Information Technology Reference
In-Depth Information
The algorithm of Section 4 matched the target pressure distribution (the norm of the
distributions is less than 2
10
-5
) using less than 22 equivalent high-fidelity model
evaluations. The direct optimization of the high-fidelity model using the pattern-search
algorithm required 401 function calls to yield a comparable matching with the target.
×
Table 1.
Numerical results for design Cases 1 and 2; axisymmetric hull direct drag
minimization and inverse design with a target pressure distribution, respectively. In both cases,
the flow speed is 2
m
/
s
and the Reynolds number is
2⋅10
6
.
All the numerical values are from the
high-fidelity model.
N
c
and
N
f
are the number of low- and high-fidelity model evaluations,
respectively.
F
is the norm of the difference between the target and the design shapes.
Case 1 (Drag minimization)
Case 2 (Inverse design)
Variable
Initial
Pattern-search
SBO
Initial
Pattern-search
SBO
a
15.0000
21.8611
20.9945
18.000
24.7407
24.7667
x
n
5.0000
5.6758
5.6676
7.0000
7.3704
6.8333
y
n
5.0000
2.7022
2.7531
8.0000
4.7407
4.5667
x
t
90.0000
98.000
96.6701
85.0000
88.1111
88.6333
y
t
5.0000
0.8214
3.0290
7.0000
5.5926
5.3000
F
-
-
-
0.0204
1.64E-5
1.93E-5
C
D
0.0915
0.0853
0.0857
0.0925
0.0894
0.0893
N
c
-
0
300
-
0
500
N
f
-
282
3
-
401
5
Total cost
-
282
13
-
401
< 22
Fig. 10.
Case 1 results for the direct hull drag minimization, showing the initial and optimized:
(a) pressure distributions (b) skin friction distributions and (c) hull shapes; Case 2 results of the
inverse design optimization with a prescribed target pressure distribution: (d) target, initial, and
optimized pressure distributions, (e) zoom on two regions of Fig. 15(d), and (f) the initial and
optimized hull shapes
Search WWH ::
Custom Search