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
Y−coordinate [m] 0.04 0.03 0.02 0.01 0 −0.01 −0.02 −0.03 −0.04 not tuned tuned −15 −10 −5 0 5 10 15 X−coordinate [m] (a) Y−coordinate [m] 0.15 0.1 0.05 0 −0.05 −0.1 −0.15 −15 −10 −5 0 5 10 15 X−coordinate [m] Figure 4-6: Mode shape comparisons, worst-case RPT AO untuned (blue solid) and tuned (green dashed): (a) Mode #2, first bending (b) Mode #3, second bending. increase in OPD in this mode. 4.1.2 Design Regimes In order to assess the impact of tuning on the design space it is applied to the worst- case PT and RPT AO models across a range of uncertainty values. The results are shown in Figure 4-7, a further evolution of the design regimes plot introduced at the end of Chapter 3. The y-axis represents the performance requirement of the system and the x-axis is the level of uncertainty in the parameters. It is assumed that the uncertainty levels range ±∆ about the nominal parameter value and are the same for both uncertainty parameters. The design regimes are the numbered areas, and the design methods that are successful in each regime are listed in the legend. The addition of hardware tuning to the design process changes the design regimes significantly from those observed with PT and RPT alone (Figure 3-12). There are now six separate regimes instead of only two due to intersecting regions, where more than one technique is applicable. Consider, for example, a performance requirement of 200µm. PT is adequate for this level of performance if the uncertainty is under 120 (b)
Performance Requirement [µm] 400 350 300 250 200 150 100 50 1 5 3 4 1: PT, RPT 2: PT tuned, RPT 3: PT tuned, RPT tuned 4: PT tuned 5: RPT 6: RPT tuned 0 0 5 10 15 20 25 Uncertainty (%) Figure 4-7: Requirement vs uncertainty for PT and RPT designs with tuning: design regimes are numbered and labelled on plot. 2%. Tuning the PT design increases the tolerated uncertainty level to just about 7%. It is interesting to note that the RPT AO method is only applicable up to 3% uncertainty at this perfomrance, and that this range is only increased to 5% by the addition of tuning. Therefore there is a regime, Region 4 in the figure, in which tuning the PT design is the only successful method. This result indicates that for this problem tailoring the system to be robust actually reduces the tuning authority available for later adjustments on the hardware. At the more stringent performance requirements it is better to performance tailor the design and then compensate for the uncertainty with tuning. This approach is somewhat worrisome because the success of the mission relies heavily on the ability to tune the hardware since the predicted worst case of the PT design is many times that of the nominal performance even at the low uncertainty levels. As the requirement is relaxed the RPT and tuned RPT designs have a great effect on the design space. At a requirement of 280µm, PT is only adequate up to 121 6 2
- Page 69 and 70: also symmetric as seen in the figur
- Page 71 and 72: Chapter 3 Robust Performance Tailor
- Page 73 and 74: through careful and experienced mod
- Page 75 and 76: described above. However, one can r
- Page 77 and 78: ic, σz(�x, �p), that is depend
- Page 79 and 80: Magnitude, OPD/F x [µm/N] Magnitud
- Page 81 and 82: % Energy 100 90 80 70 60 50 40 30 2
- Page 83 and 84: metric to the cost function. Note,
- Page 85 and 86: tion: ∂hi (z,�x, �pi) ∂�x
- Page 87 and 88: values are chosen from their statis
- Page 89 and 90: Table 3.3: Algorithm performance: a
- Page 91 and 92: Statistical Robustness The statisti
- Page 93 and 94: Performance [µm] 1400 1200 1000 80
- Page 95 and 96: (Figure 3-6(b)). The nominal perfor
- Page 97 and 98: Norm. Cum. Var. [µm 2 ] PSD [µm 2
- Page 99 and 100: energy by mode for easy comparison.
- Page 101 and 102: Y−coordinate [m] Y−coordinate [
- Page 103 and 104: RMS performance, [µm] 400 350 300
- Page 105 and 106: The requirement chosen here is some
- Page 107 and 108: Chapter 4 Dynamic Tuning Robust Per
- Page 109 and 110: on a physical truss. Since tailorin
- Page 111 and 112: Table 4.1: Tuning parameters for SC
- Page 113 and 114: m 2 [kg] J ∗ # # time y ∗ [kg]
- Page 115 and 116: m 2 [kg] 800 700 600 500 400 300 20
- Page 117 and 118: configuration than the untuned, but
- Page 119: Norm. Cum. Var. [µm 2 ] PSD [µm 2
- Page 123 and 124: is considered. 4.2.1 Hardware-only
- Page 125 and 126: and added to the objective function
- Page 127 and 128: using either a decreasing step-size
- Page 129 and 130: for tailoring, but tuning parameter
- Page 131 and 132: tained by randomly choosing paramet
- Page 133 and 134: p [GPa] y ∗ [kg] Performance [µm
- Page 135 and 136: # Func. Evals Performance RMS (µm)
- Page 137 and 138: tion changes in the updated solutio
- Page 139 and 140: Data: initial iterate, p0, performa
- Page 141 and 142: the new tuning configuration is ver
- Page 143 and 144: Performing an AO tuning optimizatio
- Page 145 and 146: Uncertainty Bounds Test �y [kg] S
- Page 147 and 148: Table 4.6: Tuning results on fifty
- Page 149 and 150: eters are discussed. The optimizati
- Page 151 and 152: Chapter 5 Robust Performance Tailor
- Page 153 and 154: MPC optimization by allowing a diff
- Page 155 and 156: where the notation yij indicates th
- Page 157 and 158: (Table 4.1), and the uncertainty pa
- Page 159 and 160: Table 5.2: Performance and design p
- Page 161 and 162: it in the worst-case uncertainty re
- Page 163 and 164: The data in Figure 5-2 indicate tha
- Page 165 and 166: configuration. The tuned configurat
- Page 167 and 168: same requirement. The effect become
- Page 169 and 170: indicating that this requirement is
<strong>Performance</strong> Requirement [µm]<br />
400<br />
350<br />
300<br />
250<br />
200<br />
150<br />
100<br />
50<br />
1<br />
5<br />
3<br />
4<br />
1: PT, RPT<br />
2: PT tuned, RPT<br />
3: PT tuned, RPT tuned<br />
4: PT tuned<br />
5: RPT<br />
6: RPT tuned<br />
0<br />
0 5 10 15 20 25<br />
Uncertainty (%)<br />
Figure 4-7: Requirement vs uncertainty for PT and RPT designs <strong>with</strong> tuning: design<br />
regimes are numbered and labelled on plot.<br />
2%. <strong>Tuning</strong> the PT design increases the tolerated uncertainty level to just about<br />
7%. It is interesting to note that the RPT AO method is only applicable up to 3%<br />
uncertainty at this perfomrance, and that this range is only increased to 5% by the<br />
addition of tuning. Therefore there is a regime, Region 4 in the figure, in which<br />
tuning the PT design is the only successful method. This result indicates that for<br />
this problem tailoring the system to be robust actually reduces the tuning authority<br />
available for later adjustments on the hardware. At the more stringent performance<br />
requirements it is better to performance tailor the design and then compensate for the<br />
uncertainty <strong>with</strong> tuning. This approach is somewhat worrisome because the success<br />
of the mission relies heavily on the ability to tune the hardware since the predicted<br />
worst case of the PT design is many times that of the nominal performance even at<br />
the low uncertainty levels.<br />
As the requirement is relaxed the RPT and tuned RPT designs have a great<br />
effect on the design space. At a requirement of 280µm, PT is only adequate up to<br />
121<br />
6<br />
2