Numerical Difference on DMC+ Model and DMC3 Builder Identification Results?
Download as pdf :
Products:
Aspen DMC3, Aspen DMC3 Builder
Last Updated:
09-May-2022
Last Updated:
09-May-2022
Versions:
Article ID:
000099765
Article ID:
000099765
Primary Subject:
Problem Statement
This article talks briefly on the differences that may exist on the identification results between DMC+ model and DMC3 Builder.
Solution
The short answer to the question is Yes, there could be numerical differences on the Identification results between both applications. Most likely these differences can be noticed on the identified Gain results. Lets take the following example of an ID case in DMC+ Model:
This is a 1MV vs 1CV Sub Space ID Case with the following parameters [T=60 L=30 SR=1 FT=180 PP=DT]
As result the gain value is:
-0.2912202979542998
If we export the same dataset and case and import it on DMC3 Builder (same case meaning same ID parameters and slices) we obtained:
The result gain value is:
-0.291220297954376
For this specific example the Dataset and Slices, along with the Case parameters are the same.
So why both applications are showing different results?
The mains reason of these changes is related to the dll use for the identification purposes. Both uses AtIdentification.dll. However, the number of changes and improvements done for this file are essentially different. So, even thought that both applications use the same technology the differences on fixes and enhancement is different (is worth to mentioned that most of these improvements has been apply on DMC3). So, it would be an expected behavior to have different numerical results as consequence of this changes in the dll file.
Moreover, other segments of DMC3 use for the Identification as Data preparation, slicing numerical accuracy, data loading, etc., has gone under some changes as well so it is highly possible that this numerical difference could be carried as results of these multiple fixes and enhancements, resulting on the final gain ID.
So, what application results should I trust??
We can argue that due all fixes and enhancements DMC3 results could be a bit more accurate that the ones identify by DMCplus (among other possible problems related to ID). However, as we could notice on the example, the results turn to be pretty much the same (around less than .0001% difference) it is expected that in any case both ID may work for the deployed controller as the will lead to small or significant changes on the MV (maybe small enough to be physically impossible to take place in the controller). So, even thought that the numerical results are different, they won’t be completely different (unless the ID and Case condition are different). So, the best suggestion would be to stick to the technology that you are working with and use that difference to verify the results in case is necessary but have the confidence that both are going to be showing pretty similar numerical results, trending and dynamic behaviors.
Keywords
DMC3, DMCplus, Identification