Leverging AI on Restraint System Design

Subjects related to the reconstruction and simulation of Occupants in vehicles and Pedestrians struck by vehicles, ATB & ATB clones #ATB #MADYMO #MultiBody
MSI
Site Admin
Posts: 2302
Joined: Thu Jun 18, 2009 12:37 pm

Leverging AI on Restraint System Design

Post by MSI »

Ph.D. research.png
Ph.D. research.png (460.12 KiB) Viewed 5148 times
From Linkedin page by Hamed Joodaki
  • For my Ph.D. dissertation, I wanted to optimize the restraint system for two different anthropometries. The way that I approached this problem was that first, I ran a limited number of simulations using finite element (FE) models of the human body and restraints. Then, I used the results of those simulations along with machine learning to develop mathematical models (metamodels) of the FE results. Those mathematical models were capable of predicting the behavior of the system, with any given set of restraint parameters, and with no need to run a full FE simulation. Finally, I applied a genetic algorithm to those mathematical models to find the global optimum solution.
    A part of this work is now published in CMBBE.
    Title: Leveraging machine learning for predicting human body model response in restraint design simulations
    Link: https://lnkd.in/eTxAziN
    • In this article, it is discussed how essential it is to avoid over-fitting and under-fitting if we want to use metamodels for optimization. On the other hand, some available and widely used commercial software for optimization via FE modeling does not have this feature. This paper shows how over-fitting and under-fitting can be avoided by tuning the hyperparameters of a metamodel.
Comment:
  • As you progress in your studies be sure to also try to include actual measured full scale testing in your research either staged for the project or from the many tests contained in the literature.
    Full scale testing is important to act as a baseline/range/reality check on simulations, especially in occupant simulation (since no two humans are alike and so most full scale testing is with ATDs which are physical models of humans).
    The marriage of full scale testing (which can be expensive) and simulation/technology (which gets more powerful and cheaper every year) have brought about significant improvements in safety and technological advances for decades.
    Now metamodels and AI will further enhance and expand our understandings of the important safety issues related to car crashes, occupant safety and many other important things.
    Congratulations on a very interesting project for you Ph.D. research!