24.08.2013 Views

NONLINEAR CONTROLLER COMPARISON ON A BENCHMARK ...

NONLINEAR CONTROLLER COMPARISON ON A BENCHMARK ...

NONLINEAR CONTROLLER COMPARISON ON A BENCHMARK ...

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

all of the physical parameters except for the motor constants can be combined into<br />

the term as described in Chapter 1 = p mc<br />

(I+mc2 )(m+M )<br />

. All of the controllers<br />

were designed for the actual value of = :2. The simulations were then run with<br />

varying values of until the controller no longer regulated the system. Thus, the<br />

more robust controllers could tolerate greater changes in epsilon than the less robust<br />

designs. In Table 4.2 below, the maximum and minimum values of that the control<br />

designs could regulate, given that they were assuming a plant value of = :2, are<br />

compared:<br />

Table 4.2: Tabular Comparison of Simulated Robustness<br />

LQR Lin. H1 PBC Backstep HJB HJI<br />

max .277 .275 .245 .251 .262 .275<br />

min .016 .015 .014 .015 .102 .093<br />

Interestingly, the linearized controls outperformed the nonlinear designs in this<br />

simulation as well: the linearized LQR control had the highest tolerance for raising<br />

the value of in the system, with the linearized H1 and the SGA to the nonlinear<br />

H1 yielding only slightly less robustness with respect to this upper bound. On the<br />

lower bound, the system value for could be lowered to = :014 before the passivity<br />

based control failed to regulate the system, over a 1000 percent change in the value<br />

of ! In fact, all of the controls gave similarly robust results for the lower bound of ,<br />

except the Galerkin approximations whose lower bounds were notably inferior.<br />

Robustness is a property that is very di cult to measure precisely, and the<br />

above simulated results only give a rough idea of how the controllers behave when<br />

there are signi cant modeling errors present. It should be noted that they all exhibit<br />

a wide range of values for , wherein the controls continue to regulate the system,<br />

and they are all, therefore, meaningfully robust.<br />

47

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!