Page 406 - Design of Simple and Robust Process Plants

9.6 Project Methodology for Operation Optimization 393
The unit models are combined into sections. The robustness of the section models
must be tested before they are combined into the total plant model. The testing is
done by variation of feed rates and compositions, but also different product specifi-
cations. Lack of robustness is often a signal for a modeling error. During testing of
the section models, prices might also be connected to the individual streams to have
the optimizer search for a quasi optimum (this is optional). Exploration of the
model constraints under extreme process conditions is a worthwhile exercise that is
not a trivial effort. The final step in the model building is the connection of the
different sections and addition of the economic section and the optimizer. At this
point it is a good time to replace some simplified models by more rigorous ones.

9.6.8
Test Process Model for Robustness on Process Conditions and Prices: Step 7M

The robustness of the optimization is crucial for final success. Failures to converge
will result in no update of the values for the DOFs and as such would undermine
the trust of operation and the success of the project. The overall model needs to be
extensively tested for variations in feed rates, feed composition, product specifica-
tion, prices and all other variables that might be subject to change. Bouncing against
the constraints is another exercise that needs to be explored extensively, as reflected
in the model. These exercises provide valuable information about modeling short-
comings (see also under the validation step).

9.6.9
Implement Data Analysis on Selected Data and Evaluate Steady-State Situations: Step 9

Implementation of the executive, with its different functions, is an effort which
leads gradually to CLO. The successive steps are implementation of steady-state
detection, followed by data reconciliation, parameter estimation, and model vali-
dation. All these steps are required to evaluate the quality of the control effort as
developed under steps 5C±8C to achieve as stable operation and the quality of the
modeling effort, steps 5M±7M.
After the selection of the SRVs (as discussed under step 4), the data must be ana-
lyzed. For each of the SRVs, analyses are performed on both raw and filtered data.
The selection of criteria for the data filtering are the elimination of high-frequency
process and instrument noise.
The selection of criteria for the average signal is impacted by the stability of the
units and the overall process. The whole project is set up for steady-state optimiza-
tion, so the plant should represent that state. The control loops as planned for BC
and MBC levels need to be closed at this step based on appropriate design and tun-
ing. Evaluation is carried out by setting stability criteria for decider-2 and tracking
process stability over time.