Volume 13, Issue 1

ollision C

Volume 13 Issue 1

The International Compendium for Crash Research

Power Loss Issues

Related To Edr Data In Motorcycles

Unmanned Aircraft Systems Photogrammetry VS. Total Station

DAsh Camera

video velocity analysis Cover Pages.indd 1

collisionmagazine.com 5/28/2019 12:30:19 PM

S I MON

SImulation MOdel Non-linear SIMON is a fully 3-dimensional dynamic simulation of the response of one or more vehicles to: • Driver Inputs (steering, braking, throttle, gear shift) • Collisions • Terrain Factors (irregular terrain, curbs, soft soil, slippery regions, hydroplaning, potholes, …)

• Aerodynamics (wind forces, airborne trajectory) SIMON was built using the latest advances in software and simulation technology, based on a new, 3-D vehicle dynamics engine developed by Engineering Dynamics Corporation.

Any number of vehicles may be included in a SIMON simulation. The user assigns initial positions and velocities for each vehicle. The user may also assign driver controls, wheel set-up conditions (tire blow-out, wheel damage, brake adjustment/failure, curb impact) and accelerometers. Using this information, SIMON calculates the forces acting on the vehicle and uses these forces to calculate vehicle position, velocity, acceleration and collision damage at user-specified time intervals. The results may be displayed numerically (in a spreadsheet format) or visually (in 3-D viewers).

SIMON has been validated against numerous well-instrumented vehicle handling studies, including combined steering and braking, severe irregular terrain traversal, rollover tests, staged crash tests and tire blow-out experiments. Program results easily meet Federal standards for the admissibility of scientific evidence.

VISUAL REALITY

Cover Pages.indd 2

Engineering Dynamics Corporation • 8625 SW Cascade Ave. Suite 200 • Beaverton, OR 97008 USA • 888.768.6216 www.edccorp.com SIMON and DyMESH are trademarks of Engineering Dynamics Corporation. All rights reserved.

5/28/2019 12:30:19 PM

bo s ch C DR p r o t o o l kit Bosch is the world leader in Event Data Recorder (EDR) information and imaging technology. Since 2000, Bosch Crash Data Retrieval (CDR) products have been trusted internationally by law enforcement, crash researchers, auto manufacturers, insurance SIUs and government agencies to access EDR information on a wide range of passenger cars, light trucks and SUVs. The CDR Pro Tool Kit is the heart and soul of the Bosch CDR Tool. This package contains all of the hardware required to perform a DLC/OBD retrieval of EDR crash data in Bosch supported vehicles.

Included in this kit are both the CDR 900 and CANplus vehicle interface modules.

This package also comes with a 1-year CDR software subscription.

The Bosch CDR Pro Kit Retrieve crash data from

over 45 automotive brands with 1 tool!

Now available from

Schedule a webinar or product demo today www.crashdatagroup.com

Cover Pages.indd 4

5/28/2019 12:30:21 PM

Volume 13 Issue 1.indd 1

5/28/2019 9:34:14 AM

Volume 13 Issue 1.indd 2

5/28/2019 9:34:14 AM

Contents

Volume 13 Issue 1

inside 8

Crash-ol-o-gy

6

Letter From the Editor

5

Advertiser Index and Digital Download Information

48

features 10

Vehicle System Forensics for Crash Reconstruction

by Wes Vandiver and Robert Anderson

16

Dash Camera Video Velocity Analysis

by Adam Cybanski

38

Power Loss Issues Related To Edr Data In 2013-2017 Kawasaki Ninja 300 And Zx-6R Motorcycles

by Edward C. Fatzinger Jr.

48

The Impact Of Nonlinear Boundary Geometry Considerations In Regards To Residual Damage Based Model Coefficients, Equivalent Barrier Speed And Internal Work Absorbed

by Jai Singh

68

Small Unmanned Aircraft Systems Photogrammetry vs.

Total Station

by Joseph Weadon

86

My Turn at the Wheel: " Driver’s Early Arrival at the Scene Caused Accident?"

by Erik Carlsson

88

Motorcycle Accident Reconstruction: Applicable Error Rates for Struck Vehicle EDR-Reported Delta-V

by Nathan Rose, William Bortles, Neal Carter

112

Documenting A High Speed, Rear End, Partial Overlap, Crash Test Of A Large Sedan & Stationary Commercial Trailer

by Craig Proctor-Parker www.collisionpumagazine.com

Volume 13 Issue 1.indd 3

68

88

112

Collision Magazine - Volume 13 Issue 1 3

5/28/2019 9:34:17 AM

ADVERTISING RATES & INFO Crash Data Group Inc PO Box 892885, Temecula, CA 92589 Toll Free: 800-280-7940 E-mail: sbaker@crashdatagroup.com Web: www.collisionmagazine.com ISSN: 1934-8681

COLLISION STAFF Scott Baker Sean Haight Tonya Baker Courtney Baker

Owner, Managing Editor Senior Editor Advertising Manager Subscription Services

All rights reserved, © 2019 Cash Data Group 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 e-mail to the editor at sbaker@crashdatagroup.com or sending a "regular mail" letter to us at PO Box 892885, Temecula, CA 92589. By

sending correspondence to either our e-mail 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 e-mail 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.

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

ADVERTISER INDEX

PAGE

4N6XPRT Systems Collision Magazine Collision Safety Institute Crash Academy Crash Data Group The Crash Hub Engineering Dynamics EDR Summit Houston Auto Appraisers Laser Technology Inc Leica Geosystems Northwestern CPS Tesla EDR Kit VirtualCRASH

Inside Back 4 61 46, 62 128, Back 110 Inside Front 47 62 37 2, 63-66 7 67 1

Included with this issue of Collision

is the 2019 EDR Summit presentation files (PDF).

Use the link and password below to access the EDR Summit presentation files that were made available.

Link: http://www.crashdatagroup.com/edr2019/ Password: 2019subaru

SUBMIT AN ARTICLE If you would like to submit an article to be published in Collision please review the following webpage for instructions: http://www.collisionmagazine.com/submit/

BACK ISSUES

If you would like purchase back issues of Collision Magazine, please visit http://www.collisionmagazine.com/back-issues/

www.collisionpumagazine.com

Volume 13 Issue 1.indd 5

Next EDR Summit Houston, Texas March 9-11, 2020 www.edrsummit.com Collision Magazine - Volume 13 Issue 1 5

5/28/2019 9:34:18 AM

Letter from the editor We would like to thank our loyal subscribers as we transitioned and merged certain companies over the past year. In the last issue's "Letter from the Editor", we were completing the merger of Collision Publishing and Crash Dtata Group Inc. We have now also finalized merging The ARC Network into Crash Data Group Inc. Although these tasks were more difficult and time consuming than anticipated, I believe we are in a position to bettter serve all of our customers and the industry in general as Crash Data Group Inc. As you hold this issue of Collision Magazine (Vol 13, Issue 1) we have peer reviewed and editorial reviewed articles lined up through the next two issues. As always, if your would like to contribute to Collision Magazine, please email us your abstract for review. All of the articles published in Collision Magazine are fresh, new articles and research.

We do not publish "I found this interesting stuff on the internet..." or articles that have already been published and circulated.

For this issue, we had planned certain topics for inclusion but due to the availibility of additional research and the number of quality articles submitted and approved, we are moving some of those articles to the next issue of Collision Magazine, Vol 13, Issue 2. For example, the Toyota recording order article planned for this issue has been bumped to the next issue to include additional information about a better defined trigger bias, brake versus acceleration data and, if all goes as planned, additional research into data which might compliment and augment Toyota crash data in a way not currently available.

Summit set a new standard for this type of conference, the data and information presented, and has become the basis for a suite of articles and papers in this and the coming issues of Collision Magazine. Having adapted our publishing schedule to the EDR User's Summit and now banking a depth chart of interesting and meaningful articles and papers for the upcoming issues, Collision Magazine will continue to be a leading source for relevant crash research information.

Another development readers of Collision Magazine might find interesting and useful is the newly developed web site https://www. thecrashhub.com/. As mentioned earlier, Crash Data Group has merged the The ARC Network (www.accidentreconstruction.com) and EDR Experts (www.edrexperts.com) into one comprehensive website - The Crash Hub. Locating local and regional experts, finding training opportunities, initiating research and other crash related topics have been brought together in one easy to use website.

A specialized experts referral database, and more make The Crash Hub a vital emerging resource for crash reconstructionists, insurance professionals and potential clients. Check it out at www.thecrashhub.com and give it a try. Join now and receive one month free.

Related to this issue of Collision, the 2019 EDR User's Summit was the single most widely attended summit for vehicle crash data of its kind ever. Highlighting new technology, the future of the EDR Tools and new technologies such the the CDR 900, the 2019 EDR User's 6 Collision Magazine - Volume 13 Issue 1

Volume 13 Issue 1.indd 6

Managing Editor

www.collisionmagazine.com

5/28/2019 9:34:18 AM

Management and Leadership Highway Safety Crime Scene and Forensic Science Police Motorcycle Community-Police Partnerships

Every Dimension of Public Safety Only at Northwestern.

More than 50 courses providing professional skills with a leadership perspective. On campus, around the country and online.

Find the right course for your department. nucps@northwestern.edu or 800-323-4011

nucps.northwestern.edu www.collisionpumagazine.com

Volume 13 Issue 1.indd 7

Collision Magazine - Volume 13 Issue 1 7

5/28/2019 9:34:18 AM

8 Collision Magazine - Volume 13 Issue 1

Volume 13 Issue 1.indd 8

www.collisionmagazine.com

5/28/2019 9:34:19 AM

crash

verb\'krash\ 1. to collide violently with an obstacle or another vehicle.

-ol·o·gy

noun\'äl-ə-jē\ 1. a subject of study; a branch of science.

-ology is a back-formation from the names of certain disciplines. The -logy element basically means "the study of ____". Such words are formed from Greek or Latin roots with the terminal -logy derived from the Greek suffix -λογια (-logia), speaking, from λεγειν (legein), "to speak". Through the years -ology and -logy have come to mean, "study of" or "science of" and either of these suffixes often utilize the form of –ologist.

crash·ol·o·gy

noun\'krash-'äl-ə-jē \ 1. The science of crashes.

So, Crashologists, welcome to Crashology – The Science of Crashes. In this new, recurring feature of Collision – The International Compendium for Crash Research, the authors plan to address a wide range of topics centered around the study of crashes and crash reconstruction. We plan to offer crash test reviews with a focus toward the validation of new or existing reconstruction methodologies and insight into the ever-increasing data we recover from vehicles. Original research and testing are planned in an effort to continue to broaden the information and data available to crash investigators and reconstructionists. Finally, there will be technical articles related to issues and concepts relevant to the collision reconstruction community.

When someone describes a project as, “It’s a work in progress,” it is often with the intent of explaining why the current state of the project might look unfinished or not conform to other examples of similar work. We proudly consider this a “work in progress” because we recognize that the research in this field is never done and we have no existing model as our guide. We intend to offer topics of interest, always with the goal of pushing forward the science of crashes, which we now call, “Crashology.”

The topic of the inaugural edition of Crashology is a recent development in the digital age of automobiles, Vehicle System Forensics for Crash Reconstruction.

www.collisionpumagazine.com

Volume 13 Issue 1.indd 9

Collision Magazine - Volume 13 Issue 1 9

5/28/2019 9:34:19 AM

Vehicle System Forensics for Crash Reconstruction Wesley Vandiver collision forensics, inc

Robert Anderson Biomechanics analysis

10 Collision Magazine - Volume 13 Issue 1

Volume 13 Issue 1.indd 10

www.collisionmagazine.com

5/28/2019 9:34:21 AM

August 2017: In the early hours of the morning, several individuals get into an altercation at a bar. The incident moves outside to the streets where an 18-year-old throws a bottle, damaging the side-view mirror of a red Ford Mustang belonging to one of the antagonists. The 18-year-old flees with a companion on a moped. Meanwhile, the Mustang owner, in a rage, pursues them at high speed, until moments later the Mustang crashes into the moped from behind, leaving one rider dead, another severely injured, and a homicide case for detectives to investigate.

There were differing stories from witnesses, gaps in the evidence gathered from closed circuit cameras, forensics from two vehicles and the crime scene itself. All left key questions about precisely what had happened during the chase. Prosecutors and police were pursuing a charge of “death by dangerous driving” instead of murder, based on the initial investigation by the Metropolitan Police Service in London.

But a relatively new source of digital evidence—the data stored inside the vehicle systems —revealed that the driver of the Mustang was accelerating as his car approached and at the precise moment of impact with the moped. Additional data was found on the vehicle systems that showed deliberate actions taken by the driver of the vehicle to harm the riders of the moped, as well as pinpointing the precise route taken from the bar to the location of impact.

With this new evidence in hand, the charge was elevated to murder…[1]

such as roadway markings, vehicle damage, and recorded data from event data recorders (EDRs). Digital forensics specialists routinely work with digital data from electronic devices, such as cell phones and computers. Digital forensics is the practice of preserving, gathering and presenting evidence from digital devices for the purposes of criminal or civil investigations. [5] Key evidence in the above case included both crash evidence and forensic data recovered from the vehicle’s infotainment and telematics system.

Data acquired from these systems as well as others in the vehicle are rapidly growing sources of critical evidence.

Crash investigators routinely acquire and analyze electronic data in their day-to-day investigations. The Crash Data Retrieval (CDR) system brought to market by Vetronix Corporation and subsequently acquired and expanded

by Bosch Corporation has, for nearly two decades, been a valuable source of crash data for investigators. Over time, the available data from supported vehicles have grown from obtaining a partial, single-axis crash pulse to modern data that include a number of pre-crash parameters. The availability of EDR data brought crash investigators into the digital forensics world and introduced them to the processes of data acquisition and data analysis.

A new and different source of automotive forensic data has emerged in recent years – infotainment/telematics system data. These data are acquired using the iVe system, developed by Berla Corporation, in Annapolis Maryland. [6] The iVe ecosystem is a collection of tools that consists of a mobile app used for identification and system removal, a hardware kit for data acquisition, and forensic software used for data analysis.

Figure 1. Crime scene photo courtesy Metropolitan Police The Metropolitan Police Service in London is one of the most technologically advanced law enforcement agencies in the world. This case required the cooperative efforts of investigators from multiple disciplines and the result was a successful murder prosecution. [2] [3] [4] Crash investigators are generally accustomed to working with evidence www.collisionpumagazine.com

Volume 13 Issue 1.indd 11

Figure 2. iVe Mobile Application

Collision Magazine - Volume 13 Issue 1 11

5/28/2019 9:34:21 AM

Figure 3. Discovering Vehicle Data Together, these components allow a user to identify supported systems, access removal instructions for those systems that require removal, establish proper hardware and software connections to the system, acquire available data, conduct analysis of the data, and generate reports.

Vehicle System Forensics is a three-step process – Identify, Acquire, and Analyze.

Figure 4. Vehicle Forensic Process What types of data can be recorded by a vehicle’s infotainment/telematics systems?

These data can be categorized as, 1) Connected Devices, 2) Location Data, and 3) Vehicle Events.

Connected devices can include phones connected to the system, including specific phone identifiers, contacts, and call logs. Other devices can include devices connected to the vehicle via USB, Wi-Fi, or Bluetooth connections.

Location data can include user-inputted destinations, saved locations, and/or tracklogs (breadcrumb trial) detailing a history of everywhere the vehicle has traveled. This information is generally accompanied by date/time stamps and a derived vehicle speed at each track point

Figure 5. Acquisition of connected devices from a system 12 Collision Magazine - Volume 13 Issue 1

Volume 13 Issue 1.indd 12

Although the frequency at which track points are recorded varies by manufacturer and system, many vehicles are equipped with systems that record the vehicle location at intervals of 1 Hz. Some vehicles are also equipped with systems that not only record

www.collisionmagazine.com

5/28/2019 9:34:21 AM

the GPS track points at 1 Hz, but also record the reported wheel speed signal from the vehicle’s CAN bus (see figure 9).

Figure 6. Map of vehicle track log

Figures 7 and 8. Map of vehicle track log showing data for specific points (vehicle was driven along same route on different dates) www.collisionpumagazine.com

Volume 13 Issue 1.indd 13

Event data can include a broad range of potential recordings, including when and where the vehicle was when devices were connected, gear shift selections were made, doors were opened or closed, etc. Instances of hard braking or acceleration can also be identified. Events associated with location data (GPS coordinates) can be mapped alongside tracklogs (when available) so that the motion of the vehicle along with recorded events can be displayed geographically (see figure 10 and 11).

Obviously, the iVe system can acquire data that often fall outside what is recorded by EDRs. For example, although vehicle speed data can potentially be acquired from both the EDR and the vehicle infotainment/telematics system for the same vehicle, infotainment/telematics system data most commonly are GPS based, so the speed data are coupled with location data, as well as possibly date and time stamps.

These track logs can, depending on the source system, include data going back weeks, months, or years. One of the most notable differences between EDR data and infotainment/telematics data is that EDR data is written to memory as a result of an event. Generally speaking, when the rate of change in the acceleration of a vehicle falls within certain parameters, the airbag system is awakened for the purpose of analyzing the event for the potential deployment of safety system devices (e.g., airbags, seat belt pretensioners, etc.). Thus, in CDR data files, we commonly deal with deployment events and non-deployment events, which are defined as events that awakened a system, qualified for recording, but did not result in the deployment of any devices.

The Federal Regulation, Title 49 Transportation, CFR Part 563 – Event Data Collision Magazine - Volume 13 Issue 1 13

5/28/2019 9:34:22 AM

Figure 9. Vehicle velocity log (CAN bus data)

Figure 10. Vehicle event data

Recorders, commonly referred to as Part 563, applies to vehicles manufactured on or after September 1, 2012. [7] For those vehicles intended to be Part 563 compliant, simply awakening the system is not a guarantee that data related to a non-deployment event will be written to memory.

Part 563 defines an “event” as, “…a crash or other physical occurrence that causes the trigger threshold to be met or exceeded, or any non-reversible deployable restraint to be deployed, whichever occurs first.” “Trigger threshold” is 14 Collision Magazine - Volume 13 Issue 1

Volume 13 Issue 1.indd 14

defined as, “…a change in vehicle velocity, in the longitudinal direction, that equals or exceeds 8 km/h within a 150 ms interval. For vehicles that record “delta-V, lateral,” trigger threshold means a change in vehicle velocity in either the longitudinal or lateral direction that equals or exceeds 8 km/h within a 150 ms interval.” Most manufacturers supported by the Bosch CDR system have adopted the upper bound of this definition, 8 km/h in 150 ms, as their threshold for recording a non-deployment event.

www.collisionmagazine.com

5/28/2019 9:34:22 AM

Figure 11. Vehicle events and tracklog on map Therefore, many low-speed collisions and low delta-V col- References lisions, (e.g., automobiles versus pedestrians and bicyclists) do not result in the recording of EDR data. Infotainment/ telematics system data are not crash initiated or related to any trigger other than the vehicle motion and occurrences of the events themselves.

While certain EDR data can be “locked,” preventing the overwriting of that data, vehicle infotainment/telematics data is not locked and its life in memory is dependent upon the specific system. Although certain systems support physical acquisitions that can recover some data that has been deleted and resides in unallocated space, this is not true of all systems. Therefore, if a vehicle under investigation continues to be operated after the incident in question, the acquisition of infotainment/telematics data should be given consideration in the early stages of an investigation.

[1] LeMere, Ben and Bollö, Joel, Vehicle Forensics, A Rapidly Growing Source of Critical Digital Evidence, Evidence Technology Magazine, Winter 2018 [2] Ford Mustang Driver Jailed for Revenge Crash Murder, BBC News, May 4, 2018, https://www.bbc.com/ news/uk-england-london-44004012 [3] Mustang Owner Drove at and Killed Teenager After Bottle Thrown at his Prized Car, The Telegraph, May 4, 2018 [4] Sharman, Jon, Driver Chased Moped Rider After Wing Mirror Broken, Ran Him Down the Beat Him as He Lay Dying, Independent, May 2, 2018 [5] Cheah, Madeline, The Need for Digital Forensics in the Automotive World, Automotive Testing Technology International, March 26, 2019

Vehicle Systems data can serve as valuable evidence in nearly any investigation involving an automobile. This is certainly true in cases involving the investigation of a [6] https://berla.co/ crash. Whether the investigator is searching for vehicle [7] https://www.govinfo.gov/content/pkg/CFR-2011-timotion data, such as location and speed, or device data, tle49-vol6/pdf/CFR-2011-title49-vol6-part563.pdf such as phone calls and/or texting, such data may recorded in the vehicle, and if so, is now able to be acquired.

www.collisionpumagazine.com

Volume 13 Issue 1.indd 15

Collision Magazine - Volume 13 Issue 1 15

5/28/2019 9:34:22 AM

DASH CAMERA VIDEO VELOCITY ANALYSIS Adam Cybanski Gyro Flight & Safety Analysis Inc.

O

verview On 20 June 2018, a vehicle performance engineer from the Office of Research and Engineering at the National Transportation Safety Board (NTSB) met with the founder of Gyro Flight & Safety Analysis in Ottawa, Canada. Dash camera refresher training was conducted and covered a variety of topics including frame extraction, frame rate calculation, lens distortion, focal length estimation, tracking, geoidentification and solving camera motion.

The author (Adam Cybanski) carried out live testing in the area. As part of a practical exercise, he drove down a stretch of road for speed testing. This was captured using a dash camera, and a racing quality GPS. An initial analysis of the dash camera footage was conducted during the refresher training. Comprehensive camera analysis followed over several months, and is now provided below in order to compare the velocities extracted from dash camera video, with the high-performance GPS data.

16 Collision Magazine - Volume 13 Issue 1

Volume 13 Issue 1.indd 16

Background

As an aircraft accident investigator, Adam Cybanski has been extracting velocity information from witness video since 2008. Cockpit cameras, similar to dash cameras, ramp cameras similar to traffic cameras, and handheld witness video were photogrammetrically analyzed in order to derive the velocities of the cameras, and the vehicles seen in the field of view as part of aircraft accident investigations.

In 2015 the author also started assisting the local police with investigations of traffic accidents that were caught on video. The techniques developed for analyzing video of aircraft accidents are now employed for traffic accident reconstruction.

Velocity analysis from witness video is based on three workflows: matchmoving, geoidentification and time-distance. Matchmoving is the calculation of camera and object position from video. Geoidentification involves identifying references from a video in real world coordinates (lat/long, UTM grid). Time-distance requires analysis of

www.collisionmagazine.com

5/28/2019 9:34:23 AM

video frame timings and consolidation with distances to calculate speed.

camera correctly to match the actual camera that captured the original imagery.

Matchmoving is a process in film making which aims to insert computer graphics into live-action footage, with correct position, scale, orientation and motion relative to the background image. This has the effect of making the CGI content blend seamlessly into the live footage, but requires careful photogrammetric analysis of the video using special software. Analysis is used to determine exactly where the camera was in 3D space, what its orientation was, and the 3D location of any objects of interest.

The end result of geoidentification and matchmoving is typically position and distance information. Analysis of video can yield up to 30 measurements per second, and the timing between each measurement is not always constant. Using detailed analysis, a precise estimation on each image's time must be calculated, then combined with distance estimates to yield velocities. The resulting data then undergoes statistical techniques in order to produce derived plots of the velocities over time.

Geoidentification involves the designation of identifiable features in video, and determining measured coordinates for them based on their real-world location. Sources for this data are typically surveys made onsite, but resources such as Google Earth can be used in their place. The aim of this information is to help the software determine the scale of the scene under analysis, and the relative location of the identifiable features so that it can orient a virtual

Speed Testing

Speed testing was conducted in Ottawa. A vehicle was fitted with two GPS receivers and a dash camera. The author drove down a nearby road at different speeds. The intent was to compare speeds derived from video velocity analysis of the dash cameras with truth data from the GPS receivers.

www.collisionpumagazine.com

Volume 13 Issue 1.indd 17

Collision Magazine - Volume 13 Issue 1 17

5/28/2019 9:34:24 AM

A high-performance racing GPS, the Qstarz BT-Q1000eX GPS lap timer recorded the vehicle movement, as did a R300 dash camera GPS module. After testing, the data from both devices were recovered and position information was imported into Google Earth to review the ground traces. Testing was captured by both GPS units, but it was evident that the BT-Q1000eX was higher resolution and precision. Speed information was exported to Excel for further analysis and comparison.

Although it was understood that the capabilities of the R300 dash camera GPS would be limited, it served to validate the data collected by the BT-Q1000eX. With the velocity plots aligned, the accelerations and decelerations showed a high degree of correlation. Both GPS units showed similar velocity ranges over the testing timeframe.

Traffic cones were placed at 20m intervals down the right side of the roadway. Velocity was calculated using the timing that each cone passed by the vehicle in the dash camera video. These speeds validated the speeds from both GPS receivers in an independent manner.

Video Velocity Analysis

A dash camera attached to the center of the windshield successfully recorded all the testing. Several passes down the road were chosen for video velocity analysis.

Accurate times were needed for each frame before velocities could be calculated. The data embedded in the video suggested it had a frame rate of 30.00030 frames per second. Duplicate frames were found and deleted. Once the duplicates were removed and the frames renumbered consecutively, frame rate analysis was conducted. A unique frame rate of 24.95 frames per second was calculated.

With this value, each frame represented a time interval of 1/24.95 = 0.04 seconds. Similar frame rate measurements were conducted on the second and third pass clips. The results were 24.98 and 24.97 frames per second respectively.

In order to compare methods, lens distortion was estimated in two different ways. For the first pass, a multiple reference straight line adjustment was made. For the second and third passes, a calibration grid was filmed and detailed distortion analysis was conducted on the footage.

Identifiable features in the imagery were designated and tracked for as long as possible. These trackers were also identified in Google Earth, and corresponding UTM coordinates were transcribed. The location information was imported back into video analysis software so that it could determine where the dash camera was located and how it was oriented for each frame of video.

18 Collision Magazine - Volume 13 Issue 1

Volume 13 Issue 1.indd 18

The dash camera position information was matched with estimated timings for the corresponding frames, and velocities were calculated in the X (N-S) and Y (E-W) directions. These velocities were added together in order to estimate speeds down the road. The results were plotted against time to show the acceleration profile.

Cubic spline interpolation was used to smooth the unrealistically noisy data. This method employs an elastic coefficient that strikes a balance between keeping the curve shape and reducing spurious data. Goodness of fit statistics can be calculated on the curve, and reasonable values sampled for the required range. The spline interpolation has the effect of improving the data to reflect real constraints such as vehicle limitations on vertical and lateral movement.

The combined X and Y data for the first speed pass were imported into mathematical analysis software. Fitting a spline onto the data with a smoothing parameter of 0.08 produced a plot that clearly kept the curve shape but eliminated spurious fluctuations. Spline fits were also produced for the other two passes.

Plotting the velocity data from the primary sources, there appeared to be very strong agreement between the analysis and truth data. Minor fluctuations were visible in the dash velocity plots.

Examining all three passes together, the velocity analysis very closely matched the underlying 10Hz GPS data, both at 60 km/h and 80 km/h. The acceleration and deceleration were captured each time, and only minor variations were noted. The overall analysis suggested that the dash camera video analysis process described in this report produced speeds within 2 km/h of the true speeds.

Findings

Over the course of the video analysis, several things were found based on the dash camera video footage from Leitrim Road: a. Analysis of the test vehicle dash camera video suggested that it accelerated to a maximum of 60 km/h on the first pass, 60 km/h on the second pass, and 80 km/h on the third pass. b. Vehicle deceleration was accurately captured on all three passes, and acceleration was also captured on the third pass analysis. c. The velocities produced through dash camera analysis were commensurate with truth data captured by a high-performance GPS speed logger, generally within 2 km/h.

www.collisionmagazine.com

5/28/2019 9:34:24 AM

Figure 1: Dash camera analysis

Figure 2: Third pass distance vs time www.collisionpumagazine.com

Volume 13 Issue 1.indd 19

Collision Magazine - Volume 13 Issue 1 19

5/28/2019 9:34:24 AM

Figure 3: Speed plot comparison Conclusions

The dash camera analysis yielded velocity data that was very much in agreement with the high-performance GPS data. This suggests that accurate and useful speed information can be extracted from video, even if an onsite survey is unavailable. Analysis is not limited to investigators in the vicinity of the accident site, but can be conducted by experts from around the globe.

Video has typically been seen as qualitative evidence for traffic accident collision reconstruction. This and previous testing has shown that when properly analyzed, video can reveal accurate quantitative data as well. Despite being a relatively new capability, video velocity analysis has been used in several court cases and at this point in time has not been contested.

Video Forensic Examiner The video forensic examiner and author of this report is Mr Adam Cybanski, a qualified accident investigator and video velocity specialist that has been leading the industry in analysis of video for velocity. Mr Cybanski holds a BSc in Computer Mathematics and gained his Investigator In Charge level 3 in 2012 at the Directorate of Flight Safety in Ottawa, Canada. Mr Cybanski has also studied photogrammetry and traffic accident reconstruction. Throughout his years of experience Mr Cybanski worked on velocity and motion extraction from video on a professional level within the aircraft and traffic accident investigation communities domestically and internationally. As founder of Gyro Flight and Safety Analysis Inc., his key responsibilities over the past five years included: audio analysis, video analysis, velocity extraction, accident reconstruction, accident visualization, accident site 3D modeling and employment of UAVs to support accident investigation. Mr. Cybanski has been recognized in court as an expert in the field of Video Velocity Analysis.

Gyro Flight and Safety Analysis Inc. provides expert video analysis and accident reconstruction services in the areas of 20 Collision Magazine - Volume 13 Issue 1

Volume 13 Issue 1.indd 20

video forensics and velocity extraction from video. Based in Ottawa, Canada, it offers an impartial, independent and specialised service to police, military and civilian prosecution and defence. The company adheres to a strict nondisclosure policy in relation to all case files worked on and abide by the APCO guidelines and the Data Protection Act 1998.

A. Cybanski BSc, IIC3 CEO Gyro Flight & Safety Analysis Inc. cybanski@gyrosafety.com The above person – hereinafter to be referred to as ‘the person concerned’ – maintained strict confidentiality of all of the information and insights he had obtained in respect of the investigation referred to above until it was cleared for release.

Annex A Velocity Reference Data Overview

The author drove a sedan down a nearby roadway at different speeds. Their passes were recorded with an onboard dash camera. In order to assess the accuracy of derived dash camera video velocity analysis, GPS data was also collected from two onboard receivers.

A standalone Qstarz BT-Q1000eX GPS lap timer, and a R300 dash camera GPS both recorded the speed testing.

After the testing, the data was downloaded from both devices and reviewed in the associated software applications.

The data was subsequently exported to Google Earth, where the track could be overlaid on the actual roadways to verify the validity of the data. Qstarz data including speed could be exported into Excel for subsequent review, but the dash camera GPS speed information had to be transcribed from the dash camera playback, as the application could not export anything other than position

www.collisionmagazine.com

5/28/2019 9:34:24 AM

information. Velocity data from both GPS units on the analyzed speed run was compared.

UTC on the same day. In total, 15,403 samples were recorded during the testing period.

Traffic cones were placed at 20m intervals down the right side of the roadway. Velocity was calculated using the timing that each cone passed by the vehicle in the dash camera video. These speeds were employed to validate the speeds from both GPS receivers in an independent manner.

Speed Passes Once cones were placed on the roadway, the test vehicle (a 2013 Hyundai Sonata) was positioned on the north side of the road heading southwest. The GPS and dash camera were confirmed on and running, then the vehicle accelerated to 60 km/h with reference to the automobile speedometer. The goal was to attain this reference speed prior to reaching the traffic cones, and maintain the speed until the cones were no longer visible.

After the first pass, the test vehicle was driven back to the starting position, the equipment checked, then a second test pass was conducted, again to 60 km/h with reference to the speedometer. The automobile returned for a third and final pass, this time at 80 km/h. Once the passes were completed, the GPS and dash camera were switched off, and the cones were retrieved.

Figure 4: GPS unit placed on the dash during testing QRacing saved the gpx data into a 6.7MB kml file employing gpx2kml.com conversion. The track captured by the BT-Q1000eX matched the Google Earth imagery nicely, and exhibited significant resolution that clearly discerned the individual passes down Leitrim Road.

Qstarz BT-Q1000eX GPS

The primary GPS reference was the BT-Q1000eX made

by Qstarz. It is designed for logging speeds for extreme sports, specifically super moto, road course motorcycles, oval or road course autos. It is easy to use with a simple on-off activation switch. The GPS is a powerful 10Hz data logger with 8MB memory size able to log up to 400,000 waypoints with a 42-hour battery life.

The BT-Q1000eX employs the MTK II chipset with high sensitivity -165dBm and 66 channel tracking. It has a built-in patch antenna with LNA an L1 frequency of 1575.42MHz and C/A Code 1.023MHz chip rate. The stated accuracy is 3.0m 2D-RMS <3m CEP(50%) without SA (horizontal) and a velocity accuracy of 0.1m/s without DGPS aid.

Once testing was completed, data was downloaded from the GPS into the supporting software, Qstarz – QRacing Version 3.60.000 Build Date 8 June 2016. The track was subsequently exported as gpx, Excel csv and Google kml files. The Excel file was comprehensive, providing an index, UTC and local times to the second, time in milliseconds, latitude, longitude, altitude, speed and heading information every one tenth of a second. The track started at 17:37:51 UTC on 19 June 2018, and ended at 18:04:16

Figure 5: Google Earth track for BT-Q1000eX The testing runs needed to be identified in the data. Although the GPS recorded 26 minutes of data, the three passes took less than 15 seconds each. The relevant passes were identified by the speed and track plots. Plotting the speed showed a smooth, high resolution (10 Hz) graph that depicted six peaks in speed. These peaks represented the three logged East to West speed passes, along with the return route back to the starting point. This was verified by the alternating GPS tracks of 60 and 240 degrees. Time was measured in seconds from the point where the vehicle first started moving. The GPS-calculated speed was in km/h.

www.collisionpumagazine.com

Volume 13 Issue 1.indd 21

Collision Magazine - Volume 13 Issue 1 21

5/28/2019 9:34:24 AM

R300 Dash Camera GPS

The back-up GPS reference was the R300 external GPS unit for the dash camera. It is an add-on for a low-cost dash camera, and very little information about it was available. The unit attaches to the camera with a long cord.

Once plugged in, position data is encoded into the dash camera video, and can be extracted using the supporting software.

Testing was conducted with the R300 dash camera attached to the windshield, and the GPS attachment tethered to the camera. The resulting video was called G2012-01-01-0024-57.avi. In order to access the embedded GPS data, the file had to be opened in the software application X2Player. exe, version 2.5.6.0 dated 2014-07-04 12:14 PM. The software allows the user to review the forward and backward camera footage with typical VCR controls. If the GPS module was attached during recording, the software will also display the instantaneous latitude, longitude, and speed in orange at the top of the screen.

measurement devices, the BT-Q1000eX could be used as velocity truth data with confidence.

Visual Measured Reference Speed Analysis

To provide a non-GPS confirmation of reference speeds, eleven orange cones were placed down the right side of the roadway at 20m intervals. On the shoulder of the road, they could be seen passing by the front of the vehicle in the dash camera footage.

Figure 7: Cones placed with 20m spacing

Figure 6: Dash camera interface software The velocity information could not be directly exported from the application. In order to extract this information, the sequence was played back in the interface software and recorded onto video. On review, all distinct changes in velocity were transcribed, along with the corresponding dash camera time stamp. This produced a course plot very similar to the BT-Q1000eX. Although it was understood that the capabilities of the R300 dash camera GPS would be limited, it served to validate the data collected by the BT-Q1000eX. Once the velocity plots were aligned they showed a high degree of correlation in both acceleration and deceleration. Both GPS units showed similar velocity ranges, over the testing timeframe. With two independent agreeing velocity

22 Collision Magazine - Volume 13 Issue 1

Volume 13 Issue 1.indd 22

By noting the frame number and timestamp as each cone left the view over the vehicle dash, and knowing the distance between each cone, the speed of the vehicle could be calculated using cubic spline interpolation. Speeds near 60 km/h for the first pass, 60 km/h for the second pass, and 80 km/h for the third resulted. The speeds at each cone passage were plotted along with the previous GPS data, and matched well, supporting the BT-Q1000ex speeds. d (m) 0 20 40 60 80 100 120 140 160 180 200

Pass1 t(s) 0.00 1.12 2.24 3.36 4.52 5.72 6.96 8.12 9.32 10.60

Pass2 t(s) 0.00 1.20 2.40 3.64 4.84 6.04 7.29 8.49 9.73 10.93 12.13

Pass3 t(s) 0.00 0.90 1.80 2.70 3.60 4.53 5.43 6.36 7.30 8.23 9.16

Figure 8: Cone passage timings

www.collisionmagazine.com

5/28/2019 9:34:24 AM

Figure 9: Combined reference speeds Speed Comparison

The 10Hz data logger GPS provided the highest resolution data, and appeared very accurate. It did not provide any data during the first pass acceleration, perhaps because it had not yet locked enough satellites, but worked well for the remainder of the testing. The dash camera GPS produced speeds at a relatively low 1 Hz, but closely aligned with the 10Hz GPS speeds. The calculations from the cone passing were of a very short duration, but again reflected the same speeds obtained from the GPS units. Comparing the data from the two GPS units and the speeds obtained through passage of the reference cones, a very high confidence was placed in the 10Hz GPS data for the purposes of this testing.

Summary

Speed data was collected from two independent GPS sources and by timing the passage of cones placed at measured distances. All three sources agreed, although the BT-Q1000eX provided much better temporal resolution at 10 position readings per second. The associated speed calculations were employed as truth values for comparison with dash camera video velocity analysis.

Annex B Dash Camera Analysis

coordinates were noted for each. Other distinct features were also manually tracked.

The video sequences were matchmoved to derive camera/ vehicle position. The On Screen Display (OSD) time information from each frame was transcribed, and compared to the frame rate timing. Distinct time values were calculated for each frame. The derived position and time information were matched and used to calculate velocity for each frame of the test passes.

R300 Dash Camera

The R300 is an inexpensive and widely available Dash Camera that was purchased off of eBay. It was bought on February 2016 for $50USD, and as of the time of this article is still available for purchase. Although it cannot be considered high quality, its output was seen as being representative of the type of dash camera footage that could be available from an accident.

The dash camera has a 2.7” LCD screen on the back, a G-Sensor, a microphone to capture cabin audio, and an external tethered GPS module. A front 140-degree wide angle lens captures video at 640x480 resolution, while a rear-facing 120-degree lens can also capture video at the same resolution. The camera includes a 12V adapter to power the unit, and a windshield mount.

Overview

An inexpensive dash camera was affixed to the test vehicle windshield in order to record the testing. The resulting video was subsequently downloaded off the camera. Three passes down the road were chosen for velocity analysis. The sequence was broken down into individual images which were sequentially numbered and documented. Duplicate frames were noted and removed.

The imagery underwent geoidentification. Identifiable features were designated and tracked. These features were also identified in Google Earth, and corresponding UTM

Figure 10: R300 dash camera www.collisionpumagazine.com

Volume 13 Issue 1.indd 23

Collision Magazine - Volume 13 Issue 1 23

5/28/2019 9:34:25 AM

The dash camera was mounted in the center of the windshield for testing. Power was applied, and the unit continuously recorded video until the vehicle returned and the dash camera was shut off. The result was an avi video file, titled G2012-01-01-00-24-57. The video ran for 15 minutes and 40 seconds. The associated file size was 1.17GB, recorded on a 32GB microSD card.

could be verified. The OSD time changed from 00:24:57 to 00:24:58 at frame00032. It changed from 00:40:33 to 00:40:34 at frame28110. Over the 936 second period, there were 28078 frames, resulting in a calculated frame rate of 28078/936= 30.0 frames per second

Individual png frames were extracted from G2012-01-0100-24-57 starting at 00:24:57 and they were sequentially named frame0000.png to frame28168.png. The images were cropped on the left, to only show the front camera view and not the back camera. Three subsets of the images were set aside, representing the three test passes. The new sequences were reviewed frame by frame, and any duplicates (identified by a lack of movement), were deleted.

The OSD time for each frame, along with its unique frame number were transcribed into Excel for subsequent temporal analysis. The frames were then renumbered so that the sequence again began at Frame0000.png, but this time at dash camera times 00:30:45, 00:32:05 and 00:34:15. The result was continuously numbered sequences of unique images covering the three test runs.

Lens distortion analysis

The calculated 30 frames per second does not tell the whole story, as duplicate frames were found and deleted prior The videos were opened in VLC media player in order to to analysis. Once the duplicates were removed and the view the codec information. The video codec was Motion frames renumbered consecutively, the same sort of frame JPEG Video (MJPG) with a resolution of 1280x480. The rate analysis was conducted. The OSD transitioned to front camera imagery was on the left side of the frame, 00:30:46 at frame 25, then transitioned later to 00:31:06 640 pixels wide, while the rear camera view was shown at frame 524. This was found to represent a rate of (524on the right. The indicated frame rate was 30.000300, 25)/20=24.95 frames per second, the unique frame rate. and the decoded format was Planar 4:2:2 YUV full scale. With this value, each frame represented a time interval of Audio was mono PCM S16 LE (araw) with a sample rate 1/24.95 = 0.04 seconds. of 16160 Hz and 16 bits per sample.

The final frame timings were not yet complete. As statThe footage was reviewed. The first suitable sequence for ed, there were duplicate frames which had to be removed video velocity analysis was at dash camera time 00:30:45. before analysis could be carried out. With five duplicate This segment started with the vehicle accelerating from the frames per second, the question was whether the dupliright side of the road, moving left into the traffic lane, go- cates were evenly spaced and each image recorded with a ing by the orange test cones, then decelerating. It fit well 1/25 second interval, or if the images were recorded with with the type of dash camera video that might be collected a 1/30 second interval, followed by an extra interval after from an accident, and also captured significant accelera- every 5th image. Previous testing had confirmed that the tion which can be more difficult to derive speeds for. initial scenario, with a constant frame rate of 24.95 frames per second was correct.

An additional two sequences were selected for dash camera video analysis, all starting from the same general area. The Similar frame rate measurements were conducted on the second sequence started at 00:32:05, and the third started second and third pass clips. The results were 24.98 and at 00:34:15.

24.97 frames per second respectively.

Temporal Analysis

Accurate times were needed for each frame before velocities could be calculated. The data embedded into the video suggested it had a frame rate of 30.00030 frames per second. By closely examining the OSD timings, this 24 Collision Magazine - Volume 13 Issue 1

Volume 13 Issue 1.indd 24

In order to compare methods, lens distortion was estimated in two different ways. For the first pass, a multiple reference straight line adjustment was made. For the second and third passes, a calibration grid was filmed and detailed distortion calculated from the result.

For the first pass, a frame showing three telephone poles and the bottom dash line were selected at a point just prior to pole passage by the vehicle. The frame exhibited significant curvature for the nearby poles and the dashboard.

A straight line was drawn from the bottom to the top of each pole, and the quadratic lens distortion coefficient was adjusted until the lines showed approximately the same amount of curvature as the poles and dash. This yielded a quadratic distortion of -0.06300, and was employed for subsequent analysis of the first pass.

For the second and third passes, a calibration grid was printed, filmed, then employed for more accurate lens distortion analysis. The 8” x 10” sheet of equally-spaced

www.collisionmagazine.com

5/28/2019 9:34:25 AM

Figure 11: Straight line adjustment www.collisionpumagazine.com

Volume 13 Issue 1.indd 25

Collision Magazine - Volume 13 Issue 1 25

5/28/2019 9:34:25 AM

Figure 12: Lens grid auto-calibration 26 Collision Magazine - Volume 13 Issue 1

Volume 13 Issue 1.indd 26

www.collisionmagazine.com

5/28/2019 9:34:25 AM

Figure 13: Feature tracking dots were filmed on the dash camera for two seconds.

This footage was extracted from the camera and imported into SynthEyes. Lens grid trackers were created, tracked through the duration of the video, corrected if they ran off, then processed. The resulting distortion map was saved and employed in the analysis of passes two and three.

While the straight-line adjustment lens distortion analysis showed unrealistic square corners in the perspective view, the auto-calibration showed more realistic curved distortion, likely indicative of a more accurate lens distortion analysis.

The road trackers were augmented by 22 other distinct objects visible in the video, such as telephone poles tree branches and bushes. Although they could not be used to geolocate the camera because precise 3D coordinates were not available, they were useful as stationary visual references and helped the software distinguish the movement of the camera. All the feature tracking was reviewed with trackers trails and in graph view in order to identify and correct any tracking mistakes or inaccuracies. The end result was a cloud of trackers that were consistent with the dash camera motion.

The first image sequence was loaded into SynthEyes and the video reviewed for suitable ground tracking references.

A total of 100 distinct features on the ground were designated, that could also be positively identified in Google Earth. They were labelled sequentially starting at Tracker 1. Each feature was manually tracked in the reverse direction so that they could be followed as long as possible.

Road trackers consisted predominantly of dashed roadway lines, along with a couple of other lines and poles. These trackers would be used as references to geolocate the camera.

Each tracked road feature was identified in Google Earth, and a placemark carefully positioned on the surface of the roadway. Although it varied with the type and clarity of the feature, the horizontal position of each feature could be discerned to a few centimetres, more than accurate enough for vehicle velocity calculations. The UTM coordinates for each were obtained and transcribed. These georeferenced road features were collected into a file that could be imported into SynthEyes as known lock points.

Once the road tracker locations were designated in SynthEyes, they were viewed in a top-down perspective. The dispersion of the trackers in SynthEyes matched those in

www.collisionpumagazine.com

Volume 13 Issue 1.indd 27

Collision Magazine - Volume 13 Issue 1 27

5/28/2019 9:34:25 AM

Figure 14: Google Earth placemarks Google Earth with no outliers. With a verified group of reference trackers, matchmoving could start. In this case, the roadway was relatively flat, and an estimated height of zero was used for all road tracker heights.

For the first pass, which employed a simple curved line adjustment for camera distortion, the camera solve resulted in an error of just over 5 pixels. Error cannot be measured in physical units such as metres or centimetres in matchmoving, because an object that is 2 pixels wide could be very small if it is near the camera (such as a crack in the roadway), or very large if it is far away (like a tree in the distance). From experience, the error should be less than 1 pixel if the matchmoving is being done for filmmaking, but can be between much higher (5 or greater) for velocity analysis purposes.

For each sequence, the software was able to figure out where the camera had to be in relation to the ground trackers, how far above the ground, and where the camera was pointed (pitch, roll, and heading) in order for the solved camera to match what was seen in the video. The software also estimated what the camera focal length was, and for this project the calculated value was approximately 7 mm.

Once solved, the application was able to show exactly where the camera was and how it was oriented for each frame of video, both graphically from different perspectives (top, front, left), and numerically.

28 Collision Magazine - Volume 13 Issue 1

Volume 13 Issue 1.indd 28

The camera position information was carefully reviewed to make sure it made sense. It showed the camera as travelling to the right of the roadway line. The calculated height above ground was approximately 1m, which is reasonable for a dash camera mounted on the windshield of a sedan.

The orientation information indicated that the camera was mounted level, pointed on an upward angle of 4 degrees, and oriented towards a heading of 243 degrees as it travelled down the road initially. A quick check in Google Earth showed that the orientation of that section of the road was 239 degrees, so with the camera oriented forwards in the sedan, it could certainly have been pointed 4 degrees right of centre, resulting in the 243-degree heading. The camera solves appeared reasonable.

The next step was to combine the timing and distance data in order to derive velocity. The camera X and Y positions from the solve were exported into Excel, and matched with the corresponding time data. Prior to additional data analysis, a rough velocity plot was assembled.

At 60 km/h, the vehicle moves approximately 0.7m per frame (0.04s). At slower speeds, it moves less. Over the time range 0-20 seconds, when the test vehicle was travelling at approximately 60 km/h, the matchmoving process calculated that the vehicle moved an average of 0.702m.

Over this range, the minimum was 0.12m and the maximum was 6.5m. The variation is due to a number of rea-

www.collisionmagazine.com

5/28/2019 9:34:26 AM

Figure 15: Instantaneous unsmoothed speed sons including the low resolution of the video, changing trackers in different positions, and dispersion of the trackers. Calculating the instantaneous speed from these values showed large fluctuations, but a very discernable trend.

This variation in distance travelled results in significant variations in calculated velocity, when the distance is measured over just 0.04s. One way to improve the velocity precision is to increase the sample size, or measure distance travelled over a longer time such as a quarter, half, or even one second. The disadvantage of this is that useful data is being thrown away. When a one-second interval is used, 24 other measurements between the first and last value are ignored. The result is a single velocity value, and no acceleration/deceleration data during that one second period.

In addition, if the one distance value used for measurement happens to be an outlier, the velocity at that point will be skewed.

A better way to employ the data is to use cubic spline interpolation. This method employs a virtual elastic ruler to join the sample points. The ruler is attached to the first point, then stretched through each point to the last, making a cohesive curve. The tightness of the elastic can be decreased or increased, taking it from the raw data through

different levels of smoothness, all the way to a straight line.

Once an elastic coefficient is chosen that strikes a balance between keeping the curve shape and reducing spurious data, goodness of fit statistics can be calculated on the curve, and reasonable values sampled for the required range.

The spline interpolation has the effect of improving the data to reflect real constraints. SynthEyes solves the camera location and orientation separately for each frame of video. As a result, the camera (and vehicle) varies in position as a result of error. The actual vehicle cannot jump laterally, because of the friction of the tires on the road.

Any changes in lateral position will be smooth as the vehicle drives forward. Similarly, the vehicle cannot jump in the longitudinal axis, because of tire friction and inertia.

As a result, the true vehicle track in the lateral and longitudinal axis would be smooth, similar to a cubic spline, and not noisy like the raw data.

The combined X and Y data were imported into mathematical analysis software. With a smoothing parameter of zero, the fit is a straight line. The resulting Sum of the Squares Error (SSE) is very high at 6.189 e+04, and the R-square error is below 1 at 0.9932. The R-Square error

www.collisionpumagazine.com

Volume 13 Issue 1.indd 29

Collision Magazine - Volume 13 Issue 1 29

5/28/2019 9:34:26 AM

Figure 16: Fits with p = 0 and p = 1 30 Collision Magazine - Volume 13 Issue 1

Volume 13 Issue 1.indd 30

www.collisionmagazine.com

5/28/2019 9:34:26 AM

Figure 17: First pass distance vs time

Figure 18: First pass dash camera speed plot is still relatively close to one because the data curve itself is quite straight. This value would decrease if there was a much greater change in velocity over the time.

Once the smoothing parameter is increased to the maximum of 1, the fit exactly reflects the data, with every bump and variation. The Sum of the Squares Error decreases to zero, and the R-Square increases to 1. This is a perfect fit of the data, but the variation/noise in the graph does not reflect the actual vehicle motion. A lower smoothing value is needed.

A data set was produced with time on the horizontal axis and both X and Y distances on the vertical. Fitting a spline onto the data with a smoothing parameter of 0.08 produced a plot that clearly kept the curve shape, but eliminated spurious fluctuations. The calculated RMSE were 0.4184 and 0.4020, while the R-square was one, suggesting good fits.

The 1st derivative of distance (in metres) and time (in seconds) produced velocity data in m/s for each frame of video. With the X (or N/S) and Y (E/W) velocity informa-

www.collisionpumagazine.com

Volume 13 Issue 1.indd 31

Collision Magazine - Volume 13 Issue 1 31

5/28/2019 9:34:26 AM

tion, the speed down the road could be calculated by the square root sum of the squares of each X and Y velocity component. These instantaneous velocities were multiplied by 3600 and divided by 1000 to change the units from m/s to km/h. The resulting data was aligned with the previously calculated speed truth data and plotted.

The derived video analysis speed data reflected the underlying GPS speed data well. The speed climbed from 57 km/h to a maximum near 65, then fell to 60, stabilized for a short period, then bled off to 15 km/h. The derived speeds did not vary from the truth data by more than 3 km/h, and was for the most part within 1 km/h. It captured both the acceleration and deceleration well. The slight inconsistency just prior to the maximum speed could have been the result of the relatively high camera solve error. Better lens distortion analysis would have helped address this.

Second Test Pass

For the second pass which employed the lens grid autocalibration for camera distortion, the camera solve resulted in an error of 1.06 pixels, a very low and accurate value.

The camera movement looked smooth in playback, again travelling to the right of the roadway line. The software estimated the camera focal length at 22mm, a narrower

field of view than the previous pass due to the cropping of the lens distortion correction. The calculated height above ground was approximately 1.1 m, and the orientation information indicated that the camera was tilted 6 degrees to the left, pointed on an upward angle of 5 degrees, and oriented towards a heading of 238 degrees as it travelled down the road initially.

A data set was produced with time on the horizontal axis and both X and Y distances on the vertical. Fitting a spline onto the data with a smoothing parameter of 0.05 produced a plot that clearly kept the curve shape, but eliminated spurious fluctuations. The calculated RMSE were 0.1896 and 0.1302, while the R-square was one, suggesting very good fits.

The 1st derivative of distance (in metres) and time (in seconds) produced velocity data in m/s for each frame of video. With the X (or N/S) and Y (E/W) velocity information, the speed down the road could be calculated by the square root sum of the squares of each X and Y velocity component. These instantaneous velocities were multiplied by 3600 and divided by 1000 to change the units from m/s to km/h. The resulting data was aligned with the previously calculated speed truth data and plotted.

Figure 19: Second pass solve 32 Collision Magazine - Volume 13 Issue 1

Volume 13 Issue 1.indd 32

www.collisionmagazine.com

5/28/2019 9:34:26 AM

Figure 20: Second pass distance vs time

Figure 21: Second pass dash camera speed plot www.collisionpumagazine.com

Volume 13 Issue 1.indd 33

Collision Magazine - Volume 13 Issue 1 33

5/28/2019 9:34:26 AM

The second pass speed data again reflected the underlying GPS speed data well. The speed measurement started near 60 km/h, was relatively stable with the exception of a short decrease then increase near the end of the pass, then bled off to 20 km/h. The derived speeds did not vary from the truth data by more than 2 km/h. It captured the speed bump and deceleration.

For the third pass which also employed the lens grid autocalibration for camera distortion, the camera solve resulted in an error of 1.01 pixels, a very low and accurate value.

The camera movement looked smooth in playback, again travelling to the right of the roadway line. The software estimated the camera focal length at 7mm, similar to the first pass. The calculated height above ground was approximately 1.1m, and the orientation information indicated that the camera was tilted 4 degrees to the right, pointed on an upward angle of 13 degrees, and oriented towards a heading of 255 degrees as it travelled down the road.

Once again, the raw speed values were calculated and plotted. Employing the more accurate lens distortion, and resulting low solve error, the variation in raw speed seemed

lower than the previous calculated for the first pass. While there were a few outliers, the data was very close to a solid line when speed was low, and varied by approximately 20 km/h at higher speeds. The first pass showed similar variation at 60 km/h, but with significantly more outliers.

A data set was produced with time on the horizontal axis and both X and Y distances on the vertical. Fitting a spline onto the data with a smoothing parameter of 0.08 produced a plot that clearly kept the curve shape, but eliminated spurious fluctuations. The calculated RMSE were very low values of 0.2895 and 0.2114, while the R-square was one, suggesting good fits.

The 1st derivative of distance (in metres) and time (in seconds) produced velocity data in m/s for each frame of video. With the X (or N/S) and Y (E/W) velocity information, the speed down the road could be calculated by the square root sum of the squares of each X and Y velocity component. These instantaneous velocities were multiplied by 3600 and divided by 1000 to change the units from m/s to km/h. The resulting data was aligned with the previously calculated speed truth data and plotted.

Figure 22: Third pass solve 34 Collision Magazine - Volume 13 Issue 1

Volume 13 Issue 1.indd 34

www.collisionmagazine.com

5/28/2019 9:34:26 AM

Figure 23: Instantaneous unsmoothed speed

Figure 24: Third pass dash camera speed plot www.collisionpumagazine.com

Volume 13 Issue 1.indd 35

Collision Magazine - Volume 13 Issue 1 35

5/28/2019 9:34:26 AM

Figure 25: Three passes combined

“

Figure 25: Three passes combined

The dash camera is a low-cost imaging device easily accessible to the general public, not a high-precision instrument. The limited resolution of 640x480 on the video hampered the analysis and results

36 Collision Magazine - Volume 13 Issue 1

Volume 13 Issue 1.indd 36

The derived video analysis speed data reflected the underlying GPS speed data very well. The speed started near 10 km/h, fell to near zero, then accelerated quickly to a maximum near 80 km/h. After a short period with slight deceleration, the derived speed fell to 10 km/h. The derived speeds did not vary from the truth data by more than 2 km/h, and was for the most part within 1 km/h. It captured both acceleration and deceleration well.

Examining all three passes together, the velocity analysis very closely matched the underlying 10Hz GPS data, both at 60 km/h and 80 km/h. The acceleration and deceleration were captured each time, and only minor variations were noted. The overall analysis suggested that the dash camera video analysis process described in this report produced speeds within 2 km/h of the true speeds.

Summary

The dash camera is a low-cost imaging device easily accessible to the general public, not a high-precision instrument.

The limited resolution of 640x480 on the video hampered the analysis and results. Even with this limitation, dash camera video velocity analysis was able to extract precise high-resolution velocity data from the image sequence.

This data was compared to a high sample rate GPS designed to accurately measure speeds for racing, and found to be comparable. Dash camera video velocity analysis has proven to be a precise and useful capability for the modern accident investigator.

www.collisionmagazine.com

5/28/2019 9:34:27 AM

Volume 13 Issue 1.indd 37

5/28/2019 9:34:27 AM

POWER LOSS ISSUES RELATED TO EDR DATA IN 2013-2017 KAWASAKI NINJA 300 AND ZX-6R MOTORCYCLES Edward C. Fatzinger Jr., MS, PE

Momentum Engineering Corp.

38 Collision Magazine - Volume 13 Issue 1

Volume 13 Issue 1.indd 38

www.collisionmagazine.com

5/28/2019 9:34:27 AM

S

tarting in 2013 Kawasaki Heavy Industries, Ltd. (KHI) has installed Event Data Recorders (EDR) on select US sold motorcycles. The two motorcycles covered in 2013 are the Ninja 300 and the ZX-6R. On both these models an EDR event is triggered when the motorcycle is tipped over and goes into an emergency shutdown (ES). The emergency shutdown is a safety feature that involves shutting off the fuel pump relay, fuel injectors, and ignition system when the motorcycle senses it has fallen. The rest of the electronics will remain active. Additionally, to trigger an EDR event the rear wheel must be in motion or gone through a sudden deceleration in the several seconds prior to ES. On the Ninja 300 and ZX-6R the time between tip-over and ES is approximately 1.3 seconds. However, as studies have shown this time can be significantly increased if the motorcycle is still sliding/bouncing creating considerable “noise” in the tip-over sensor. The data parameters captured in an EDR event can be seen in Table 1. In addition to the data parameters, the EDR will capture the ECU runtime and key cycles at the event, as well as the elapsed ECU runtime and key cycles since the event. 1,2 2 Hz data Front wheel speed* Rear wheel speed Gear Position Inlet Air Temperature Coolant Temperature Battery Voltage DTC’s Power (P-Mode)* KTRC Mode*

10 Hz data Throttle Position Engine RPM Clutch In/Out Fuel Injector Pulse Timing BTDC Fuel Cutout *ZX-6R only

Table 1: EDR data parameters for Ninja 300 and ZX-6R The EDR data is stored in the Engine Control Unit (ECU) on three non-volatile memory chips (EEPROMs). An image of the three EEPROMs on the Ninja 300 can be seen in Figure 1. The power supply critical to triggering and subsequently writing an EDR event to the ECU comes in two forms; switched and unswitched.

The switched power comes from the ECU relay and the unswitched power comes straight from the battery. Similar to your typical automotive radio, there is a constant power feed (unswitched) which stores the clock, presets, and other user data on the radio, and a switched power feed that generally powers up the radio. The ECUs in the Ninja 300 and ZX-6R work in a very similar fashion. Both the switched and unswitched power to the ECU is necessary for the motorcycle to go into ES and trigger an EDR event. Once in ES and the EDR event is triggered, only the unswitched power is necessary to save the data to the EEPROMs. On the Ninja 300, all the motorcycle power comes through the main fuse located on the starter relay assembly (see Figure 2). From the main fuse, unswitched power is fed directly to the ECU through the fuel injection fuse. Additionally, power is fed directly to the ignition switch from the main fuse. When energized, the ignition switch powers the switched side of the fuse block. This switched side of the fuse block is responsible for energizing the ECU relay. Unfortunately, if the main fuse of the Ninja 300 is compromised, the ECU will lose both the switched and unswitched power assuming the engine has stalled. However, if the main fuse is blown or there is a sudden battery loss and the engine has not stalled, the motorcycle’s alternator power will keep the engine running and supply the switched and unswitched power to the ECU. www.collisionpumagazine.com

Volume 13 Issue 1.indd 39

Collision Magazine - Volume 13 Issue 1 39

5/28/2019 9:34:28 AM

Figure 1: Ninja 300 ECU EEPROMs This is not entirely the case on the ZX-6R. The switched and unswitched power to the ECU is provided in a slightly different manner (see Figure 3). The ZX-6R contains two fuses incorporated into the starter relay circuit, the main fuse and the ECU fuse. A simple way of looking at it would be that the motorcycle operates using two “main” fuses. One fuse supplies power to the motorcycle and the other fuse supplies power only to the ECU. The ECU fuse is the same as the fuel injection fuse on the Ninja 300, but due to relocating it to the starter relay, it is now not reliant on the main fuse. In this case, if the main fuse blows for whatever reason, the ECU will still have its unswitched power. Similar to the Ninja 300, if the main fuse is blown or there is a sudden battery loss and the engine has not stalled, the motorcycle’s alternator power will keep the engine running. The alternator will still supply the switched and unswitched power to the ECU. However, if the ECU fuse is compromised regardless of alternator or battery power, the engine will immediately shut down.

On both the Ninja 300 and ZX-6R, if for any reason the switched or unswitched power is lost prior to ES (and assuming engine stall), no EDR event will be saved. This power loss would keep the bike from going into ES, which is necessary for an EDR even to be saved. A classic example of this would be a motorcycle colliding into another vehicle and fracturing the ignition switch or the ignition switch wiring prior to the motorcycle falling and going into ES. This is not necessarily the case if there is an unswitched power loss that occurs after the ES. Additionally, the engine stop switch (kill switch) on the handlebar has no effect on the switch or unswitched power to the ECU. The ECU can still command an ES and trigger and EDR event regardless of the stop switch position.

In a normal/non-power loss situation, once the motorcycle goes into ES, the ECU will write the EDR data to the nonvolatile memory. The ECU will then “wait” for the key cycle to end in order to write the appropriate ECU runtime and key cycles for that event. An example of this can be 40 Collision Magazine - Volume 13 Issue 1

Volume 13 Issue 1.indd 40

seen in Tables 2-3. Table 2 depicts a baseline download of a Ninja 300 ECU which contains two events. Table 3 depicts a test with the same ECU where the motorcycle rear wheel was elevated on a wheel stand with the motorcycle idling in 5th gear. The tip-over sensor was tilted, and the ES was commanded. Several seconds after ES with the ignition switch still on, the main fuse was pulled. As shown, the data wrote to the non-volatile memory just fine, but the ECU runtime, key cycles, and number of ES events parameters did not update. This was due to the ECU “waiting” for the key cycle to end, but it never did due to the total power loss before the key was turned off. An example of this situation occurring would be where the motorcycle falls in a low-side-type accident and slides long enough to go into ES. Subsequently, the motorcycle collides with another object severing the main fuse, fuel injection fuse or battery power (all of which provide an unswitched power loss). Similarly, if a motorcycle fell, went into ES, and was subsequently struck by another vehicle severing the main fuse, fuel injection fuse or battery power.

It is unclear as to what Kawasaki will do if this situation is present when they perform an ECU download. The author has seen in the past where KHI only provides data plots for two events. When data for only two events is supplied, it is presumed a third event never occurred. The question becomes how does KHI determine if there are only two events? If they are looking solely at the ECU runtimes and key cycles, they could overlook data present from a power loss after ES. Of note in the previous Ninja 300 example, the third event that occurred during power loss is an “unlocked” event. If the criteria are met for another EDR event, all the data in the third event will get overwritten and the ECU runtime and key cycles will populate as normal assuming no power loss issues.

This anomaly is also present on the ZX-6R. Similar to the Ninja 300, if there is unswitched power loss after ES, the event will likely be an “unlocked” event with no ECU runtime or key cycle information. On the ZX-6R the unswitched power loss would come from a blown ECU fuse or severed battery power. Remember on the ZX-6R, the unswitched power does not go through the main fuse, so compromising the main fuse would essentially yield the same result as switching the ignition off and saving an EDR event normally.

Another anomaly resulting from power loss after ES is a partial recording. As mentioned previously, once the motorcycle goes into ES, if the EDR recording criteria is met, the data will begin writing to the non-volatile memory immediately after ES. However, if the unswitched power is lost during the writing process, only part of the EDR data

www.collisionmagazine.com

5/28/2019 9:34:29 AM

Figure 2: Ninja 300 Fuse Wiring 3

Figure 3: Ninja ZX-6R Fuse Wiring 4

www.collisionpumagazine.com

Volume 13 Issue 1.indd 41

Collision Magazine - Volume 13 Issue 1 41

5/28/2019 9:34:29 AM

“

Table 2: Ninja 300 Baseline Download

When inspecting a Kawasaki motorcycle with possible EDR recording capability, be certain to inspect the ignition switch and wiring in detail

42 Collision Magazine - Volume 13 Issue 1

Volume 13 Issue 1.indd 42

www.collisionmagazine.com

5/28/2019 9:34:29 AM

Table 3: Ninja 300 Power Loss After ES Occurrence will be saved. An example of this can be seen in Tables 4-5. Table 4 depicts a baseline download of a ZX-6R ECU with two events. Table 5 depicts a test with the same ECU where the motorcycle rear wheel was elevated on a wheel stand with the motorcycle idling in 1st gear. The tip-over sensor was tilted, and the ECU fuse was pulled approximately 1.5 seconds after ES. As can be seen in the table for Event #3, the data was saved up to and including gear position. Similar to the Ninja 300 example, the ECU runtime, key cycles, and number of ES events parameters did not update, and this event is “unlocked”. It was found on several other tests performed, that nearly 4 seconds was required to save a complete EDR event.

In the unlikely event the main fuse is compromised, or the battery power becomes severed, but the motorcycle still runs up to ES, an event may still be recorded. Both the

Ninja 300 and ZX-6R will continue to run under their own power via the alternator after the main fuse or the battery power becomes severed. For example, if the main fuse on the ZX-6R is compromised and the motorcycle continues to run, the ECU will still have switched and unswitched power. Upon ES, the switched power will cease, effectively turning the ignition switch off and the ECU will save the EDR data due to the unswitched power still being supplied through the ECU fuse. Likewise, if the battery power becomes severed and the motorcycle continues to run, upon shutdown both the switched power (from the main fuse side) and the unswitched power to the ECU (through the ECU fuse) will cease. This will likely lead to not recording an EDR event or an “unlocked” event may be recorded with minimal data. If the ignition switch becomes disabled, ignition fuse is compromised or ECU fuse is compromised prior to ES, the motor will immedi-

www.collisionpumagazine.com

Volume 13 Issue 1.indd 43

Collision Magazine - Volume 13 Issue 1 43

5/28/2019 9:34:29 AM

Table 4: Ninja ZX-6R Baseline Download ately shut off and no EDR event will be recorded. Since the Ninja 300 does not have an ECU fuse, if the main fuse is compromised or the battery power becomes severed prior to ES with the motor continuing to run up to ES, both the switched and unswitched power to the ECU will cease after the motor stalls and an EDR event will unlikely be recorded or an “unlocked’ event may be recorded with minimal data.

When inspecting a Kawasaki motorcycle with possible EDR recording capability, be certain to inspect the igni44 Collision Magazine - Volume 13 Issue 1

Volume 13 Issue 1.indd 44

tion switch and wiring in detail. Also, be familiar with where the fuse block(s) are and make note of any open or compromised fuses. Inspect the battery and cables to make sure they are intact. Make note of the key position, whether it’s in the “on” or “run” position. Is the key broken off inside the switch? If the key is broken and the ignition was never switched off, there is a high probability the battery will fully discharge. This would be comparable to the battery power being severed after ES and likely lead to an “unlocked” event. If you suspect a possible power loss after ES, be sure to get all three data plots from KHI

www.collisionmagazine.com

5/28/2019 9:34:29 AM

Table 5: Ninja ZX-6R ECU Fuse Pull ~1.5s After ES to ensure you have all available data. Again, it is unknown how Kawasaki Heavy Industries, Ltd. handles the data of an “unlocked” or partial event.

R

eferences 1. Fatzinger, Edward, Landerville, Jon, “An Analysis of EDR Data in Kawasaki Ninja 300 (EX300) Motorcycles,” SAE Technical Paper 2017-01-1436, 2017.

2. Fatzinger, Edward, Landerville, Jon, “An Analysis of EDR Data in Kawasaki Ninja ZX-6R and ZX-10R

Motorcycles Equipped with ABS (KIBS) and Traction Control (KTRC),” SAE Technical Paper 2018-011443, 2018.

3. Kawasaki Heavy Industries, Inc., “Ninja 300, Ninja 300 ABS Motorcycle Service Manual,” Part No. 99924-1460-05.

4. Kawasaki Heavy Industries, Inc., “Ninja ZX-6R, Ninja ZX-6R ABS Motorcycle Service Manual,” Part No. 99924-1462-04.

www.collisionpumagazine.com

Volume 13 Issue 1.indd 45

Collision Magazine - Volume 13 Issue 1 45

5/28/2019 9:34:29 AM

Online Class:

How To Use the Bosch CDR Tool

The “How To Use the Bosch CDR Tool” is an online course offered at our elearning portal website. Once registered, you will have 14 days to complete the class. From start to finish, this class typically takes 4 hours to complete, giving you plenty of time to review the material and take the course at your leisure.

After each learning module, you will take a short quiz to advance to the next learning module. Don’t worry, you can take the quizzes as many times as needed. Once you have completed all 9 learning modules, you will take the final exam. You can take the final exam a maximum of two times. A passing grade on the final exam generates a certificate that you can download.

Once logged in to the course, you can download the course material as a guide while you take the course. The course material is yours to keep as a reference.

Course Outline

1. Introduction Video 2. Module 1: CDR Introduction and History + Quiz 3. Module 2: CDR Hardware Components + Quiz 4. Module 3: CDR Software Installation and Activation + Quiz 5. Module 4: CDR Software Operation + Quiz 6. Module 5: CDR Help File and Vehicle/Cable Lookup + Quiz 7. Module 6: Regulation and Privacy (USA) + Quiz 8. Safety Video (no quiz) 9. Module 7: Performing a DLC In-car download + Quiz 10. Module 8: Performing a Direct-tomodule download + Quiz 11. Module 9: Troubleshooting + Quiz 12. NEW! Module 10: CDR 900 + Quiz 13. Final Exam

Register online 24/7 at:

www.crashdatagroup.com/bosch-cdr-class-online/ Volume 13 Issue 1.indd 46

5/28/2019 9:34:30 AM

Mark your calendar!

March 9-11, 2020 - Houston, Texas The EDR Summit will deliver the next steps in advanced EDR technology for vehicle crash analysis. Most importantly, this summit brings together industry experts from around the world to present on timely topics and case studies.

The presentations will focus on EDR data found in light trucks, passenger cars, SUVs, motorcycles, heavy commercial vehicles, active safety systems, autonomous driving, and vehicle infotainment systems.

The EDR Summit is open anyone who has an interest in learning more about event data recorders and how they are used in vehicle safety and crash investigation. As a result, typical attendees of this Summit include law enforcement, insurance (SIU and claims), government, legal and collision reconstructionists.

Therefore, if you are involved with the collection and/or analysis of EDR data from vehicle crashes or forensics you won’t want to miss the EDR Summit!

www.crashdatagroup.com/edr-summit www.collisionpumagazine.com

Volume 13 Issue 1.indd 47

Collision Magazine - Volume 13 Issue 1 47

5/28/2019 9:34:31 AM

The impact of nonlinear boundary

geometry considerations in regards to residual damage based model coefficients, equivalent barrier speed and internal work absorbed Jai Singh, BS, MS, MA, ACTAR Biomechanical Engineering Analysis & Research, Inc.

48 Collision Magazine - Volume 13 Issue 1

Volume 13 Issue 1.indd 48

www.collisionmagazine.com

5/28/2019 9:34:32 AM

A

bstract Extant closed-form analytic solutions for quantifying residual damage based model parameters, equivalent barrier speed (EBS) and internal work absorbed (IWA) that use the residual damage profile present after a collision are predicated upon the employment of global or piecewise linear interpolation at the level of the residual damage profile.

The subject work focuses, primarily, on the theoretical evaluation of this predicate. This evaluation is approached first by defining the residual damage depth function as the difference between the reference and damaged boundary geometry functions. It is shown that the extant formulation is reproducible when both boundary geometry functions are separately linear (interpolated or otherwise) over a mutual domain. It is also shown that the presence of non-linearity in either boundary geometry function also appears in the residual damage depth function and thereby changes the form of the equations that are currently employed for determining the relevant parameters. The case in which the interpolation function for the reference boundary geometry consists of a general polynomial function is detailed in depth. A worked example is provided in which other forms of interpolation functions are considered for both the reference and damaged boundary geometries.

I

ntroduction Equations (1-2) are the most commonly employed mathematical relationships that serve as foundational for the determination of collision severity as a function of the residual damage present to a motor vehicle following involvement in an impact (Sharma et al., 2007). The first of these equations, proposed by Campbell (1972, 1974), is an empirical relationship between the depth of residual damage (c) and the equivalent barrier speed (EBS). The second equation, proposed by McHenry (1976), relates the residual damage depth to the peak collision force magnitude, |F|, normalized to the reference configuration direct damage contact width (L) and is one of the critical relationships in the third iteration of the Calspan Reconstruction of Accident Speeds on the Highway (CRASH3) damage analysis algorithm. This relationship is analytic for the closure phase of a coaxial collision when the structural response characteristics of the collision partners are modeled as linear elastic (Noga and Oppenheim, 1983) but becomes empirical in consider-

ation of collisions with a separation phase for which the coefficient of restitution is not zero-valued.

EBS = b0 + b1c F L−1 = A + Bc

(2)

The internal work absorbed (IWA) during a collision can be directly related to the EBS and the mass of the collision partner (m) as shown by equation (3).

IWA =

1 mEBS2 2

(3)

The relationship shown by equation (2) is not a timeparametric force-deflection response nor does it represent a time-parametric force-deflection response shifted along the abscissa. The force-deflection data generated from any singular collision test, for example, maps to a single point along the response curve. In practice, however, this relationship is treated as being equivalent to a time-parametric force-deflection response in regards to the determination of the IWA. Traditional solution for IWA for uneven damage profiles In the consideration of residual damage profiles for which the depth of residual damage varies as a function of location along the width of the direct damage portion of the profile, the traditional approach has been that of using a discretization scheme consisting of the piecewise employment of the trapezoidal rule using two, four or six equally spaced nodal points. This was extended to the use of any N number of equally spaced nodal points (Singh et al., 2003) and then for the use of any N number of unequally spaced nodal points (Singh, 2005a). The latter presentation, being the most general, is considered herein. In this regard, the independent variable, l {l: 0 ≤ l ≤ L}, denoting the location along the width of the direct damage portion of the residual damage profile, in the plan view and mapped to the reference configuration, is introduced. The residual damage depth function may then be stated as c(l). The ith {i: 1 ≤ i ≤ N} nodal value of the residual damage depth, located at li, is denoted as ci. The linear interpolation function between successive nodal values derives from the simple definition of a two-dimensional line.

www.collisionpumagazine.com

Volume 13 Issue 1.indd 49

(1)

Collision Magazine - Volume 13 Issue 1 49

5/28/2019 9:34:33 AM

c( l ) =

ci + 1 − ci ( l − l i ) + ci l i +1 − l i

l {l : l i ≤ l ≤ l i + 1 }

(4)

Following the traditional approach to determining the IWA, equation (2) is integrated over the residual damage domain. l i +1 c( l )

l i +1

li

li

∫ ∫ ( A + Bc) dcdl = ∫ ( A c ( l ) + 0.5B ( c ( l ) ) ) dl 0

2

(5)

Substitution of equation (4) into equation (5) followed by evaluation of the final integral yields the following result: A B ( ci + ci + 1 )( l i + 1 − l i ) + ( ci 2 + ci ci + 1 + ci + 12 ) ( l i + 1 − l i ) 2 6

(6)

Unlike a time-parametric force-deflection relationship, the ordinate intercept of the response function, shown by equation (2), does not fall at the origin. A consequence of this result is that the abscissa intercept occurs at a value of co=-AB-1. The area bounded to the left of the ordinate, above the abscissa and below the response curve can be determined by integration. l i +1

0

l i +1

A2 

A2 





∫ ∫ ( A + Bc) dcdl = ∫  2B  dl =  2B  ( l i + 1 − l i ) = G ( l i + 1 − l i )

l i − A B−1

li





(7)

The traditional solution for the IWA for each zone (i.e. the region between successive nodal points) is obtained by adding equations (6) and (7). Summation of the result over all zones from i = 1 to i = N-1 yields the IWA for the entire residual damage profile.

IWA =

A B α + β + GL 2 6

(8)

Where: N −1

N −1

i =1

i =1

2

i =1

(9)

The traditional approach when considering a collision force vector that acts at an angle θ, with respect to the relevant vehicle principle axis, is to employ trigonometric relationships such that the force considered is F sec(θ) and the residual damage is c sec(θ). The resultant form of equation (8), as a result, contains the multiplicative term sec2(θ), which is equal to 1 + tan2(θ).

50 Collision Magazine - Volume 13 Issue 1

Volume 13 Issue 1.indd 50

(10)

Traditional solutions for the model coefficients The traditional solution for the model parameters proceeds by equating the IWA based on equation (3), following substitution for the EBS from equation (1), to the solution for the IWA from equation (8). Both equations are second order polynomial equations with respect to the residual damage depth and are only equal when the coefficients, from each equation, for each order of the polynomial, are equal.

This approach allows for the development of the following relationship between the coefficients of the two models.

A=

b 0 b 1m L

B=

b 12 m L

G=

b02 m 2L

(11)

Substitution of these equivalencies into equation (8), followed by algebraic rearrangement, results in the following quadratic form for the Campbell b1 model coefficient. b12 + 3b0αβ−1b1 + 3β−1L ( b02 − 2m −1IWA ) = 0

(12)

In practice, the Campbell b0 model coefficient, representing the maximum EBS at which no residual damage depth is produced, is estimated and the IWA is written in terms of the EBS. b12 + 3b0αβ−1b1 + 3β−1L ( b02 − EBS2 ) = 0

(13)

The positive valued solution for equation (13) fits the physical nature of the modeling problem under consideration. b1 =

(

−3b0αβ−1 + 9b02 α 2β−2 − 12β−1L ( b02 − EBS2 )

)

0.5

2

(14)

N −1

α = ∑ ( ci + ci + 1 )( l i + 1 − l i ) β = ∑ ( ci + ci ci + 1 + ci + 1 ) ( l i + 1 − l i ) L = ∑ ( l i + 1 − l i ) 2

B A  IWA =  α + β + GL  ( 1 + tan 2 ( θ ) ) 6 2 

When θ is not zero-valued, this solution becomes: b1 =

(

−3b0αβ−1 + 9b02 α 2β−2 − 12β−1L ( b02 − EBS2 cos2 ( θ ) ) 2

)

0.5

(15)

Back-substitution of equation (15) into the three relationships listed under equation (11) allows for the determination of the CRASH3 model coefficients. This approach is typically employed in the case in which one seeks to quantify the model parameters, on a vehicle platform-specific

www.collisionmagazine.com

5/28/2019 9:34:34 AM

basis, from controlled collision testing in which the testing configuration consists of a colinear full-width engagement impact between a test vehicle and a fixed, rigid, massive barrier (FRMB). For testing in which both collision partners are deformable (Neptune et al., 1992; Neptune and Flynn, 1998; Singh, 2005b) or for field studies for which the model coefficients of only one collision partner are known (Hull and Newton, 1993; Neptune and Flynn, 1994; Long, 1999; Chen et al., 2005), the solution procedure is based on implementing Newton’s Third Law, with the procedure referenced in the previous literature as the force balance method. Using the subscript A to denote the collision partner for which the model coefficients are known, the zonal force is expressed as: B   Fi →i + 1 =  A A ,i →i + 1 + A ,i →i + 1 ( cA ,i + cA ,i + 1 )  ( l i + 1 − l i ) sec ( θA ) 2  

(16)

The zonal force equation for the opposing collision partner, referenced using the subscript B, can be expressed as: B   Fi →i + 1 =  A B,i →i + 1 + B,i →i + 1 ( cB,i + cB,i + 1 )  ( l i + 1 − l i ) sec ( θB ) 2  

(17)

When the model coefficients are constant across all of the zones of direct damage, the zonal indication associated with the model coefficients in equation (17) can be dropped. The total force is obtained by summing equation (17) over all zones.

N B −1

∑F i =1

i →i + 1

 N B −1  B N B −1 = F =  A B ∑ ( l i + 1 − l i ) + B ∑ ( cB,i + cB,i + 1 ) ( l i + 1 − l i )  sec ( θB ) 2 = = i 1 i 1  

(

)

(18)

The value of the collision force, in equation (18), however, must be equal to the value of collision force obtained by summing equation (16) over all zones of direct damage.

Substitution from the relationships under equation (11) into equation (18) followed by algebraic rearrangement results in the following quadratic equation: b1,B2 + ( 2b0,Bα B−1L ) b1,B − 2m B−1Fα B−1L cos ( θB ) = 0

(19)

The solution for the Campbell b1 model coefficient is again the positive valued solution for the quadratic and with the A, B and G model coefficients determined by back-substitution. b1,B = −b0,Bα B−1L + ( b0,B2 α B−2 L2 + 2m B−1Fα B−1L cos ( θB ) )

0.5

(20)

The structural response approaches detailed above and considered herein are uniaxial compressive. The models do not contain an explicit mechanism for the inclusion of deformation and residual damage due to transverse shear or out of plane bending. In theory, the uniaxial compressive response can be taken with respect to an arbitrary, inplane, orientation with respect to the principle axes of a vehicle when considered with respect to the plan projection (i.e. with respect to the plane based upon the longitudinal and lateral axes). In practice, however, the vast majority of controlled collision test data is reported with the residual damage depth information taken along the relevant principle axis and with respect to the reference configuration.

While this is a limitation, it does provide for simplifications in regards to the focus of the subject work, which is detailed as follows.

Objectives The residual damage present to a vehicle, following a collision, clearly represents the difference between the reference geometry of a vehicle and the deformed (i.e. damaged) geometry of the same. Equations (1-2) are clearly linear functions of the residual damage depth and the subsequent derived quantities are clearly predicated upon this linear presumption. The extant literature includes a number of alternative linear static models such as the constant force model, the saturation force model, the bilinear stiffness model (Strother et al., 1986; Kerkhoff et al., 1993) and the multilinear stiffness model (Singh and Perry, 2008) as well as the nonlinear static power law model (Nilsson-Ehle et al., 1982; Woolley, 2001; Singh and Perry, 2008). One area that has not been explored, however, and represents the objective of the subject work, is the evaluation of the relationship between the reference and deformed geometries and the assumption of linearity in regards to their difference, defining the residual damage profile, and the derived quantities that are based upon the difference.

T

heory Axes mapping and generalized coordinates

For the plan-projected view the relevant vehicle principle axes are the longitudinal axis (x) and the lateral axis (y). The mathematical expressions for the front and rear aspects of the geometry boundary models may be expressed as x = f(y) whereas the left and right sides of the geometry boundary models may be expressed as y = f(x). Both cases can be expressed in a singular compact form by denoting the independent variable as ξ (where ξ = y for the end case and ξ = x for the side case) and the dependent variable as η = f(ξ). There are two important consequences that arise from the assumed uniaxial compressive structural response model. The first is that the

www.collisionpumagazine.com

Volume 13 Issue 1.indd 51

Collision Magazine - Volume 13 Issue 1 51

5/28/2019 9:34:34 AM

vehicle axes in the reference and deformed configurations remain coincident. The second is that the mapping of l ↔ ξ and c(l) ↔ η (ξ) can readily be related to the reference configuration. Denoting the reference boundary geometry as ηr(ξ) and the deformed boundary geometry as ηd(ξ), the residual damage depth function can readily be written as: c ( l ) ↔ c ( ξ ) = ηr ( ξ ) − ηd ( ξ )

(21)

Linear boundary geometry functions The simplest plan-view boundary geometry model is that of the rectangular approximated boundary geometry model for the reference geometry. This model is parameterized by the maximum overall width and length of a vehicle and is geometrically represented by a rectangle. Any boundary region of the model is therefore defined by a zero-valued slope with respect to the relevant principle axis. One may readily consider the more general form of this model (i.e. for non zero-valued slopes). Using the axis mapping discussed above, for the case in which both the reference and damaged geometry boundaries are linear functions over a mutual domain {l: li ≤ l ≤ li+1}, the equations for the segments are: − ηr ,i  η ηr ( l ) =  r ,i + 1  ( l − l i ) + ηr ,i  l i +1 − l i  − ηd ,i  η ηd ( l ) =  d ,i + 1  ( l − l i ) + ηd ,i  l i +1 − l i 

− ηd ,i + 1 − ηr ,i + ηd ,i  η c ( l ) =  r ,i + 1  ( l − l i ) + ( ηr ,i − ηd ,i ) l {l : l i ≤ l ≤ l i + 1 } l i +1 − l i  

(23)

The nodal values of the residual damage depths are: ηr ,i − ηd ,i = ci

(22)

The boundary geometry functions are shown schematically in Figure 1.

(24)

Substitution of these definitions from equation (24) into equation (23) results in the following: c( l ) =

ci + 1 − ci ( l − l i ) + ci l i +1 − l i

l {l : l i ≤ l ≤ l i + 1 }

(25)

As expected, equation (25) exactly matches equation (4).

It therefore follows, again, as expected, that the case of the linearly modeled boundary geometry functions for both the reference and damaged geometries, over a mutual domain, produce the linear interpolation function for l-c(l) that is employed in the extant formulation. This will clearly have a direct impact on the IWA formulations for which c(l) is a limit of integration. It can also be shown that the form of c(l) impacts the relationship between collision force and the depth of residual damage. l i +1

ξ ↔ l {l : l i ≤ l ≤ l i + 1 }

ηr ,i + 1 − ηd ,i + 1 = ci + 1

(

)

F = ∫ A + B ( c ( l ) ) dl li

(26)

Substitution of equation (25), which again derives from the case in which both the reference and damaged boundary geometries are linear functions over the mutual domain, into equation (26) followed by evaluation of the integral results in the following solution.

B   F =  A + ( ci + ci + 1 )  ( l i + 1 − l i ) 2  

(27)

Non-linear boundary geometry functions

Figure 1: Reference and damaged linear boundary geometry functions over a mutual domain.

Substitution of the two equations under (22) into equation (21) results in the following solution for the residual damage depth function:

52 Collision Magazine - Volume 13 Issue 1

Volume 13 Issue 1.indd 52

The number of potential forms and formulations for alternative functions for modeling the boundary geometries is legion and the exploration of all such cases is beyond the scope of this work. The argument, however, can be made that nonlinearity in either boundary geometry function, that is not removed by the subtraction operation of equation (21), becomes manifest in the residual damage function and therefore in the collision force and IWA relationships. If ηr(l) = f(λr1, λr2) and ηd(l) = f(λd1, λd2) where λr1 and λd1 are linear terms and λr2 and λd2 are nonlinear terms then the following holds:

www.collisionmagazine.com

5/28/2019 9:34:34 AM

ηr ( λ r 1 ) − ηd ( λ d 1 ) → c ( l ) = f ( λ r 1 , λ d 1 )

relationship between the collision force and the depth of residual damage becomes:

ηr ( λ r 1 ) − ηd ( λ d 1 , λ d 2 ) → c ( l ) = f ( λ r 1 , λ d 1 , λ d 2 )

 1 2 1  F =  A + B  ard 2 ( l i + 1 + l i ) + ard 1 ( l i + 1 + l i ) + ard 0   ( l i + 1 − l i ) sec ( θ ) 2 3  

ηr ( λ r 1 , λ r 2 ) − ηd ( λ d 1 ) → c ( l ) = f ( λ r 1 , λ d 1 , λ r 2 )

ηr ( λ r 1 , λ r 2 ) − ηd ( λ d 1 , λ d 2 ) → c ( l ) = f ( λ r 1 , λ d 1 , λ r 2 , λ d 2 )

(28)

The functional dependencies shown for c(l), in the relationships shown under equation (28), hold to the extent that λr1 ≠ λd1 and λd1 ≠ λd2. The presence of the same category of terms (i.e. linear or nonlinear) in both boundary geometry models would generally be expected to result in the same category of terms appearing in the difference given that the points used to generate each boundary geometry model are not mutually inclusive. The theoretical development for one family of functions is considered below.

The linear functions considered thus far are simply first order polynomial expressions. A simple approach, first, for consideration of the impact of nonlinearity, is by increasing the order of the polynomial function used to model one of the boundary geometries by a single order. Either one or both of the boundary geometry functions may be modeled in this way. For this presentation, the damaged boundary geometry function is left unchanged and the resulting functions over the mutual domain may be written as: ηr ( l ) = ar 2 l 2 + ar 1l + ar 0 ηd ( l ) = ad 1l + ad 0

ξ ↔ l {l : l i ≤ l ≤ l i + 1 }

(29)

In equation (29), the terms associated with l and l2 are presented without an offset term (i.e. in the form of l – li). For the case in which a polynomial is used, the function becomes interpolating rather than approximating when the number of nodal points used in generating the polynomial is one greater than the order of the polynomial. Therefore, the functional domain of the second equation under (29) can fully be defined by the nodal points located at l = li and l = li+1, whereas the functional domain for the first equation, in the interpolation sense, would have the two nodal points within its domain. The residual damage function is obtained as before. c ( l ) = ar 2 l 2 + ( ar 1 − ad 1 ) l + ( ar 0 − ad 0 ) = ard 2 l 2 + ard 1l + ard 0

l {l : l i ≤ l ≤ l i + 1 }

(30)

The polynomial coefficients in equation (30) encode the nodal values for both the reference and damaged boundary geometries as well as the residual damage depths, which can be retrieved by solving c(l) at l = li and at l = li+1. The

(31)

Equation (31), either for a single zone of direct damage or for the entire width of direct damage, would result in a differing quantified value not only for the total collision force magnitude, but also for the downstream quantified model coefficients for the cases when the force balance approach is employed. For the sake of comparison, rewriting equation (25) in an equivalent first order polynomial form and rewriting equation (27) leads to the following form:  1  F =  A + B  ard 1 ( l i + 1 + l i ) + ard 0   ( l i + 1 − l i ) sec ( θ ) 2    

The zonal form of equation (10), for the IWA, for the first order polynomial for c(l), can be rewritten as:  1    A  ard 1 ( l i + 1 + l i ) + ard 0  +  2   ( l − l ) ( 1 + tan 2 θ ) IWA =   1  i +1 i 1 1 2 2 2 2  B  ard 1 ( l i + 1 + l i + 1l i + l i ) + ard 0ard 1 ( l i + 1 + l i ) + ard 0  + G  2 2   6 

(33)

For the second order polynomial form for c(l), the zonal IWA becomes:  1 1   2 2  A  ard 2 ( l i + 1 + l i + 1l i + l i ) + ard 1 ( l i + 1 + l i ) + ard 0  +  2   3   1 1 1   2 2 2 2 3 3 a l − l + a a l − l + a l − l + ( ) ( ) ( )   rd 0 i + 1 i rd 0 rd 1 i +1 i rd 1 i +1 i   2 6 IWA =  B  2  +  ( 1 + tan 2 θ ) 1 1  1 3 3 4 4 2 5 5     3 ard 0 ard 2 ( l i + 1 − l i ) + 4 ard 1ard 2 ( l i + 1 − l i ) + 10 ard 2 ( l i + 1 − l i )      G ( l i +1 − l i )     

(34)

The use of polynomial functions of arbitrary order further supports the view that uncancelled nonlinearity in the difference between the reference and damaged geometries becomes manifest in the residual damage function and the terms that derive from it. If the order of the polynomial function for the reference boundary geometry is denoted as p and that of the damaged boundary geometry is denoted as q then the maximum polynomial order under consideration is n = Max(p, q). Both boundary geometry functions can be expanded to this maximum order by setting the coefficient terms associated with each order beyond the normative order equal to zero. The boundary geometry functions may then be written as: p

n

j=0

j=0

ηr ( l ) = ∑ arj l j = ∑ arj l j

www.collisionpumagazine.com

Volume 13 Issue 1.indd 53

(32)

q

n

j=0

j =0

ηd ( l ) = ∑ adj l j = ∑ adj l j

ξ ↔ l {l : l i ≤ l ≤ l i + 1 }

(35)

Collision Magazine - Volume 13 Issue 1 53

5/28/2019 9:34:34 AM

The residual damage function may then be written as: n

n

n

j=0

j=0

j=0

(

n

)

c ( l ) = ∑ arj l j − ∑ adj l j = ∑ arj − adj l j = ∑ ardj l j

ξ ↔ l {l : l i ≤ l ≤ l i + 1 }

j =0

(36)

The peak collision force magnitude, in closed form, for this case, becomes: n  a  rdj F = A ( l i + 1 − l i ) + B∑  l i + 1 j + 1 − l i j + 1  sec ( θ ) j=0  j + 1 

(

)

(37)

There does not appear to be a simple method for expressing the integral in equation (42) in an evaluated form as it relates to the product operator in the integrand. The zonal IWA, for this case, is:  n  aj   l i + 1j + 1 − l i j + 1  + A ∑  + j 1 = j 0     2   ( 1 + tan θ ) l i +1   2!  n B  k t tk t  ardt l   dl  + G ( l i + 1 − l i )   2 k +k ∑  k !k !L... k n ! l∫  ∏ ... t =0   i  0 1 +L + k n = 2  0 1 

(

)

(

)

(43)

W

orked Example The Vehicle Crash Test Database (VCTB) of the A closed form solution for the IWA, for this case is only National Highway Traffic Safety Administration partially determinable. The term associated with the A co(NHTSA) was queried in a semi-random manefficient is: ner for a Federal Motor Vehicle Safety Standard (FMVSS) l n  a 208 or high-speed frontal New Car Assessment Program  j A ∫ ( c ( l ) ) dl = A ∑  l i + 1j + 1 − l i j + 1 )  ( (NCAP) test conducted for vehicles falling within the last j=0  j + 1 l  (38) decade of model years. Test No. v08781 (NCAP test for a 2015 Hyundai Genesis four door all wheel drive sedan) The solution for the term associated with the B coefficient was selected from the resulting set of tests. The as-testproceeds by employment of the multinomial theorem, ed vehicle mass (2192.5 kg) and test vehicle speed at the start of closure (56.3 KPH) were taken directly from the which can be written as: contractor’s report for the collision test. Residual damage γ  γ  n k   n  γ profile information for collision testing of this type fol( x 1 + x 2 + x 3 + L... + x n ) =  ∑ x i  = ∑   ∏x  K ,k n  t = 1 t   i = 1  k + k +L... + k =γ   k 1 ,k 2 ,... lows one of two formats. The first is reporting the distance (39) aft from the reference centerline location. The second is The term to the right of the equality in equation (39) can reporting the distance from a datum reference at the rear surface of the vehicle. In regards to the second method, alternatively be expressed as: it is assumed that the datum reference plane is tangent to n the rear surface of the vehicle at the centerline. The sec  γ! xt k   ∑ ∏ ond method was used in regards to the reporting of the ... k + k +L...+ k =γ  k 1 !k 2 !L k n ! t = 1  (40) longitudinal dimensional data at the damage profile distance six (DPD6) locations. This dimensional data was In equation (40), the kt terms are the coefficients associ- converted into x-axis coordinate data by subtracting the ated with the terms in the polynomial expansion. For ex- reference configuration values for the rear overhang (Example, for a third order polynomial raised to the second pert AutoStats v.5.5.1; 4N6XPRT Systems; La Mesa, Calipower, the expansion would consist of six terms, each of fornia), the reported wheelbase and adding the reported which would involve k1, k2 and k3 being set to zero, or two longitudinal distance of the as-tested center of mass aft of with the sum of the kt values for each term being equal to the front wheels center. The y-axis coordinate data for the same six points was determined by noting that the location two. For the subject case: of the direct damage width midpoint was at the centerline 2 n  n    test2!vehicle nin its kreference  of the configuration and that 2! j t k  ardt l )  = ardt l tk )  (  ∑ ardj l  = (  spacing between ∑ ∑ ∏ ∏ ... the consecutive nodal locations was at a k !k ! k ! k !k ! k !

L L ... k + k +L + k = 2  0 1 n t =0 n t =0   k + k +L + k = 2  0 1  j=0  distance of 0.2 L (L = 1.520 m).

A seventh point, that n n   2! k tk t k  which was located at the reference centerline of the test  ∏ ( ardt l )  = k + k ∑ ∏ ( ardt l )  ...

L k n ! t =0 +L...+ k = 2  k 0 !k 1 !L k n ! t = 0   (41) vehicle, was also considered for inclusion. The inclusion of this point, however, proved problematic in that the resulThe term associated with the B model coefficient is only tant plan-view projected boundary geometry of the front partially reducible, in a closed form sense, for this case, as: of the test vehicle was pointed, in an exaggerated man() () ner, at the centerline, and inconsistently so with respect  B 2!    ∫ ∫ ( Bc) dcdl = ∫ ∫ ( Bc) dcdl = 2 ∑...  k !k !L... k ! ∫  ∏ ( a l )  dl  to the observable frontal geometry boundary shape as per   (42) qualitative examination of the pre-test photographs. As a consequence, only the DPD6 locations were considered i +1

i

t

1

2

n

t

1

2

n

t

0

1

0

t

t

0

1

l i +1 c l

li

li

0

k 0 + k 1 +L + k n = 2

54 Collision Magazine - Volume 13 Issue 1

Volume 13 Issue 1.indd 54

1

t

n

t

n

l i +1 c l 0

t

n

0

1

n

l i +1

n

li

t =0

rdt

k t tk t

www.collisionmagazine.com

5/28/2019 9:34:34 AM

for both the reference and damaged boundary geometry modeling tasks. The reference coordinate data points, in units of meters, were determined to be (-0.760, 2.162), (-0.456, 2.273), (-0.152, 2.315) and with symmetry about the test vehicle centerline, in the reference configuration, for the remaining three data points. The damaged coordinate data points, in units of meters, were determined to be (-0.760, 1.833), (-0.456, 1.831), (-0.152, 1.858), (0.152, 1.806), (0.456, 1.814) and (0.760, 1.801).

ηr ( l ) = 2.31962 + ( 1.22823 ⋅ 10−15 ) l − ( 1.968 ⋅ 10−1 ) l 2 − ( 2.61725 ⋅ 10−15 ) l 3 − ( 1.31722 ⋅ 10−1 ) l 4

(45)

The resultant reference boundary geometry model for this case is shown in Figure 2.

The following tasks were undertaken (all evaluations were conducted using Mathematica v.11.2; Wolfram Corporation; Champaign, Illinois): 1. Comparing the extant methodology, as per equation (14), for which the use of piecewise linear interpolation is employed for the residual damage function, against the formulation, based on the use of equations (21) and (22), which employs piecewise linear interpolation at the level of the reference and damaged boundary geometries. Showing an equivalency of the b1 model coefficient is sufficient for showing equivalency for the A and B model coefficients when the damage onset EBS (b0), the width of direct damage (L) and the mass of the vehicle (m) are held constant.

2. Comparing the results of the extant methodology against that based on the employment of the use of simple polynomial interpolation for both the reference and damaged geometries.

3. Comparing the results of the extant methodology against the results based upon the use of other interpolation methodologies (spline interpolation and Hermite interpolation).

Figure 2: Reference boundary data points and the resultant fourth order polynomial interpolation fit through the data. Note that the origin shown is not at a zero-value along the ordinate Fitting of the damaged boundary data points with a singular polynomial function proved to be highly problematic. The adjusted R2 values for fits from the first through fourth order were, 0.301698, 0.221018, 4.53928∙10-3, and -0.932776, respectively. The fifth order polynomial interpolation, through this data, exhibited the undulations between the data points that are characteristic of Runge’s phenomenon. This is shown in Figure 3.

For the first task, the use of the as-reported nodal residual damage depths resulted in α = 1.3449 m2 and β = 9.01624∙10-1 m3. The solution for the b1 model coefficient, for both the extant method and the use of piecewise linear interpolation at the level of the reference and damaged boundary geometries resulted in the following solution (with b0 expressed in units of m/sec and b1 expressed in units of sec-1): b1 = −2.23746b0 +

1 20.0249b02 − 20.2302 ( b02 − 244.575) 2

(44)

The use of polynomial interpolation for the reference data resulted in the generation of an exact fit (adjusted R2 = 1), using a fourth order polynomial function, and with no exhibition of Runge’s phenomenon. The interpolation equation was:

Figure 3: Data points for the damaged boundary geometry and the fifth order polynomial interpolation function for fitting the data Polynomial interpolation for the second task, due to the problematic nature of the fit, was not used. Instead, for the second task, the damaged boundary geometry was modeled in a piecewise linear manner while the fourth order polynomial function shown by equation (45) was

www.collisionpumagazine.com

Volume 13 Issue 1.indd 55

Collision Magazine - Volume 13 Issue 1 55

5/28/2019 9:34:34 AM

used for the reference boundary geometry model. The residual damage function that was generated consisted of five piecewise fourth order polynomial functions. The resultant solution for the b1 model coefficient is: b1 = −2.2128b0 + ( 2.25413 ⋅ 10−3 ) 2.37887 ⋅ 108 − ( 8.9918 ⋅ 103 ) b02

(46)

For the case of spline and Hermite interpolation, the inbuilt Interpolation function was used for both the reference and damaged boundary geometries. The interpolation method was specified within the function call as was the interpolation order. Visual inspection of the resultant, for the spline function, revealed no significant change in the visual appearance of the resultant beyond an interpolation order of two for both the reference and boundary geometries. A similar approach for the Hermite interpolation function revealed an interpolation order of three for the reference boundary geometry and two for the damaged boundary geometry. In the latter case, a higher order interpolation

resulted in the exhibition of Runge’s phenomenon. The residual damage function was then determined for each pairwise case for the reference (piecewise linear, polynomial, spline and Hermitian) and damaged (piecewise linear, spline and Hermitian) boundary geometry models. The use of the inbuilt interpolation function resulted in the generation of Interpolation Objects for c(l). The Interpolation Objects for both c(l) and (c(l))2 were numerically integrated and with the numerical results serving as a portion of the coefficient values in the quadratic form of the Campbell b1 model coefficient equation. That being: N − 1  l i +1    2b02 ∑  ∫ ( c ( l ) ) dl   2 2    i =1  l i  + ( b0 − EBS ) L = 0 b 12 + b 1   l   N − 1  l i +1  2 2  N − 1  i +1 c l dl   ∑  ∫ ( c ( l ) ) dl   ∑  ∫ ( ( ) )       i =1  l i   i =1  l i

(47)

Figure 4: Residual damage functions based upon the use of piecewise linear (left), spline(center) and Hermite (right) modeling of the damaged boundary geometry. For each plot, piecewise linear (blue), polynomial (red), spline (green) and Hermite (orange) models were used for the reference boundary geometry Reference Linear Polynomial Spline Hermite Linear Polynomial Spline Hermite Linear Polynomial Spline Hermite

Damaged Linear Linear Linear Linear Spline Spline Spline Spline Hermite Hermite Hermite Hermite

a1 -2.23746 -2.21280 -2.21412 -2.21319 -2.23796 -2.21317 -2.21450 -2.21356 -2.23988 -2.21519 -2.21654 -2.21559

a2 1.94214 ⋅ 10-8 8.76205 ⋅ 10-16 1.94070 ⋅ 10-15 9.01541 ⋅ 10-15 1.28996 ⋅ 10-14 1.66971 ⋅ 10-14 3.90565 ⋅ 10-14 3.12938 ⋅ 10-14 5.47228 ⋅ 10-09 3.29728 ⋅ 10-14 1.93164 ⋅ 10-15 6.80645 ⋅ 10-15

a3 3.27937 ⋅ 1018 1.57442 ⋅ 1033 3.21420 ⋅ 1032 1.48785 ⋅ 1031 7.43620 ⋅ 1030 4.33706 ⋅ 1030 7.93881 ⋅ 1029 1.23528 ⋅ 1030 4.14007 ⋅ 1019 1.11430 ⋅ 1030 3.25185 ⋅ 1032 2.61623 ⋅ 1031

a4 -1.36082 ⋅ 1014 -5.95107 ⋅ 1028 -1.25724 ⋅ 1028 -5.69091 ⋅ 1026 -3.08591 ⋅ 1026 -1.64136 ⋅ 1026 -3.10763 ⋅ 1025 -4.72989 ⋅ 1025 -1.73912 ⋅ 1015 -4.26375 ⋅ 1025 -1.28560 ⋅ 1028 5.47228 ⋅ 10-09

Table 1: Constant terms for the form of the Campbell b1 model coefficient based upon the form of equation (48) and as a function of the type of boundary geometry models used for the reference and damaged boundary geometries 56 Collision Magazine - Volume 13 Issue 1

Volume 13 Issue 1.indd 56

www.collisionmagazine.com

5/28/2019 9:34:35 AM

The form of equations (44) and (46) serve as examples of a convenient generalized form for expressing the solutions for the Campbell b1 model coefficient. This form is given by equation (48), in which the terms a1 through a4 are constants that are expected to differ based upon the forms of the reference and damaged boundary geometry models that are used to generate the residual damage function. b1 = a1b0 + a2 a3 + a4 b02

(48)

The residual damage functions are shown in Figure 4 and the constant coefficients are shown in Table 1.

The response shown by equation (48) is that of a linear response of the Campbell b1 model coefficient in regards to the Campbell b0 model coefficient. This linear response was shown by plotting each function, as shown in Figure 5, over a reasonable domain of b0 values.

For a nominal b0 value of 1.1176 m/sec (2.5 MPH), the minimum b1 value was 32.2930 sec-1 (polynomial reference boundary geometry and linear damaged boundary geometry) and the maximum value was 32.7063 sec-1 (linear reference boundary geometry and Hermite damaged boundary geometry). For the same nominal b0 value, the minimum and maximum A model coefficient values occurred, as expected, for the same cases as for the Campbell b1 model coefficient, and with respective values of 5.20584 . 104 N/m (297.261 lb/in) and 5.27246 . 104 N/m (301.066 lb/in). The graphical results are shown in Figure 6. The same models provided, again, as expected, the minimum and maximum values for the B model coefficient. For the same nominal b0 value, the minimum and maximum values, respectively, were 1.50423 . 106 N/m2 (218.17 lb/in2) and 1.54297 . 106 N/m2 (223.789 lb/in2).

The graphical results are shown in Figure 7.

Figure 5: Campbell b1 model coefficient for the cases of piecewise linear (left), spline (center) and Hermite (right) modeling of the deformed boundary geometry. For each plot, piecewise linear (blue), polynomial (red), spline (green) and Hermite (orange) models were used for the reference boundary geometry

Figure 6: CRASH3 model A coefficient values for the case of piecewise linear (left), spline (center) and Hermite (right) modeling of the deformed boundary geometry. The color coding for each plot follows that of Figure 5

Figure 7: CRASH3 model B coefficient values for the case of piecewise linear (left), spline (center) and Hermite (right) modeling of the deformed boundary geometry. The color coding for each plot follows that of Figure 5 www.collisionpumagazine.com

Volume 13 Issue 1.indd 57

Collision Magazine - Volume 13 Issue 1 57

5/28/2019 9:34:35 AM

The percent relative difference in the EBS as well as the IWA, between each of the residual damage depth functions and the extant model were calculated over the domain of b0 {b0: 0 m/sec ≤ b0 ≤ 2.2352 m/sec} and c {c: 0 m ≤ c ≤ 0.508 m (20 in)}. For every case under consideration, the maximum percent relative difference for the EBS as well as the IWA was found at b0 = 0 m/sec and c = 0.508 m. The results are shown in Table 2 and the surface plots, over the domain, are shown in Figures 8-9.

D

iscussion The rectangular approximation of the plan projected reference boundary geometry is oft present, at least implicitly by means of depiction in figures and diagrams, in the pioneering scientific literature. The linear nature of the regional boundary geometry functions, that being straight lines, is quite apparent.

When coupled with a piecewise linear model for the plan projected deformed boundary geometry, the resulting residual damage function is itself piecewise linear. Both by expectation and as shown by the subject work, the residual damage depth function is linear when both the reference and damaged boundary geometry functions are linear over a mutual domain. Piecewise linearity holds when both boundary geometry functions are linear over all subdomains of the direct contact length. For these cases, the use of the residual damage depth function as an antecedent step reproduces the extant approach, which is predicated

Reference Polynomial Spline Hermite Linear Polynomial Spline Hermite Linear Polynomial Spline Hermite

Damaged Linear Linear Linear Spline Spline Spline Spline Hermite Hermite Hermite Hermite

% RD EBS -1.15 % -1.07 % -1.12 % 1.79 . 10-2 % -1.13% -1.05 % -1.11 % 1.11 . 10-2 % -1.03% -9.59. 10-1 % -1.01%

% RD IWA -2.23 % -2.13 % -2.24 % 3.57 . 10-2 % -2.25% -2.10% -2.20% 2.29 . 10-1 % -2.06 % -1.91 % -2.01 %

Table 2: Maximum EBS and IWA percent relative difference values for each residual depth function formulation (the reference geometry boundary function is shown in the first column and the damaged boundary geometry function is shown in the second column) when compared to the extant formulation 58 Collision Magazine - Volume 13 Issue 1

Volume 13 Issue 1.indd 58

upon the application of linear or piecewise linear interpolation to the residual damage profile, itself.

Plan projected vehicle reference boundary geometries, however, are rarely linear to the extent that a single constant valued function suffices for accurately modeling an entire region of the same. Instead, the curvature of the relevant vehicle structures in ℜ3 maps into curved planar functions in ℜ2. These curved shapes are inherently nonlinear and the introduction of nonlinearity in regards to the reference boundary geometry introduces nonlinearity into the residual damage depth function. The same argument holds when the deformed boundary geometry is modeled using nonlinear functions. The presence of nonlinearity in the residual damage depth function produces a deterministic inaccuracy when the extant closed form solutions are employed for quantifying the collision force magnitude and the IWA in that both are predicated upon the use of linear interpolation of the residual damage depth function, itself. There exist a vast number of interpolation approaches that can reasonably be considered as having utility when it comes to modeling either the reference or the deformed boundary geometries or both. The development of compact, closed-form, analytic solutions for quantifying the model coefficients, the collision force magnitude and the IWA may or may not be readily achievable depending on the nature of the nonlinear functions that are employed. This was readily evidenced in the subject work for the case of a piecewise quadratic model for the reference boundary geometry coupled with the case of a piecewise linear model for the damaged boundary geometry, which produced a piecewise quadratic model for the residual damage depth function. Closed-form analytic solutions were readily derived for the collision force magnitude and the IWA. When the model was extended an arbitrary order polynomial for both boundary geometry functions, the compact, closed-form analytic solution for the IWA, specifically involving the CRASH3 B model coefficient, could not be fully elucidated.

The presence of a deterministic inaccuracy and the inaccuracy being significant are not mutually inclusive. The example that was developed is constructive in this regard.

The use of piecewise linear, fourth order polynomial, second order spline, or third order Hermite interpolation functions for the reference boundary geometry and piecewise linear, second order spline or second order Hermite interpolation functions for the deformed boundary geometry produced minimal differences in the minimum and maximum values of the resultant Campbell b1, CRASH3 A and CRASH3 B model coefficients. The resultant maximum relative percent differences, when compared with the extant methodology, over a reasonable domain of Camp-

www.collisionmagazine.com

5/28/2019 9:34:35 AM

Figure 8: Percent relative difference in the EBS for the cases of the deformed geometry boundary component of the residual damage depth function being generated from the piecewise linear (left), spline interpolated(center) and Hermite interpolated (right). The extant model serves as the reference case for the percent relative difference comparison. For each graph, the reference geometry boundary component of the residual damage depth function are linear (orange), polynomial (red), spline (blue) and Hermite (green) generated.

Figure 9: Percent relative difference in the EBS for the cases of the deformed geometry boundary component of the residual damage depth function being generated from the piecewise linear (left), spline interpolated(center) and Hermite interpolated (right). The extant model serves as the reference case for the percent relative difference comparison. For each graph, the reference geometry boundary component of the residual damage depth function are linear (orange), polynomial (red), spline (blue) and Hermite (green) generated. bell b0 vales and residual damage depths, was -1.15% for the EBS and -2.25% for the IWA. The minimum nature of these differences, for the example case, can readily be explained, first, by the fact that the measurement protocol for the collision test that was used as the example, accounts for the backset at the nodal locations that were lateral of the test vehicle centerline. In turn, the use of piecewise linear interpolation at the level of the residual damage depth not only did not overestimate the nodal damage depths but also produced the same result as that which was determined by employing piecewise linear interpolation, separately, for the reference and damaged geometry boundaries and then taking their difference for the development of the residual damage depth function. The second reason that can be cited for the small relative percent errors is that the set of pairs for the two boundary geometry functions produced very similar residual damage depth functions.

One may reasonably say that the values for the percent relative errors would have been greater if the rectangular approximation was used for the reference boundary geometry (i.e. all points on the front of the reference boundary

geometry being located at a constant longitudinal distance from the center of mass of the test vehicle and with this distance being equal to the centerline distance) secondary to an overestimation of the residual damage depth lateral to the centerline.

The findings determined for the single case shown in the example are for the single example itself and are limited to the scope of the geometry boundary models that were considered. It would be grossly inappropriate to say that they are representative or not representative of a broader norm simply due to the fact that the broader norm, itself, has neither been established nor developed in the subject work. Furthermore, there are a number of other interpolation modeling approaches that were not considered and that could in theory lead to different results for the single test case that was provided. While the theory was developed for the cases in which a balance of forces is employed, the impact of the theory was not shown by example. A final point of consideration for future work is the mapping of the location of the centroid of the damaged area

www.collisionpumagazine.com

Volume 13 Issue 1.indd 59

Collision Magazine - Volume 13 Issue 1 59

5/28/2019 9:34:35 AM

from the l-c(l) space into the local coordinate system of the vehicle itself. In this regard, the extant methodology does employ the rectangular approximation for the placement of the l-axis.

R

eferences

1. Campbell K (1972) Energy as a basis for accident severity – a preliminary study. Dissertation: University of Wisconsin, Automotive Engineering.

2. Campbell K (1974) Energy basis for collision severity.

Society of Automotive Engineers Technical Paper No. 740565.

3. Chen F, C Tanner, P Cheng and D Guenther (2005) Application of force balance method in accident reconstruction. Society of Automotive Engineers Technical Paper No. 2005-01-1188.

4. Hull W and B Newton (1993) Estimating crush stiffness when reconstructing vehicle accidents. Society of Automotive Engineers Technical Paper No. 930898.

5. Kerkhoff J, S Husher, M Varat, A Busenga and K Hamilton (1993) An investigation into vehicle frontal impact stiffness, BEV and repeated testing for reconstruction. Society of Automotive Engineers Technical Paper No. 930899.

6. Long T (1999) A validation study for the force balance method in determination of stiffness coefficients.

Society of Automotive Engineers Technical Paper No. 1999-01-0079.

7. McHenry, RR (1976) Extensions and refinements of the CRASH computer program, part II, user’s manual for the Crash computer. US Department of Transportation, National Highway Traffic Safety Administration (DOT HS 801 838).

8. Neptune J, G Blair and J Flynn (1992) A method for quantifying vehicle crush stiffness coefficients. Society of Automotive Engineers Technical Paper No. 920607.

11. Nilsson-Ehle A, H Norin and C Gustafsson (1982) Evaluation of a method for determining the velocity change in traffic accidents. Society of Automotive Engineers Technical Paper No. 826081.

12. Noga T and T Oppenheim (1981) CRASH3 user’s guide and technical manual. US Department of Transportation, National Highway Traffic Safety Administration (DOT HS 805 732).

13. Sharma D, S Sterns, J Brophy and E Choi (2007) An overview of NHTSA’s crash reconstruction software WinSMASH. Proceedings: 20th International Conference on the Enhanced Safety of Vehicles, Lyon, France. Paper No. 07-0211.

14. Singh J, J Welcher and J Perry (2003) N-point linear interpolation of motor vehicle crush profiles applied to various force-shortening models. International Journal of Crashworthiness 8(4), 321-328.

15. Singh J (2005a) Closed form implementation of an N-point unequally spaced approximation of vehicle shortening applied to various force-shortening models. International Journal of Impact Engineering 31(10), 1235-1252.

16. Singh J (2005b) The effect of residual damage interpolation mesh fineness of calculated side impact stiffness coefficients. Society of Automotive Engineers Technical Paper No. 2005-01-1205.

17. Singh J and J Perry (2008) Multilinear and nonlinear power law modeling of motor vehicle force-deflection response for uniaxial front impacts. Accident Reconstruction Journal 18(6), 30-38, 62.

18. Strother C, R Woolley, M James and Y Warner (1986) Crush energy in accident reconstruction. Society of Automotive Engineers Technical Paper No. 860371.

19. Woolley R (2001) Non-linear damage analysis in accident reconstruction. Society of Automotive Engineers Technical Paper No. 2001-01-0504.

9. Neptune J and J Flynn (1994) A method for determining accident specific crush stiffness coefficients.

Society of Automotive Engineers Technical Paper No. 940913.

10. Neptune J and J Flynn (1998) A method for determining crush stiffness coefficients from offset frontal and side crash tests. Society of Automotive Engineers Technical Paper No. 980024.

60 Collision Magazine - Volume 13 Issue 1

Volume 13 Issue 1.indd 60

www.collisionmagazine.com

5/28/2019 9:34:35 AM

www.collisionpumagazine.com

Volume 13 Issue 1.indd 61

Collision Magazine - Volume 13 Issue 1 61

5/28/2019 9:34:36 AM

62 Collision Magazine - Volume 13 Issue 1

Volume 13 Issue 1.indd 62

www.collisionmagazine.com

5/28/2019 9:34:36 AM

Volume 13 Issue 1.indd 63

5/28/2019 9:34:36 AM

Volume 13 Issue 1.indd 64

5/28/2019 9:34:36 AM

Volume 13 Issue 1.indd 65

5/28/2019 9:34:36 AM

TE

Volume 13 Issue 1.indd 66

5/28/2019 9:34:37 AM

branding@teslamotors.com.

1

TESLA BRAND STANDARDS MANUAL

EDR Retrieval Kit for Tesla Vehicles Introducing the EDR Retrieval Hardware Kit for Tesla vehicles. This kit contains the cables and adapters required to download the Event Data Recorder (EDR) data that may be stored in Tesla Model 3, Model S and Model X vehicles.

The Tesla EDR Hardware Kit contains genuine Tesla cables and is available only at

crashdatagroup.com Order your kit today.

www.collisionpumagazine.com

Volume 13 Issue 1.indd 67

Collision Magazine - Volume 13 Issue 1 67

5/28/2019 9:34:37 AM

Small Unmanned Aircraft Systems Photogrammetry vs.

Total Station Sergeant Joseph Weadon, III Missouri State Highway Patrol

I

M

ntroduction The use of small unmanned aircraft systems (sUAS) for crash scene mapping has taken the reconstruction community by storm. Many crash reconstructionists have begun using these aerial photography platforms in conjunction with photogrammetry software to capture scene evidence and roadway characteristics. Most, it seems, have relied on what they were told about the accuracy of the process in a short course and simply accepted its validity. We took a different approach. To determine the validity of the data and any potential error rate, we compared our results to another long-accepted method of measurement; the total station.

ethodology The testing took place at the Emergency Vehicle Operations Course (EVOC) of the Missouri State Highway Patrol, located in Jefferson City, Missouri. We selected four sites along the course, which provided varying degrees of grade and super-elevation. Additionally, the sites were composed of a variety of surfaces. After the sites were identified, a total of 100 test points were painted on the different surfaces at each site. Test points were painted using a template comprised of offset 0.3-foot squares, which touched only at one corner. White paint was used to paint most test points and all points were numbered. Figure 1: Test Points shows a sample of the test points painted on asphalt and grass.

68 Collision Magazine - Volume 13 Issue 1

Volume 13 Issue 1.indd 68

www.collisionmagazine.com

5/28/2019 9:34:38 AM

Site one was exclusively asphalt. It was a large, banked curve with extreme super-elevation. The site was approximately 900 feet long. Site two was in an “S” shaped curve, had a change in grade, and incorporated a negative super-elevation. It was approximately 550 feet long. Of the 100 test points at site two, 64 were on asphalt and 36 were on grass. The grass was approximately 0.2-0.3 feet tall.

Site three was mostly level and was approximately 450 feet long. There was a 0.35-foot curb on the asphalt that paralleled the roadway. At site three, 70 points were painted on asphalt and 30 points were painted on the varying height grass surface.

The grass ranged from 0.2 feet tall to over 2.5 feet tall. Furrows were created in the taller grass using a string trimmer(weed-eater) and test points were painted on the ground in the furrows. Site four was approximately 700 feet long and consisted of both asphalt and gravel roadway. The asphalt road was

The sUAS used for this testing was a DJI Inspire 2 with a Zenmuse X4S camera. Each site was photographed in two separate ways. The first photographs were taken with the camera looking straight down at a 90-degree angle to the surface (ortho mode).

The second group of photos were taken with the camera gimbal slightly angled (oblique mode). The oblique mode angle varied site to site but was between approximately 75 degrees and 90 degrees.

This simulated flying a crash scene with moving traffic on the road without flying directly over the traffic. Existing federal regulations prohibit flying a sUAS above moving traffic.

For each mode used, the following guidelines were adopted. One pass was flown at approximately 100 feet above ground level (AGL). During this pass, photographs were taken with an overlap of approximately 80 percent. During this pass, each point along the test site should have shown up in

After documenting each site with the total station it was flown and photographed with a sUAS. mostly level. The gravel sloped down away from the asphalt. The gravel area was approximately the width of a one lane road. At site four, 80 test points were painted on the asphalt surface and 20 points were painted on the gravel.

In addition to the test points, ground control points (GCPs) were painted at each site. The ground control points were painted at each end of the test points and throughout the center. Site three, which had the “furrows” in the tall grass, had ground control points painted in the furrows and grass. Each site was mapped with a Sokkia iX robotic total station, using Magnet Field as the data collection software.

To mitigate pole sway, the range pole was held by one person and the data collector was held by another. When the test points were documented, the total station averaged three shots to improve its accuracy. After the 100 test points were documented, the ground control points were documented. After documenting each site with the total station, it was flown and photographed with a sUAS.

approximately five photographs. The second pass was flown at approximately 50 feet AGL. During this pass photographs were taken with an overlap of approximately 75 percent. This means each point should have shown up in approximately four photographs. During this pass if the entire width of the test area was not able to be seen, then two passes were flown at 50 feet and they overlapped in the center of the test area. The last pass was flown at approximately 25 feet AGL. During this pass there should have been 50 percent overlap of each photograph. This means each point should have shown up in approximately two photographs. Again, if the width of the test site cannot be seen in each photograph, then the photographs were overlapped near the center of the test site and two or more passes were flown. The only difference in the oblique passes was the sUAS was not flown over the “lanes of travel.”

www.collisionpumagazine.com

Volume 13 Issue 1.indd 69

Collision Magazine - Volume 13 Issue 1 69

5/28/2019 9:34:38 AM

After the test sites were documented with the total station and the sUAS, the photographs were loaded into PIX4D. PIX4D was the photogrammetry software used to process the scenes. Each site was processed separately with the ortho mode photographs and the oblique mode photographs. Each mode was developed with 5 or 6 ground control points and again with 10 or 12 ground control points.

Once the software finished processing the scene, the point clouds were turned off and test points were marked on the 3D mesh in PIX4D. The marked points were then exported in a comma separated value list. After this data was exported, each point was adjusted using the raw photographs in PIX4D. The data was again exported in a comma separated value list. Both lists were compared to the data collected with the total station.

The data was compared in the simplest form. The x,y,z coordinates documented with the total station, were compared to the x,y,z coordinates of each point documented

with PIX4D. Then the straight-line difference was calculated between the two points. This distance was called the slope distance (SD). The following formula was used.

This yielded a real-world difference that was easily explained.

R

esults When comparing the data sets, there was no appreciable difference identified between the orthogonal and oblique photography methods.

There also appeared to be little difference between marking the test points only on the triangle mesh and using raw photographs to mark the location of each test point.

The exception to this was in the grass and likely any other object that was not at ground level. When the test point was in the grass it had slightly less difference after it was

Figure 1: Test points 70 Collision Magazine - Volume 13 Issue 1

Volume 13 Issue 1.indd 70

www.collisionmagazine.com

5/28/2019 9:34:41 AM

marked on raw photographs. (This is also likely true for three dimensional objects like signs, vehicles, etc. but that was not tested).

Site One Site one was first processed with 5 ground control points.

The location of the ground control points can be seen in Figure 4: Site 1-5 GCP X,Y. Notice the ground control points were placed in a colinear fashion. Site one is shown in Google Earth image Figure 2: Site one. The numbered push pins were also the location of the test points and correspond to the point number.

The data collected with the total station was compared to the data developed and documented in the 3D model using Pix4D. The results of that comparison are shown in the line graph Figure 3: Site 1-5 GCP. Looking at this data, the points near ground control points were more accurate.

However, near the center of the curve, the difference grew to over 0.8 feet. When this site was initially processed, the PIX4D 3D model appeared accurate. There was no indication the test points near the center of the curve were off by nearly a foot.

The data sets were also shown in a X,Y scatter chart with the test points plotted according to their X and Y coordinates (Figure 4: Site 1-5 GCP X,Y). Since the location of the test points was located with the total station in an X,Y,Z configuration it matched the shape of the test site.

The test points in the scatter chart are color coded according to the length of the slope distance difference. The color code is shown at the top of the chart. This is useful because you can see how ground control points affected the accuracy of surrounding points.

Site one was processed again with 10 ground control points. This time more widely spaced and non-colinear.

The same method was used to compare the data. These results are shown in Figure 5: Site 1-10 GCP. The location of the ground control points can be seen in scatter chart Figure 6: Site 1-10 GCP X,Y. The largest slope distance difference was less than 0.2 feet.

The data was again plotted in a X,Y scatter chart with points color coded to indicate the slope distance difference. The color code is shown at the top of the chart. This was shown in chart Figure 6: Site 1-10 GCP X,Y.

Figure 2: Site 1 www.collisionpumagazine.com

Volume 13 Issue 1.indd 71

Collision Magazine - Volume 13 Issue 1 71

5/28/2019 9:34:43 AM

Figure 3: Site 1-5GCP

Figure 4: Site 1-5GCP X,Y 72 Collision Magazine - Volume 13 Issue 1

Volume 13 Issue 1.indd 72

www.collisionmagazine.com

5/28/2019 9:34:43 AM

Figure 5: Site 1-10GCP

Figure 6: Site 1-10GCP X,Y www.collisionpumagazine.com

Volume 13 Issue 1.indd 73

Collision Magazine - Volume 13 Issue 1 73

5/28/2019 9:34:43 AM

The average final data for site one is shown in Table 1.

Site One # of Photographs 218 5 GCP Average 0.35 Minimum SD Difference 0.04 Median SD Difference 0.35 Largest SD Difference 0.83

10 GCP 0.09 0.01 0.09 0.17

at site two was approximately 0.2-0.3 feet (as shown in Figure 8).

Site two was processed using 5 ground control points and again with 12 ground control points. The location of the ground control points can be seen in their respective scatter charts.

Table 1: Site 1 average final data Site Two Site two was comprised of points on asphalt and grass. It had gradual elevation and super-elevation changes. It was shown in Google Earth Image Figure 7: Site 2. The numbered push pins were also the location of the test points and corresponded to the test point number.

Site two data points one through 64 were painted on asphalt. Data points 65 through 100 were painted in the grass off the edge of the roadway. The height of the grass

Figure 8: Grass height

Figure 7: Site 2 74 Collision Magazine - Volume 13 Issue 1

Volume 13 Issue 1.indd 74

www.collisionmagazine.com

5/28/2019 9:34:46 AM

The data in Figure 9: Site 2-5 GCP shows the slope distance difference near point one and near point 63 (near the ground control points) is smaller. This illustrated the effectiveness of ground control points.

and density of the grass. Photogrammetry cannot see into the grass to determine where the ground was or how deep the grass was. Therefore, what the photogrammetry software determined was the “surface” was influenced by the height of the grass and its density. Photogrammetry could Figure 10: 2-5 GCP X,Y shows the test points plotted on only build a model to what it perceived was the “surface.” an X,Y scatter chart. The same analysis performed in the This was most likely the top of the grass. The third chalsite one scatter chart, was performed in this chart. The lenge was being able to consistently pick the exact same points were color coded according to the size of the slope spot painted on grass with the range pole and in the phodistance difference. The color code is shown at the top of tographs. the chart. This graph helped illustrate the grass accounts for most of the error. However, there was a small section The data from site two with 12 ground control points is near the center of the test points that has a slightly larger shown in Figure 11: Site 2-12 GCP. First, notice the scale difference when compared to each end of the site. It was was the same for both line graphs. here the ground control points were painted adjacent to the test points and not close enough to the super-eleva- Like the data from site one, the average asphalt difference tion change to help build the model. Granted, the differ- was lower on the model with 12 ground control points. ence was small. Grass and soft soil present a few unique This was because of the placement of ground control challenges when comparing total station data to the 3D points near the center of the test site. However, notice the model built using photogrammetry. The first problem in- average difference in the grass, which starts at point 65 on volved the range pole and its tendency to sink into soft both line graphs is larger. soil. That could not be accounted for using photogram- The model with 12 ground control points had a larger metry. The second issue we identified involved the depth slope distance difference in the grass when compared to

Figure 9: Site 2-5 GCP (Notice the “Y” scale of this graph is different than it was on the site one graph) www.collisionpumagazine.com

Volume 13 Issue 1.indd 75

Collision Magazine - Volume 13 Issue 1 75

5/28/2019 9:34:46 AM

Figure 10: Site 2-5 GCP X,Y

Figure 11: Site 2-12GCP 76 Collision Magazine - Volume 13 Issue 1

Volume 13 Issue 1.indd 76

www.collisionmagazine.com

5/28/2019 9:34:47 AM

the total station data than the model built with five ground control points. The larger difference is shown on the line graph. It appeared to be largely caused by the elevation difference. The slope distance difference almost mirrored the elevation difference. X and Y values remained very accurate.

Notice when comparing the two “X,Y” scatter charts for site two, the area near the center no longer has the difference. The difference was negligible; however, ground control points made the model more accurate. The only area that had a slope distance consistently above 0.05 feet was grass.

Another item worth noting, which highlighted the capabilities of photogrammetry, was found at point 64. In line graph Figure 11: Site 2-12 GCP, you can see point 64 was an asphalt point, yet its “Z” difference appears larger than the other asphalt points. While laying out the test site, point 64 fell on a crack in the asphalt. This is shown in Figure 13. While mapping this point with the total station the tip of the range pole went into the crack approximately 0.05 feet. The PIX4D software could not see this small hole in the asphalt and developed the triangle mesh at the surface of the asphalt. The larger slope distance difference

Figure 13: Point 64 located on a crack in the asphalt for point 64 can be accounted for because the range pole went into this crack.

The data gathered from site two with 12 ground control points was shown in Table 2: Site 2 average final data.

Figure 12: Site 2-12 GCP X,Y www.collisionpumagazine.com

Volume 13 Issue 1.indd 77

Collision Magazine - Volume 13 Issue 1 77

5/28/2019 9:34:48 AM

Site 2 # of Photographs 421 Grass Average 0.13 Minimum SD Difference 0.03 Median SD Difference 0.13 Largest SD Difference 0.21

Asphalt 0.03 0.00 0.03 0.09

Overall 0.07 0.00 0.03 0.21

Table 2: Site 2 average final data Site Three Site three lay in a relatively level area. The points numbered 1-70 were on the asphalt and the points numbered 71-100 were in the grass. Site three is shown in Google Earth image Figure 14: Site 3. The numbered push pins were also the location of the test points and correspond to the point numbers.

Approximately 12 feet from the east edge of the pavement there was a curb. The 12-foot shoulder was lower than the rest of the road. This curb was approximately 0.35 feet tall and was shown in Figure 15.

The grass ranged from 0.2 feet tall to approximately three feet tall. Grass taller than 0.2 feet had mock furrows cut

Figure 15: Curb height

Figure 14: Site 3 78 Collision Magazine - Volume 13 Issue 1

Volume 13 Issue 1.indd 78

www.collisionmagazine.com

5/28/2019 9:34:51 AM

into it with a string trimmer. The “furrows” were cut down to the soil. This did not replicate tires rolling across grass.

Like the other scenes, the 3D model of this scene was first developed with five ground control points and then a second time with 12 ground control points.

In the model built with five ground control points four of those points were painted on the asphalt and one was painted at the far-east end of the grassy area, on bare soil, in one of the simulated furrows. The location of the ground control points can be seen on the respective scatter charts.

The scene was processed through PIX4D. Figure 16: Site 3-5 GCP shows the results of that data compared to the total station data. Note the scale of this graph is slightly different than the previous graphs.

Examining the data revealed the slope distance difference on the asphalt surface was less than 0.1 feet. However, the points in the grass had a larger slope distance difference.

That difference remained less than 0.25 feet. The points in the grass started at number 71. In Figure 17: Site 3-5 GCP X, Y, the test points were plotted on a X-Y chart.

The points were color coded according to the difference in slope distance. The color code is shown at the top of the chart.

The data looked very similar to the data from site two.

There was very little difference in the slope distance difference on the asphalt. The difference in the slope distance

in the grass was mainly introduced by the elevation difference.

Site three was processed again with 12 ground control points. The results of that are shown in Figure 18: Site 3-12 GCP. In Figure 19: Site 3-12 GCP X, Y, the points were plotted on an X-Y scatter chart color coded according to the difference in slope distance. The color code is shown at the top of the chart.

The additional ground control points on the asphalt slightly increased the accuracy. It was however, minimal.

Looking at the data from the two 3D models generated by Pix4D there was not a significant difference. The data for site three with 12 ground control points is shown in table three.

Site 3 # of Photographs 261 Grass Average 0.08 Minimum SD Difference 0.01 Median SD Difference 0.08 Largest SD Difference 0.22

Asphalt 0.04 0.01 0.03 0.06

Overall 0.05 0.01 0.04 0.22

Table 3: Site 3 average final data

Figure 16: Site 3-5 GCP www.collisionpumagazine.com

Volume 13 Issue 1.indd 79

Collision Magazine - Volume 13 Issue 1 79

5/28/2019 9:34:51 AM

Figure 17: Site 3-5 GCP X,Y

Figure 18: Site 3-12 GCP 80 Collision Magazine - Volume 13 Issue 1

Volume 13 Issue 1.indd 80

www.collisionmagazine.com

5/28/2019 9:34:51 AM

Figure 19: SITE 3-12 GCP X,Y

Site Four Site four consisted of asphalt and gravel. The gravel was the approximate width of a one lane road. The asphalt was nearly level and the gravel road sloped down away from the asphalt. Site four can be seen in Google Earth image Figure 20: Site 4 The test points at site four started at the north end on the asphalt and continue toward the south on the asphalt to point 80. Point 81 was painted on the gravel at the edge of the asphalt and continued east to point 100. I used the data gathered using the “oblique” mode for this site. The 3D model was processed using five ground control points and a second time using ten ground control points.

A comparison of the data gathered with the total station and the 3D model developed using only five ground control points is shown in Figure 21: Site 4-5GCP. Ground control points were painted near point 1, point 80, and point 100. The location of the ground control points can be seen on the respective scatter charts. While the overall difference between the 3D model and the total station data was small, you can see ground control points appear

to make the difference even smaller. Examining the line graph, it was evident the difference in slope distance was smallest near points 1, 80 and 100.

In Figure 22: Site 4-5 GCP X,Y, the points were plotted on an X-Y table and color coded according to the difference in slope distance when compared to the total station data.

Again, the difference at each end of the asphalt near the GCPs is smaller. The color code is at the top of the chart.

The 3D model developed with 10 ground control points and compared to the data gathered with the total station is shown in Figure 23: Site 4-10 GCP. This data showed the difference decreased slightly. This is negligible for most crash reconstruction applications.

However, it illustrates the importance of ground control points. This can also be seen in Figure 24, Site 4-10 GCP X,Y. The points were plotted on an X-Y chart and color coded according to the difference in slope distance when compared to total station data. The color code is located at the top of the chart.

www.collisionpumagazine.com

Volume 13 Issue 1.indd 81

Collision Magazine - Volume 13 Issue 1 81

5/28/2019 9:34:51 AM

Figure 20: Site 4

Figure 21: Site 4-5 GCP 82 Collision Magazine - Volume 13 Issue 1

Volume 13 Issue 1.indd 82

www.collisionmagazine.com

5/28/2019 9:34:53 AM

Figure 22: Site 4, 5 GCP X,Y

Figure 23: Site 4-10 GCP www.collisionpumagazine.com

Volume 13 Issue 1.indd 83

Collision Magazine - Volume 13 Issue 1 83

5/28/2019 9:34:53 AM

Figure 24: Site 4-10 GCP X,Y Table 4 shows the data from site four using ten ground control points.

foot altitude photographs and was more difficult to mark.

It was also a much smaller number of photographs.

Site 4 # of Photographs 761 Gravel Average 0.08 Minimum SD Difference 0.03 Median SD Difference 0.07 Largest SD Difference 0.15

The data from the model made with photographs taken at 100 feet AGL was compared to the data from the model using the “three-pass” method. When the “three-pass” method was used it increased the accuracy slightly and produced a much more detailed 3D model.

Asphalt 0.06 0.02 0.06 0.09

Overall 0.06 0.02 0.06 0.15

Table 4: Site 4 average final data

Table 5 displays the overall averages for all sites. They were calculated using weighted averages for each surface. There may be slight differences because of rounding.

Average

All site averages Grass Gravel Asphalt Overall 0.107 0.06 0.05 0.06

Table 5: All site average data

After the initial processing and comparison of the data, sites one and two were reprocessed using only the photographs taken at 100 feet AGL. Site one used 54 photographs and 10 ground control points. Site two used 38 photographs and 12 ground control points. The line graphs from those tests were shown in Figure 25: Site 1-100AGL and Figure 26: Site 2-100AGL. You can see the average slope distance difference increased for both 3D models. This could be because the 3D model was not as clear with only the 10084 Collision Magazine - Volume 13 Issue 1

Volume 13 Issue 1.indd 84

C

onclusions Employing the methods described above, the use of small unmanned aircraft systems, in conjunction with modern photogrammetry software, is a cost-effective option for documenting crash scenes. However, the placement of ground control points is paramount to the development of an accurate 3D model. Through this testing, it was discovered ground control points shouldbe placed in a non-colinear fashion and widely dispersed throughout the scene. If increased accuracy is desired these six locations should be considered:

• • • • • •

Each end of the scene Near the center of the crash scene At significant elevation changes critical to the crash investigation Along significant curves Near areas of crucial evidence When evidence is in the grass, in furrows or large gouges

Using the data from site one it was discovered without a way to check the 3D model it could be inaccurate in some

www.collisionmagazine.com

5/28/2019 9:34:53 AM

Figure 25: Site 1-100AGL

Figure 26: Site 2-100AGL areas. To combat this, we placed “check points” in a few areas between ground control points. The location of these “check points” was recorded, but not used to build the 3D model like a ground control point. After the model was built, the location previously recorded was compared to the location of the “check point” in the 3D model. This method worked well for validating the 3D model.

have larger errors than demonstrated here unless the soil is visible and ground control points can be placed on the soil directly. The photography and modeling process used in this testing was in accordance with the training received by the Missouri State Highway Patrol Major Crash Investigation Unit. I have not used any other method or attempted to validate any other method of photogrammetry.

The data demonstrated the proximity to ground control points increased the accuracy of points within the 3D model. Of note regarding the testing was the problem posed by grass. The tests conducted involved grass, that ranged from approximately 0.2 feet to approximately three feet. Grass taller than approximately 0.2 feet was trimmed to the soil to simulate furrows. The surrounding grass was not trimmed or laid down. When processing a 3D model with tall grass it must be understood it will most likely

About the author: Master Sergeant Joseph Weadon, III is a 22-year veteran of the Missouri State Highway Patrol. He is currently assigned to the Major Crash Investigation Unit and is the Team Leader of Team #4.

Acknowledgements: Aerial Metrics, Stan Taylor and Iain Lopata. Sergeants Paul Meyers, and Glen Ward; and Trooper Dan Yingling.

www.collisionpumagazine.com

Volume 13 Issue 1.indd 85

Collision Magazine - Volume 13 Issue 1 85

5/28/2019 9:34:53 AM

eel h W The t a rn u T My

Driver’s Early Arrival at the Scene Caused Accident?

Erik Carlsson

T

his is a short story, with no pictures or illustrations, but I believe it is nevertheless offering very interesting reading for anyone who works with attorneys or insurance companies with the purpose of determining the cause of traffic accidents.

A car driver on his way to an assignment traveled westbound on Interstate 80 in New Jersey. He turned off the highway onto an exit ramp on his right that led up to a two-way north/south local street that passed over the highway. There was no traffic approaching from the left at the end of the ramp, so the driver made a right turn and continued north. About 200 yards north of the intersection is a restaurant on the eastern side of the street. Just as the driver passed the entrance driveway to the restaurant, the driver of a southbound car who was on his way to the restaurant started his left turn, and his car collided with the northbound car.

The accident reconstruction firm presented a detailed report of the accident and the damage to the vehicles, and posited that the plaintiff was the sole cause of the accident. Had he passed the restaurant’s driveway a few seconds later, the defendant’s car would have entered the driveway before the plaintiff’s car reached the scene!

What I said above may sound like bad joke, but it is the opinion presented by the accident reconstruction firm in its report. As expected, the writer of the report explained his reason for his opinion. The defendant was meeting a friend of his at the restaurant when the accident happened. The friend arrived a few minutes before the defendant and had stepped out of his car.

He was looking on the road, expecting to see his friend arriving, and saw the collision. But he had also been looking the other way and had seen the plaintiff’s car coming up the exit ramp from the highway. He saw, he testified, that the plaintiff had a red light at the end The left-turning driver stated at the police investiga- of the exit ramp, and he saw that the plaintiff made tion that followed that he didn’t see the oncoming his turn to the right without stopping. In its report, car because he was looking to his left, focusing on the the accident reconstruction firm considered that statedriveway that led into the restaurant’s parking lot. The ment by the defendant’s friend as evidence that the northbound driver stated that he was merely traveling plaintiff violated a New Jersey regulation that states down the straight road when the driver of the oncom- that a driver may not make a right turn on red without ing car suddenly started to turn left, causing the cars first stopping the vehicle. Thus, the expert opined, had to collide. the plaintiff not violated that regulation, but made a The northbound driver naturally was rather upset, and full stop before entering the street, he would have arsubsequently sued the turning driver, claiming inju- rived at the scene a few seconds later, and this would ries, lost wages, reduced value of his new car, etc. The have been enough for the defendant to complete his defendant, or rather his attorney, retained an accident left turn onto the restaurant’s driveway ahead of the reconstruction firm that I believe had been recom- approaching car. Therefore, the plaintiff was the sole cause of the accident. mended by the defendant’s car insurance company.

86 Collision Magazine - Volume 13 Issue 1

Volume 13 Issue 1.indd 86

www.collisionmagazine.com

5/28/2019 9:34:54 AM

Some thoughts: Not surprisingly, the case did not go to court, but the parties settled the case. One may wonder, though, would the attorney for the defendant really have been prepared to try the case in court? Would the defense expert have been prepared to tell a jury or a judge that the plaintiff caused the collision because he arrived at the scene a few seconds “too early”, while the defendant, who made a left turn on a two-way road without even looking forward, was not the one who caused the accident, but merely an innocent victim?

he began his right turn, which of course is what he could be expected to say.)

One can make more “what if ” questions. What if the plaintiff had arrived at the intersection a few seconds earlier and made a full stop and then started?

He may then have arrived at the scene of the accident at the very same moment as he now did? Or what if a northbound car had passed the intersection with the exit ramp just before the plaintiff made his right turn? That car may have arrived at the accident scene at the moment the defendant began his left turn and could have been the car that collided with the defendant’s car.

After all, it is very common that drivers make a right turn on red without stopping completely. Furthermore, could the defendant’s friend really have seen that the plaintiff had a red light some 200 yards Anyone who has some thought about this case is away considering that the traffic signal was at about welcome to contact me at erikcarlsson@live.com.

90 degrees angle from the witness’ position? (The plaintiff claimed that the light turned green just as

www.collisionpumagazine.com

Volume 13 Issue 1.indd 87

Collision Magazine - Volume 13 Issue 1 87

5/28/2019 9:34:55 AM

Motorcycle Accident Reconstruction:

Applicable Error Rates for Struck Vehicle EDR-Reported Delta-V Nathan Rose William Bortles Neal Carter

I

ntroduction A common motorcycle crash scenario occurs when a passenger vehicle equipped with an Event Data Recorder (EDR) turns left across the path of a motorcycle and is struck by the motorcycle.

The EDR data on the passenger vehicle will often be accessible with either the Bosch Crash Data Retrieval (CDR) system or the Global Information Technology (GIT) system. In these instances, pre-crash EDR data can be useful for establishing the specific characteristics of the left turn that preceded the collision. This data may include speed, throttle or accelerator pedal percentage, brake applications, and steering angles for the struck vehicle. In addition to that, an EDR-reported change in velocity (∆V) from the struck vehicle can potentially be used to infer the ∆V and impact speed of the motorcycle. This article reviews and summarizes the literature related to error rates for EDR-reported ∆Vs under various impact conditions and assesses which of these error rates are most applicable when analyzing impacts between motorcycles and passenger vehicles. This lays the groundwork for the companion article [Rose, 2019], which illustrates the application of these error rates within the context of reconstructing real-world intersection collisions involving motorcycles and EDR-equipped passenger vehicles. The companion article also covers the accuracy and application of the pre-crash data from the struck vehicle to these collisions. This article focuses only on the ∆V. There are potential problems that can arise when using the struck vehicle ∆V to infer the ∆V and impact speed of the motorcycle, particularly related to the large weight ratio that often exists between the motorcycle and the struck vehicle. Newton’s 2nd and 3rd laws together (conservation of momentum) dictate that, during a collision between two vehicles, the ratio of the mass of Vehicle #1 (m1) to the mass of Vehicle #2 (m2) is equal to the ratio of the change in velocity experienced by Vehicle #2 (∆V2) to the change in velocity experienced by Vehicle #1 (∆V1), as follows:

88 Collision Magazine - Volume 13 Issue 1

Volume 13 Issue 1.indd 88

www.collisionmagazine.com

5/28/2019 9:34:56 AM

www.collisionpumagazine.com

Volume 13 Issue 1.indd 89

Collision Magazine - Volume 13 Issue 1 89

5/28/2019 9:34:57 AM

This equation can also be written with the vehicle weights, instead of the masses. As an example, the companion article covers a case study in which the ratio of the weight of the struck vehicle (including the driver) to the weight of the motorcycle (including the rider) was approximately 7.15. The EDR-reported resultant ∆V of the struck vehicle was 12.1 mph. Applying Equation (1) yields a ∆V for the motorcycle of 86.5 mph, as follows:

However, there is some level of error that could be present in the EDR-reported ∆V. Given the weight ratio, every 1 mph of potential error in the ∆V would produce 7.15 mph of uncertainty in the calculated ∆V for the motorcycle. Assume for the sake of illustration that the potential error within the ∆V could be ±3 mph. If this were the case, then the potential range on the ∆V for the motorcycle would be 65.1 to 108.0 mph, a 42.9 mph spread. This is such a large range, that it may have little practical value. On the other hand, if the potential error within the ∆V were ±1 mph, then the potential range on the ∆V for the motorcycle would be 78.6 to 92.9 mph, a 14.3 mph spread and a range more likely to be useful for assessing the cause of the crash. Of course, the error could also be asymmetric, and perhaps we could say that the EDR-reported ∆V is likely to be an under-estimate within some range. If this were the case, then the EDR-reported ∆V could be used as-is to obtain a conservative (low) estimate of the motorcycle’s ∆V and impact speed. At any rate, these calculations illustrate the sensitivity in conservation of momentum analysis caused by the weight ratio of the motorcycle to the passenger vehicle and emphasizes the need to understand what the potential error in an EDR-reported ∆V is likely to be in any given case.

The Accuracy of the Struck Vehicle EDR-Reported ∆V For the scenario where an EDR-equipped passenger vehicle turns left across the path of a motorcycle, the motor-

cycle is likely to strike the passenger vehicle on the side, but to cause the vehicle to experience both a longitudinal and a lateral ∆V. The component of the collision force that causes the lateral ∆V could be oriented from left-to-right or right-to-left, depending on the intersection geometry and the specific maneuver the passenger vehicle was in the midst of before being struck. The component of the collision force that causes the longitudinal ∆V, on the other hand, will typically be oriented front-to-back on the vehicle and produce a negative direction ∆V. The specifics of the EDR-reported ∆V will vary from system-to-system.

Some vehicles will report only the longitudinal ∆V experienced by the struck vehicle. Others will report both the longitudinal and lateral ∆Vs. The following error sources can affect the accuracy of the EDR-reported speed change (∆V): Error Source #1: During an impact, there can be a ∆V that occurs prior to the accelerations exceeding the event-triggering threshold (algorithm enable or AE). This is an error source that contributes to underreporting of the ∆V.1 Prior studies have indicated that the triggering threshold for most passenger vehicle EDRs is in the 1 to 2g range.

Error Source #2: The recording window for the ∆V may be too short to capture the full ∆V. This error source contributes to under-reporting of the ∆V. This error source can be ruled out if the EDR-reported ∆V reaches a maximum and begins to decrease prior to the end of the recording window.

Error Source #3: The recording window for the ∆V may be too long and a ∆V experienced by the vehicle due to post-impact tire and dragging forces could be recorded. This error source could contribute to overreporting of the ∆V. This error source can be ruled out if the EDR-reported ∆V reaches a maximum prior to the end of the recording window and then decreases for the remainder of the recording window. This error source would be recognizable if the EDR-reported ∆V reached a local maximum, then began to decrease, but eventually began to increase again prior to the end of the recording window.

Error source #1 can be exacerbated for frontal impacts by the fact that many EDRs reportedly have a built-in positive offset in the accelerometers that can contribute to under-reporting for frontal impacts and over-reporting for rear impacts. Ruth notes that this offset can vary from vehicle to vehicle and EDR generation to generation [Ruth, SAE Course #1210, November 2018]. See also, Wilkinson [2013] and Ruth [2016]. Xing et al [2016] note that this acceleration bias is not a complete explanation for discrepancies between the actual and EDR-reported ∆Vs, observing that “a previous study of Toyota ACMs proposed a constant acceleration bias model to explain the observed differences between reference speed change and ACM recorded speed change. However, our full regression models suggest these differences are not constant across the range of speed changes we tested, but rather that they also depend on the peak acceleration and pulse duration.”

1

90 Collision Magazine - Volume 13 Issue 1

Volume 13 Issue 1.indd 90

www.collisionmagazine.com

5/28/2019 9:34:57 AM

Error Source #4: The peak accelerations during a collision can exceed the capabilities of the accelerometer in the airbag control module (ACM) where the EDR resides, in which case the system will not capture the peak accelerations. This is referred to as clipping. This error source contributes to under-reporting of the ∆V. This error source will be recognizable through a flatline portion of either the EDR-reported ∆V curve or the EDR-reported acceleration curve.

Error Source #5: The ACM may reside some distance from the center of gravity (CG) of the vehicle. This is relevant because accident reconstruction calculations often calculate and utilize the CG ∆V. In instances where the collision induces significant rotation, the ∆V of the struck vehicle – which is measured at the ACM location – may need to be adjusted to accurately reflect the ∆V at the CG of the struck vehicle. Bundorf [1996], Marine and Werner [1998], Rose [2007], and Haight [2013a] describe methods for making this adjustment.

Error Source #6: Physical damage to or displacement of the ACM can affect the accuracy of the reported ∆V. A vehicle involved in a collision can be split into two regions – a deforming or crushing region and a non-deforming region [Emori, 1968]. The occupants would ideally be contained within the non-deforming region of the vehicle and it is the ∆V of this region of the vehicle that the accident reconstructionist would be calculating in their analysis. Varat and Husher [2000] have observed that “care must be exercised in the analysis of accelerometer data to ensure that the instrumentation output accurately represents the vehicle under study. Varying locations within the vehicle may have unique kinematic time histories. A significant factor for rigidly mounted accelerometers is whether the instrument is mounted in or out of the crush zone.” When a collision is severe enough that the ACM ends up being within the crushing region of the vehicle, the accelerations and ∆V reported by this module will not be representative of the accelerations and ∆V experienced by the non-deforming region of the vehicle. This error source can contribute to significant under-reporting or over-reporting of the ∆V. Under-reporting is an indication that deformation caused the ACM to become pitched relative to the longitudinal axis of the vehicle. Over-reporting is an indication

that, due to the crush, the ACM experienced accelerations in excess of what the CG experienced.

Error Source #7: The EDR could lose power before the collision is complete and some of the data may not get recorded. When this occurs, the system would under-report the ∆V. Typically, the EDR report will indicate whether recording of a reported event was complete. If the report states that the recording was complete, then this error source can be ruled out.2 None of these are measurement errors, per se. The first is inherent to any event triggered system/algorithm. Error sources two through four are hardware limitations and their effects will often be detectable through analysis of the EDR report. A reconstructionist will sometimes be able to correct for them. The fifth error source is simply something that the reconstructionist will need to account for in cases where there would be a significant discrepancy between the ∆V experienced at the ACM and that experienced at the CG. This will typically be instances when there is significant post-impact rotation. The sixth and seventh error sources amount to physical damage occurring to the measuring device itself or the power source. Errors in the reported ∆V in instances where there is actual damage to the ACM are not surprising and the reported ∆V may not be useable in those instances. For reconstructing motorcycle collisions into EDR-equipped vehicles, it will be important to determine the extent to which these 7 error sources are detectable and can be corrected for since, given the large weight discrepancy typically present between the motorcycle and the struck vehicle, calculations of the motorcycle’s change in velocity will be sensitive to any errors in the struck vehicle ∆V. The reconstructionist can view this list as a checklist of error sources that would ideally be ruled out or accounted for through analysis. If these error sources can be ruled out through physical inspection of the struck vehicle and through analysis of the EDR report, then the potential error in the EDR-reported ∆V can be limited just to measurement error.

Some of these error sources are unlikely in collisions between motorcycles and passenger vehicles. For example, because of the typically large weight ratio between the vehicles in most motorcycle versus passenger vehicle collisions, the ∆V experienced by the passenger vehicle will often be less than 10 mph. At this level of severity, the ACM on the passenger vehicle is unlikely to be physically damaged or displaced (error source #6). Similarly, the ac-

Other error sources could be added to this list. During the questions after my presentation of this material at the 2019 EDR Summit, Rick Ruth raised the issue of tire forces present during a collision that could potentially add to underestimation of the EDR-reported ∆V. We will address this and other potential error sources in a future article.

2

www.collisionpumagazine.com

Volume 13 Issue 1.indd 91

Collision Magazine - Volume 13 Issue 1 91

5/28/2019 9:34:57 AM

celerations experienced by the struck vehicle are unlikely to exceed the full-scale value of the accelerometer (error source #4). Error source #3 seems generally rare. On the other hand, in some motorcycle-to-passenger vehicle collisions, the passenger vehicle can experience significant rotation, which can cause the ACM to experience a different ∆V than the CG. This can cause errors (error source #5) if it is not accounted for in the analysis.

Error source #2 could be an issue for motorcycle collisions, but it will become less of an issue with time now that the Code of Federal Regulations (49 CFR 563) requires passenger vehicle EDRs to report cumulative ∆V over a duration of 250 milliseconds and to monitor and report the maximum ∆V over a period of 300 milliseconds. Part 563 also contains criteria under which the recording can be terminated short of these times, but these criteria are designed to ensure full capture of the collision. This requirement has been in place since September 1, 2012. Many EDRs manufactured prior to this requirement report the ∆V over a significantly shorter interval than this and this error source is most relevant to those older EDRs. Error source #1 is likely to be present and a contributor to under-reporting for many EDRs, and this error source is likely to produce a higher percentage error in the EDRreported ∆V for motorcycle collisions where the struck vehicle ∆V is 10 mph or less. The acceleration level necessary to trigger an event will vary from EDR-to-EDR, but for the sake of illustration, consider an impact between a motorcycle and a passenger car that produces a 10 mph ∆V for the car. Producing this level of ∆V for the struck car will require a significant impact speed on the part of a motorcycle (perhaps around 70 mph). For the sake of this illustration, assume that the collision produces a haversine crash pulse, which can be described with the following equation [Varat and Husher, 2000]:

In this equation, a is the acceleration at any instant during the collision, P is the peak acceleration, ∆t is the impact duration, and t is the time at any instant during the impact. The acceleration pulse defined with this equation can be integrated to calculate the cumulative ∆V over the duration of the pulse. For a given impact duration, a given ∆V will have a particular peak acceleration associated with it. Figure 1 is a graph showing 4 different haversine collision pulses of varying durations – 75, 100, 125, and 150 milliseconds. Each of these pulses produces a ∆V of 10 mph, and so they each have a different peak acceleration – 12.15, 9.11, 7.29, and 6.07 g, respectively. Figure 2 shows

92 Collision Magazine - Volume 13 Issue 1

Volume 13 Issue 1.indd 92

the cumulative ∆V curves that result from each of these pulses.

For the purpose of this illustration, consider an EDR that has a triggering threshold of 2 g. For the 75-millisecond pulse, this level of acceleration is achieved 10 milliseconds into the collision at a time when the cumulative ∆V is 0.15 mph. Thus, for a 10-mph ∆V with this pulse, error source #1 would only account for a 1.5% under-reporting of the ∆V. For the 100-millisecond pulse, a 2 g acceleration is reached approximately 16 milliseconds into the collision at a time when the cumulative ∆V is 0.26 mph (2.6%).

For the 125-millisecond pulse, a 2 g acceleration is reached approximately 22 milliseconds into the collision at a time when the cumulative ∆V is 0.34 mph (3.4%). For the 150-millisecond pulse, a 2 g acceleration is reached approximately 30 milliseconds into the collision at a time when the cumulative ∆V has reached 0.49 mph (4.9%).

This error source can become a larger percentage of the EDR-reported ∆V at lower severity levels. Consider for instance, a collision where the struck vehicle experiences a ∆V of 5 mph. With haversine collision pulses of 75, 100, 125, and 150 milliseconds that produce a 5 mph ∆V, the peak accelerations would be 6.07, 4.55, 3.64, and 3.04 g, respectively. Again, consider an event-triggering threshold of 2 g. For the 75-millisecond pulse, this level of acceleration is achieved approximately 15 milliseconds into the collision at a time when the cumulative ∆V is 0.24 mph (4.8%). For the 100-millisecond pulse, a 2 g acceleration is reached approximately 24 milliseconds into the collision at a time when the cumulative ∆V is 0.4 mph (8.0%). For the 125-millisecond pulse, a 2 g acceleration is reached approximately 34 milliseconds into the collision at a time when the cumulative ∆V is 0.57 mph (11.4%). For the 150-millisecond pulse, a 2 g acceleration is reached approximately 45 milliseconds into the collision at a time when the cumulative ∆V has reached 0.74 mph (14.8%).

These two examples illustrate that the significance of error source #1 depends on the magnitude of the ∆V and on the impact duration. Assuming that the recording window of the EDR is adequate to capture the entire pulse, the impact duration is potentially something that can be quantified from the EDR data. Sometimes the peak acceleration will be reported as well. In such instances, the reconstructionist can use these to quantify the magnitude of this error source for the specific case being analyzed.

Full-Overlap Front and Rear Collisions Now consider some of the literature related to the error rates of EDR-reported ∆Vs. Figure 3 plots the error in the EDR-reported ∆V for a number of instances reported in

www.collisionmagazine.com

5/28/2019 9:34:57 AM

Figure 1: Haversine Crash Pulses that Produce a 10 mph ∆V

Figure 2: Cumulative ∆V Curves from the Corresponding Haversine Crash Pulses the literature.3 The error was quantified in the cited studies by comparing to the ∆V calculated from on-board, laboratory grade accelerometers with which the vehicles were instrumented. As long as the laboratory accelerometers are outside of the crushing region of the vehicle and the col-

lision does not induce significant rotation, the accelerometers would not be subject to any of the 7 error sources listed for the EDR-reported ∆V. The data in Figure 3 is for full-overlap, front or rear impacts where the vehicles experienced insignificant yaw rotation following the col-

www.collisionpumagazine.com

Volume 13 Issue 1.indd 93

Collision Magazine - Volume 13 Issue 1 93

5/28/2019 9:34:57 AM

lision. The ∆V calculated from the accelerometers on the test vehicle is plotted on the horizontal axis and the error in the EDR-reported ∆V is plotted on the vertical axis. A negative ∆V on the horizontal axis is a frontal impact and a positive ∆V is a rear impact. The errors reported on the vertical axis are calculated from the ∆V magnitudes, such that a negative error is always an under-reporting of the magnitude of the ∆V and a positive error is always an overreporting of the magnitude of the ∆V.4 The dashed black lines represent a window of ±10% error, the required accuracy per 49 CFR part 563.85 and the rule of thumb first found in the literature based on a study by Chidester’s examination of early General Motors Sensing and Diagnostic Modules [1999]. A significant number of the data points in Figure 3 are from the NHTSA’s New Car Assessment Program (NCAP) 35-mph frontal barrier crash tests, which yield ∆Vs around 40 mph. Figure 3 appears to show increasing absolute error in the EDR-reported ∆Vs with increasing ∆V, consistent with the underlying implication of the 10% error rule. Much of the data does lie with the 10% error band, but there are also a number of points that fall outside of this window. In particular, for the frontal collisions with ∆Vs around 40 mph there are a number of points with considerably higher error than 10%. Ideally, we would be able to identify the sources of these high errors and rule these sources out for motorcycle collisions into passenger vehicles. On the other end of the spectrum, for frontal collisions with ∆Vs lower than 10 mph, almost all of the points lie outside of the 10% error window. For these lower speed collisions, it may be more useful to think in terms of absolute error, rather than percentage error.

the three crashes where ACM mounts broke, the accelerations tracked well early in the [impact].”

The 20-mph under-reporting was from the EDR in a test of a 2007 Lexus ES-350. This vehicle impacted the barrier at a speed of approximately 50 mph, producing a ∆V of approximately 54 mph. During this test, the mounting flanges of the electronic control module were fractured due to deformation of the floor pan underneath the module.

This led to inaccurate acceleration measurements within the ACM. In addition to that, other instrumentation on the test vehicle measured accelerations exceeding 50 g.

Since the accelerometer in the ACM could only measure accelerations up to 50 g, the accelerations recorded by the module were clipped. Thus, the ∆V was under-reported due to the module being in the deforming region and reoriented off the longitudinal axis of the vehicle and from peak acceleration clipping at the hardware level (error sources #4 and #6).

Another one of the tests, with a ∆V just over 40 mph, resulted in an EDR-reported ∆V that was 5.8 mph low (14.3%). This was a test of a 2003 Toyota Camry. Exponent reported that “the rearmost of the three flanges used to bolt the airbag control module to the vehicle fractured…this was the result of deformation to the vehicle floor pan underneath the ACM and occurred after the vehicle had experienced significant crush and after the airbags had deployed.” In addition, “the accelerometers mounted near the vehicle CG also recorded accelerations in excess of 50 g’s for more than 10 ms, so it is probable that the accelerations recorded by the ACM’s accelerometers were truncated.” This data point is in the same severity range as many of the other tests in Figure 3 that exhibited Exponent [2011]: The largest under-reported ∆V (-20.0 significant under or over-reporting of the ∆V. The other mph) in Figure 3 was published by Exponent [2011]. This five of the seven full-scale crash tests reported by Exponent data point is represented in the lower left corner of Figure had EDR-reported ∆Vs within 7% of those obtained from 3. For their study, Exponent generated and evaluated 231 other instrumentation.

EDR records in 24 Toyota and Lexus vehicles. Seven of the EDR records were generated in full-scale frontal crash Tsoi [2013]: The next largest under-reported longitudinal tests of sufficient severity to cause airbag deployment. The ∆V (-11.7 mph) in Figure 3 was published by Tsoi, Hinch, remaining 224 EDR records were non-deployment events Ruth, and Gabler [2013]. These authors evaluated the acgenerated by tapping the ACM from the rear. Three of the curacy of 41 EDRs extracted from 2012 General Motors, full-scale crash tests were severe enough to damage the at- Ford, Honda, Mazda, Toyota, and Volvo vehicles that had tachment points of the ACM. The authors reported that been tested in the NCAP 35 mph (56 km/h) full-overlap the “EDRs accurately recorded accelerations within the +/- frontal collisions into a rigid, non-moving barrier. Tsoi 50g limit of performance for which they were designed. In noted that these tests “are very severe, demonstrated by the fact that they are representative of the 99th percentile of all frontal crashes.” They reported that “the average absoBortles et al presented a prior version of Figure 3. This present article offers an updated version of this figure.

This is an intuitive, but inconsistent way of reporting the errors. This sign convention was a feature of the original version of Figure 3, and we have chosen to retain that feature here.

5 https://www.law.cornell.edu/cfr/text/49/563.8.

3 4

94 Collision Magazine - Volume 13 Issue 1

Volume 13 Issue 1.indd 94

www.collisionmagazine.com

5/28/2019 9:34:57 AM

Figure 3 : Error in the EDR-Reported ∆V for Full-Overlap, Frontal Impacts lute error was 4.20 kph (6.6%) for final longitudinal ∆V and 4.32 kph (6.6%) for maximum longitudinal ∆V. Our results show that EDRs underreport the reference instrumentation ∆V in the vast majority of cases.” Unfortunately, Tsoi et al report several instances with EDR-reported ∆V errors significantly greater than 6.6%.

Consider each of the 7 potential error sources for this test. Error Source #1 – Xing et al [2016] reported that the triggering acceleration threshold for this generation of Toyota EDR was approximately 2 g. Based on the data from the on-board laboratory accelerometers, this acceleration threshold would have been met approximately 4 milliseconds into the collision, at a time when the vehicle had The specific test reported by Tsoi et al with the most sig- experienced only a 0.03 mph ∆V. Thus, the ACM missing nificant under-reported ∆V (-11.7 mph) was an NCAP this ∆V would be an insignificant contributor to the sigtest of a 2012 Toyota Sienna (NHTSA Test #7615) with nificant overall under-reporting of the ∆V. Ruth indicates an impact speed of 35.0 mph (56.3 km/h). We examined that Toyota Generation 2 EDRs have a +0.39 g to +0.87 g the test and CDR reports for this test, along with four ac- [SAE Course #1210 and 2016-01-1496] offset in the acceleration channels from the test instrumentation, all of celerometers. He calculated that over 150 milliseconds the which measured the longitudinal accelerations from the 0.39 g will contribute approximately 1.3 mph to underarea of the rear seats (Channels 93, 94, 99, and 100), a reporting of the ∆V. The +0.87 g offset would contribute position likely to be outside the crushing region of the ve- about double this. Thus, while this error source could be a hicle. These signals were filtered, averaged, and then inte- factor in this test, it does not explain the substantial undergrated to obtain the ∆V. The CDR report indicated that reporting. Error Source #2 – The recording window for the ACM on this vehicle was an 06EDR (Generation 2) the ∆V was 200 milliseconds and the maximum reported and the report listed a maximum ∆V of 28.8 mph. The ∆V occurred at approximately 110 milliseconds. The deactual maximum ∆V was 40.5 mph, so the EDR under- celerations obtained from the other on-board acceleromreported the ∆V by 28.9%. www.collisionpumagazine.com

Volume 13 Issue 1.indd 95

Collision Magazine - Volume 13 Issue 1 95

5/28/2019 9:34:57 AM

eters show that the collision pulse had essentially ended by 90 milliseconds. Therefore, the recording window was long enough to capture the full ∆V. Error Source #3 – In the CDR report, the reported cumulative longitudinal ∆V decreased from the time it reached its maximum to the end of the recording window; therefore, external forces other than the collision forces do not appear to be increasing the reported ∆V, and so, the recording window was not too long. Error Source #4 – The CDR report did not contain any graphical or tabular reporting of the underlying accelerations experienced by the ACM. There was no evidence of acceleration clipping from examination of the CDR report and the accelerations reported by the on-board accelerometers did not exceed 50 g. Error Source #5 – Review of the test video did not reveal any significant yaw rotation of the vehicle during or following the collision. Error Source #7 – The CDR report indicated that recording of the event was complete. This leaves error source #6 – deformation of or around the ACM. Error Source #6 – Figure 4 shows the cumulative ∆V for this test, both as reported by the EDR and as calculated from the other on-board accelerometers. This graph shows the timeframe up to 125 milliseconds, so it does not include all of the EDR data. It does include the full crash pulse, though. The EDR data has been time shifted by 9 milliseconds to visually align the initial portions of these curves. The two curves track reasonably well until approximately 60 milliseconds. At that point, the curves diverge.

The EDR curve levels off while the cumulative ∆V from the on-board accelerometers continues to increase significantly above the other curve. This pattern is consistent with a change in alignment of the ACM relative to the vehicle due to localized deformation around the module such that the longitudinal accelerometer in the ACM stops capturing a significant portion of the collision accelerations. The ACM was not photographically documented either before or after this test, so this interpretation cannot be directly confirmed. However, the ACM on this vehicle was located underneath the center stack, and deformation to the ACM’s mounting location can be inferred from deformation to the center stack (the area behind/under the radio).

Examination of the photographs in Figure 5 reveal deformation of the vehicle floor in this area underneath the center stack. The photograph on the left is a pre-test photograph of the passenger’s side footwell and the photograph on the right is the same area post-test. In the post-test photograph, it is apparent that the space in the footwell has been significantly reduced due to the impact. The panels covering the lower portion of the center stack are displaced and deformed and their mounting clips have come out. The deformation evident in this area of the vehicle 96 Collision Magazine - Volume 13 Issue 1

Volume 13 Issue 1.indd 96

is confirming of the interpretation that the ACM was reoriented due to deformation during the collision. Thus, the inaccuracy in the EDR-reported ∆V can be attributed to error source #6. This is a detectable error source that a reconstructionist would be able to document for a vehicle involved in a real-world collision.

The magnitude of the error in the EDR-reported ∆V exceeded 9 mph in two additional tests examined by Tsoi et al. First, the CDR report for NHTSA Test #7622 – a 35 mph NCAP test of a 2012 Honda CR-Z – reported a maximum longitudinal ∆V of 30 mph. The actual maximum ∆V was 41.3 mph. Thus, the EDR under-reported the ∆V by 11.3 mph (27.4%). We examined the test and CDR reports for this test, along with four acceleration channels from the test instrumentation, all of which measured the longitudinal accelerations from the area of the rear seat (Channels 93, 94, 99, and 100). These signals were filtered with a Butterworth filter with a cutoff frequency of 400 Hz and then averaged. The resulting acceleration pulse is plotted in Figure 6. In this figure, time is plotted on the horizontal axis and deceleration on the vertical axis. Deceleration has been plotted with a positive sign and acceleration with a negative sign. This figure also contains the decelerations reported in the CDR report.

Several observations can be made from examination of Figure 6. First, the CDR report listed a peak deceleration of 49 g. The on-board accelerometers measured peak decelerations exceeding 50 g during portions of the impact. Even though the ACM accelerometer did not capture these higher accelerations, it is unlikely that this had a significant influence on the EDR ∆V calculation. In conducting our analysis, we also experimented with filtering the accelerations from the on-board accelerometers with a cutoff frequency of 60-Hz. This brought the peak deceleration down to 49 g and made little difference in the ∆V calculation. Second, the maximum recording window for this EDR was 250 milliseconds. The data limitations in the CDR report indicate that the recording of ∆V will terminate prior to 250 milliseconds if the change in longitudinal or lateral velocity equals or falls below 0.8 km/h (0.5 mph) over a 20-millisecond timeframe. During this crash test, the longitudinal ∆V leveled off around 100 milliseconds and the EDR stopped reporting the ∆V at 130 milliseconds. This makes sense with the decelerations reported by the on-board accelerometers (Figure 6), and so, it can be concluded that the under-reporting of the ∆V in this instance was not due to a recording window that was too short.

Finally, consider the accelerations reported by the EDR. The EDR-reported accelerations were aligned to the de-

www.collisionmagazine.com

5/28/2019 9:34:57 AM

Figure 4: Comparison between Cumulative ∆V from On-Board Instrumentation and from CDR Report

Figure 5: Comparison between Passenger’ Side Footwell Pre and Post Test (Test #7615)

Figure 6: Comparison between Deceleration from On-Board Instrumentation and from CDR Report www.collisionpumagazine.com

Volume 13 Issue 1.indd 97

Collision Magazine - Volume 13 Issue 1 97

5/28/2019 9:34:58 AM

celerations from the other on-board instrumentation by plotting the cumulative ∆V calculated from the on-board instrumentation with the cumulative EDR-reported ∆V (see Figure 7 for this alignment). This resulted in a 6-millisecond shift of the EDR-reported accelerations. Overall, the area under the EDR deceleration curve is smaller than the area under the curve generated with the on-board accelerometers. This is consistent with the EDR-reported ∆V under-estimating the actual ∆V. In addition, the decelerations reported by the ACM begin dropping significantly in magnitude at approximately 56 milliseconds. The decelerations reported by the on-board accelerometers do not begin dropping significantly until perhaps 15 milliseconds later. This indicates that at this point in the collision, the ACM has started to measure different decelerations than the accelerometers further back on the vehicle. This has the look of localized deformation around the ACM that caused the ACM to become misaligned relative to the vehicle.

This interpretation is confirmed by examination of Figure 7, which shows the cumulative ∆V through the collision, both as reported by the EDR and as calculated from the other on-board accelerometers. The two curves track very closely until approximately 55 milliseconds into the collision. At that point, the curves begin diverging. The EDR curve levels off while the cumulative ∆V from the onboard accelerometers continues to increase, significantly above the other curve, a pattern similar to that observed for the Toyota Sienna test.

The ACM was not photographically documented either before or after this test, so this interpretation cannot be confirmed directly. However, the dynamic deformation that occurred during this test was of great enough magnitude that it could have deformed the area where the ACM was mounted (underneath the center stack). Figure 8 is one frame from the test video that shows the overall dynamic deformation during this test.

Deformation of the center stack can be confirmed indirectly by examining the photographs included in Figure 9 and Figure 10. The photographs in Figure 9 compare the driver’s side footwell pre and post-test. In the post-test photograph, some floor deformation is evident beneath the center stack. The panels covering the lower portion of the center stack are displaced and their mounting clips have

Figure 8: The Instant of Maximum Deformation for NHTSA Test #7622 (2012 Honda CR-Z)

Figure 7: Comparison between Cumulative ∆V from On-Board Instrumentation and from CDR Report 98 Collision Magazine - Volume 13 Issue 1

Volume 13 Issue 1.indd 98

www.collisionmagazine.com

5/28/2019 9:34:58 AM

come out. The photographs in Figure 10 compare the passenger’s side footwell pre and post-test. On the passenger side, there appears to be more significant floor deformation than what was present on the driver’s side, and again, the panels covering the lower portion of the center stack are displaced and one of the mounting clips has come out.

These observations are confirming of the interpretation that the ACM was reoriented due to deformation during the collision. Thus, the inaccuracy in the EDR-reported ∆V can be attributed to error source #6.

As an additional example, the CDR report for NHTSA Test #7566 – a 35 mph NCAP test of a 2012 Mazda 6 – indicated a maximum longitudinal ∆V of 31.7 mph (51 km/h) occurring at 125 milliseconds. The actual maximum ∆V was 41.1 mph (66.2 km/h). Thus, the EDR under-reported the ∆V by 9.4 mph (15.2 km/h). As with the previous tests, we examined the test and CDR reports for this test, along with four acceleration channels from the test instrumentation, all of which measured the longitudinal accelerations from the area of the rear seats (Channels 93, 94, 99, and 100). These signals were filtered and then averaged. The resulting acceleration pulse is plotted in Fig-

ure 11. This figure also contains the decelerations reported in the CDR report. As with the previous tests, the accelerations from the ACM and the on-board accelerometers were aligned by visually aligning the cumulative ∆V curves (see Figure 12).

In this case, the accelerations measured by the ACM do appear to reach the upper limit of the accelerometer and to get clipped at 50 g. However, the decelerations from the laboratory accelerometers do not ever reach 50 g. This is an indication that the ACM is in the deforming region of the vehicle and experiencing higher accelerations than what the non-deforming region of the vehicle experienced.

Also, similar to the previous 2 tests, the decelerations begin dropping back to zero approximately 10 milliseconds too early, indicating likely misalignment of the ACM relative to the longitudinal axis of the vehicle due to deformation.

There are other indications in the Tsoi data that the larger reported errors are due to deformation around the area where the ACMs are mounted. For example, Tsoi et al presented the data in Table 1, which lists the average absolute error for EDR-reported ∆Vs, segregated by manufacturer.

Figure 9: Comparison between Driver’s Side Footwell Pre and Post Test (Test #7622)

Figure 10: Comparison between Passenger’ Side Footwell Pre and Post Test (Test #7622) www.collisionpumagazine.com

Volume 13 Issue 1.indd 99

Collision Magazine - Volume 13 Issue 1 99

5/28/2019 9:34:59 AM

Figure 11: Comparison between Deceleration from On-Board Instrumentation and from CDR Report

Figure 12: Comparison between Cumulative ∆V from On-Board Instrumentation and from CDR Report This table shows significant differences from manufacturer to manufacturer. As an illustration, consider the difference between the average error for the 6 tests with Fords (1.88%) and that for the 5 tests with Hondas (10.47%).

Table 2 compares these tests. The ACMs for the Fords are all located on the center tunnel, whereas the ACMs for the Hondas are further forward under the center stacks. The vehicles in these tests experience comparable length reductions due to the damage, so the further forward the ACM is, the more likely it is to be involved in the deforming region of the vehicle.

100 Collision Magazine - Volume 13 Issue 1

Volume 13 Issue 1.indd 100

Ruth and Tsoi [2014]: The largest over-reported ∆V (+5.5 mph) in Figure 1 was published by Ruth and Tsoi [2014] and was from a 2012 Hyundai Accent (NHTSA Test #7504). The tests reported by Ruth consisted of Kia and Hyundai vehicles from model years (MY) 2010 to 2012.

This was prior to Kia and Hyundai’s full compliance with CFR Part 563, and Ruth observed clearly anomalous data in some of the tests he examined for these model years.

That said, for these full-overlap, frontal impacts, 19 out of 20 of the EDR-reported ∆Vs were within 10% of the ∆V determined from the other instrumentation on the vehicle (≅ ±4 mph).

www.collisionmagazine.com

5/28/2019 9:34:59 AM

7475 7478 7495 7623 7624

Ford Mustang Ford Focus Ford Explorer Ford F250 Super Crew Ford Expedition

Vehicle Weight (lb) 3915 3373 5329 7602 6450

7579 7619 7622 7627 7732

Honda Fit Honda Civic IMA Honda CR-Z Honda Civic Honda CR-V

2888 3240 3135 3093 3863

Test No. Make/Model

7625

Ford F150 Supercab

6325

Pre-Test Ve- Max hicle Length Crush (in) (in) 187.8 20.2 178.3 18.7 197.2 25.2 263.4 26.1 205.9 20.4 250.2

161.7 177.4 160.5 175.0 178.2

23.0

20.0 21.9 11.3 17.7 22.9

Percent ACM Length Location Reduction 10.8% Center Console 10.5% Center Console 12.8% Center Console 9.9% Center Console 9.9% Center Console 9.2%

12.3% 12.3% 7.0% 10.1% 12.9%

Center Console

Under Dashboard, Center Under Dashboard, Center Under Dashboard, Center Under Dashboard, Center Under Dashboard, Center

Table 2: Comparison Between the Fords and the Hondas in the Tsoi [2013] Dataset Average Number Absolute of Tests Error Chrysler 12 2.93% Ford 6 1.88% GM 7 6.42% Honda 5 10.47% Mazda 1 22.95% Toyota 9 10.79% Volvo 1 3.99%

Vehicle Make

Table 1: Accuracy of EDR-reported ∆V by Vehicle Make

Figure 13: Comparison between Cumulative ∆V from OnBoard Instrumentation and from EDR Report

NHTSA Test #7504 was an NCAP test of a 2012 Hyundai Accent with an impact speed of 34.9 mph (56.2 km/h).

We examined the test and EDR reports for this test, along with four acceleration channels from the test instrumentation, all of which measured the longitudinal accelerations from the area of the rear seats (Channels 93, 94, 99, and 100). These signals were filtered, averaged, and then integrated to obtain the ∆V. Figure 13 shows the cumulative ∆V for this test, both as reported by the EDR and as calculated from the other on-board accelerometers.

The two curves track reasonably well until approximately 55 milliseconds. At that point, the curves diverge. The

EDR curve levels off while the cumulative ∆V from the on-board accelerometers continues to increase significantly above the other curve. This pattern is consistent with the ACM experiencing higher accelerations than the laboratory accelerometers due to localized deformation around the module. The ACM was not documented photographically either before or after this test, so this interpretation cannot be directly confirmed. However, the ACM on this vehicle was located underneath the center stack and deformation to the floor under the center stack was evident in the posttest photographs from this test.

www.collisionpumagazine.com

Volume 13 Issue 1.indd 101

Collision Magazine - Volume 13 Issue 1 101

5/28/2019 9:34:59 AM

Low Severity Tests: The previous discussion has confirmed, at least provisionally, that the higher errors for the higher severity tests in Figure 3 are likely related to the severity of these collisions and the proximity of the ACM to the crushing region of the vehicle. The more severe the collision and the closer the ACM is to the collision, the greater the likelihood that the ACM or surrounding structure will be damaged or deformed. If such deformation occurs, increased error in the EDR-reported ∆V will be the result. The highest errors occurred around a ∆V of 40 mph. Using the 7.15 weight ratio that was used in the example in the introduction, a motorcycle would have to be traveling about 286 mph to produce a 40 mph ∆V for the struck vehicle. This is obviously improbable, and so the high error rates exhibited by some of the higher severity collisions included in Figure 3 are unlikely to be applicable to motorcycle collisions with passenger vehicles. That said, high-speed motorcycle collisions can sometimes produce deformation patterns similar to pole (or other narrow object) impacts, and it is possible that the ACM could be damaged or displaced during a motorcycle collision, at least in rare instances. Assuming that the reconstructionist can rule this out, though, the error rates reported in studies of EDR accuracy in low-severity collisions are more likely to be applicable to what the struck vehicle experiences during a motorcycle collision.

Correia, Iliadis, and McCarron [2001] reported 12 lowseverity rear end vehicle-to-vehicle staged collisions using two different GM bullet vehicles – a 2000 Chevrolet Malibu and a 1997 Chevrolet Cavalier. These vehicles struck one of two target vehicles – a 1984 VW Rabbit or a 1994 Honda Accord. The SDM on the Malibu did not report ∆V, thus no comparison between EDR-reported ∆V and the actual ∆V could be made for the 5 Malibu tests. The SDM on the Cavalier did report ∆V data. Of the 7 tests run with this vehicle, 4 did not cause an event to be recorded. Of the remaining 3 tests where a non-deployment event was recorded, the EDR-reported ∆Vs were between 1.3 and 2.2 km/h low (0.8 to 1.4 mph). The authors attributed the under-reporting to a recording window that was too short to capture the full collisions. These three data points are included in Figure 3.

Lawrence, Wilkinson, King, and Heinrichs [2002] examined the accuracy of EDR-reported ∆Vs from the sensing and diagnostic modules (SDMs) on MY 1996-1999 General Motors vehicles for low-severity collisions. Two SDMequipped vehicles were subjected to 260 staged frontal collisions with speed changes below 11 km/h (6.8 mph).

Of the 260 collisions, 105 of them were vehicle-to-barrier tests and 155 of them were vehicle-to-vehicles tests, with the front bumper of the SDM-equipped vehicle striking 102 Collision Magazine - Volume 13 Issue 1

Volume 13 Issue 1.indd 102

the rear bumper of a 1984 Volvo GL. Airbags did not deploy in any of the tests. The authors reported that “in all of the vehicle tests, the speed change reported by the SDM underestimated the actual speed change of the vehicle…

The difference between the SDM-reported speed change and the actual speed change was as high as 4 km/h [2.5 mph] at low speed changes and decreased to a maximum of 2.6 km/h [1.6 mph] at a speed change of 10 km/h [6.2 mph].” They attributed this difference to the portion of the ∆V that occurred prior to the threshold acceleration necessary to wake-up the module. They determined that this threshold acceleration was between 1.2 and 1.4 g for the modules in their study. The authors also found that the error in the SDM-reported ∆V was sensitive to pulse shape and duration. For the same ∆V, shorter collisions pulses generated more accurate SDM-reported ∆Vs, since more of the pulse was above the threshold acceleration. The authors observed that “long duration and low peak acceleration pulses have larger areas excluded from the integration and will result in larger errors in the SDM-reported speed change than short duration or high peak acceleration pulses of the same general shape.” Finally, they found that “the highest speed change that did not generate a near-deployment event was 6.2 km/h [3.9 mph] and the lowest speed change that did generate a near-deployment was 4.9 km/h [3.0 mph].” These authors did not report tabular data for their tests, and so these tests are not included in Figure 3.

However, the results of these tests are consistent with other low severity tests that are included in the figure.

Wilkinson, Lawrence, Heinrichs, and Siegmund [2004] examined the accuracy of the EDR-reported ∆Vs from Ford restraint control modules (RCMs) in low-severity collisions. They conducted 84 frontal barrier collisions with two RCM-equipped vehicles with actual speed changes as high as 13.5 km/h (8.4 mph). These authors reported that the accuracy of the EDR-reported ∆Vs ranged from an underestimate of 1.8 km/h (1.1 mph) to and overestimate of 0.3 km/h (0.2 mph). They attributed these errors both to the portion of the ∆V that occurred prior to the threshold acceleration necessary to wake-up the module and, in many instances, to a recording window that was too short to capture the full collision pulse. In a follow-up study in 2005, Lawrence and Wilkinson reanalyzed the Ford RCM data with an updated version of the CDR software. They found differences between the ∆Vs reported by the two versions of the software. When analyzed with the updated version, more accurate speed changes resulted for one of the vehicles, but less accurate for the other vehicle. Overall, the accuracy of the EDR-reported ∆Vs was between -1.3 km/h (-0.8 mph) and 0.4 km/h (0.2 mph). These authors did not report tabular data for their tests, and so these tests are not included in Figure 3. However, the results of these

www.collisionmagazine.com

5/28/2019 9:34:59 AM

tests are consistent with other low severity tests that are included in the figure.

Wilkinson, Lawrence, Heinrichs, and King [2005] examined the accuracy of EDR-reported ∆Vs from the sensing and diagnostic modules (SDMs) on MY 2003 and 2004 General Motors vehicles for low-severity collisions. Three MY 2004 SDM-equipped vehicles (Chevrolet Cavalier, Impala, and Trailblazer) were subjected to 136 vehicle-tobarrier and vehicle-to-vehicle frontal collisions with speed changes up to 8 km/h (5 mph). Of these tests, 65 of them were vehicle-to-vehicle and 71 were vehicle-to-barrier. The SDMs were also tested on a linear sled that allowed for replicating the crash pulses from the vehicle tests and also for applying similar pulses of greater severity to the SDMs. The authors reported that “in all of the tests, the speed change reported by the SDM underestimated the actual speed change. The speed change underestimates ranged from 0.2 to 2.9 km/h [0.1 to 1.8 mph] except for several anomalous tests in which the underestimate was as high as 12.3 km/h [7.6 mph].” In comparing their results to their 2002 publication with earlier model year SDMs, the authors state that “the newer model GM SDMs, and in particular the 2004 Cavalier SDM, appear to be more accurate in reporting speed change than their predecessors.

The improved accuracy is at least partially explained by the lower threshold acceleration in the new Cavalier SDM (1.1g versus 1.3g).” These authors did not report tabular data for their tests, and so these tests are not included in Figure 3. However, the results of these tests are consistent with other low severity tests that are included in the figure.

Wilkinson, Lawrence, Nelson and Bowler examined the accuracy of EDR-reported ∆Vs from MY 2005 to 2008 Toyota Corolla EDRs for low-severity collisions. They utilized vehicle-to-barrier tests and ACM sled tests for their evaluation. The frontal impact barrier tests ranged in speed from 2.0 to 5.2 km/h (1.2 to 3.2 mph) with ∆Vs between 3.5 and 9.0 km/h (2.2 and 5.6 mph). The lowest ∆V that produced a recorded event was 4.5 km/h (2.8 mph) with a peak acceleration of 2.1g. The authors reported that “in all in-vehicle tests, the speed change reported by the ACM underestimated the actual speed change for frontal collisions and overestimated the actual speed change for rearend collisions. The speed change underestimates ranged from 1.3 to 2.6 km/h [0.8 to 1.6 mph] and the speed change overestimates ranged from 0.6 to 2.2 km/h [0.4 to 1.4 mph] … Threshold accelerations required to initiate the recording of an event were found to be between 2.0 and 2.1g for all of the ACMs tested.” The authors also reported that “integrating the sled acceleration pulses after a 2g threshold was achieved and adding a constant bias of 0.4g reproduced the temporal shift in the cumulative

speed change values reported by the ACM relative to the reference speed change…it also generated maximum speed change estimates (∆Vmodel) that were close to the ACM speed changes (∆VACM) especially for frontal crash pulses.

These authors did not report tabular data for their tests, and so these tests are not included in Figure 3. However, the results of these tests are consistent with other low severity tests that are included in the figure.

Xing, Lee, Flynn, Wilkinson and Siegmund [2016] compared the response of 19 Generation 1, 2, and 3 Toyota EDRs from Toyota Corollas, Camrys, and Priuses for lowseverity collisions. They used a sled to subject the EDRs to frontal and rear haversine crash pulses of varying duration (80, 120, 160, and 200 ms), peak acceleration (0.17 to 4.59 g), and ∆V (0.9 to 13.0 km/h). They found that acceleration necessary to trigger an event was approximately 2 g for Generation 1 and 2 EDRs. The acceleration necessary to trigger an event for a Generation 3 EDR appeared to vary, but events were consistently triggered for ∆Vs around 8 km/h. Xing et al theorized that “this new and different behavior for the Gen3 ACMs could be the results of software changes that simply do not report events that fall below this 8 km/h threshold despite an internal trigger that remains at 2 g.” These authors further reported that “in most front impacts, the ACMs underestimated the reference speed change…” Xing et al presented regression equations for each EDR generation/vehicle combination that would related the EDR-reported ∆V to the actual ∆V. For individual cases involving Toyotas, a reconstructionist can apply these equations. For the purpose of our discussion here, it is sufficient to observe that for frontal impacts, the EDR-reported ∆V errors were between a 0.72 km/h (0.45 mph) over-estimate and a 3.81 km/h (2.37 mph) underestimate. The vast majority of the EDR-reported ∆Vs under-estimated the actual ∆Vs. Xing et al also noted that the magnitude of errors in the EDR-reported ∆Vs could be reduced to ±1 km/h with the use of their regression equations. This suggests that testing of individual ACMs could be useful for achieving high levels of accuracy on a specific case. Some cases will warrant such testing and others will not.

In a follow-up study in 2017, Lee, Xing, Yang, Lee, Wilkinson, and Siegmund expanded their examination of Toyota ACMs to include mid-severity crashes and to confirm that sled testing of ACMs could be applied to the analysis of vehicle-to-barrier and vehicle-to-vehicle crashes. These authors reported that “in the frontal vehicle-to-barrier tests, the ACM-reported speed changes consistently underestimated the reference speed change…the ACM-reported speed change consistently overestimated the reference speed change in rear-end vehicle-to-barrier

www.collisionpumagazine.com

Volume 13 Issue 1.indd 103

Collision Magazine - Volume 13 Issue 1 103

5/28/2019 9:34:59 AM

collisions. In vehicle-to-vehicle collisions, the same pattern of underestimating in frontal collisions and overestimating in rear-end collisions was observed…Overall, the ACMs underestimated frontal speed change by 1.25 km/h and overestimated rear-end speed change by 0.79 km/h.” For the frontal impact vehicle tests, the range in the underreporting for the EDR-reported ∆Vs varied between 0.8 and 2.5 km/h (0.5 to 1.6 mph). These authors presented additional regression modeling with which the EDR-reported ∆V could be corrected and the accuracy improved.

The vehicle tests from this study are included in Figure 3.

These studies related to the accuracy of EDR-reported ∆Vs in aligned, low-severity collisions consistently show the EDR-reported longitudinal ∆Vs under-reporting the actual longitudinal ∆V, generally by a magnitude of 2 mph or less – most of the time by a magnitude of 1.5 mph or less. Based on these studies, a reasonable range on the actual longitudinal ∆V for the struck vehicle would have a low-end of the longitudinal ∆V reported by the EDR and a high-end of 1.5 to 2 mph greater in magnitude than the longitudinal ∆V reported by the EDR. So, for example, if the EDR-reported longitudinal ∆V was -5 mph, the reconstructionist could apply a range for the actual longitudinal ∆V between -5 and -6.5 mph. Now, consider additional studies of the EDR-reported ∆V accuracy for other impact configurations.

Partial-Overlap Frontal Collisions Haight [2013b] analyzed 12 high-severity, small overlap collisions involving MY 2013 vehicles, where 25% of the front of the vehicle was engaged in the collision with a barrier. These tests, which involved the vehicles impacting the barrier at a speed of approximately 40 mph (64 km/h), were conducted by the Insurance Institute for Highway Safety (IIHS). Each of the vehicles in these tests had available EDR data, accessible either with the Bosch Crash Data Retrieval (CDR) Tool or the GIT/Snap-On EDR tool, and Haight evaluated the accuracy of the EDR-reported ∆Vs. These EDRs reported both longitudinal and lateral ∆Vs. Haight found that all of the EDR-reported lateral ∆Vs, measured at the location of the ACM, under-reported the ∆Vs measured at the location of the accelerometer array installed on the vehicle for the test. The average difference was 8 mph (13 km/h). Haight attributed the difference to the difference in location between the ACM and the accelerometer array, along with the significant yaw rotation induced by the collisions (error source #5). He presented a method for adjusting the ∆V to reflect a different location in the vehicle than where it was initially measured, and he successfully employed this method to reasonably reconcile the EDR-reported ∆Vs with the ∆Vs obtained 104 Collision Magazine - Volume 13 Issue 1

Volume 13 Issue 1.indd 104

from the accelerometer array. Based on our review of the post-test photographs from these tests, there could also be some error in the EDR-reported ∆Vs for these tests due to deformation of or around the ACMs. Other authors have examined the EDR-reported ∆Vs for partial-overlap frontal collisions, but they did not make corrections for yaw rotation experienced by the test vehicles and discrepancies in position between the ACMs and the laboratory accelerometers. Comeau [2004] examined 2 partial-overlap frontal collisions, one involving a 1998 Chevrolet Cavalier and the other a 2002 Chevrolet Impala, in which 40% of the front of the vehicle impacted a non-moving, deformable barrier face at approximately 40 km/h (25 mph). The EDRs on these vehicles only reported longitudinal ∆Vs. For the Cavalier, the EDR-reported ∆V was about 2 km/h (1.3 mph) different than that determined from laboratory instrumentation. For the Impala, the EDR under-reported the magnitude of the ∆V by about 14 km/h (8.7 mph). The authors attributed the error for this second test to limited recording capabilities of the EDR on the Cavalier. Yaw rotation of the vehicle did not appear to be a significant factor.

Niehoff et al [2005] reported analysis of 9 IIHS highseverity, moderate overlap (40%) collisions involving MY 2000-2004 vehicles. Most of the EDRs in their study only recorded longitudinal ∆V. Two of the vehicles did record lateral ∆Vs as well, and a comparison to the actual lateral ∆Vs were made in these two cases. The authors reported significant errors in the EDR-reported lateral ∆Vs for these two cases. However, no method was presented, and no analysis appears to have been done to correct for discrepancies between the location of the ACM and the laboratory accelerometers. The authors stated that “the EDR crash sensor and the crash test accelerometer were not positioned at the same locations in the car. This may complicate this comparison is some types of crashes. In full frontal barrier crash tests, there should be no difficulty as the EDR accelerometer and a crash test accelerometer located in the occupant compartment should experience the same acceleration. In other types of crash tests such as frontal offset or angled impacts, however, the impact may be characterized by significant vehicle rotation. In these cases, the EDR and crash test accelerometer may experience a different acceleration due to this rotation.” Since the authors did not account for this error source or make any attempt to isolate if from other error sources, the errors reported for the partial overlap collisions in their study cannot be considered an assessment of the measurement error.

Instead, they are more representative of incomplete analysis and comparing apples (the ∆V at the EDR location) to oranges (the ∆V at the accelerometer array location).

www.collisionmagazine.com

5/28/2019 9:34:59 AM

The Niehoff et al study also included a single vehicle-to-vehicle crash test between a 2000 Cadillac Seville and a 1997 Honda Accord that involved the Cadillac impacting the Accord at an angle of 330 degrees. This collision primarily engaged the front of the Cadillac, although given the impact configuration, the vehicle likely experienced a lateral ∆V as well. The EDR on this vehicle reported longitudinal ∆Vs only. Also, the EDR had a limited recording window that was insufficient to capture the full crash pulse. Nonetheless, up to the time at which recording terminated, the EDR-reported cumulative ∆V tracked closely with the cumulative ∆V determined from the other instrumentation on the vehicle.

Wilkinson, Lawrence, and King [2007] reported on the accuracy of the maximum SDM-reported ∆Vs for General Motors vehicles involved in NHSTA frontal impact crash tests. They examined 23 tests, 21 of which were full-overlap, frontal impacts, 1 of which was an offset impact into a rigid barrier, and 1 of which was an offset and angled vehicle-to-vehicle collision. The offset impact into a rigid barrier was a 2003 Chevrolet Suburban, which impacted the barrier with 40% overlap at a speed of approximately 40 km/h (25 mph). The authors calculated the longitudinal ∆V for the test vehicle at the five locations in the vehicle (the front and rear sills on the left and right sides and the CG). There was a discrepancy between the left and right side ∆Vs that was consistent with the vehicle experiencing yaw rotation from the collision. The authors reported that the EDR-reported ∆V underestimated these ∆Vs by as much as 5.4 km/h and overestimated them by as much a 0.9 km/h. They report that lateral acceleration data was not available for this test. This would have precluded them from calculating the rotation rate from accelerometer data and resolving their calculated ∆V to the ACM location.

The offset and angled vehicle-to-vehicle collision involved the front of a 1997 Honda Accord traveling 56.6 km/h impacting the front of a 2000 Cadillac sedan traveling 55.9 km/h. The Cadillac was angled at 30 degrees relative to the Accord and there was approximately 50% overlap between the vehicles. For this test, both longitudinal and lateral accelerations were available and the total ∆V was calculated at 5 locations in the test vehicle (the front and rear sills on the left and right sides and the CG). The authors reported that the EDR-reported longitudinal ∆V under-reported the CG longitudinal ∆V determined from the accelerometers by 10.4 km/h (6.5 mph) and the CG resultant ∆V by 10.8 km/h (6.7 mph). This error was partially attributed to an EDR recording window that was too short. The authors also noted the effects of yaw rotation and a discrepancy in position between the CG and the ACM – the ACM was mounted under the right front seat of the Cadillac – but

they did not correct for the effects of this in their comparison of the ∆Vs. Gabler [2008] examined the accuracy of the EDR data downloaded from 48 crash-tested vehicles. Four of these were partial-overlap frontal impacts – three impacted a deformable barrier at approximately 40 mph with 40% overlap of the front end and the other impacted a pole at approximately 40 mph, with approximately 15% overlap.

Gabler noted that for 14 out of the 48 tests that he examined, the recording window for the EDR was insufficient to capture the entire collision. The three test vehicles involved in the 40% overlap tests were among those with an insufficient recording window and this resulted in the EDRs significantly under-reporting the ∆Vs. In addition, these collisions induced significant yaw rotation of the test vehicles. Gabler did not account for discrepancies in position between the CG and the ACM, but in this instance, the short recording windows appeared to be a dominant factor that would not have been overcome through consideration of the yaw rotation. The EDR for the 15% overlap pole impact test had a sufficient recording window and the EDR-reported longitudinal ∆V in this test was within 6% of the actual.

These studies demonstrate that a discrepancy in position between the ACM and the vehicle CG, mixed with significant yaw rotation of the vehicle that results from the collision, can lead to significant errors (error source #5) if these factors are not incorporated into the analysis. This error source is likely to be encountered at times when reconstruction motorcycle collisions and the methods outlined in the following references should be utilized to account for them: Bundorf [1996], Marine and Werner [1998], Rose [2007], and Haight [2013a]. Since this error source can be accounted for in the analysis, it does not need to be included in the range on the ∆V for an uncertainty analysis. Thus, the presence of this error source does not necessitate any change to the proposed range on the ∆V. Side Impacts Haight, Gyorke, and Haight [2013b] reported analysis of a single side impact crash test of a 2013 Kia Rio, which was run by the IIHS. In this test, the stationary Kia was struck on the driver’s side at a 90-degree angle by a deformable barrier traveling 50 km/h (31.1 mph). Lateral accelerations were measured on the test vehicle, near the A and B pillars on the non-struck side of the vehicle. This impact did not induce significant yaw rotation of the Kia. The authors reported that “the lateral delta-V reported by the EDR Tool in this side impact test reasonably compares to the delta-V calculated using the IIHS accelerometer. The maximum delta-V calculated from the IIHS accelerometer

www.collisionpumagazine.com

Volume 13 Issue 1.indd 105

Collision Magazine - Volume 13 Issue 1 105

5/28/2019 9:34:59 AM

is about 26.7 km/h (16.6 mph) while the maximum reported delta-V reported by the EDR tool is 25 km/h (15.5 mph).” This is an under-reporting by the EDR of 1.1 mph and an approximately 6.4% error. Thus, this is an instance where the test procedure does not induce significant rotation and the ∆V accuracy is well within 10%.

Tsoi, Johnson, and Gabler [2014] evaluated the accuracy of 75 EDRs from MY 2010 to 2012 Chrysler, Ford, General Motors, Honda, Mazda, and Toyota vehicles in side impacts. These vehicles were each subjected to side impacts with a moving deformable barrier (MDB) as a part of the NHTSA Side-Impact New Car Assessment Program (SINCAP). This test procedure involved the MDB impacting the stationary test vehicle at 62 kph (39 mph). The heading angle of the MDB is 90 degrees relative to the test vehicle, but the wheels are crabbed, such that the velocity of the MDB is angled 27 degrees relative to its heading.

The reference ∆Vs were calculated at the vehicle CGs from the laboratory accelerometers on the vehicle. In describing this analysis, Tsoi et al noted that the vehicles in these tests experienced yaw rotation and they describe the equations they used to account for this rotation during calculation of the reference ∆Vs. Tsoi et al reported that the “EDRs underreported the reference lateral delta-v in the vast majority of cases, mimicking the errors and conclusions found in some longitudinal EDR accuracy studies.

For maximum lateral delta-v, the average arithmetic error was -3.59 kph (-13.8%) and the average absolute error was 4.05 kph (15.9%).” Unfortunately, Tsoi et al do not appear to have used the laboratory accelerometer data to calculate the ∆Vs at the ACM positions for the vehicles, and so their reference ∆Vs are not directly comparable to the EDRreported ∆Vs. Tsoi et al acknowledged this as a possible error source in their study, but they believe the error would be small. However, given the results in Haight et al [2013a and 2013b], it seems to be an important error source to account for prior to reporting the “accuracy” of the EDRreported ∆Vs. For now, this study simply serves as another reminder to account for error source #5. Additional research could render additional insight and explanation for the results obtained by Tsoi et al.

varied as much as 20%. The CG sensor was chosen as the most comparable, but it should be noted that the principle direction of force (PDOF) vector does not pass through the CG in this test. The vehicle rotates counter clockwise during the test. An alternative measurement was evaluated, taking the momentum vector of the impacting cart, and the relative weights of the cart and test vehicle, and calculating the expected lateral Delta V. This correlated more closely with the EDR than the CG sensor.” The authors do not appear to have used the laboratory accelerometer data to calculate the ∆Vs at the ACM positions for the vehicles, and so their reference ∆Vs – whether from other sensors or from momentum analysis – are not directly comparable to the EDR-reported ∆Vs. Without such analysis being completed, not a lot of stock can be put on the differences between the EDR-reported and “actual” ∆Vs reported in this study. Instead, this study can be taken as another reminder of the importance of accounting for error source #5.

Carr, Rucoba, Barnes, Kent, and Osterhout tested passenger car EDRs on a HYGE crash simulation sled in various orientations designed to represent different principal directions of force (PDOF), not limited to those typical of standard crash test configurations [2015]. This is a useful study because it eliminated any error arising from yaw rotation, ACM location, deformation to or around the ACM, acceleration clipping, or an inadequate recording window. They performed direct comparison of the EDR-reported and actual longitudinal and lateral ∆Vs and also examined the possibility of accurately reconstructing the PDOF orientation from the combined longitudinal and lateral EDRreport ∆Vs. They reported that the maximum percentage error in the ∆V was less than 10%, with the average error magnitude for the various EDRs ranging between 0.3% to 4.3%. The magnitude of the errors was between 0 and 1 mph. They reported that the maximum PDOF angle error magnitude was 2.0 degrees.

The specific EDRs tested by Carr et al were from a 2012 Chevrolet Malibu, a 2012 Dodge Durango SXT, and a 2012 RAM 1500. Their test rig allowed the orientation of the EDRs to be swept through a range of yaw angles between -90 degrees (driver side leading) to +90 degrees The Ruth and Tsoi study from 2014, which examined the (passenger side leading). The EDRs were tested throughaccuracy of EDR-reported ∆Vs for Kia and Hyundai vehi- out this range in angle increments of 22.5 degrees. They cles from the model years (MY) 2010 to 2012, included 19 also tested the EDRs with pitch angles between 5 and 20 side impact tests. Again, this was prior to Kia and Hyun- degrees in each direction. Each module was subjected to dai’s full compliance with CFR 563. The side impact tests 13 simulated crash pulses with an actual ∆V of approxiinvolved a moving deformable barrier impacting the vehi- mately 25 mph (40 km/h). This study appears to show that cles at a speed of approximately 62 km/h (38.5 mph). The the magnitude of errors in the EDR-reported ∆Vs is not authors noted that “for the MDB tests, the center of grav- heavily dependent on direction of the collision force, at ity (CG), far side front and rear door sill, and rear floor- least for the modules in the study. One difference between pan sensors were reviewed. The values from these sensors the lateral and longitudinal ∆Vs for these tests was that the 106 Collision Magazine - Volume 13 Issue 1

Volume 13 Issue 1.indd 106

www.collisionmagazine.com

5/28/2019 9:34:59 AM

errors in the EDR-reported longitudinal ∆Vs were mostly in the direction of under-reporting, whereas the errors in the lateral ∆Vs appeared more distributed about zero, with both positive and negative errors occurring. This is consistent with these EDRs having a small positive acceleration bias along the longitudinal axis.

is confirmed by the results from Beck et al [2006], which were from motorcycle collisions with EDR-equipped vehicles. The results from Beck were within the same range of EDR-reported ∆V errors as the other low-severity studies.

Based on these studies, a reasonable range on the actual longitudinal ∆V for the struck vehicle would have a lowend of the ∆V reported by the EDR and a high-end of 1.5 Collisions Involving Motorcycles to 2 mph greater in magnitude than what is reported by the Beck [2006]: Beck, Casteel, Phillips, Eubanks, and Eng- EDR. So, for example, if the EDR-reported longitudinal lish reported three staged collinear collisions involving an ∆V was -5 mph, the reconstructionist could apply a range EDR-equipped passenger vehicle and a motorcycle. The for the actual ∆V between -5 and -6.5 mph. A reconstrucpassenger vehicle was a 2002 Chevrolet Cavalier weighing tionist could also use the EDR reported longitudinal ∆V 2,650 lb. This vehicle was driven into a stationary, upright as reported and this would result in a conservative (low) 1989 Kawasaki EX500 weighing 415 lbs and being “rid- estimate of the motorcycle’s ∆V and impact speed. Based den” by a 160 lb dummy (a combined weight of 575 lb). on the Carr et al study, a similar range could be applied to The weight ratio between the Chevrolet and Kawasaki was the EDR-reported lateral component of the ∆V, though approximately 4.6:1, including the weight of the dummy perhaps the range should be centered on zero (±1 mph, for on the motorcycle. In the first test, the Chevrolet had an instance). When considering all of the error sources disimpact speed of approximately 12 mph. The peak longitu- cussed in this article, the actual resultant ∆V can be higher dinal acceleration experienced by the Cavalier in this test or lower than the resultant EDR-reported ∆V. was -1.7 g and the impact duration was approximately 186 In addition, the analysis reported in this article has led us ms. No event was recorded by the Cavalier’s EDR in this to the following considerations for incorporating struck impact. In the second test, the Chevrolet had an impact vehicle EDR data into the reconstruction. These steps are speed of approximately 27 mph. The peak longitudinal ac- specific to incorporation of the EDR data and are not all celeration experienced by the Cavalier in this test was -6.6 of the steps in a complete reconstruction. These steps will, g and the impact duration was approximately 180 ms. A of course, be interwoven with the rest of the investigation non-deployment event was recorded by the EDR. In the and reconstruction process. Not all of these steps will be third test, the speed of the Chevrolet at impact was ap- necessary for every case. proximately 37 mph. The peak longitudinal acceleration experienced by the Cavalier in this test was -12.7 g and the 1. If possible, during an inspection of the struck vehicle, impact duration was approximately 96 ms. A deployment physically examine the ACM and the surrounding strucevent was recorded by the EDR. Comparing the ∆V from ture (the floor or center tunnel, for instance) for damage the EDR to that calculated from other instrumentation on or deformation. If the vehicle is not available, examine the vehicle, Beck et al found that the EDR under-reported photographs and other information about the damage the actual magnitude of the ∆V by 0.86 mph and 0.53 to the vehicle to determine if damage to or around the mph for the second and third tests, respectively. These tests ACM is probable. Many motorcycle-to-car collisions were setup in such a way that they focused on the longiwill be of insufficient severity for the struck car to cause tudinal axis of the vehicle. The results of these two tests deformation or damage to or around the ACM. are consistent with the results for other low-severity tests a. If there is no damage to the ACM or its attach(low-severity for the passenger vehicle, that is) in that the ment points and the surrounding structure is undeEDR-reported ∆Vs under-report the actual ∆Vs, but they formed, then this can be ruled out as an error source are within the previously proposed potential error band in the EDR-reported ∆V (error source #6). of 1.5 to 2 mph. These two tests are included in Figure 3.

C

onclusions The studies related to the accuracy of EDR-reported ∆Vs in aligned, low-severity collisions are likely to be the most applicable to motorcycle-passenger vehicle collisions. These studies consistently show the EDRreported longitudinal ∆Vs under-reporting the actual longitudinal ∆V, generally by a magnitude of 2 mph or less – most of the time by a magnitude of 1.5 mph or less. This www.collisionpumagazine.com

Volume 13 Issue 1.indd 107

b. If there is damage to the ACM or its attachment points, or if the surrounding structure is deformed, then there could be significant error in the EDRreported ∆V. While this error source is detectable, it is unlikely that the reconstructionist would be able to correct for it, and so, the EDR-reported ∆V will likely be unusable for the reconstruction. The precrash data can still be used in the reconstruction.

Collision Magazine - Volume 13 Issue 1 107

5/28/2019 9:34:59 AM

2. Document the position of the ACM within the vehicle, if possible. Measure its position relative to an undamaged portion of the vehicle – the front or rear axle for instance. This measurement can be used in later analysis to determine the position of the ACM relative to the vehicle CG. 3. Examine the EDR report for the collision. Again, the steps listed below are not necessarily a comprehensive analysis of the EDR report, but just steps that specifically relate to the error sources discussed in this article. a. Does the EDR report indicate that there was complete recording of the event? If so, then power loss prior to full recording can be ruled out (error source #7). b. If they are available in the EDR report, examine the graphical and tabular reporting of the longitudinal and lateral accelerations and cumulative ∆V. Do any of these exhibit flatlining at the peak acceleration? This may be an indication that the longitudinal or lateral accelerations exceeded the capabilities of the accelerometer in the ACM. If this flatlining is not present, then error source #4 can potentially be ruled out. c. Now, examine the recording window for the longitudinal and lateral accelerations and cumulative ∆V and the shape of the curves leading up to the end of this recording window. Error source #2 can be ruled out if the EDR-reported accelerations and ∆V reach a maximum and begins to decrease prior to the end of the recording window. If they continue to decrease through the end of the recording window, then error source #3 can be ruled out.

4. Analysis could be performed to estimate the probable ∆V that occurred prior to AE. This analysis could potentially use an impact duration and peak acceleration from the EDR report. This analysis would allow the EDR-reported ∆V to be corrected for error source #1. That said, if a 1.5 to 2 mph range on the ∆V is a sufficiently tight range, then there is not a need to conduct this analysis.

This range does not include the effects of yaw rotation of the vehicle and a discrepancy between the ACM position and the CG. 5. Incorporate a reasonable range for the struck vehicle ∆V with other analysis methods – conservation of momentum, analysis of the car and motorcycle deformation, and analysis of the struck vehicle’s post-impact translation and rotation. This more comprehensive ap-

108 Collision Magazine - Volume 13 Issue 1

Volume 13 Issue 1.indd 108

proach is described extensively in the companion paper to this one.

R

eferences Beck, R., Casteel, D., Phillips, E., et al., “Motorcycle Collinear Collisions Involving Motor Vehicles Equipped with Event Data Recorders,” Collision Magazine 1(1): 82-96, 2006.

Bortles, W., Biever, W., Carter, N., and Smith, C., “A Compendium of Passenger Vehicle Event Data Recorder Literature and Analysis of Validation Studies,” SAE Technical Paper 2016-01-1497, 2016, https://doi. org/10.4271/2016-01-1497.

Bundorf, R.T., “Analysis and Calculation of Delta-V from Crash Test Data,” SAE Technical Paper 960899, 1996, https://doi.org/10.4271/960899.

Carr, L., Rucoba, R., Barnes, D., Kent, S., et al., “EDR Pulse Component Vector Analysis,” SAE Technical Paper 2015-01-1448, 2015, doi:10.4271/2015-01-1448.

Chidester, A., Hinch, J., Mercer, T., and Schultz, K., “Recording Automotive Crash Event Data,” Proceedings of the International Symposium on Transportation Recorders, Arlington, Virginia, 1999.

Code of Federal Regulations, 49 CFR 563 – Event Data Recorders, in effect as of September 1, 2012.

Comeau, Jean-Louis, German, Alan, Floyd, Donald, “Comparison of Crash Pulse Data from Motor Vehicle Event Data Recorders and Laboratory Instrumentation,” Proceedings of the Canadian Multidisciplinary Road Safety Conference XIV, Ottawa, Ontario, June 27-30, 2004.

Comeau, Jean-Louis, Dalmotas, Dainius, German, Alan, “Evaluation of the Accuracy of Event Data Recorders in Chrysler Vehicle in Frontal Crash Tests,” Proceedings of the 21st Canadian Multidisciplinary Road Safety Conference, Halifax, Nova Scotia, May 2011.

Comeau, Jean-Louis, Dalmotas, Dainius, German, Alan, “Event Data Recorders in Toyota Vehicles,” Proceedings of the 21st Canadian Multidisciplinary Road Safety Conference, Halifax, Nova Scotia, May 2011.

Correia, J., Iliadis, K., McCarron, E., et al., “Utilizing Data from Automotive Event Data Recorders,” presented at the Canadian Multidisciplinary Road Safety Conference XII, June 10-13, 2001.

Emori, R., “Analytical Approach to Automobile Collisions,” SAE Technical Paper 680016, 1968, https://doi. org/10.4271/680016.

Exponent Failure Analysis Associates, “Testing and Analysis of Toyota Event Data Recorders,” https://pressroom. toyota.com/article_download.cfm?article_id=3196, October 2011.

www.collisionmagazine.com

5/28/2019 9:34:59 AM

Gabler, H. Clay, Thor, Craig P., Hinch, John, “Preliminary Evaluation of Advanced Air Bag Field Performance Using Event Data Recorders,” DOT HS 811 015, August 2008.

German, Alan, Dalmotas, Dainius, Comeau, Jean-Louis, “Crash Pulse Data from Event Data Recorders in Rigid Barrier Tests,” Paper No. 11-0395, 22nd ESV Conference, Washington, D.C., June 2011.

[Haight, 2013a] Haight, S., Haight, R., “Analysis of Event Data Recorder Delta-V Reporting in the IIHS Small Overlap Crash Test,” Collision 8(2): 8-23, 2013.

[Haight, 2013b] Haight, R., Gyorke, S., Haight, S., “Hyundai and Kia Crash Data, The Indispensable Compendium: Section 2 – Crash Testing Involving Hyundai and Kia Vehicles,” Collision 8(2): 77-86, 2013.

Lawrence, J., Wilkinson, C., King, D., Heinrichs, B. et al., “The Accuracy and Sensitivity of Event Data Recorders in Low-Speed Collisions,” SAE Technical Paper 2002-010679, 2002, doi:10.4271/2002-01-0679.

Lawrence, J. and Wilkinson, C., “The Accuracy of Crash Data from Ford Restraint Control Modules Interpreted with Revised Vetronix Software,” SAE Technical Paper 2005-01-1206, 2005, doi:10.4271/2005-01-1206.

Lee, F., Xing, P., Yang, M., Lee, J., Wilkinson, C., Siegmund, G.P., “Behavior of Toyota Airbag Control Modules Exposed to Low and Mid-Severity Collision Pulses,” SAE Technical Paper 2017-01-1438, 2017, doi:10.4271/201701-1438.

Marine, M.C., Werner, S.M., “Delta-V Analysis from Crash Test Data for Vehicles with Post-Impact Yaw Motion,” SAE Technical Paper 980219, 1998, https://doi. org/10.4271/980219.

Niehoff, Peter, Gabler, Hampton C., Brophy, John, Chidester, Chip, Hinch, John, Ragland, Carl, “Evaluation of Event Data Recorders in Full System Crash Tests,” Paper No. 05-0271, 19th ESV Conference, Washington, D.C., June 2005.

Rose, Nathan A., Beauchamp, Gray, Bortles, Will, “Quantifying the Uncertainty in the Coefficient of Restitution Obtained with Accelerometer Data from a Crash Test,” SAE Technical Paper 2007-01-0730, 2007, https://doi. org/10.4271/2007-01-0730.

Rose, N., Bortles, W., Carter, N., Randolph, M., McDonough, S., “Motorcycle Accident Reconstruction: Incorporating EDR Data from the Struck Vehicle,” forthcoming from Collision, 2019.

eration 04 EDR,” SAE Technical Paper 2016-01-1496, 2016, doi:10.4271/2016-01-1496.

Ruth, R., “Applying Automotive EDR Data to Traffic Crash Reconstruction,” SAE Course #C1210, Course Slides.

Tsoi, A., Hinch, J., Ruth, R., and Gabler, H., “Validation of Event Data Recorders in High Severity FullFrontal Crash Tests,” SAE Int. J. Trans. Safety 1(1):2013, doi:10.4271/2013-01-1265.

Tsoi, A., Johnson, N., Gabler, H., “Validation of Event Data Recorders in Side-Impact Crash Tests,” SAE Technical Paper 2014-01-0503, 2014, https://doi.org/10.4271/201401-0503.

Vandiver, W., Anderson, R., Ikram, I., Randles, B., Furbish, C., “Analysis of Crash Data from a 2012 Kia Soul Event Data Recorder,” SAE Technical Paper 2015-011445, 2015, doi:10.4271/2015-01-1445.

Varat, M. and Husher, S., “Vehicle Impact Response Through the Use of Accelerometer Data,” SAE Technical Paper 2000-01-0850, 2000, https://doi.org/10.4271/200001-0850.

Wilkinson, C., Lawrence, J., Heinrichs, B., and Siegmund, G., “The Accuracy of Crash Data Saved by Ford Restraint Control Modules in Low-speed Collisions,” SAE Technical Paper 2004-01-1214, 2004, doi:10.4271/2004-01-1214.

Wilkinson, C., Lawrence, J., Heinrichs, B., and King, D., “The Accuracy and Sensitivity of 2003 and 2004 General Motors Event Data Recorders in Low-Speed Barrier and Vehicle Collisions,” SAE Technical Paper 2005-01-1190, 2005, doi:10.4271/2005-01-1190.

Wilkinson, Craig C., Lawrence, Jonathan M., King, David J., “The Accuracy of General Motors Event Data Recorders in NHTSA Crash Tests,” Collision Vol 2, Issue 1, 2007.

Wilkinson, C., Lawrence, J., Nelson, T., and Bowler, J., “The Accuracy and Sensitivity of 2005 to 2008 Toyota Corolla Event Data Recorders in Low-Speed Collisions,” SAE Int. J. Trans. Safety 1(2):420-429, 2013, doi:10.4271/2013-01-1268.

Xing, P., Lee, F., Flynn, T., Wilkinson, C., et al, “Comparison of the Accuracy and Sensitivity of Generation 1, 2 and 3 Toyota Event Data Recorders in Low-Speed Collisions,” SAE Int. J. Trans. Safety 4(1):172-186, 2016, doi:10.4271/2016-01-1494.

Ruth, R. and Tsoi, A., “Accuracy of Translations Obtained by 2013 GIT Tool on 2010-2012 Kia and Hyundai EDR Speed and Delta V Data in NCAP Tests,” SAE Technical Paper 2014-01-0502, 2014, doi:10.4271/2014-01-0502.

Ruth, R. and Muir, B., “Longitudinal Delta V Offset between Front and Rear Crashes in 2007 Toyota Yaris Gen-

www.collisionpumagazine.com

Volume 13 Issue 1.indd 109

Collision Magazine - Volume 13 Issue 1 109

5/28/2019 9:34:59 AM

Introducing The Crash Hub

The Crash Hub is a new vehicle crash expert directory site. Both the ARC Network and EDR Experts have been merged into a single site giving the end user the ability to search, review, and retain vehicle crash experts around the world.

The Crash Hub is also much more than an expert directory site because each member of The Crash Hub has the ability to add meaningful content to the site in the form of articles, events, training classes, photo albums, and videos. The Crash Hub can also be used as job search site and a classified ads directory.

The Crash Hub is also the place to search, review and learn about products available to the vehicle crash investigator.

To learn more, visit at www.thecrashhub.com

Volume 13 Issue 1.indd 110

5/28/2019 9:34:59 AM

Search Engine

The Crash Hub has a powerful search engine allowing you to locate experts around the world with a simple click of your mouse. You can search by expertise, location, name, or any keyword.

Search Results

The Crash Hub search results profile the globe and pin the location of every active member. Simply move your mouse over the pin and that expert will be displayed.

The results map has full “zoom” capabilities to zero in on the location you are looking for.

Member Dashboard

Each member has full access and control of their professional profile in The Crash Hub. Add your photo, company logo, C.V., full contact information, specialties, publish articles, post events, job listings, photo albums, videos, and even post classified ads. You can even check how many people have viewed your professional profile.

Join Today www.thecrashhub.com Volume 13 Issue 1.indd 111

5/28/2019 9:35:00 AM

Documenting A High Speed, Rear End, Partial Overlap, Crash Test Of A Large Sedan & Stationary Commercial Trailer Craig Proctor-Parker

Accident Specialist, Durban, South Africa

A

bstract Crash scenarios involving a rear end impact are common place internationally. Some of the most devastating are often where a sedan collides into the rear of a commercial trailer, partial overlap. This type of crash is almost always with serious or fatal consequences.

With a high number of these identified in a recent high profile Major Crash Investigation (MCI) project, a real life high speed test of this scenario was undertaken. Obtaining data from such a crash in a controlled environment for future comparative analysis is rarely presented. This paper presents a brief overview of the setup and results of the high speed rear end, sedan to stationary commercial trailer.

Keywords & phrases - Rear-end crash; head-rear crash; Under- ride crash, Rear end accident, High speed rear end impact, Partial overlap impact.

I

ntroduction: This research is based on a practical crash test session undertaken by Accident Specialist as a combination of research and awareness. Three different crash scenarios were presented, two of these remote controlled1. This particular crash being one of the three and is available online at: https://www.youtube.com/watch?v=dsT1VCL18Uk

Internationally, rear end crash scenarios are well represented in crash statistics, this too is a statistic that is prevalent in South Africa. Such crash scenarios are typically as a direct result of major traffic congestion, vehicles broken down at hazardous locations and/or simply due to driver error in their negligence either as the vehicle being struck (Target), or the vehicle striking (Bullet). As with opposite 112 Collision Magazine - Volume 13 Issue 1

Volume 13 Issue 1.indd 112

direction head-on, type impacts, high speed rear end partial overlap impacts present an even greater risk for a number of reason, not least of all the reduced contact area and therefore reduction in available material to absorb and dissipate the energies. Such crash scenarios typically result in fatal or at least serious injuries and damages. The investigation of such a crash scenario almost always centres around certain interrelated key issues that are almost always raised in litigation 35, an issue that will be highlighted later in the paper, however in brief are typically: •

Where was the target vehicle • •

Precise point of impact

It`s lateral position in respect of the road layout

•

Why was the Target vehicle at that position

•

What was pre-crash visibility

•

What was the speed of the vehicles

•

Mechanical issues?

•

The line of sight of the Bullet vehicle

•

Usually the focus is on the Bullet vehicle, but may also be the Target vehicle.

The nature of a high speed, partial overlap rear end crash is such that massive destruction of the Bullet vehicle is typical. This is especially so where massive disparity between the general structure types, size and strengths are the case as where a sedan strikes a commercial vehicle. Such disparities are already well identified internationally 13,14,15. If not from the crash itself, the extent of decimation to the Bullet vehicle

www.collisionmagazine.com

5/28/2019 9:35:00 AM

is typically aggravated by further rescue cutting and recovery process. These damages all too often contaminate the original damages and on occasion, telemetry that may have been available 12. On occasion, conflagration results and too, decimates vital evidential factors. There is no doubt that in time, with advancement in technology, detailed telemetry will be and is already to some extent available, be secure and easily downloadable from vehicles post-crash.

It may be that such data is streamed live time and stored off-site of the vehicles 12,37a,37b. Nonetheless, there will almost always be some need to do a physical inspection and consideration of certain parameters of a crash.

The contributions of this paper are: •

Identify the characteristic correlation with the results of the vehicles after a crash of this manner

• • • •

Identify possible weaknesses on the Bullet and the Target that could be improved

Identify common evidence after such a crash has occurred

Determine to what extent resulting evidence can be used to answer some of the many questions that arise from this type of crash Provide documented record of the specific crash for further analysis if and when required

L

iterature review & research: The subject matter of rear end impacts is a relatively well researched well documented one 3,4,6,8,17 et al.. Rear end crash scenarios in general are extremely prevalent and in the USA during 1999 they accounted for nearly one third of all crashes 7. During 2000 they once again accounted for as much as 29.7% of all crash cases 2.

www.collisionpumagazine.com

Volume 13 Issue 1.indd 113

Collision Magazine - Volume 13 Issue 1 113

5/28/2019 9:35:01 AM

In South Africa, it is well documented that there is a serious problem in the recording of crash statistics 26, as such, it is impossible to given any reasonable indication of the level of the problem. This noted, Accident Specialist was one of the specialist service providers on the Major Crash Investigation (MCI) (Specific criteria guided the classification of the major crash classification) project undertaken by the Road Traffic Management Corporation (RTMC) between 2009 and 2016. All MCI cases nationally were attended to by the services providers, resulting in 726 cases attended to. Of these, some 66 (9%) were some type of rear end impact. Further, it is fair to assume that the percentage of rear end crash cases nationally, covering all cases not just MCI level crash cases, would not be vastly different to that of international norms and therefore presents a sizeable percentage of all crash cases.

Much of the current automotive media hype centres directly around the mitigation of such crash scenarios with the use of a range of technologically driven approaches.

These include vehicle radar speed evaluation, already in use and available in the market place, however only currently on executive vehicles. These systems are typically referred to as Adaptive Cruise Control (ACC) or Collision (Crash) Warning Systems (CWS), in some iteration of the two 5,7.

Other technology such as vehicle to vehicle live reporting, GPS positioning and vehicle volume reporting are all technology driven mitigation aspects that are key items in reducing this type of crash.

It is often said that as far as reasonably possible, problems should be engineered away, this is not always possible, nor practical. A typical comment of prohibitively expensive 6,7 is often made. Interestingly, there is also much researched and written about the standards of driver training and driver awareness through media, assisting to mitigate these types of incidents 2,5,7.

Many research papers 2,5 highlight the majority of rear end crash scenarios being at congested intersections, congested roads, urban areas and in the vicinity of traffic backup locations such as at off ramps and onramps. These scenarios can be serious, but in general statistics suggest moderate to minor injuries, as speeds are somewhat lower at these locations. It is not these scenarios that is primarily considered here, it is the somewhat higher speeds that is considered.

Interestingly this exact consideration of the more serious crash is referenced to with the specific conclusion in one paper, more severe when commercial vehicles are involved 7 ; and another comment, Big trucks are not in any way crash-friendly 33.

Parameters indicated in various research, both Automotive and Aeronautical, across the various directions of force, 114 Collision Magazine - Volume 13 Issue 1

Volume 13 Issue 1.indd 114

suggest G- Force tolerance levels between 20 – 45G at 0.1 second crash pulse for a restrained occupant being the guiding limits. This noted, there are various detailed factors that should be considered carefully when dealing with this type of crash as forces are typically very high due to the severe differences 9,10,11.

All research indicates that the specific issue of rear end crash scenarios reflects the same high risk age group of 18 to 29 years of age being the high risk category 2,6, as is also evident with all crash types. This therefore supports the issue of driver training and/or education and general public awareness efforts being a key factor in mitigation.

Although this is by no means a legal analysis of this type of crash, understanding and appreciating the basic nuances of such a crash scenario from a legal perspective is important.

Perhaps not so in general research, but particularly in crash investigation circumstances where there may be need to testify as a witness on some aspect.

As has already been alluded to, such crash scenarios inevitably reach some form of litigation, especially where the more serious crash occurs. This may be criminal, civil or perhaps interdepartmental. To this end, understanding both the legal requirements of vehicles in respect of the minimum standards of such items as taillights, indicators, chevrons, warning triangles, underride bars and headlights among many other items is important. Locally these would form part of the South African Bureau of Standards (SABS/SANS) specifications. Likewise appreciating that research may need to be undertaken on USA, European, Australian, Society of Automobile (SAE) or International Standards Organisation (ISO) specifications needs to be kept in mind.

Appreciating the laws of the land is important as these are interrelated to the requirements of vehicles, the roads and the drivers. As an example, The South African National Road Transport Act (NRTA) 36 dictates. Common references there from that would pertain to the scenarios of a rear end crash may be, Regulations, Rules of the road, 304–Stopping of vehicles; 319- Hindering or obstructing traffic, 298–Passing of vehicles (Overtaking); 214–Emergency warning signs (Triangles), these serve as guidance.

Appreciating the rulings that have arrived from court proceedings, referred to as Judgements 35, is another common avenue of research and literature review. There are many key judgements that are referenced in ligation cases, some perhaps only vaguely relevant and serving as a basic guide.

Perhaps others may be of direct relevance to a particular rear end crash. By example, a rather detailed judgement (35b) concerning two heavy commercial vehicles that col-

www.collisionmagazine.com

5/28/2019 9:35:01 AM

lided in a rear end type crash where one was stationary refers. The case considers many factors such as width of lanes, positions of vehicles, layout of the scene, approach path sighting distances, warning triangles, reflective chevrons, passing vehicle headlight glare, braking distances and others. The judgement provides some interesting insight to the intricacies of legal considerations in such crash cases.

Rather interestingly, it appears that there has been littleto-no specific research, nor actual crash testing examples located that deals directly with the issue of rear end, partial overlap, high speed impacts where sedans or similar passenger vehicles and commercial vehicle are involved.

T

est parameters & methodology The intention of the crash scenario was to document and consider all of the evidential results of a rear end, partial overlap type of crash orientation with a high speed impact.

The test parameters involved the positioning of a standard flat deck, double axle Swift trailer, 2006 model, as the Target vehicle. The trailer basic specifications of 9.0 meters in length, 2.4 meters wide and a mass of 6440 kilograms (Tare).

The rear axles being solid axles, 92 mm square, with a 9000 kg per axle maximum load rating, Swift manufactured axles. Tyres fitted at the right rear impact location were typical commercial Bridgestone, Radial, V-Steel MIX 857 11.00 R20 (Regroovable) 830 kpa, in fair to good condition.

A rear bar / underride was fitted, mounted by means of two flat surfaced brackets welded to the underside of the rear cross member of the upper deck chassis. The underride bolted to this with two bolts per bracket (Figures 7; 10 & 16).

The Target setup at a position 50% across the width of a normal freeway lane, to represent a vehicle stationary and protruding partially into a lane. The Target was unladen and left standing on the trailer landing legs, brakes engaged.

The Bullet vehicle a good condition 1996 model Ford Falcon sedan, the basic specifications of 4.906 meters in length, 1.861 meters in width and a mass of 1457 kg (Tare). The Bullet vehicle was remote control driven, the remote control system designed & fitted in house 1.

The target speed of 100 km/h was appointed as the majority of regional routes where the majority of such commercial vehicles operate between main freeways and urban areas are sign posted at a maximum of 100 km/h, National routes at 120 km/h as a maximum.

The testing facilities made use of were the Toyota South Africa, Eston test site, an outdoor testing facility. The specific section of the facilities being a high speed figure-ofeight, tar (Bitumous) road. The section between the two ends being a 1000 m straight, flat, good condition section.

The specific area of positioning of the Target, at a section that has been further widened with an extra two lanes on one side, creating a 200 meter long layby area.

Extensive use of High Definition (HD), high speed cameras was made to allow detailed analysis. The Target vehicle fitted with a view down the right side, looking towards the rear and therefore directly at the impact (Figure 1). A view from the internal of the Bullet vehicle, from the rear parcel shelf, looking towards the front of the vehicle (Figure 2). A view at right angle to the overall crash, from the left side of the Target and bullet vehicle (Figure 3)

Figure 1: Target - Impact overlap – external view www.collisionpumagazine.com

Volume 13 Issue 1.indd 115

Collision Magazine - Volume 13 Issue 1 115

5/28/2019 9:35:02 AM

Figure 2: Bulllet - Impact overlap – Internal view Axis orientation for the vehicles based on the SAE right hand coordinate system, X, fore aft through the longitudinal, Y, left right through and Z vertical through (24).

The Target vehicle was fitted with basic speed and force reading telemetry, being a Geotab GO6 product. The device secured to the target at the chassis members ahead of the axles (Figure 17) and self-powered. The bullet vehicle fitted with a Geotab GO6 product, providing both force readings and GPS position speed readings. The device internal at the rear foot well area, secured to the steel floor structure and self powered.

The Bullet vehicle was followed by a chase vehicle, the bullet vehicle remote control accelerated and steered. Impact was targeted at around a 50% overlap and target speed 100 km/h.

T

he test Although the target 100 km/h was not attained, the speed reached at 87 km/h (GPS speed) (Figure 18) was high enough to be considered high speed and result in a very serious crash scenario that would likely see a fatality and if not serious injuries and damages. An impact overlap of approximate 70% front end width of the Bullet vehicle (Figure 1 & 2). The front end, left side of the bullet vehicle colliding into the rear end right side of the target vehicle.

The Target vehicle was propelled forward a distance of some 2.5 meters along the direct centre of mass movement.

A forward movement of some 2.1 m A lateral displacement towards the left side of some 0.6 meters. Some clockwise rotation of approximately 3.9 degrees. The rear wheels of the Bullet elevating to a height of 0.52 meters, determined through scaling from the right angle vide footage.

116 Collision Magazine - Volume 13 Issue 1

Volume 13 Issue 1.indd 116

Figure 3 – External left side view – crash sequence

www.collisionmagazine.com

5/28/2019 9:35:02 AM

Figure 4: Target – Movement parameters The Bullet vehicle moved forward a distance of some 4.0 meters along the direct centre of mass movement. A lateral displacement towards the right of some 2.5 meters. A forward movement of some 2.9 meters. Some clockwise rotation of 19.6 degrees. The visible forward movement of Target vehicle over the 2.5 meters occurred over a video footage time frame of 1.02 seconds, determined through time frames of the high speed video footage.

Figure 6: Pre & Post impact positions With or without the rear underride bar that has detached from the Target vehicle and remained attached to the front end of the Bullet vehicle, the overlap contact areas can be matched to a very fine level of accuracy, perhaps to within a few millimetres of orientation. This is an important consideration where the positioning of the Target and Bullet vehicles in respect of the layout of the road and specific pre impact positioning (Point Of Impact) is called to question.

Figure 5 – Bullet - Movement parameters

R

esults & discussion on evidence Post impact positions

All positions, pre and post vehicle and other evidential factors were recorded with the use of a Nikon NPR352 (Figure 6). The post impact positions are somewhat expected, where, although there is an approximate 4.4:1 mass ratio difference (against the Bullet), some forward displacement, rotation and or lateral displacement of the vehicles would be expected where the mass of the Bullet at 1457 kg at 87 km/h (24.16 m/s) indicates a Kinetic Energy of some 425464J, massive energy.

Damage to the Bullet vehicle (evidence from the vehicle): The damage to the Bullet vehicle is simply catastrophic and is expected given the parameters. Notably, modern vehicles are designed to absorb impacts and to collapse, however to maintain the general integrity of the occupant compartment, this appears to have occurred. www.collisionpumagazine.com

Volume 13 Issue 1.indd 117

Figure 7: Damage to the Bullet vehicle Collision Magazine - Volume 13 Issue 1 117

5/28/2019 9:35:05 AM

Damage to the target vehicle (evidence from the vehicle) The damage to the Target vehicle is arguably minor to moderate and is somewhat expected given the parameters. Notably, the underride bar has totally separated.

The left side two mounting bolts between the lower end of the chassis bracket and underride bar sheered. The right side chassis bracket sheered off at the mounting point between the main mounting bracket and chassis itself, at the point of the welded connection. The entire structure of the underride bar separating from the Target vehicle as a complete unit (Figure 7). The rear axle has sheered off at a position on the outer side (right side) of the spring mounting brackets (Figure 8), separating the complete right side dual wheel combination. Minor damage to the rear right taillight housing was incurred. Besides these damages, very little damage to the trailer appears evident.

Damages to the respective vehicles allows the primary analysis of the manner in which the two vehicles have collided, or phrased differently, the angle and overlap orientation of the vehicles relative to one another. This type of analysis is well document and pertains to the issue of the Principal Direction of Force (PDOF) 22,23. The particular relevance of this analysis is of understanding the positioning of the vehicles on the scene relative to one another and likewise allowing considerations of the movement of the vehicles through the phase of the crash and likewise the occupants.

The consideration of the damages to the vehicles in respect of measuring the damages, could provide further insight into the speed of the vehicle. This noted, ”crush analysis“ is a sensitive subject matter of its own. More so on modern vehicles were multiple strength materials are used, glues and not welding and specifically designed to deform. Although crush is not dealt with herein, it is reminded that this is an avenue of consideration and should therefore see the Bullet vehicle carefully measured as is required 22,23.

The Bullet vehicle damages measured right to left at 0,55 m from ground level to bumper leading edge height, with measurements of; 1 – 0,53 m; 2 – 0,56 m; 3 – 0,60 m; 4 – 0.64 m; 5 – 0,69 m; 6 – 0,74 m, Figure 9 guides.

Liquid spill – spatter & soak liquid spill, typically as a result of vessel damage, such as the radiator, sump, water bottle and gearbox or even CV joints almost always results from the Bullet vehicle. It is rare that liquid evidence will allow an accurate positioning of the Target or the Bullet vehicle as liquid is dynamic.

Nonetheless its presence and position should be noted as it does provide indication of the immediate area of impact.

Although not common practice, such liquid debris could

118 Collision Magazine - Volume 13 Issue 1

Volume 13 Issue 1.indd 118

Figure 8: Damage to the target vehicle also be used as a source of matching to determine from which vehicle it came.

The initial impact of the sump with the road surface causing rupture of the vessel and therefore the initial liquid (oil) spill at the immediate area of the sump and road surface contact. The movement of the Bullet from impact to rest sees the continued deposit of the liquid debris along this path. The resting position allowing the further draining and development of the liquid soak patch at that point.

All factors located and evident to a large extent at Figure 10.

www.collisionmagazine.com

5/28/2019 9:35:07 AM

Figure 11: General debris Figure 9: Bullet vehicle damage measurement

Specific debris: The total separation of the entire right side dual rear wheels to a position ahead of the bullet vehicle and to the right of both the target and bullet vehicle is notable. The wheels have separated as a result of the snapping of the axle, at a position on the outer side (right side) of the spring mounting brackets. Noting the orientation of the Bullet vehicle laterally (partial overlap) at impact and to the rear of the target, sees that the approximate centre line of the bullet vehicle and therefore the engine, would have been in the general centre line of the two rear right side wheels.

Although damages to the tyre and even the rim are not uncommon where impact occurs directly to the wheel, this does not appear to be the case here. The wheel combination has separated and essentially become a large section of vehicle debris from the target vehicle. However the wheel combination has largely remained unscathed, there were evident impact markings on the tyre, however the tyres remained inflated (Figure 12).

Figure 10: Liquid spill from rupture sump and other vessels

General debris

The position of the wheel is noted and recorded and serves as a general indication of some of the debris location in respect of the area of crash. Little specific information is gleaned from the location of the wheel as a piece of debris itself.

Some shards of glass, paint and small Bullet vehicle component pieces were noted as strewn around the general area. Although these may not be specifically recorded in each of their positions or specification, their presence is notable and should generally be recorded as this forms part of determining the general area of a crash. In certain cases these could serve to assist in determining what vehicles were involved if for example there was a colour matching or glass fragment matching undertaken.

Figure 12 – Specific debris www.collisionpumagazine.com

Volume 13 Issue 1.indd 119

Collision Magazine - Volume 13 Issue 1 119

5/28/2019 9:35:09 AM

Scratches and scrapes It is well documented that almost all crash scenarios will almost always result in some evidence on the road surface, scratches and scrapes often forming part thereof. Scratches and scrapes are on occasion overlooked, with emphasis placed on tyre mark evidence.

In cases where the scene may only be visited at some later stage, vehicles already removed, scratch and scrape marks may have greater longevity than that of other marks and therefore may be the only evidence notable.

In this particular scenario, the trailer landing legs left very distinct marks highlighting their point of origin, their movement as the target was propelled and subsequently their position of rest (Figure 13). In this particular case, this evidence is extremely accurate evidence as to the exact distance moved and the angle the target moved. This may be crucial in determining the exact positioning of the target if there was some dispute as to the target having been impeding the normal path of travel of the bullet vehicle.

This known would also form part of the distances required for possible calculative processes undertaken.

A relatively severe combination of scratch and scrape marks was created by the longitudinal orientation of the sump of the engine (Inline, six cylinder) of the Bullet. The sump engaging with the road surface as the vehicle bottomed out (Figure 10 & 14). The associated liquid spill of the engine oil is evident where the sump (Vessel) ruptured, liquids largely following the post impact path of the Bullet. This is a notable evidential factor that allows the extremely accurate positioning of the Bullet vehicle.

Figure 13: Scratches and scarapes

Tyre marks As the Target vehicle has moved, so the engagement of the tyres with the road surface from their point of origin to rest have created impact scrub marks. This is not unusual in such severe crash cases. This type of evidence is important and is arguably one of the first pieces of evidence that most scene investigators will look for.

In the same process as the Scratch and Scrape marks, this may be crucial evidence in determining the exact positioning of the Target vehicle (The tyre positioning and therefore the entire vehicle); if there was some dispute as to the target having been impeding the normal path of travel of the bullet vehicle. Tyre marks would be more common as landing legs would not normally be employed. The detailing of the tyre marks would also form part of the distances required for possible calculative processes undertaken.

120 Collision Magazine - Volume 13 Issue 1

Volume 13 Issue 1.indd 120

Figure 14: Scratches and scrapes Perhaps somewhat surprisingly, no tyre marks at all were notable from the Bullet vehicle. Given the severity of the impact and the resulting deflated front right tyre (The left side remained inflated), it would reasonable have been assumed that severe impact loading tyre scrub marks would have been evident. Tyre scrub marks from both vehicles would normally be searched for. The lack of tyre marks in this particular scenario may be (among other factors) as a result of cold tyres, the particular type of tyre and the con-

www.collisionmagazine.com

5/28/2019 9:35:14 AM

mated / separated. Closer inspection revealed the primary welding of the two main mounting brackets connected to the rear main structure cross member of the trailer (Target) was poor. Welding was only along one edge of each bracket interface to the chassis and likewise with poor penetration and overall concentration 38.

The underride separated with the left side bracket remaining secure to the chassis at the primary mounting point (welded). The lower section of this left side bracket separating from the underride where the two mounting bolts sheered. The right side main mounting bracket sheering from the main chassis mooting point on the Target trailer chassis (Welded), the lower section of this left side bracket remained connected (two bolts) to the underride (Figure 16)

Figure 15: Tyre scrub marks from point of origin to rest dition of the road surface, however this is not considered any further herein.

S

pecific evidence: Broken rear axle: Although not unseen, it is somewhat unusual to see the Axle having sustained such catastrophic damages (Figure 8 & 15). A more common result being the laceration and deflation of the tyre/s and perhaps even the damaging of the rims and even the damaging of the hub and various wheel mounting components. The Axle determined as a Swift, 92 mm square solid bar axle (9 ton).

Axle strength along with tyres are major factors considered during the design of a trailer, the axle and tyre combinations being the major factors in the permissible load mass of the trailer. The particular axle on this trailer being that of a now defunct axle manufacturer, Swift Axles, as such no specific detailing could be obtained for the particular model. Recording specific details of the axle could allow for calculative indications of sheer force and therefore the detailed photography, measuring and data plate specifications should be carefully noted.

Separation of underride bar The total separation of the underride bar as a complete unit is somewhat unusual. It is perhaps more common that severe bending of these is incurred or part of it is deci-

Figure 16: Underride bar & mounting bracket separation Telemetry & data therefrom The telemetry in the form of standard GPS based vehicle tracking systems (Geotab GO6)(37A)(37B) with on board crash sensing mounted to the Target and the Bullet vehicle was utilized. The detailed accuracy thereof not interrogated at this stage and accepted as being reasonably accurate in respect of the basic data. The Bullet vehicle telemetry selforientating as the vehicle travelled pre testing, therefore provided accurate and correct axis direction speed readings as the vehicle progressed. (Figure 17) Unfortunately, impact caused a power loss and further force readings and specific crash telemetry readings to the unit from impact were not held.

The positioning of the telemetry on the Target (Figure 18) did not allow for self-orientating as the device was not

www.collisionpumagazine.com

Volume 13 Issue 1.indd 121

Collision Magazine - Volume 13 Issue 1 121

5/28/2019 9:35:16 AM

Figure 17: Speed and time of Bullet vehicle driven“ pre testing. As such the telemetry is read in the device standard orientation mode, effectively in reverse.

Nonetheless, the Target vehicle readings (Figure 19) peaking at some 5 g (Series s1 - fore and aft). Such high reading not surprising given the high speed impact and largely direct inline impact and largely forward displacement of the Target. An approximate 1.4 g lateral (Series s2 - Side/side).

Such reading not surprising given that there was lateral displacement of the Target. An approximate 3.0 g upward (Series s3 – Up/Down). Such reading not surprising given that the front end of the Bullet vehicle is relatively low and largely collided at a level at or just lower than the centre line of the rear tyres diameter and into the rear underride bars.

The recorded time frame specifications of the data indicate movement recorded from impact to rest of the Target over the distance of 2.5 meters at 2.1 seconds. This does however include residual movement of the Target where it may be stationary in its displacement, but with some residual movement still recorded by the device.

The known static position of the Target and the telemetry indicated speed of the Bullet vehicle (87 km/h – Figure 17) provides a reasonable basis against which norms can be considered.

Other evidence Although easily available in a ”setup“ session of this nature, it is reminded that during actual crash scenes, noting the manufacturers details and the specifications of the Target and Bullet vehicle is crucial. Necessary details would typically be sourced from a combination of data plates, licence discs and registration number detailing. These allow, among other factors, details of weights, load limits, 122 Collision Magazine - Volume 13 Issue 1

Volume 13 Issue 1.indd 122

Figure 18: Telemetry mounted to the Target vehicle size, axle specifications and brake specifications. Recording details of tyres on both vehicles is also important.

Details of loads on the Target and Bullet vehicle should also be recorded, typically from load manifest, however visual and photographic inspection thereof is highly recommended. Similarly, detailing loads in the Bullet vehicle, such as number of occupants, any baggage/luggage and perhaps careful consideration of large aftermarket fitments like large sound system speaker boxes often placed in the boot, could add substantial overall weight.

It may be necessary in certain considerations to know specific weight distribution on the Target vehicle between the wheels and landing legs. Similarly, it may be necessary to know weight distribution on the Bullet vehicle. Such values may be needed in determining Centre Of Mass. This may not be possible in a real crash situation post-crash due to damage and or load movement. If not possible, it may

www.collisionmagazine.com

5/28/2019 9:35:17 AM

S1 - Fore/aft

s2 - Side/side

S3 - Up/down

Figure 19: Target vehicle telemetry be necessary to locate an exemplar vehicle and undertake necessary testing.

Gradients and super elevation, road surface coefficient, advanced warnings if any and sighting obstructions, either

permanent or momentarily (weather / sunblind and the like) and medical conditions may be other factors to consider. None of these factors were of necessity in this particular session.

www.collisionpumagazine.com

Volume 13 Issue 1.indd 123

Collision Magazine - Volume 13 Issue 1 123

5/28/2019 9:35:17 AM

C

onclusions: The opportunity to undertake a full scale, controlled environment, crash test of this nature is far and few between. Although in time, improvements in technology will create modelling a crash and considering actual crash telemetry even more accurate, a real life crash test will always provide the ultimate results. The documenting of such scenarios is import in the preservation of the data, so that this may be considered in further development.

Even where there will no doubt be a time in the future where vehicles are largely fully automated and self-driving, where things may go wrong there will always need to be an understanding of what happened. This type of high speed crash will remain a demanding safety concern.

Likewise it is these historic understanding and parameters that provide for insight into the development and use of such systems and procedures in the future.

It is immediately noted that the session presented highlights the plethora of evidence that can be created and needs to be recorded for useful analysis to some extent or another. Although the majority of information needed can and usually is collated on scene, the appropriate removal and storage of the vehicles themselves can be a massive assistance in real cases.

Similarly, the careful identification of the evidence on scene and accurate recording thereof is crucial, keeping in mind that if not identified and recorded, this evidence is typically ephemeral.

Although there are many characteristics that will be similar in most rear end crash scenarios, the particular incident of a high speed partial overlap, sedan or similar vehicle into a commercial trailer, will present with characteristics that are distinct to and/or far more pronounced.

Presented with such a scenario to consider, where there is no doubt that an investigator would find themselves faced with just this at some stage, requires careful consideration and meticulous detailing. With the careful consideration and accurate recording of all the necessary data, any questions arising surrounding the crash could be answered in certainty, or at least to a very strong balance of probabilities.

All rear end crash scenarios are hazardous, however the particular scenario considered, suggest that had there been occupants, that there is a very high probability that fatalities, or at least very serious injuries would have been incurred.

This therefore is in strong correlation to other research highlighting the dangers of vehicle incompatibilities and in particular research pertaining to human force limits.

124 Collision Magazine - Volume 13 Issue 1

Volume 13 Issue 1.indd 124

The scenario presented certainly highlights and confirms the opinions already set out in many research papers and general commentary in articles, that incompatibility between vehicles is a major problem. The session supports the scope for further research and improvement in design of compatibilities between vehicles and arguably stronger regulation of designs to improve compatibility. This session also provides impetus for improvements in a range of safety aspects such as improved structural strength, improved impact absorbing qualities. Likewise, although already in use, further justification for research and improvements in technology driven safety mitigation products.

Further and once again an issue that has been highlighted in other research and commentary, this scenario again highlights the need for careful consideration of underride devices. The regulation there of in respect of the requirement to be fitted and to meet minimum standards and improvement thereof.

The identification of the many specific evidential characteristics between the vehicles themselves and evidence on the scene itself, however most notably those that allow the positioning of either vehicle on the road surface at pre impact position is almost always the crucial factor in such real life cases. The positioning or orientation of the vehicles relative to one another at impact are key to answering the likely question of ’where were the vehicles positioned at impact“ and therefore many other questions that cascade. All of the evidence carefully collected will either individually or as a whole provide the answers to questions raised. Tyre marks from the bullet vehicle may not necessarily be located, therefore other evidence as documented may be heavily relied upon.

Just as in the case here, the reliance on telemetry to provide answers should not be the case. The identification and collection of evidence should always serve to bolster or validate telemetry evidence.

A

cknowledgement: The authors would like to acknowledge the assistance and work that was undertaken by the staff of Accident Specialist as part of the development and setup of the test as well as post event. Likewise to those that volunteered to assist in the many pre event, event and post event tasks. The authors thank Klein Janse Van Rensburg for the extensive and persistent development work undertaken on the remote steering system developed and used for the test. Thank you to the various sponsors that provided funding, equipment or facilities, without which the testing would have been difficult to undertake.

www.collisionmagazine.com

5/28/2019 9:35:17 AM

R

eference & bibliography: 1. Steer Induced Loss of control of a minibus on a wet surface: K Setty; C Proctor-Parker; R Stopforth; S Davraj / 19 May 2017

2. Driver attributes and rear-end crash involvement propensity: DOT HS 809 540 March 2003 – Technical Report - Published By: National Centre for Statistics and Analysis Advanced Research and Analysis 3. Underride in rear-end fatal truck crashes: Submitted to National Highway Traffic Safety Administration prepared by - Daniel Blower / Kenneth L. Campbell - Centre for National Truck Statistics - The University of Michigan Transportation Research Institute -October 1999

12. Analysis of Event Data Recorder Survivability in Crashes with Fire, Immersion, and High Delta-V: H. Tsoi / John Hinch / H Gabler 13. The crash compatibility of cars and light trucks: Hampton C. Gabler - Department of Mechanical Engineering Rowan University Glassboro, NJ 08028 gabler@rowan.edu / William T. Hollowell National Highway Traffic Safety Administration NRD-11 400 Seventh Street, S.W. Washington, DC 20590 thollowell@nhtsa.dot.gov 14. NHTSA’S Vehicle aggressivity and compatibility research program: Hampton C. Gabler / William T. Hollowell - U.S. National Highway Traffic Safety Administration United States Paper No. 98-S3-O-01

4. Heavy-vehicle crash data collection and analysis to characterize rear and side underride and front override in fatal truck crashes: DOT HS 811 725 March 2013 - Blower, Daniel; John Woodrooffe

15. Vehicle Weight, Fatality Risk and Crash Compatibility of Model Year 1991-99 Passenger Cars and Light Trucks: DOT HS 809 662 October 2003 NH TS A Technical Report

5. Optimization of Rear Signal Pattern for Reduction of rear-End Accidents during Emergency Braking manoeuvres: Federal Highway Research Institute - Dr. rer. nat. Jost Gail / Dipl.-Ing. Mechthild Lorig / Dr. phil. Christhard Gelau / Dipl.-Phys. Dirk Heuzeroth / Dr.-Ing. Wolfgang Sievert / Bergisch Gladbach, November 2001

16. Analysis of Four Staged Crashes of Passenger Vehicles into a Semi-Trailer: Jeremy Daily / Russell Strickland / Institute of Police Technology and Management Special Problems in Traffic Crash Reconstruction 25-29 April 2005

6. Rear end crashes: Centre for automotive safety research / Te University of Adelaide – Australia – Mr J Baldock, AD Long, VL Lindsay, AJ McLean - CASR REPORT SERIES / CASR018 September 2005 7. Vehicle and infrastructure-based technology for the prevention of rear-end collisions: National Transportation Safety Board Washington, DC. 20594 special investigation report pb2001-917003 NTSB/sir-01/01 8. Commercial Motor Vehicle Crash Investigation: David Brill / Published by: Institute of Police Technology & Management – University of North Florida – USA 9. Kinematic behaviour of the human body during deceleration: 62-3 Federal aviation agency aviation medical service Aeromedical research division civil Aeromedical research institute protection and survival branch Oklahoma City, Oklahoma 10. Human tolerance to abrupt accelerations a literature review: Dynamic Science report 70-13.

11. Human Tolerance and Crash Survivability: Dennis F. Shanahan, M.D., M.P.H. Injury Analysis, LLC 2839 Via Conquistador Carlsbad, CA 92009-3020 USA

17. Crush Analysis with Under-rides and the coefficient of Restitution: Jeremy Daily, Russell Strickland, John Daily 27 April 2006 18. Heavy-Vehicle Crash Data Collection and analysis to Characterize Rear and Side Underride and front Override in Fatal Truck Crashes: DOT HS 811 725 March 2013 / Authors: Blower, Daniel; John Woodrooffe 19. Factors contributing to commercial vehicle rear-end conflicts in China - A study using on-board event data recorders: Journal of safety research Giulio Bianchi Piccinini, Johan Engström, Jonas Bärgman, XuesongWang 20. The crash compatibility of cars and light trucks Authors: Hampton C. Gabler Department of Mechanical Engineering Rowan University Glassboro, NJ 08028 gabler@rowan.edu William T. Hollowell National Highway Traffic Safety administration NRD11 400 Seventh Street, S.W. Washington, DC 20590 thollowell@nhtsa.dot.gov 21. Variation of crash severity and injury risk depending on collisions with different vehicle types and object Authors: Helena Stigson, Anders Ydenius, Anders

www.collisionpumagazine.com

Volume 13 Issue 1.indd 125

Collision Magazine - Volume 13 Issue 1 125

5/28/2019 9:35:17 AM

Kullgren - Karolinska Institutet, Sweden, Folksam Research, Sweden 22. Measuring protocol for quantifying vehicle damage from an energy basis point of view: Society of Automobile Engineers (SAE) International, 1988.

23. Surface vehicle standard - Collision Deformation Classification: (SAE J224), Socienty of Automobile Engineers International.

24. Surface vehicle information report – sign convention for vehicle crash testing: SAE J1733 / December 94 25. Rear Impact Guards, Rear Impact Protection; Proposed Rule: Department of Transportation National Highway Traffic Safety Administration 49 CFR Part 571 - Vol. 80 Wednesday, No. 241 December 16, 2015 Part III 26. Profiling the safety needs of the South African truck transportation sector: C Roberts Key comments: “Upon examination of these results, it became clear that most legislation is old, strategies are not always executed and followed up, and very little effort is spent on research and development.

During 1996 the White Paper on National Transport Policy (SA, 1996:3) has been accepted by the Government and during 2007 a Road Safety Conference was held in Ghana, where a road safety strategy with some goals was discussed (see Par. 2.3). To date no significant progress has been made, thus there are concerns that the goals set by this sector, will not be met.

Labuschagne (2008:1) alludes to the fact that improved access to accident data by researchers greatly benefits general research and development in the field of road safety.

The human factor is by far the biggest reason for road accidents. Drowsiness, substance abuse or poor judgment, are just some of the reasons why there are so many accidents in South Africa (Radebe, 2010). For the purpose of this paper, the following three main categories will be discussed: The human factor; indirect non-human factors; the effect of company policy and procedure on the occurrence of road accidents within the LHHV transportation sector.

In a study done by the CSIR on driver fatigue on the N3 between Warden and Villiers, it has been highlighted that driver fatigue plays a major role in road accidents (Venter, 2013). According to Venter, the results of this decline in performance are reduced vigilance, reduced attention or awareness and an increase in drowsiness and fatigue.

126 Collision Magazine - Volume 13 Issue 1

Volume 13 Issue 1.indd 126

Venter (2013) further alludes to the fact that the consequences of fatigue include slow reaction time, slow control of movements, decreased tolerance for other road users, poor judgement when overtaking, and loss of situational awareness. It is also considered an internal distraction for the driver that leads to poor decision making and cognitive impairment. In this study, specific reference has been made to fatigue related accidents. In order to compile the report, a total of 790 accidents that occurred from 2007 to 2010 have been analysed. Of these 790 accidents no less than 346 involved long haul heavy vehicles. The time of the day the accidents has occurred, is highlighted in the study. It has been found that during the period in question, there have been 213 head-to-tail accidents, 195 accidents in which a single vehicle was involved, and 142 vehicles left the road causing an accident, leaving no signs where brakes were applied. In other words, there was an absence of skid marks or other signs of harsh braking. Police reports also indicated that the driver, for a few seconds, could clearly see the road ahead. In the majority of accidents the weather was good, and in some cases witnesses reported the vehicle drifting from the one lane to the other, also called ‘lane drifting’, prior to the accident, indicating that fatigue might have had a causal relationship to the accident 27. The Traffic Accident Investigation Manual – At Scene Investigation and Technical Follow Up: Northwestern University, Traffic Institute / University North Florida, IPTM. 28. Traffic Accident Reconstruction: Northwestern University, Traffic Institute / University North Florida, IPTM. 29. https://www.consumerreports.org/cro/2011/08/ crash-test-101/index.htm (Online - 2011) IIHS rear-impact evaluations Key comments: “Though common, not many rear-impact crashes are fatal. But they do cause many injuries, especially whiplash trauma to the neck. The IIHS evaluates rear impacts with physical inspections and crash testing.

The crash test, which is conducted with the vehicle seat attached to a moving sled, simulates a rear-end crash about equivalent to a stationary vehicle being struck at 20 mph (32 km/H) by a vehicle of the same weight.

30. http://www.thetruthaboutcars.com/2011/02/has-thetime-come-for-rear-crash-testing/ (Online 2011) Key comments: “No government requires rear-crash testing, but in the wake of several accidents, Germany’s AutoBild magazine decided to look into what exactly happens

www.collisionmagazine.com

5/28/2019 9:35:17 AM

when a car is hit from behind at 64 km/h… and the results are not encouraging.”

31. https://trid.trb.org/view.aspx?id=673828 (National Academy Of Sciences – US / Society of Automotive Engineers) High speed rear impact crash test

partment typically crushes as the truck body intrudes into the vehicle safety cage.”

33. https://www.trucks.com/2016/05/06/traffic-expertsdebate-how-to-prevent-deadly-truck-crashes/ Key comments: Big trucks “are not in any way crashfriendly,” said Robert Molloy, director of highway safety at the National Transportation Safety Board.

Key comments: “Produced for the Society of Automotive 34. http://www.nphm.com/wp-content/upEngineers, this video demonstrates the impact of a heavy loads/2014/10/Piercing-The-Passenger-Compartment1. truck into the rear of a four-door, compact sedan. This pdf video demonstrates the plans, photography, instrumentation, techniques and procedures used in a crash test. The Key comments: No matter how safe the car may actually supplemental notebook contains pre and post-crash test be, the safety features are only effective if there is good photos, time histories of accelerations and displacements structural interaction (Crash Compatibility) between coland the test data analysis. lision partners. This means there is a geometric match up of the crush structure of both the striking vehicle and the SUMMARY: This NPRM proposes to upgrade the Fedvehicle being struck. eral motor vehicle safety standards that address rear underride protection in crashes into trailers and semitrailers. 35. Judgements: NHTSA is proposing to adopt requirements of Transport A. Paul Dumisani Dlamini v Road Accident Fund – ApCanada’s standard for underride guards, which require peal No. A12/2012 - Free State High Court, Bloemfonrear impact guards to provide sufficient strength and entein, Republic of South Africa. Judgement – Molefe, AJ ergy absorption to protect occupants of compact and subcompact passenger cars impacting the rear of trailers at B. Ravenhill Transport LTD Vs Hultrans Ltd Case 56 kilometres per hour (km/h) (35 miles per hour (mph).

NHTSA is issuing this NPRM in response to a petition No. 7179/89 – Supreme Court Of South Africa Coast Division 17th September 1989 (Bristowe, J.) for rulemaking from the Insurance Institute for Highway Safety (IIHS), and from Ms. Marianne Karth and the 36. https://www.jp-sa.org/wp-content/uploads/2017/06/ Truck Safety Coalition (TSC). This is the second of two National-Road-Traffic-Act-Regulations-20150725.pdf documents issued in response to the 37A. http://gpstrackingcanada.com/geotab-user-guideKarth/TSC petition. Earlier, NHTSA published an ad- geotab-go6-hardware-specifications.html vanced notice of proposed rulemaking requesting comment on strategies pertaining to underride protection af- B. https://www.geotab.com/wp-content/themes/geotabtemplate/resources/doc/GO6-Support-Doc-Rev14.pdf forded by single trucks.”

32. http://www.iihs.org/iihs/news/desktopnews/underride-guards-on-big-rigs-often-fail-in-crashes-institutepetitions-government-for-new-standard IIHS News | March 1, 2011

38. Welding standards / specifications: ASME – American Society Of Mechanical Engineers - section IX; AWS - American Welding Society D1.1; ISO – International Standards Organisation – ISO3834

Key comments: Underride guards on big rigs often fail in crashes; Institute petitions government for new standard: “Rear guards are the main countermeasure for reducing underride deaths and injuries when a passenger vehicle crashes into the back of a tractor-trailer. In 2009, 70 percent of the 3,163 people who died in all large truck crashes were occupants of cars or other passenger vehicles.

Underride makes death or serious injury more likely since the upper part of the passenger vehicle's occupant comwww.collisionpumagazine.com

Volume 13 Issue 1.indd 127

Collision Magazine - Volume 13 Issue 1 127

5/28/2019 9:35:17 AM

Get the facts of the crash get answers quickly

Helping you understand the crash evaluate causation

validate physical evidence

corroborate witness statements

With proven admissibility in court, the Bosch CDR Tool is the worldwide standard used by law enforcement for obtaining unbiased, critical crash data from vehicles. Equip your investigators with the trusted tool to get answers quickly, evaluate causation & physical evidence and corroborate witness statements producing accurate incident reports.

In addition to selling the Bosch CDR Tool, Crash Data Group also assists with training needs, assuring you understand how to use the tool properly.

The CDR Tool allows you access to crash data from the majority of vehicles on the road today. Because of the impact the data provides to the investigation process, this tool has become the essential tool for law enforcement and accident inviestigators.

www.crashdatagroup.com

Volume 13 Issue 1.indd 128

5/28/2019 9:35:17 AM