37# Constra<strong>in</strong>t # Constra<strong>in</strong>t1. h 1 ≥ h m<strong>in</strong> 6. L 2 ≥ 10h 22. h 2 ≥ h m<strong>in</strong> 7. X ≥ 2L 2 + 2h 1 + b m3. b m ≥ h m<strong>in</strong> 8. Y ≥ 2L 1 + h m4. h m ≥ 2h 2 + h m<strong>in</strong> 9. k y ≥ αk x5. L 1 ≥ 10h 1 10. β max ≥ βTable 5.2: <strong>Design</strong> constra<strong>in</strong>ts for crab-leg resonator.enforce that the stress, β, at the jo<strong>in</strong>t (where the two beams jo<strong>in</strong> to form a leg) doesnot exceed a maximum. The nonl<strong>in</strong>ear expression for β is given byβ =12Eh 3 1h 3 2L 1 Dh 2 2L 2 1(4h 3 1L 2 + h 3 2L 1 )where D is the maximum allowable deflection <strong>of</strong> the structure <strong>in</strong> the x-direction.Once aga<strong>in</strong>, through simple algebraic manipulation the constra<strong>in</strong>ts can be written aspolynomials <strong>in</strong> the form g i (x) ≤ 0. Table (5.3) conta<strong>in</strong>s the values <strong>of</strong> constants usedfor this example. All <strong>of</strong> the design parameters were chosen to be realistic.<strong>Design</strong> Parameter - DescriptionValuet - thickness <strong>of</strong> pro<strong>of</strong> mass and beams 2.0 µmρ - density <strong>of</strong> silicon (2330 kg/m 3 ) 2.3 × 10 −12 gm/µm 3E - Elastic Modulus <strong>of</strong> silicon (160 GPa) 1.6 × 10 8 gm/µms 2h m<strong>in</strong> - m<strong>in</strong>imum dimension 2.0 µmD - maximum deflection <strong>in</strong> the x-direction 2.0 µmβ max - maximum allowable stress (1.6 GPa) 1.6 × 10 6 gm/µms 2X - maximum size <strong>of</strong> structure <strong>in</strong> the x-direction 600µmY - maximum size <strong>of</strong> structure <strong>in</strong> the y-direction 600µmα - m<strong>in</strong>imum stiffness ratio k y /k x 16Table 5.3: <strong>Design</strong> parameters for crab-leg resonator.5.2.1 Determ<strong>in</strong><strong>in</strong>g the set <strong>of</strong> feasible resonant frequenciesCurrently we have a rational polynomial that describes the resonant frequency <strong>of</strong>the crab-leg and a set <strong>of</strong> constra<strong>in</strong>ts. There is no notion <strong>of</strong> uncerta<strong>in</strong>ty yet, but the
38problem is well-def<strong>in</strong>ed. Before a designer tackles the robust problem, a good firstquestion to address is “what range <strong>of</strong> performance can be achieved subject to theconstra<strong>in</strong>ts?” For this example, we would like to f<strong>in</strong>d lower and upper bounds on theachievable resonant frequencies.F<strong>in</strong>d<strong>in</strong>g the lower-bound requires m<strong>in</strong>imiz<strong>in</strong>g w 2 n subject to the constra<strong>in</strong>ts. Conversely,the upper-bound can be found by maximiz<strong>in</strong>g w 2 n. Maximiz<strong>in</strong>g the resonantfrequency however is equivalent to m<strong>in</strong>imiz<strong>in</strong>g 1 , or −ww n. 2 Solv<strong>in</strong>g for the two boundsn2requires m<strong>in</strong>imiz<strong>in</strong>g a rational polynomial subject to polynomial constra<strong>in</strong>ts; whichis precisely the type <strong>of</strong> problem we designed our optimization algorithm for.The lower, w m<strong>in</strong> , and upper, w max , bound were found to be 13 kHz and 11.8MHz, respectively. Computationally, the calculations typically took less than 1500iterations, or 12 seconds, to compute on a P4 1.8Ghz L<strong>in</strong>ux workstation.If thetarget resonant frequency is outside <strong>of</strong> the feasible range, then either the design orconstra<strong>in</strong>ts need to be modified.5.2.2 Choos<strong>in</strong>g a target frequencyWith an understand<strong>in</strong>g <strong>of</strong> the system’s achievable performance, given the designand constra<strong>in</strong>ts, we chose a target resonant frequency, w target , <strong>of</strong> 200 kHz. Beforetackl<strong>in</strong>g the robust problem, we would like to f<strong>in</strong>d the set <strong>of</strong> designs that have aresonant frequency <strong>of</strong> w target when there is no uncerta<strong>in</strong>ty. There are two importantreasons which motivate us to explore this set. First, <strong>in</strong> study<strong>in</strong>g this set we hopeto demonstrate that there is considerable freedom <strong>in</strong> how the design variables arechosen. Secondly, by f<strong>in</strong>d<strong>in</strong>g a set <strong>of</strong> designs that achieve the desired performanceunder nom<strong>in</strong>al conditions (δ = 0), we will have a set for which we can compare therobust design. We will refer to this set as ΦΦ = {x : wn 2 = wtarget, 2 g i (x) ≤ 0 for i = 1, ... , m} (5.9)Tak<strong>in</strong>g w target to be 200 kHz, the set Φ conta<strong>in</strong>s an <strong>in</strong>f<strong>in</strong>ite number <strong>of</strong> designs. Whileequation (5.9) is easily written, f<strong>in</strong>d<strong>in</strong>g the set Φ is an impractical task. Instead,we chose to f<strong>in</strong>d a subset, which we will refer to as Φ 1000 , that conta<strong>in</strong>s 1000 uniquedesigns.