Cereals processing technology

• Model-based predictive control
• Rule-based control
• Fuzzy control
• Adaptive control.
Wheat, corn and coarse grains milling 39
Model-based predictive control
Model-based predictive control (MBPC) is an advanced control technique that is
applied where time delay compensation is required, complex process dynamics,
and/or multivariable interactions exist. A prerequisite is that suitable process
models are available. There are a number of software packages available that
encompass model and control strategy development, and system implementation.
A common application of MBPC is in feed forward control of processes. An
example in the flour milling context is the feed forward moisture addition
system that the Buhler TM company of Switzerland market. This controller uses
MBPC to predict the post-addition moisture content of the grain being treated,
based on the pre-addition moisture content.
The technique requires the construction of a dynamic model that is capable
of predicting future process behaviour. The employment of this model within
a predictive control strategy involves calculating optimal settings for the plant
to ensure the controlled variables are kept at their set points. This may not
always be possible since the optimum value for the process may change in a
dynamic fashion because of varying raw material properties or some other
variables.
The model of the process is developed from historical data by applying
statistical techniques. The least squares method is the favoured method, although
other modelling methods may be used where least squares cannot satisfactorily
manipulate the data. Preferably a continuous model is used in MBPC, but the
higher powered microcomputers now available can deal with discontinuous
functions.
MBPC has a number of useful features where the control of processes is
considered. Process response is predictable and so operation at optimum levels is
straightforward. The basic nature of the control strategy also means that the
computing demand of a controller based on MBPC is quite low. Other
advantages include rapid controller response to changing process conditions and
the ease with which process dead times can be handled.
On the negative side, MBPC must have models available that consider all
process variables. If any variable is not considered, the result can be significant
errors in the control system’s response. The models employed are usually not
adaptable, except through human intervention, thus in the situation where the
process model changes with variable values or even lapsed time the controller
error becomes large.
Mill processes experience this kind of model evolution. For example, as the
fluting wears off the rolls used in flour mills, the particle size distribution
produced by the roller mills changes for a given roll setting. Thus a rigid