I am a final year PhD student in Mechanical Engineering. My PhD comprises only numerical work with a Finite Element Method software, and structural optimization with a developed MATLAB code. I have already a good number of publications, and a few innovations in my project, like structures with a new configuration. However, my supervisor tells me that I need to validate my numerical results with an experimental test. Do I really need to validate my results experimentally to have a "safe" work, I mean, work with results that I can trust? I did a mesh sensitivity analysis for every geometry I have tested.I am sure the mesh is fine enough to obtain converged results. Is it safe to assume the accuracy of the numerical results? And if numerical work alone is not enough, is there any alternative to experimental testing, that would delay my thesis defense and is expensive (allocation of funds is very difficult right now due to crisis in EU?"

I am asking this because I don't have any experience in experimental work, and this will delay my thesis defense 1 year or so. I don't plan on working with anything experimental after my PhD graduation. I would like to have arguments to convince my supervisor that I can do my PhD without experimental work, presenting results in my thesis that are accurate, and that anyone can trust.

I am asking for help of Mechanical Engineering PhD graduates, specialists in numerical work. I would also like to know if you did experimental testing to prototypes in your work, and if not, the arguments you had to defend yourselves in the thesis examination, if you confronted with questions like :"why haven't you done experimental work?"

P.S. I know someone that did his PhD without experimental work, but it was in a different field, like Biomedical Engineering, or so.

  • Sounds like you have leaned much heavier towards applied mathematics than straight up mechanical engineering. This is totally acceptable, and many people across many fields are doing this. That being said, your advisor knows best about your specific situation. And if he/she says you need an experimental validation, then that's probably a good idea. Dec 10, 2015 at 17:08
  • 10
    My supervisor tells me that I need to do X. Do I really need to do X? — As a general rule, unless you want to find a new supervisor or just not finish, yes, you do.
    – JeffE
    Dec 10, 2015 at 17:55
  • You need to determine whether your advisor wants you to do some experimental work because he believes it is good for you to have experience conducting experiments, or whether you just need the results to verify your simulations (further evidence that your findings are correct). In the latter case, I would see if you can get away with running some simulations to reproduce some other experimental results from the literature. Dec 11, 2015 at 16:59

5 Answers 5


I'm an expirmentalist, and I have worked with FE modeling for a long time now. I have many publications based on tests, validated models and purely numerical "without validation". I can tell you this much, the ones that keeps me awake at night and the hardest ones to publish are those with "numerical models without any experimental tests" to back the numerical predictions up or at the least include some sort of validation.

My MSc (structural engineering) was all numerical (but based on full scale tests done by others), so I did not have to test my own specimens (validated my models against experiments that were already published). I can tell this much, without a validation, your model is just a model! As you know FEA is just an approximation with multiple simplistic assumptions of a real phenomenon.

However, I tend to agree that whether needing to validate your model against measured data (solely by doing your own experiment or simply against other published work) is case specific. I do not know your "thesis problem" and the fact that you have published a good number of articles (if published in high quality journals) and your advisor was fine with that "until now!" leads me to think that you may be able to get away to a "numerical" thesis.

Perhaps you can discuss the following ideas with your advisor/committee;

  1. If they insist on an experiment (although they are tight on funding as you have mentioned [not a very good sign in your case since you are trying to avoid that!]), try to find a somehow similar published work, use your model to validate it. I understand that this may not be a "direct validation" but at least your advisor/committee can be assured that your model is capable of simulating a real problem. In other words, your model is "not-directly" more than an FE model.
  2. Ask if you can validate your model with some other ways i.e., analytically, hand calculations, published methods etc. (I hope you have done this by now!). You can always do a parametric study (I assume you did already), pick parameters that can relate to an actual problem (either from literature or industry) and see how your model performs. If possible/applicable, contact an industry firm (that is willing to share) and see if they have anything on hand that you can use to validate the model.
  3. Discuss with your advisor why did not he ask you to do this long time ago ("usually" one will do an experiment and then develop a model! [that's what I always do, keep in mind this is different that finding a problem and trying to get some funding for in which you will develop a model first to show the feasibility of your problem rather than spending money on an experiment that turns out to be "worthless"]). I understand this is too late to ask now, but if you know the "inside" of whats going on "with your committee (perhaps?)", you will know how to tackle this issue more specifically.

Remember this, you do not have to prepare a 6 months, $ 100k experiment to validate your model. Sometimes, all what you need is a well organized set-up, few strain gauges, loading-inducing actuators and a specimen (of course this is a simplification - but you get my point).

  • I would like to thank "The Fire Guy" for the very helpful insights into the matter. Thank you!!
    – KingBaboon
    May 2, 2016 at 21:15
  • @KingBaboon I'm glad My answer was of help!
    – The Guy
    May 2, 2016 at 22:28

In an industry setting, numerical simulations are often used to rule out bad designs before the prototype phase to reduce costs. But the end product of that process is a prototype, not a thesis.

Validating FE simulation with experimental data is standard for academic works in that field. It's going to be hard to argue, on a scientific level, that it is not useful to run experiments.

my supervisor tells me that I need to validate my numerical results with an experimental test

This is a strong reason to believe it's necessary in order to write a thesis that will be recognized as valid in your field. Beside the fact that you probably want to graduate, the purpose of the PhD is to give you the necessary background and credential to pretend to good academic positions.

allocation of funds is very difficult right now due to crisis in EU

If your adviser requires you to do more work, then (s)he should arrange funding. You shouldn't be worrying about that bit.


If your adviser says that you need to do, you don't have a whole lot of options. After all, the first one who needs to sign off on all pieces of paper you need to graduate is your adviser, and if your adviser doesn't sign it, none of your other committee members will either.

In either case, it's a topic to discuss with your adviser.


It's definitely possible. My Aerospace engineering PhD was entirely numerical, and only included some peripheral comparisons to the experimental work of others. ME and ASE are pretty close in a lot of ways, but this is going to depend more on what your advisor and your committee think than anything else. It's definitely possible to do a numerical-only PhD in engineering, though.

  • The point is: Does a commerical Finite Element software ANSYS Mehanical APDL, if used properly, provide reliable results?? This is what I would like to know. German companies are a lot focused on FEM engineers, but there is still the thinking that "numerical is an approximation", and the real results are experimental. Can you comment on this ideas? And can you send me the reference of your thesis, so I can have arguments to my teacher? Many thanks for your reply!!
    – KingBaboon
    Dec 10, 2015 at 17:07
  • 1
    @KingBaboon, there are plenty of companies that use or are attempting to use such software to completely eliminate physical tests. However, if that's your real question, maybe you should edit your posted question with this updated information. My answer is unrelated to this update.
    – Bill Barth
    Dec 10, 2015 at 17:16
  • Hi Bill, No, that's not my real question. However, it's a question whose answer can give me arguments to discuss my position with my supervisor. Thanks for all the help. Can you post a link to your PhD thesis, if it's published online?
    – KingBaboon
    Dec 10, 2015 at 17:42
  • @KingBaboon, it's not published online.
    – Bill Barth
    Dec 10, 2015 at 17:59

In real-life mechanical engineering, numerical simulations are generally used to reduce the number of physical prototypes that need to be built and tested. In some situations, the simulation results are so well trusted that no physical prototype is ever built -- people go directly to production. But, where human lives are at stake (passenger aircraft, automobile crash testing, etc.) physical prototypes are necessary.

There are a few situations where it's not practical (or it's prohibitively expensive) to test physical prototypes. Examples are spacecraft that are designed to function in zero gravity. For those situations, like it or not, numerical simulation is all you have.

But you're not doing real-life engineering, you're writing a thesis. So, in your situation, as others have said, you probably have to do what your advisor says. It's worth arguing for a while to see if you can change his/her mind, but ultimately you are powerless, I suspect.

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .