Environmental Engineering Reference
In-Depth Information
Data reconciliation techniques are involved at two different levels when auto-
matic control and real-time optimization are simultaneously implemented. The two
observers must be different since the dynamics of the two loops are quite different,
optimization being performed with slower dynamics than automatic control. This
implies that the variances of the measurement errors and of the accumulation rates
(node imbalances) must be tuned differently. Also, the models as well as the state
variable of the two observers may be slightly different. Even if the same stationary
reconciliation structure is used for both observers, the variances in the reconcilia-
tion criterion must be different. As the measured values in the optimization observer
are obtained through an averaging technique involving a moving time window with
several samples, their variances are usually divided by the number of samples in
the window. Also, the accumulation rate variances are necessarily lower because
the averaging process in the time window decreases the magnitude of the node im-
balance variations, by attenuating the signal dynamics. Ultimately, the optimization
observer could be steady-state.
In addition to the two observers, Figure 2.25 shows peripheral tools for data pre-
processing. Sensor failures or abnormal process behavior must be detected before
feeding the reconciled values to the optimizer or the controller. As the optimization
observer is assumed working in stationary regime, it is important to test that the
process variable means are statistically constant before reconciliation. Also when
persistent mean changes are detected, it might be helpful to adapt model parameters,
when, for instance, permanent changes occur to operating conditions such as ore
grindability and grade, tonnage, chemical reagent type. If an adaptation procedure
is integrated into the loops, it should be activated only when permanent changes due
to persistent disturbance means or set-point changes are detected.
Flotation plant example: to illustrate the concept of control- optimization-
reconciliation coupling, a simple example for a flotation plant is depicted in Fig-
ure 2.26. Data reconciliation is performed only at the optimization level and the
control loop limited to a single-input-single-output system, where the collector ad-
dition is the manipulated variable and the concentrate grade the controlled variable.
The grade set-point is supervised through the maximization of an economic index.
Although there is no documented study of the performance of such a real-time opti-
mization strategy, the concept has certainly a potential that should be investigated.
2.14 Conclusion
The objective of this chapter was to point out a problem that faces most metal-
lurgical engineers and mineral processors who are involved with metal production
and willing to understand and optimize the processes they are dealing with: the
available measurements are uncertain, incomplete, and inconsistent with process
behavior prior knowledge. The emphasis is put here on data reconciliation with
mass conservation constraints, however this topic belongs to the universal problem
of matching raw data and prior theoretical knowledge. The subject is superficially
Search WWH ::
Custom Search