Chapter 5 Robust Performance Tailoring with Tuning - SSL - MIT
Chapter 5 Robust Performance Tailoring with Tuning - SSL - MIT
Chapter 5 Robust Performance Tailoring with Tuning - SSL - MIT
You also want an ePaper? Increase the reach of your titles
YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.
Multiple Model<br />
The multiple model RPT optimization is run on the SCI sample problem using both<br />
SA, SQP and MC SQP algorithms. The optimization performance metrics and re-<br />
sulting designs are listed in Table 3.4. The optimal costs listed in the table are much<br />
Table 3.4: Algorithm performance: multiple model, βi =1/npv.<br />
J ∗ iter fevals time x ∗ [m] x ∗ [kg]<br />
Alg. [µm] # # [min] d1 d2 m1 m2<br />
SA 273.51 64 2059 12.79 0.0440 0.0515 0.8464 19.5131<br />
SQP 271.78 20 48 3.83 0.0432 0.0529 0.0 22.613<br />
MC SQP 271.78 238 545 41.06 0.0432 0.0529 0.0 22.613<br />
lower than those of the AO designs (Table 3.3), since the multiple model objective<br />
is the weighted sum of the performance values at each of the uncertainty vertices,<br />
and not the worst-case performance. In the results presented here the weighting βi<br />
is the same at each vertex and equals 1/npv, so that the cost is the average of the<br />
performance values at the uncertainty vertices.<br />
The SQP and SA designs are nearly equivalent as the SA cost is only 0.6% higher<br />
than SQP. The individual runs in the MC SQP algorithm result in a few different<br />
optimal solutions <strong>with</strong> slight variations in the diameter of the inner array segments<br />
and the design mass values. All designs, however, are at the maximum mass con-<br />
straint, and the best design is found <strong>with</strong> either MC SQP or the combination of SA<br />
and SQP. As seen previously <strong>with</strong> AO, using SA in conjunction <strong>with</strong> SQP finds the<br />
global optimum much more quickly (16.62 minutes) than performing ten randomly<br />
started MC SQP searches (41.06 minutes). The optimal cross-sectional diameters are<br />
similar to those in the anti-optimization designs, in that the diameters of the inner<br />
truss segments are larger than those of the outer segments. However, the MM design<br />
has 22 kg of design mass on the positive-x arm of the interferometer while the AO<br />
design has no lumped mass at all.<br />
90