The results are discrete values, where two parameters are varied for each reading (load and speed). Each reading is calculated only once (no replications).
Please advise if I can use R^2 (R squared) as a measure for accuracy of results in comparison to the experimental results and advise how can I do calculations.
Any suggestions for other measures/ parameters to determine accuracy?
Best regards!
- Suppose you can compare with different load & different speed with different materials then in the Result of like Diformation & stress of higher value or upward side colors shows results its use for comparisons with any products.
Hi
There are ways to evaluate Test/FE matching. One such measure is the so called Modal Assurance Criterion (MAC), where shape A is compared with shape B and a MAC value larger than 0.8 implies that A=B. MAC is robust, ouputs a scalar number and reveals differences very efficiently.
You find a discussion on this topic here
http://qringtech.com/learnmore/why-simulate-measure-correlate-automate/
and an example paper containing Test/Test, Test/FE and FE/FE model correlation here
https://www.researchgate.net/publication/252864317_SM98-124626_Force_Measurements_during_Vibration_Tests_with_Sinusoidal_Base_Excitation?ev=prf_pub
and a paper on model correlation here
http://www.ias.ac.in/sadhana/Pdf2000June/Pe872.pdf
Some other thoughts on CAE and modelling is found here
http://qringtech.com/wp-content/uploads/2014/01/AnalysisTechnology.pdf
Sincerely
Claes
Article SM98-124/626 Force Measurements during Vibration Tests with ...
Thanks Claes!
Thanks for your detailed answer. Can I use MAC to choose the best trial among a set of trials. I am attaching a snapshot for a sample data for one trial.
Please advise.
Hi Gerges
MAC values work on vectors, i.e. all the data output from your FE model that matches your test data. It compares how well the shape is matched between Vector A and Vector B, i.e. MAC is very useful when you have data measured at many positions. .
For a single value, comparsion becomes harder as you can get the correct answer for the wrong reason when you have equations with more variables than outputs.
So, my advice is to pick positions across your structure and measure at many positions, more positions that cases you wish to compare data for and then use the MAC. If you want to read a bit more about such measurements, take a look here
http://qringtech.com/learnmore/recommended-reading/
Sincerely
Claes
What I attached is a sample for how my data is classified. I have 5 trials till now.
I want to choose the one that matches the best with the experimental results (in terms of trend while minimizing the error).
The best option you have, in my opinion is to use a software like CORA. This program compare results in magnitude, shape and phase and find the best correlation based on a weighted analyse.
http://www.pdb-org.com/en/information/18-cora-download.html
Till now, I couldn't find a way to select the best data set. Please help guys..
Hi Gerges
The problem in comparing simulation with test is that identifcation, in essence, is that of solving an inverse problem, i.e. a problem of the kind where you know the answer (the test result) to be four and your task is to identify what is the question.
For obvious reasons, as long as you have more unknown factors (you simulation model) that known answers, you can always find one or more models that give the right answer for the wrong reason.
This is why one must add check points to the correlation.This usually is called working on first principles and this is what you should try to add when having fewer test results with which to compare.
https://en.wikipedia.org/wiki/First_principle
OK, if you cannot measure more responses than the one you have, surely, you are able to verify more mundane things like weight, centre of gravity, dimensions? The more extra checks you do of independent data, the more trustworthy becomes your model (usually, but not always).
Speed, I guess, can be translated to frequency? If so, perhaps, you can get spectral data, i.e. response that varies with frequency?
Keep in mind this quote from G.E.P. Box "essentially, all models are wrong, but some are useful".
Hope this helps
Claes
- Badges
- Science topic