Page 410 - Design of Simple and Robust Process Plants

9.6 Project Methodology for Operation Optimization 397
measured performance. The spread should be of the same order as the spread of the
measurement.
The operational margin inherently provides a good overall image of the model
when mapped against the measured function, since it represents values implicitly
related through the equations of the model to be validated. The next activity is to
collect a set of measured and simulated performance data over the operational range
of the process for the same price set. Next, a plot is made between the simulated
and measured performance over its operational range for a fixed price set (see
Figure 9.16). A similar presentation was used for validation of individual measure-
ments (see above). It is a requirement to evaluate the process over the entire
operational range. The diagonal line represents the perfect fit between simulation
and measurement. Lines are plotted in the same graph for the 95% confidence lim-
its (or 1.96 times the standard deviation). A significant mismatch between the slope
and intercept, together with a large spread, are indicators for the overall process
invalidity.

Modeling mismatch During model validation it will become clear that there are ±
and there always will be ± differences between the actual process and the theoretical
model. Mismatch might be caused by errors in the model development, but also by
nonideality in the process. We might regard these mismatches as being acceptable
for the optimization, but they must be included in the decision criteria of decider-3.
Some of the mismatches might be caused by the following reasons:

. Availability of inertia, which has an effect on the dew point of vapors.
. Composition of a heavy stream; often, heavies are determined in the feed
stream or at the reactor outlet as a cluster of compounds. The composition of
the stream might change, resulting in a deviation in vapor pressure.
. Accumulation of impurities (not included in the model)in process streams
that might result in differences in temperature and pressure.
. Fixed parameters which depend on the capacity of the unit or fouling/aging;
in that case, the parameter should be updated.
. Insufficiently described and validated unit models
The model mismatch does not necessarily have a notable impact on the overall opti-
mization results, but it will have an effect on the difference between individual pro-
cess measurements and simulation results. It is important to recognize these effects
and to eliminate these measurements as SRVs and as part of decision criteria in
decider-3. After a simultaneous data reconciliation/parameter effort for the total pro-
cess model, the consistent outliers due to model mismatch need to be removed to
optimize the fit. These deviations will have a negative effect on the updating of the
simulation before optimization.