How to troubleshoot model convergence

I watched the video about Handling errors.

If I understand correctly, gross errors are detected when 'the quality of data reconciliation', a summary value given in the output, becomes zero.
Your last example in the video also shows that even a flow with a small k-value could be a gross error. In that case, the value tolerance is estimated way too low, forcing the calculation to spread the errors over other flows. I guess a missing flow would give a similar result.

Where could I find more information on how quality of data reconciliation is calculated, and how flows with likely gross errors are selected?

I my example, I get 8 flows with likely gross errors.
If I understand correctly, it might be that only one or two flows cause the whole error, and not necessarily the flows with the highest k-values.
Is there a systematic procedure to get to the root of the gross error(s)?

Note that in case of a missing flow, the flow couldn't be detected to be the cause of the problems because it is simply not included in the model!

Here is a selection of books that cover the topics of data reconciliation and gross error detection:
o Chemical Process Structure and Information Flows (Mah, 1990)
o Process Plant Performance (Madron, 1992)
o Material and Energy Balancing in the Process Industries (Veverka, Madron, 1997)
o Data Processing and Reconciliation (Romagnoli, Sanchez, 2000)
o Data Reconciliation and Gross Error Detection (Narasimhan, Jordache, 2000)
o Process Plant Instrumentation (Bagajewicz, 2001)
o Smart Process Plants (Bagajewicz, 2010)

Additionally, information about the nonlinear data reconciliation algorithm used in STAN (and much more) can be found in my PhD thesis, which is available for free under:
www.wien.gv.at/umweltschutz/nachhaltigkeit/pdf/cencic-2019.pdf
(see paper 1 in the appendix)

Unfortunately, there is no systematic procedure that I could recommend. Just rechecking the values of the suspicious variables will often do the job.

Hi adminST,

Thank you for all this information.
I'll dig into it to get a better understanding of gross error detection.


Best regards,

Geert