I trying to develope am OPEN LOOP databased model (using system identification methods) for "Tenessee Eastman Challenge problem" system. TE problem in Brief: The process has a reactor, a cooler, a seperator and a stripper (in the same order). Unconverted reactant from stripper is recycled to reactor. There are total four feed (Input) streams, out of which three (A,D,E) are fed to reactor (which is at starting of process) while one stream (C) is fed to stripper (which is at end of process).
For open loop system identification, I am introducing simulataneous "Psuedo Random Binary Signal" in all four input streams flowrates. I am developing Output Error type of model.
The problem I am facing is: The identified model has opposite gain for few IP/OP pair when compared to actual process gain.
I was told that this may be because of delayed recycle effect of stream C. But couldn't get elaborate information on it. Can anyone of you help me???