[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [EnergyPlus_Support] Why does E+ lose accuracy for lower values of the space discretization constant?





Hi again, I should also tell you that the proposed equation is
C_c = sum(abs (T node_i^m - T node_i^m-1) / nx),
where nx stands for the number of nodes.
The reason of changing "abs(sum...)" for "sum(abs...)", is that we think the form abs(sum...) (as it is stated in the source code) could not sum up properly, or carry out the substractions between the levels or iteration of the nodes temperatures correctly. 

Regards,
Miguel Casas


De: saleh saadi <salehsnjsaadi404@xxxxxxxxx>
Para: "EnergyPlus_Support@xxxxxxxxxxxxxxx" <EnergyPlus_Support@xxxxxxxxxxxxxxx>
Enviado: Viernes, 24 de mayo, 2013 19:51:57
Asunto: Re: [EnergyPlus_Support] Why does E+ lose accuracy for lower values of the space discretization constant?


Based on my previous check of the E+ code (version 7), the convergence criteria is calculated as such:

C_c = abs(sum (T node_i^m- T node_i^m-1) / sum(T node_i^m) ), where m stands for iteration level.

As the number of nodes increases, it is good to use small convergence criteria as well as small time step. This seems to work well for me when Gauss-Seidell solver is used with the equation of convergence criteria mentioned above . 

Are you modeling Phase Change Materials (PCM) or massive structure? If you are not modeling any of these structures, then go for CTF algorithm.

regards
Saleh

From: Miguel Casas <miguelcasasarredondo@xxxxxxxxx>
To: "EnergyPlus_Support@xxxxxxxxxxxxxxx" <EnergyPlus_Support@xxxxxxxxxxxxxxx>
Sent: Friday, May 24, 2013 5:37 PM
Subject: Re: [EnergyPlus_Support] Why does E+ lose accuracy for lower values of the space discretization constant?


Hi Saleh. I am glad you answered.
Yes, I am sure the number of nodes increase by decreasing the Space Discretization Constant (C). I quote from the Engineering Reference Documentation: "Lower values for the Space Discretization Constant yield more nodes, with higher values yield fewer nodes".
I already modified the Convergence Criteria, using the lowest value. My results got better, but still, the same tendecy came out: lower values for C would generate lower accuracy. Why could this be happening?
When I read the Engineering Reference Documentation I found kind of a definition for the Convergence Criteria. I quote: 
Because the solution is implicit... The Gauss-Seidell iteration loop is the inner-most solver and is called for each surface. It is limited to 30 iterations but will exit early when the sum of all the node temperatures changes between the last call and the current call, normalized by the sum of the temperature values, is below 0.000001C. This convergence criteria is typically met after 3 iterations...