So this is very general, but... How are simulations for vehicles done? I know that computational fluid dynamic simulations are used, but why are repeated tests of actual hardware done too? Obviously this is because there are some things that are found through actual physical testing. So this leads into my question. Are simulations refined from the actual test data? how are phenomena from the actual test translated into simulation? Eyeballing things? just gathering data, making a hypothesis, and then running the simulation again from the top? I guess what my question ultimately boils down to is: What does a simulation cycle look like?

  • $\begingroup$ Are you familiar with FEM? $\endgroup$
    – user14407
    Nov 16, 2018 at 12:28
  • $\begingroup$ I am not. This isn't my main field of study but it's something I'm interested in $\endgroup$
    – lhubbard01
    Nov 16, 2018 at 12:40
  • $\begingroup$ This is a very broad question, for a general answer look at the detailed answer below, but i give specific application, crash test, in simulation we use FEM to predict the final results, twenty years ago, it wasn't really reliable, so engineers tried to make the simulation as consistent as possible with the actual crash test data. These days, FEM is really close to reality for a simple geometries and stress conditions, the results are much consistent with reality. Some years ago actual test data used as a source for simulation these days things are going to be in reverse, but not yet. $\endgroup$
    – user14407
    Nov 16, 2018 at 14:16

1 Answer 1


That is a very broad question. It can be as "simple" as visually comparing the simulation and test results, and deciding on the basis of experience how the tweak the simulation if it isn't accurate enough (whatever "accurate" means, in a particular situation). After the model has been changed to agree with the test data, you can then use it to simulate other conditions which were not tested, with more confidence that the results will be meaningful.

At the other extreme, this may be treated as a mathematical optimisation problem. You set up a model with some parameters (geometry, material properties, etc) that can be varied, tell the software what the "answers" should be (as measured in the tests) and the software then runs the model repeatedly, changing the parameters to get the "best" agreement.

The problem with a fully automatic approach is that the initial model may be completely unrealistic (for example because some important feature was left out altogether, either by mistake, or because nobody thought it was important enough to be worth including!) but the software managed to make the answers line up with the test data by using unrealistic values for the parameters, and getting to a situation where several "large errors" cancelled out but the final results "looked about right".

It is fairly routine to use "automatic model generation" in a more limited way. For example in finite element analysis, you might define the geometry of the component, and the software automatically generates the finite element mesh, and then refines it by subdividing some of the elements in regions where the numerical solution hasn't converged, according to some predefined measure of "converged". This is usually much quicker (and more accurate) than generating the complete mesh by hand.

The amount of work involved in validating models can cover a huge range, from one engineer working for part of a day up to several person-years of work for a large, complex, and critically important model.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.