Information Technology Reference
In-Depth Information
Figure 6.2
Parameter Selections for Regression.
The most commonly employed activity coefficient equations are the van Laar,
Wilson, NRTL, Uniquac, and their variants. The input for all equations and the iden-
tification of the parameters is virtually the same as in the example above.
A very large collection of experimental data for many binary systems, from many
sources, fit to the activity coefficient models is incorporated within Aspen Plus. An
example of Wilson-2 parameters for an ethanol - water system is shown in Figure 6.3.
These are available for user development of Aspen Plus simulations. Note that the tem-
perature range of applicability is shown. These may not, however, satisfy requirements
for many applications. For example, temperatures and pressures may not be appropri-
ate to the modeling needs of specific applications. An important task for a user is to
locate suitable data and customize (i.e., fit) the data to a model that lends itself to the
application. If the required binary data are not available, the generalized correlation,
Unifac, can be used to estimate the VLE data, which may then be fit to any of the
activity coefficient equations. In such a situation the accuracy of a simulation may be
compromised, and caution is advised.
6.3 BASIC IDEAS OF REGRESSION
Frequently, engineers and scientists are faced with sets of data that are proposed to fit
a linear equation. The typical first pass at fitting the data is an eyeball plot on linear
coordinates. This plot will usually differ when several persons attempt to plot the same
data, and for the model
y
=
ax
+
b
(6.17)
