What is Validation for vehicle to vehicle collision models?

General questions on the SMAC, msmac3D and other Collision Simulation programs
MSI
Site Admin
Posts: 2302
Joined: Thu Jun 18, 2009 12:37 pm

What is Validation for vehicle to vehicle collision models?

Post by MSI »

For convenience here's a quick link to the latest post in this topic

For vehicle to vehicle collisions methods validation, obviously cars are mass produced and so full scale tests can be performed from which validations of computer models can be ascertained (RICSAC tests, etc).
There was a group at SAE 15 years or so ago that I participated in which was trying to come up with a validation comparison of computer models.
  • First item: What Tests to use for validation'?
    • Some of the more recent full scale tests (MacInnis, various NYSTAR, etc) have problems with post-impact speeds, etc due to post-impact arbitrarily steering and braking which were not documented (with limited test space they needed to brake/steer the vehicles to prevent errant vehicle problems and did not keep track of the extent and magnitude of post impact steering and/or braking).
    So with the RICSAC set of data (and in consideration of the results of our 1997 paper) the next requirement for a validation of a computer program would be:
  • Second Item: What are the inputs required for each technique to reconstruct the collision?
    • Many techniques have inputs which can be somewhat subjective, particularly when performing a validation run (since the answer is known in most validation tests), and so the question is:
      • How can the technique be applied in an objective way, what are the inputs required, such that a dozen people applying it to a particular collision will get the same results?
      In 2003 we wrote an SAE paper 2003-01-0486 "SMAC2003 - The Automatic Iteration of SMAC" which was a demonstration of that with a ‘hands off’ reconstructions of the RICSAC tests, starting with linear momentum results (from CRASH) and then allowing a computer algorithm to automatically iterate mSMAC to try to match the evidence, that mSMAC can produce results within 10% of the 'truth' with mSMAC for both vehicle impact speeds and impact speed changes (DeltaVs).
That brings up the next phase of a validation process:
  • Third Item: HOW to present the validation results?
    • The use of ‘combined impact speeds’ and ‘95% confidence’ as suggested by Terry Day and Engineering Dynamics in their validation of the EDCRASH program Further Validation of EDCRASH Using the RICSAC Staged Collisions, SAE paper 89-0740 is an attempt to hide the over 40% errors which can occur in some of the individual reconstructions included in the validations. It is misleading and inappropriate.
      I shudder when I wonder how many people have been wrongly convicted by police using EDCRASH or similar techniques while the police claimed accuracy which isn’t really there.
      Sure in many cases, CRASH and momentum techniques are reasonably accurate.
      • However how do you know if the case you are examining is one of the 40% error cases or one which is reasonably accurate?
        You should always test and refine any momentum/CRASH type results by application of simulation or other technique to verify the results
Please read the introduction in our 1997 SAE Paper 97-0949 which includes a discussion of Validation which also sheds light on the topic.

This topic has 2 more posts with additional information

To Read more, Please login and/or register. 2024 NOTE: Soon ALL Technical Sections will be ONLY for registered users. Optionally you can email us forum@mchenrysoftware.com your Name, Company, Location, a Username, and a Password (which you can change) and we will register you and send you a confirmation email.


Register Login