Online soft sensors

It is important to remark that the control scheme proposed in Figure 4 is catalogued as adaptive (according to the definition by Bastin and Dochain (1990)), because it has the potential to adapt itself (due to the online estimations given by the soft sensors) to variations in the kinetics of the process. 

Online applications of models describing reactor or plant behavior are used for operator information. The most interesting applications include those parameters that are difficult to measure or can only be measured with time delay such as melt flow index and polymer density. Soft sensors eliminate the time delay that is extremely critical during plant start-up and grade transitions. Even the so-called online analysis with automatic sampling has a certain time delay, as shown in Figure 4.10. 

Density soft sensors are first-principle models comprising kinetics, thermodynamics, and mass balances. They use online information from the plant, such as peroxide flow rates, modifier, and cooling water, as well as the measured temperature profile. Model execution can be organized on a separate computer connected to the DSC and the results can be trended in the DCS for operator use. Figure 4.10 shows the density soft sensor results for a grade transition toward a lower polymer density. 

Figure 4.10 Time gain in accessing transition data for product density, comparing online model application (soft sensor)) versus online measurements (OCS).

MFR soft sensors are often data-driven models that use information from the reactor and from the extruder. Figure 4.11 shows the comparison with an online... 

Figure 4.11 Comparison of soft sensor results for MFR with online analytics for a grade transition.

---