Industrial Products for Advanced Control of Mineral Processing Plants - Advanced Control and Supervision of Mineral Processing Plants - page 301

∑ i [ b i y i ( t + j )+ a i ( y i − w i )] ∑ j [ d j u j + c j ( u j − m j )]

2 2

J = + , (7.4)

where y i is a controlled variable, w i is the corresponding reference, u j is a manip-
ulated variable, and a i , b i , c j , d j ,and m j are parameters. Both error and profit
optimization can be performed.
Profit ® Controller uses the range control algorithm, which determines the small-
est process moves required to simultaneously meet control and optimization objec-
tives, see U.S. Patent No. 5,572,420 [53]. Its handling of control through “funnels”
rather than specified trajectories gives the controller additional degrees of freedom
to enhance dynamic process optimization. RMPCT uses a finite impulse response
(FIR) model form, one for each controlled manipulated pair; model identification is
accomplished using open-loop tests on the process [54]. RMPCT provides robust-
ness to plant-model mismatch using singular-value thresholding and a min-max de-
sign. In the latter, the optimizer determines the control effort over all possible plants
realizations (worst case).
Profit ® Optimizer solves multi-unit and plant-wide optimization on a minute-
by-minute basis. It leverages measured process relationships and dynamic process
information residing in underlying Profit Controller applications. By accounting for
constraints and economics between operating units, Profit Optimizer can implement
optimal operating conditions without violating global plant constraints. In addition,
it dynamically coordinates the introduction of optimal set-points for a smooth tran-
sition to optimal operation using distributed quadratic programming.
Profit ® Bridge integrates non-linear process models with Profit Controller and
Profit Optimizer applications. By frequently updating the linear models embedded
in these applications with information obtained from non-linear process models,
Profit Bridge delivers dynamic non-linear control and optimization at a fraction of
the cost of a full-scale, rigorous optimization system.
Profit ® Max provides a first-principle modeling and optimization system within
an on-line execution environment. This solution is typically used on optimization
projects requiring mixed-integer programming, or on highly non-linear processes.
UniSim ™9 Design is an interactive process modeling offering that enables en-
gineers to create steady-state and dynamic models for plant design, performance
monitoring, troubleshooting, operational improvement, business planning and asset
management. With this product, steady-state models can be extended to transient
models, allowing consistent model configurations to be used for process transient
analysis, controllability studies, and operator training applications.
Honeywell has developed various MPC applications in mineral processing plants,
one of the first being the SmartGrind ball grinding application installed at the BHP
plant in Pinto Valley, Arizona (see Figure 7.6). TotalPlant SmartGrind multivariable
control integrates MPC and neural networks to predict anomalous situations such as
mill overloads. In this ball mill example, a projected throughput increase of 4.18%
in the desired product (smaller than +65 mesh particle) was obtained. Performance

9

UniSim is a registered trademark of Honeywell International, Inc.