Process Observers and Data Reconciliation Using Mass and Energy Balance Equations - Advanced Control and Supervision of Mineral Processing Plants - page 16

the performance is frequently expressed as an economical index which embeds all
these aspects. When the operating conditions drift away from the range of optimal
performance, the plant experiences economic losses. The ability of a plant to remain
in the vicinity of its optimum operation is related to real-time decision making pro-
cesses, i.e. , to production supervisory systems, real-time optimization systems, and
automatic control strategies. Regardless of the strategy used for maintaining a plant
close to an optimum performance, the variability around the optimum value relies in
the first place upon an efficient evaluation of the performance index-in other words
upon the variance of its estimate. The lower the variance, the better is the plant
performance.
A plant performance index observer uses the available measurements of the pro-
cess variables. Typically, in a metallurgical plant, these variables are overall mate-
rial, phase, and metal flowrates, material chemical compositions, energy flowrates,
temperatures, consumed power, etc . As any other observer, the plant performance
observer simultaneously uses measured values and process models. These models
are required to cope with common data processing problems such as measurement
uncertainties - which are quite large in a metallurgical operation - lack of measure-
ment availability for critical variables (obviously a performance index is usually not
directly measurable), limited knowledge of the process behavior (a difficult prob-
lem, particularly in extractive metallurgy), and information redundancy in the avail-
able measurements and prior process knowledge. As the process model uncertainties
are very large in metallurgical industries, it is common practice to use only con-
straints, i.e. , sub-models - in the sense that they are not causal models as assumed
in traditional model-based control and observation. Since the level of confidence in
these sub-models must be high to prevent distorting the data set information con-
tent by uncertain models, the selected constraints are essentially laws of mass and
energy conservation. In the metallurgical, and more generally chemical industries,
these observation methods are called reconciliation methods, in the sense that they
reconcile the measurement data with the laws of mass and/or energy conservation.
Estimation of process states is required for process performance audit, process
modeling, monitoring, supervision, control, and real-time optimization. Whatever
the process scale, laboratory, pilot, or full industrial scale, the first step of state esti-
mation is to collect experimental data. Unfortunately, and this is particularly true at
the industrial scale, measurements are extremely difficult and inaccurate in the met-
allurgical engineering field. Production units treat materials that are multi-phase and
usually contain extremely heterogeneous particulate phases [1]. The data is highly
inaccurate and incomplete, and requires to be improved before being used in the
above mentioned applications. The usual statement “Garbage in, garbage out” is
particularly true in this context, use of poor data leading invariably to poor models,
poor decisions, and improperly designed and operated systems. Therefore, using ad-
ditional information to the experimental data through process prior knowledge leads
to better state estimates. Mathematical models are usually the most efficient way to
encapsulate process behavior knowledge. Unfortunately, in metallurgical processes,
the knowledge is frequently fuzzy and less accurate than in mechanical and chemi-
cal industries.