406
P. Z. Korondi et al.
12.4.4 RSM Construction on Noisy Data
In the previous section, the problem was considered uncertainty-free. In this section
the applicability of RSM to noisy data is investigated. RSMs on noisy response data
have to be treated carefully and the devoted literature should be studied [16]. Here,
RSM training on stochastic data is not dealt with in its full deepness; only a limited
sense of this wide topic is presented for the reader.
To mimic a stochastic problem, an additional random term was assumed
stemming from various sources of uncertainty:
C L
C D
=
a(α − α L=0 )
C d +
a 2 (α−α L=0 ) 2
π AR
+ ,
(12.39)
where takes values between −0.5 and 0.5 randomly.
Here, analogously to previous case, the lift to drag ratio is the only measured
value. Similarly, the resources are limited and 100 data points are available for
Fig. 12.9 Lift to drag ratio (
C L
C D
) as a function of the angle-of-attack (α): analytical model response
(a) and RSMs (b)–(d)
P. Z. Korondi et al.
12.4.4 RSM Construction on Noisy Data
In the previous section, the problem was considered uncertainty-free. In this section
the applicability of RSM to noisy data is investigated. RSMs on noisy response data
have to be treated carefully and the devoted literature should be studied [16]. Here,
RSM training on stochastic data is not dealt with in its full deepness; only a limited
sense of this wide topic is presented for the reader.
To mimic a stochastic problem, an additional random term was assumed
stemming from various sources of uncertainty:
C L
C D
=
a(α − α L=0 )
C d +
a 2 (α−α L=0 ) 2
π AR
+ ,
(12.39)
where takes values between −0.5 and 0.5 randomly.
Here, analogously to previous case, the lift to drag ratio is the only measured
value. Similarly, the resources are limited and 100 data points are available for
Fig. 12.9 Lift to drag ratio (
C L
C D
) as a function of the angle-of-attack (α): analytical model response
(a) and RSMs (b)–(d)