Chapter 5 Robust Performance Tailoring with Tuning - SSL - MIT

tained by randomly choosing parameters from the uncertainty distribution. In order
for a system to be a candidate for tuning the hardware performance must be worse
than the requirement. For this example, the RPT AO design at ∆ = 10% uncer-
tainty is used to generate a hardware model, and the requirement is set at 280µm.
The model parameters and performance data for the resulting hardware model is
listed in Table 4.3. Recall from Chapter 3 that in RPT design all of the mass is given
Table 4.3: Hardware Model Data.
Tailoring, x 0.0486 0.0581 [m]
Uncertainty, p 76.6 65.4 [GPa]
Tuning, y 0.0 60.01 [kg]
HW Performance σHW 290.83 [µm]
Tuned Performance σHWt 262.75 [µm]
to the cross-sectional diameters and the design masses are not used at all. However,
10% of the allowable mass is reserved during the design stage as margin and is now
available for tuning. The uncertainty parameters listed in the table are chosen ran-
domly from the uncertainty model. The performances, σHW and σHWt, are the system
performance variances before and after tuning, respectively. The tuned performance
is obtained by directly tuning the hardware with the known uncertainty values as in
the previous section. This result serves as a baseline for the other tuning schemes
and confirms that this system can be tuned below the requirement.
Model Tuning
To begin, the four model tuning schemes, nominal model tuning (NomMod), AO
model tuning (AOMod), worst-case model tuning (WCMod), and optimal model
tuning (OptMod) are applied to the hardware simulation. We start with model-only
tuning in the example since it is the easiest to implement and requires no additional
hardware testing. In a sense, these tuning methods are “free” and, at the very
least, provide starting points for the hardware tuning schemes. The results of these
schemes are presented in both tabular and graphical form in Figure 4-9. The table
(Figure 4-9(a)) includes the uncertainty parameters used in the model, the tuning
131