09.07.2015 Views

Robust Optimization: Design in MEMS - University of California ...

Robust Optimization: Design in MEMS - University of California ...

Robust Optimization: Design in MEMS - University of California ...

SHOW MORE
SHOW LESS

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

19where x k−1 was the current m<strong>in</strong>imum at the last iteration. The s<strong>in</strong>gle variable optimizationproblem ism<strong>in</strong>tN(t)D(t)s.t. g i (t) ≤ 0 (3.6)We will refer to this problem as a l<strong>in</strong>e search <strong>in</strong> IR n . This rational polynomial m<strong>in</strong>imization<strong>in</strong> one variable, subject to polynomial constra<strong>in</strong>ts is easy to solve <strong>in</strong> closedform.3.3 Exact L<strong>in</strong>e Search M<strong>in</strong>imizationThe solution to the problem posed <strong>in</strong> equation (3.6) is presented here. We beg<strong>in</strong>by def<strong>in</strong><strong>in</strong>g the set <strong>of</strong> feasible po<strong>in</strong>ts, Ω i , for each constra<strong>in</strong>t, g i (t), asΩ i := {t : g i (t) ≤ 0} (3.7)Comput<strong>in</strong>g Ω i is done by f<strong>in</strong>d<strong>in</strong>g the real roots <strong>of</strong> g i (t), and then check<strong>in</strong>g the <strong>in</strong>tervalsbetween the roots to identify where g i (t) is negative.Assum<strong>in</strong>g that the roots <strong>of</strong> g i (t) can be computed numerically, the real roots,{λ r1 , λ r2 , . . . , λ rp }, satisfy g i (λ rk ) = 0 for k = 1, ..., p. Once the real roots are identified,the set Ω i can be found by simply check<strong>in</strong>g the sign <strong>of</strong> g i (t) on either side <strong>of</strong> λ rkfor k = 1, ..., p. Double roots are easily handled because the algorithm computes thesgn(g i (t)) on both sides <strong>of</strong> λ rk . If no real roots exist, then the function g i (t) is eitherstrictly positive, <strong>in</strong> which case Ω i is the empty set, or strictly negative which impliesΩ i ∈ IR. In this case, evaluat<strong>in</strong>g g i (0) is sufficient to determ<strong>in</strong>e the sign <strong>of</strong> g i (t) for allt. Stor<strong>in</strong>g Ω i is can be done by ma<strong>in</strong>ta<strong>in</strong><strong>in</strong>g a list <strong>of</strong> numbered pairs that representthe <strong>in</strong>tervals <strong>of</strong> feasible regions.Once all <strong>of</strong> the feasible sets have been found, the next step is to f<strong>in</strong>d their <strong>in</strong>tersection.We will def<strong>in</strong>e the <strong>in</strong>tersection to beΩ := Ω 1 ∩ Ω 2 ∩ ... ∩ Ω p (3.8)Ω is the set <strong>of</strong> feasible po<strong>in</strong>ts for which all the constra<strong>in</strong>ts are satisfied. If Ω is theempty set, then the search has failed because no where along the l<strong>in</strong>e x = x k−1 + tv

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

Saved successfully!

Ooh no, something went wrong!