| T O P I C R E V I E W |
| mc |
Posted - 06/29/2004 : 11:49:48 AM
Hello world,
I have two sets of data.
One is experimental and one is obtained from simulations.
I need to compare those two to see if the simulation is a good approximation of the experiment. (Not by eye, obviously).
One way seems to be "fit compare" but that assumes a fit of the data I have and which does not seem to work well enough.
Any ideas?
Thanks a lot,
MC |
| 4 L A T E S T R E P L I E S (Newest First) |
| mc |
Posted - 07/01/2004 : 03:18:25 AM
Hello everyone,
Coming back on this, I could not find a fit function to work reliably on hysteresis, unless you redefine it all the time, but this could be my fault.
Of course, the least mean square method built by hand works nicely, but I think it would be a nice feature to include in Origin relaying on interpolation of the data rather than fit. Helps the experimental and simulation guys.
Enjoy your work,
MC
Edited by - mc on 07/01/2004 03:19:22 AM
Edited by - mc on 07/01/2004 03:19:58 AM |
| mc |
Posted - 06/30/2004 : 09:13:13 AM
Hello,
Thank you both for your help.
The simulations are produced by numerical calculation using c and represents the hysteresis of a rather complicated physical system
Both of you have been helpful and I will try using both the fit comparison if I can manage to find an approximate function, as well as separate application of a least square method.
Thank you again,
MC |
| easwar |
Posted - 06/29/2004 : 2:25:54 PM
Hi MC,
You need to define/determine what you mean by comparing the simulation and experimental data.
What Mike is suggesting is that you make use of the Fit Comparison tool to fit a model (function) to both datasets (epxeriment and simulation), and then based on a F-value computed with the two fit results, the tool will output a statistical statement, based on this F-vallue, as to whether the two datasets are different or not.
The above process works well if you pick a model that is appropriate for your data.
On the other hand, if you are just looking for a simple measure of how closely the experimental data follows the simulation data, you could:
1> Normalize one or the other dataset so that the y scales match - they align well in y scale. Can use the Normalize menu item which works on worksheet columns, or can scale one dataset or the other by hand so that they both match, say, at a particular x value
2> Create a difference dataset. Say your simulation and data are in columns Data1_B,C. Create another column D, and set the column value of this to be col(c)-col(b)
3> Plot col D - this would be a "residual" plot that then shows where the two datasets deviate, and by how much
4> Instead of just difference, you could compute
col(D)=(col(b)-col(c))^2
and then sum up column D to obtain a "Sum of Squares" estimate of the deviation of one from the other. You could then try to minimize that number by running the simulation with different parameters and comparing with experimental data again, to get a good "match", thus leading to better simulation parameters.
Easwar
OriginLab
|
| Mike |
Posted - 06/29/2004 : 1:27:50 PM
Hi MC,
The Fit Comparison tool was designed to do just what you are asking for. I am going to suggest that maybe you need to work on your model a bit. Did you try fitting your experimental data with a few different models (using the Advanced Fitting tool)? Be aware that you can define your own function in the NLSF and use it to compare your experimental and simulation data in the Fit Comparison tool.
Mike
OriginLab |
|
|
