Environmental Engineering Reference
In-Depth Information
•
Unmeasured process variables
F
2
and
F
3
are estimated.
•
Process variable measured values are corrected.
•
Reconciled values are consistent with the law of mass conservation.
•
Plant performance, such as copper recovery, calculated with reconciled data be-
comes independent of the calculation path (using copper data to estimate the
flowrates gives a recovery of 95.4%, and, using zinc data, gives an absurd value
of 515%, whereas the reconciled value gives 95.9%). Furthermore results are
more accurate than values calculated by any other methods.
•
Models that are subsequently estimated from reconciled data (for instance flota-
tion kinetic models) are much more reliable.
•
Decisions made from reconciled data are necessarily more efficient than deci-
sions made from raw data.
Reconciliation of mass and energy balance with raw data is a technique which
has been known for a long time in mineral processing (see, for instance, [2-4]), but
it was already mentioned in chemical engineering as early as 1961 [5]. Crowe wrote
a good survey paper in 1996 [6], but the first organized topics addressing the chem-
ical process reconciliation topic appeared only at the end of the 20th century [7, 8].
However, the contributions of the MMP community to the field of data reconcilia-
tion is mentioned only in the first one. Off-line reconciliation methods for steady-
state processes are now quite mature and various computer packages are available.
For example, in the MMP field, although they are not at all limited to these appli-
cations, one can mention: Bilmat™ and Metallurgical Accountan™
1
(Algosys) [9],
Bilco™ and Inventeo™
2
(Caspeo) [10], JKMultibal™
3
(JKTech) [11], Movazen™
(Banisi) [12] and more chemical process oriented: Sigmafine™
4
(OSIsoft) [13],
Datacon (IPS)[14], Advisor™
5
(AspenTech) [15], and VALI™
6
(Belsim) [16].
The most usual reconciliation techniques are based on the minimization of
quadratic criteria, therefore assuming that uncertainties mainly belong to Gaussian
distributions. This chapter focuses only on this type of approach. However, alter-
native reconciliation methods based on artificial neural networks have also been
proposed. But those do not offer either the same rigorous statistical and physical
background or the same result reliability analytical evaluation tools as in the ap-
proach presented here (see, for instance, [17-20]). Linear matrix inequality (LMI)
methods have also been proposed by Mandel
et al.
[21].
Steady-state methods are applied off-line to mineral processes such as com-
minution [22, 23], flotation [59], gold extraction [25-28], hydrometallurgy [29] and
[30], pyrometallurgy [31-33], and cement preparation [34]. On-line applications to
steady-state processes are actively used, while stationary-state methods that make a
1
Bilmat and Metallurgical Accountant are registered trademarks of Algosys, www.algosys.com
2
Bilco and Inventeo are registered trademarks of Caspeo, www.caspeo.net
3
JKMultibal is a registered trademark of JKTech, www.jktech.com.au
4
Sigmafine is a registered trademark of OSIsoft, www.osisoft.com
5
Advisor is a registered trademark of AspenTech, www.aspentech.com
6
VALI is a registered trademark of Belsim, 174k rue De Bruxelles, 4340 Awans, Belgium,
www.belsim.com
Search WWH ::
Custom Search