Automatic Control of Flotation Columns - Advanced Control and Supervision of Mineral Processing Plants - page 274

Imaging systems can provide froth characteristics such as color, form, speed and
size. Eventually, they could also be used as secondary variables for control pur-
poses. For instance, Liu et al. [70] illustrate a new way for modeling and control of
flotation processes based on froth appearances. Froth health monitoring is also dis-
cussed in Chapter 3. Examples where froth speed is measured to improve flotation
performance in cells are described in the literature [71-73].
When controlling a multivariable process such as a flotation column, two main
approaches are possible: decentralized or multivariable control. Decentralized con-
trol consists in using single-input single-output controllers. It requires a good selec-
tion of input-output pairings. Controller tuning must take into account interactions
between the control loops, which may be difficult to accomplish. Despite consid-
erable advances in multivariable control theory, decentralized PI controllers still
remain the standard for most industries. They have fewer tuning parameters, are
easier to understand, and are more failure tolerant. Multivariable controllers are
rather complex, require more engineering manpower and have a lack of integrity,
all which may result in operator non-acceptance. For the flotation column case a
feasible choice of control variables is froth depth, bias rate and gas hold-up whereas
the possible manipulated variables are the tailings, wash water and air flow rate set
points. In the case a decentralized control approach is considered, the following
pairings are suggested: froth depth - tailings flow rate, bias rate - wash water flow
rate and gas hold-up - air flow rate. For a two-phase system, it has been shown that
the resulting model is then an upper triangular matrix [33], indicating that decen-
tralized control with these pairings is very appropriate. The main weakness of this
approach is the difficulty of adequately taking into account constraints, as will be
discussed in the next paragraph and in Section 6.4.3.
Model predictive control (MPC) is, by far, the most widely accepted multivari-
able control algorithm used in the process industry. Several good topics describing
MPC are now available [74-76]. Several commercial MPC technologies allowing
model identification and control with constraints can be purchased and successful
applications are found in many process areas [77]. The main advantage of MPC is
its ability to handle multivariable processes while taking into account various con-
straints. Indeed, the plant operating point that satisfies the overall economic goals
of the process, usually lies at the intersection of constraints. The main concepts of
MPC are described in Sections 5.5.2 and 7.3.2. Model development and identifica-
tion (Chapter 4 and Section 5.5.1) are the most difficult, critical and time consuming
steps in the design of a MPC.
Two comprehensive survey papers about control have recently been presented by
Bergh and Yianatos [64] and Bouchard et al. [65]. Table 6.1 summarizes some
control applications reported in these surveys as well as more recent developments.