Sunteți pe pagina 1din 3

3.3.

6 Experimental Results of Optimal ILC


Control using a propotional valve requires replacement of the servo-valves and
modification of the hydraulic system. For a fair comparison, it is shown here using
software to simulate the slow-response proportional valve, and the experiments can
be done on one injection molding machine. As shown in Fig. 3.9, a first-order process
with unit gain and a large time constant, Gadd, was added after the controller. The
dynamics of the control valve are thus changed from Gv to G|v = Gadd .Gv.

Figure 3.9 Simulation of the slow-response valve


The optimal iterative learning control was first applied to the injection velocity
control experimentally using the material HDPE. The weighting matrices Q and R
were both selected to be 1, as in the simulation case. The injection velocity is
controlled to follow a step-change profile. The initial input signal, that is, the first
cycle’s control input, is arbitrarily set to be 10%, as shown by the short-dashed line in
Fig. 3.10(b). The control results are plotted in Fig.3.10, where Fig. 3.10(a) shows the
injection velocity responses (output) and Fig. 3.10(b) shows the corresponding servo-
valve openings (input). It can be observed that the control responses became
oscillatory with the increase of the cycle number k, contradioting the simulation
result obtained earlier. The poor experimental control performance is attributed to the
accumulation of initialization uncertainties and disturbances with the strong feed-
forward action as chosen. In the optimal ILC design, a linear time-invariant model is
used to approximate the dynamics of injection velocity, which is a nonlinear and
time-varying process, so there inevitably exists a significant model mismatch. Due to
the nature of the electrohydraulic system, the initial injection velocity response
cannot be repeated exactly, resulting in the uncertainty of initialization error of the
injection velocity control. Furthermore, there are disturbances during the molding
process from different sources, such as variation of the material or operating
conditions. With the existence of disturbances and model mismatch, a large p leads to
a strong feed-forward action and a weak feedback action. As a consequence, the
error-rejection ability of the proposed learning controller is reduced.
Figure 3.10 Experimental results using optimal learning control with system Eq. 3.36
and constant weighting matrices (HDPE):
(a) injection velocity, and (b) corresponding valve opening

The method proposed above is implemented. Thus, the controller is modified with a
varying p to ensure the system convergence and to enhance the robustness of the
optimal ILC. For the first cycle, the control input is set to be a constant value, 10%,
the same as the last experiment. The gain matrix S (t) and the feed-forward term in
Eqs. 3.31-3.33 are then calculated with p = 1.0, to ensure a rapid control response
convergence. For the following cycles, p is set to decrease exponentially with the
increase of the cycle number k, in a relation of p =0.6k-1. The material HDPE is used
as in the last experiment, and the velocity is controlled to follow the same step-
change profile as in the previous case. The resulting velocity responses are given in
Fig. 3.11(b)
It can be observed that the velocity response of the second cycle converges rapidly, as
shown by the dotted line in Fig. 3.11(a). The control of the Sixth cycle has already
converged rapidly, as plotted by the dash-dotted line. The solid line shows the result
of the tenth cycle: the velocity tracks the set-point trajectory closely, despite the delay
in the initial injection stage that was caused by the flow and charge of the hydraulic
system. It is obvious that the control oscillation has been eliminated after the
implementation of the proposed modification. The control response converges
quickly, and the control system in stable with an increase in the cycle numbers.

Figure 3.11Experimental results using optimal learning control with system Eq. 3.36
and changing weighting matrices (HDPE):
(a) injection velocity, and (b) corresponding valve opening

S-ar putea să vă placă și