where, as stated before, T(z-1) is the observer The selection of the GPC tuning factors such as N1,
polynomial, and its effect in the model identification N2 and NU was under the following criteria: N1 is
can be clearly seen if (17) is rearranged in the form: directly related to the process time-delay, but it is
usually set 1 with no loss of generality; N2 was
A( z −1)∆y (t ) B ( z −1)∆u (t − 1) chosen to exceed the degree of polynomial B(z-1), so
= + ζ (t ) (18) that the value of this factor can be formulated as N2 ≈
T ( z −1) T ( z −1)
2n-1 (n = order of A(z-1) + 1). It is known that when
NU equals 1, it usually gives good results, while
From the above expression it is clear that the input –
greater values lead to a more activated control signal
output data is “band-pass” filtered by the T-
(Clarke, et al., 1987a). Further, λ was set to 0, and for
polynomial, which affects only the disturbance
the polynomials T(z-1) and P(z-1), they were selected
rejection properties of the system (Clarke, et al.,
to be of the following form: T(z-1) = (1-0.9z-1)2 and
1987b). Therefore, once the data is filtered, the one-
P(z-1)=(1-0.8z-1)/0.2.
step model prediction is:
30 Rolling 30
RPM
RPM
0 Br increased 600%
Rolling 0
-30
-30
Controller's output Controller's output
200
100 100
Nm
Nm
0 0
-200
Load torque Model Parameters
2000 1 a2
Nm
0 0
a1
-2000
-1 b1 b2
0 5 10 15 20 25 30 -2
0 5 10 15
Time (s) Time (s)
Fig. 4. Performance under a multi-pass hot-rolling.
Fig. 5. Performance under varying mill’s parameters.
After 3 seconds, the roll friction was increased up to
600% its rated value. The controller identified this
new operating mode by adapting its parameters in (a) (b)
such a way that the rolling speed was kept constant.
This can be seen in Fig. 5 in which a sudden change
in the parameter estimation is observed. Finally, after
7 seconds, the motor inertia was also incremented up
to 800%. The effect of this change can only be seen
in the transient state when at 10s the speed reference
was decreased to 30 rpm. Although the plant’s rising
time is increased, no overshoot in the rolling speed
was observed.
(c)
5. REAL-TIME HOT ROLLING EXPERIMENTS
60 60
RPM
RPM
0 0
-60 -60
100 100
0 0
%
%
-100 -100
10 15 20 25
Model parameters
Time (s)
1
a2 Fig. 8. Real-time performance using PID control.
0
Rolling speed
8
a1 b2
-1 b1 6
RPM
4
-2
2
10 15 20 25 30 35
0
Time (s)
Controller's output
Fig. 7. Real-time performance using GPC. 100
6. CONCLUDING REMARKS 50
%