Emerson Exchange 365

Execution time for module containing neural network block

I've built a neural network that has a historian sampling rate of 300 seconds and a Time-to-steady-state of 15000. Any reason that module needs to be on a 1 second execution rate?

• 14.3.1
• neural

John Rezabek

• 8 Jun 2020 5:50 PM
• Cancel

2 Replies

• JDWheelis

  

  • 3 Nov 2020 9:06 PM

  John,
  It appears that no one has responded to your question, which I came across and thought I'd look for an answer.
  DeltaV Books Online does not contain a reference to a restriction on the module or function block execution rate, and there is nothing mentioned in the Educational Services courseware on execution rate restrictions. A test to confirm the hypothesis that an module/FB execution rate different from 1 second could be set up on a neural net known to work satisfactorily - possibly on a simulated process - by creating two modules with NN blocks and different execution rates and comparing their results.
  Contact me to discuss, if you'd like.
  J.D.

  • Cancel
• Andre Dicaire

  

  • 3 Nov 2020 9:30 PM

  In reply to JDWheelis:

  Good on you JD to drop a comment.
  
  By sampling at a slower rate, the results of the block might be influenced by aliasing. So running a test like JD suggests would be prudent to determine if the slower execution rate leads to different results.
  
  What is the execution time on this module? EXEC_TIME. You can trend this with Watchit and set Sync Reads. Is it a major CPU load? If EXEC_TIME is less than 5000 microseconds, slowing it from 1 to 2 or to 5 seconds will not buy a lot of Control capacity. If it runs 10,000, 15,000 or more and you want to reduce loading on the controller, slowing it down will help.
  
  When you slow the module down, you also slow the communication update rate, as that will never be faster than the module scan rate. based on the suggested test with a second instance at say 10 seconds, you can make a more informed decision.

  Andre Dicaire

  • Cancel