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
Create successful ePaper yourself
Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.
and efficiency. The resulting RPTT design is compared to the PT and RPT designs<br />
presented in the previous chapters. The modal energy distributions are explored to<br />
gain physical insight into the tailored and tuned designs. The RPTT methodology is<br />
run over a range of uncertainty bounds, and the design regimes plot presented in the<br />
previous chapters is updated to include the RPTT design. It is shown that RPTT<br />
further extends the uncertainty level that can be tolerated for a given performance<br />
requirement. Finally, random hardware simulations are generated and the entire<br />
tailoring-tuning process is applied using the PT, RPT and RPTT designs. It is<br />
shown that only the RPTT designs can consistently meet aggressive performance<br />
requirements over a large range of uncertainty bounds.<br />
5.1 RPTT Formulation<br />
<strong>Robust</strong> <strong>Performance</strong> <strong>Tailoring</strong> for <strong>Tuning</strong> extends <strong>Robust</strong> <strong>Performance</strong> <strong>Tailoring</strong> by<br />
designing the system to be tunable across the entire uncertainty space. Recall that the<br />
tailoring process chooses design variables, �x, to optimize the predicted performance<br />
of the final system. In RPT, �x is chosen such that the performance predictions are<br />
insensitive to the uncertainty parameters, �p, given fixed, nominal values for the tun-<br />
ing parameters, �y. The key feature of RPTT is that the prediction accounts for the<br />
fact that the tuning parameters can be adjusted after the hardware is built and the<br />
uncertainty is fixed (and known). The goal of the optimization is to choose �x to<br />
find a design that is tunable across all possible hardware realizations. In the imple-<br />
mentation, the tuning parameters change depending on the uncertainty parameters.<br />
This consideration of future tuning adds extra degrees of freedom to the problem in<br />
the form of additional �y realizations, and, as a result, the optimization is less con-<br />
strained and may have better solutions. <strong>Tailoring</strong> for tuning idea is similar to an<br />
optimization problem presented by Scokaert and Mayne for application to receding-<br />
horizon model predictive control (MPC)[104]. The MPC formulation accounts for the<br />
fact that future control decisions are made <strong>with</strong> more information than is currently<br />
available. Scokaert’s min-max problem introduces the notion of feedback into the<br />
152