diverse in order to accommodate large variations in the parameters.

8. Bruce F. McNally, Wade Bartlett, 20th Annual Special Problems in Traffic Crash Reconstruction at the Institute of Police Technology and Management, University of North Florida, Jacksonville, Florida, April 15-19, 2002 9. Adamson, Kelley S.; Alexander, Peter; Robinson, Ed L.; Burkhead, Claude I.; McMannus, John; Anderson, Gregory C.; Aronberg, Ralph; Kinney, J. Rolley; Sallmann, David W.; Johnson, Gary M.; Seventeen Motorcycle Crash Tests Held at WREX2000, WREX2000 Conference Results, 2002.

10. O. Masory, N. Putod, “Determination of Impact Force and Crush Energy Using Abductive Networks”, Proc. 20th Florida Conference on Recent Advances in Robotics, May 31-June 1, 2007.

11. AIM User’s Manual, AbTech Corporation, 1990.

Appendices for Part I and Part II are located on the data DVD included with this issue of Collision Magazine.

www.collisionpublishing.com

Expert AutoStats®

4N6XPRT BioMeknx™

Expert AutoStats® is a computer program containing over 42,000 vehicles cars, vans, utility vehicles, and light pickups. 1945-2012 model years are represented.

This program can be used alone or with our Expert Qwic Calcs® and Expert VIN DeCoder® programs. Also accessible from within REC-TEC.

(Lap top & D esktop W indow s 3 2 & 64 -bit O S C om p atible)

Enter:

M anufacturer and Year

Select:

M odel from pop-up window

Output:

Wheelbase Overall Length Height Wt. Distrib

Curb Weight Width Drive Wheels Front/Rear Track

M id-60's to present output also includes (when available): Front Overhang Rear Overhang Bumper Heights Hood Height Turning circle Wheel radius Bumper-to-Hood Bumper-to-Window

Expert AutoStats® is currently owned by over 700 private individuals and 300 Law Enforcement agencies in the U.S.A., Canada, and South America. Additional information, brochure, and order from can be downloaded from:

4N6XPRT BioMeknx is designed as a collection of the Biomechanical data of importance to the Accident Investigator into one easily accessible reference location. It is designed as a program to be useful to ALL accident investigators, not just the Biomechanical expert.

The program includes: 1. Over 65 printable illustrations and charts 2. 10 Separate categories or sections of information related to forensic biomechanics 3. Information on walking speeds of people from 17 months to > 65 years 4. Calculation modules for: - segment length from total body height, - segment weights and Centers of M ass (CoM ) based on total body weight, - Predicted M aximum Walking Speed based on total height 5. Information was obtained by review of over 65 scientific texts, several of which are now "out of print"

www.4n6xprt.com/4n6as.htm

www.4n6xprt.com/4n6bm.htm

PROGRAM ORDER FORM (Please Print) Contact Name: ________________________________________________ Company: ____________________________________________________ Street: ______________________________________________________ City:____________________ State:____ Zip:__ __ Country: __________ Phone: (_____) _______________ Fax: (_____) ___________________ E-M ail: ______________________________________________________ Expert AutoStats® 4N6XPRT BioM eknx™ Expert Qwic Calcs® 4N6XPRT Ped & Bike Calcs® 4N6XPRT StifCalcs® Expert VIN DeCoder® (pkg) Expert TireStuf® Order Processing

- $ 625.00 ea.

- $ 495.00 ea.

- $ 275.00 ea.

- $ 375.00 ea.

- $ 650.00 ea.

- $ 550.00 ea.

- $ 85.00 ea.

= ________ = ________ = ________ = ________ = ________ = ________ = ________

- Check = $5.00 / Credit Card = $10.00 / G ovt. P.O . = $15.00 = _______ Notarized Affidavit Requirement $25.00 per required Notarized Signature = _______

Q Deliver via electronic download (Email address Required) Q Ship on USB - $35.00 per disk per program California Sales Tax

- 8.50%

TOTAL (U.S. Dollars)

= 0.00 = ______ = ________

= ________

Enclosed is: Visa ___ MasterCard ___ Am. Ex. ___ Exp ____/____ - Check ___ - Gvt PO ___

Card Number: Name on Card: Signature:

_______________________________________________ _______________________________________________ ________________________________________________

Please M ake Checks*/Purchase Orders Payable to:

4N6XPRT Systems® 8387 University Ave - La M esa, CA 91942-9342 - U.S.A. FAX: (619) 464-2206 Phone: (619) 464-3478 Web: http://www.4n6xprt.com C hecks * M U ST * be draw n on a U.S. bank. O rders are shipp ed within 10 working days of receipt.

Prices subject to cha ng e without notice.

M ultip le orders and Package discounts availab le. Please call for details.

CRASH SITE EVIDENCE

A 1968 Mustang GT, driven by Lt. Frank Bullitt, is chasing a 1968 Dodge Charger R/T, driven by a man known only as “The Killer,” down Taylor Street on Russian Hill in San Francisco.

We have the STIFFNESS data you need for your calculations!!

Now includes a FORCE BALANCE calculation module!!

4N6XPRT StifCalcs® puts the NHTSA Crash Test database on YOUR 32 or 64 bit MS-Windows computer, automatically calculates A-B-G values Front, Side, and Rear - and Test Specific Crush Factors based on the NHTSA test data, allows the creation of “CLASS” vehicles and determine the “CLASS” stiffness values, PLUS conduct FORCE BALANCE calculations when needed, all of this with no need for you to get on the internet, and all for the LOW price of $650.00*!

Can you do this yourself “for free”?? Only if YOUR TIME = $0.00!!

For more information, program order form, and sample printouts please visit our web site at http://www.4n6xprt.com.

4N6XPRT StifCalcs® is provided by 4N6XPRT Systems

For a full reconstruction of this 46 second chase sequence, visit www.edccorp.com/bullitt

8387 University Avenue - La M esa, CA 91942 - USA Phone: 800-266-9778 / (619) 464-3478 Fax: (619) 464-2206

Individual Vehicle Stiffness Data

HVE directly answers the following questions using Newton’s laws of motion:

Questions

Answers

Speed on Taylor St.

30 – 37 mph (both vehicles)

Driver Inputs

Mustang: 2-3 lb brake pedal force, 200 deg left steer onto Filbert, then 25-65 %WOT Charger: 3-5 lb brake pedal force, 175 deg left steer onto Filbert, then 70-75 %WOT

Maximum Undercarriage Impact Force

Mustang: Fx= -1730 lb, Fy= -13 lb, Fz= -2120 lb Charger: Fx= -8610 lb, Fy= -120 lb, Fz= -11100 lb

Maximum Suspension Force

Mustang: 4330 lb @ L/F; 3270 lb @ R/R Charger: 4810 lb @ R/F; 3530 lb @ L/R

Hubcaps Lost

Mustang: N/A (American Racing Mags) Charger: Five

Other software may claim to answer these questions. Have you closely evaluated their claims and the basis for their results?

So, you want Stiffness data for a vehicle, QUICKLY & CONVENIENTLY, but you don’t want to buy a program? Order your data on a vehicle-by-vehicle basis from us at only $40.00 per test!

Year: _________ Make: _____________________ Model:____________________________________ Impact on (circle one): Front - Side - Rear Impact speed (circle one): High - Low YOUR CONTACT INFORMATION Name: ________________________________________ Company: _____________________________________ Address: ______________________________________ City/State/Zip: _________________________________ Phone: _______________________________________ Fax: __________________________________________ Email: ________________________________________ Visa/MC/AmEx: ________________________________ Exp.:____/____ Case Reference: ___________________

To learn more about HVE and all of the available tools for crash investigators, call 503-644-4500 or visit www.edccorp.com. Our professional sales staff will help you get started using HVE today!

Send Order to:

4N6XPRT Systems Fax: 619-464-2206

KNOWLEDGE = POWER Engineering Dynamics Corporation • 8625 SW Cascade Blvd., Suite 200 Beaverton, OR 97008 USA • Phone (503) 644-4500 • www.edccorp.com

Ph: 800-266-9778 8387 University Avenue / La Mesa, CA 91942 / USA E-mail - 4n6@4n6xprt.com

2014 ARC--CSI DATA DVD

C

2014 Vol.9 Iss.2

ollision C The International Compendium for Crash Research

Volume 9 Issue 2

FALL 2014

Collision: The International Compendium for Crash Research

Using Inequality Contraints in the Probability Analysis of

Motor Vehicle Accidents

Volume 9, Issue 2 - FALL 2014

Ignition Switch Data in CDR Reports

“Low Velocity” Testing collisionmagazine.com