Copyright
Yasufumi Yamanouchi.

A review of statistical studies of seakeeping qualities online

. (page 17 of 20)
Online LibraryYasufumi YamanouchiA review of statistical studies of seakeeping qualities → online text (page 17 of 20)
Font size
QR-code for this ebook


(From Roberts. 59 )



The equivalent linearization theory is abbreviated as EL theory. In this figure, only the
perturbation theory appears different from the rest, but here we have to remember that in
the perturbation method this is the first order approximation and we can improve the ap-
proximation by increasing the order of the approximation, as indicated in the discussion
in Section 10.2.2.

12.63 Nonlinear Oscillation in Nonwhite Excitation

Roberts 61 further extended the scope of applicability of this method and solved for
nonlinear rolling in nonwhite excitation. The equation of motion is now



I<j> + PC(</)) + k(<p) = M(t)



(12.96)



or



<P+PF(<f>) + G(<f>)=x(t)



(12.96')



where x(t) is a nonwhite excitation, and can have a colored spectrum.
Again, he adopted the total energy envelope as Eq. 12.86

V = £-+C7(0)



where



U(<j>)=lG£)d£



and considered that this was slowly varying.



350



He set



^- = -Vsin 2



u(<p) = v cos 2 e



<p=-j2V±sine



u(<pp = V2 C ose



(12.97)



introducing the phase 6 as shown in Fig. 12.7. After manipulation of the relations, the
equation of motion, Eq. 12.96', was inverted into the equation of V and<p. When the
phase process 6 was modified into a process k{t), and the joint process Z = (V, A) was
made to converge into a Markovian process, the transitional distribution function
P(Z\Zq\ t) was found to be governed by a Fokker-Planck equation of second order. The
stationary solution of the Fokker-Planck equation p(Z) was obtained, and from this
expression V and X were found to be uncoupled, and p(V) was calculated. With the
relations of V, <p and0, p(V) was modified into p{<p,<p).




[U(<»]2

u.v vs.e,<t>



Fig. 12.7. U, Vvs.6,<p.
(From Roberts. 61 )

For nondimensionalized, nonlinear rolling expressed as

' + a4> + b\$\ <j> + <p - <j> 3 = x(t),



(12.98)



Roberts calculated the probability density function of the nondimensionalized amplitude
of rolling A by the ME theory and compared it with the Rayleigh distribution obtained by
the linear theory as in Fig. 12.8. He 61,86 also compared his results with the nonlinear sim-
ulated data obtained by J.F. Dalzell 85 to show the validity of his method. Examples are
shown in Figs. 12.9 through 12.11 for a variety of damping coefficients a and b, where
Q = (o p / (Do, (t) p is the peak frequency of the excitation, anda w is standard deviation

of the input process x(t). Process 3 in the figures is a wide banded excitation for this
example. He showed many other results of comparison for other types of excitation with



351



different bandwidths, named Process 1 and Process 2, used in the simulation by J.F.
Dalzell. 85 The deviation from the Rayleigh distribution in Figs. 12.8 to 12.11, and the
relation of the standard deviation of roll Or to that of input o w in Fig. 12.12 shows the
extent of nonlinearity. The applicability of the method was also discussed.



P(A)



10 1 F

p(A!
10°

10- 1
10-*



10" 3




PRESENT THEORY



WITH LINEAR RE- ^^4
STORING MOMENT



0.1 0.2 0.3 0.4



Fig. 12.8. Probability density function for amplitude A:
a = 0, b = 1 , a w = 0.036, Q = 0.90,
Process 3. (From Roberts. 61 )



0.999

0.99

0.95
0.90

d ,ai 0-80
P(A) o.70
0.60
0.50
0.40
0.30
0.20
0.10
0.05
0.02




i 7T — ; — i r



>— RAYLEIGH



DALZELLS SIMULATION
ESTIMATES
a«0
* b-1



0.1 02 0.3 0.4 0.5 0.6 A



Fig. 12.9. Cumulative probability distribution for amplitude A:
a = 0, b = 1 , a w = 0.036, Q = 0.90,
Process 3. (From Roberts. 61 )



352



P(A)



0999



099 -



095
090

080
O70

(WO
0-50
0*0
030
020

OW-
005-




DALZELL'S SIMULATION

ESTIMATES

♦ b= 01

«b=10



002

001

0005



01 0-2 03 0-4 0-5 0-6 A

Fig. 12.10. Cumulative probability distribution for amplitude A:

a = 0.01, o„ = 0.036, Q = 0.90, b = 0.1 and 1.0;
Process 3. (From Roberts. 61 )



0999


-


i i 1

[,


I i — r 1

/_ PRESENT

I THEORY /


099




V

1/


b=3 /

/^-PRESENT


095


-


t


Y THEORY


09


-




T/l b=OI


08


-


0/
0/


>*


PMJ07




©/


yr


06




°/


/•


05






/^


04


-




-


03


-




DALZELUS SIMULATION-


02






ESTIMATES
I brM


OW






\ b=3


005






-


002


-




o =003


001


-




-


0005





' i


■ i i i



02 03 04 05



06 A



Fig. 12.11. Cumulative probability distribution for amplitude A:

a = 0.03, a„ = 0.036, Q = 0.90, b = 0.1 and 3.0;
Process 3. (From Roberts. 61 )



353



Or



DALZELL'S SIMULATION



0.20


. b .= 1 .0

• b= 03
o b = 1-0


*Z


0.15




/ $S ~


0.10




s^°


0.05




a = 0.01



0.01 0.02 0.03 0.04

Fig. 12.12. Variation of standard deviation of roll o R with standard deviation of wave
input o w : a = 0.01; Q = 0.90; b = 0.1, 0.3, and 1.0; Process 3.
(From Roberts. 61 )

12.7 PROBABILITY DENSITY FUNCTIONS OF AMPLITUDES, EXTREME
VALUES IN RELATION WITH THE FUNCTIONAL POLYNOMIALS

12.7.1 Narrow-Banded Case

As was mentioned in Section 11.1, when a nonlinear response z(t) was expanded by
the Voltera expansions (or functional expansions)



2(0



= X ■ • • 8n(ri,V2 ■ ■ ■ ■ T„) X(t-Ti)x(t-T2)



. . . . x{t-x n) dxi dr 2 . . ■ ■ dx n , (12.99)



the terms for n>2 can be considered as the modifying terms of the Taylor expansion of
this process around its linear term for n=\. If we take until n=2, and using small e, some-
times the response z(t) is expressed as

m = J gid) x(t-x) dx + e\ J g 2 (r 1 ,x 2 )x(t-x 1 )x(t-x 2 ) dx x dx 2 . (12.100)

The derivatives in terms of time is
i(0= J gi{x)x(t-x)dx+2e\ J ^ 2 (Ti,T 2 )ic(r-TiMr-T 2 ) dx x dx 2 , (12.101)



— 00 —00



354



thinking of the symmetry character of g2(t\, tz) - gifti^d- On the basis of this expres-
sion T. Vinje 87 formulated a general method for getting the probability distribution
function of the extreme values, under the assumption that the response z{t) is narrow
banded and the input x{t) is Gaussian with variance cr. His method is also based on the
assumption that the probability distribution function of extreme values can be obtained by

the joint distribution function p(z,z) only.

Now, we start from the statistical moment generating function to get p{z\,zi)

[here z\ = z{f), zi = z{t) z], that is, from the double Fourier transform of p(zi,Z2),

0(0 1 ,0 2 ) = 4 ex P{'(01*l + 02Z2)}].






expO'0iZi + i02Zz)p{zi, z-i)di\dz2. (12.102)



Then expanding exp {iQ\z\ + 1Q2Z2) into a series gives

exp (i0 lZ i + 102*2) = 1 + X -TT^lA^. (12.103)



m,n=V



assuming m, n are positive integers whose sum is greater than zero.
Inserting Eq. 12.103 into Eq. 12.102 gives



Here



0(01,02) = 1 + X-^fVc^irc^r. (12.104)



fi mn = E[z?,z% =| J i%z\ pb^zddzxdz* (12.105)



From the Fourier inverse transform of Eq. 12.102,



00 00
p(z h z 2 ) = — — 2 I J 0(0i,02)exp{-i(0izi + 2 22)}^0i^02. (12.106)



Putting Eq. 12.104 in Eq. 12.106



355



M J j Zn m - n -

— x — oe

(12.107)

This Eq. 12.107 shows that p(z h z 2 ) can be calculated if we know all the m,n\h order
moments of this probability function.

On the other hand, by the definition of the cumulant km* and the cumulant
generating function K(i6\,i02),

K(i6 1 je 2 ) = iog<p(e 1 ,e 2 ) = j^;(iei)(ie2)+^(iei)(ie2) 2 + . . .

= X-TT(^l) m (^2) n . (12.108)

mini

77! ,71

Therefore



(pWi,6 2 ) = exp(*r(i0i,i0 2 )} = expj J^^iieiTWiT

I m,n m - n -



1! ^ m!w! 2! _ f"!«!



+ — { .... } 3 + ... . (12.109)



When we insert this $(6 uOj) into Eq. 12.106,



DC CO

pfri, *2) = 7A2 f f «p{*(# 1, i0 2 ) - i(A 1Z1 + 02^)}^ id0 2
(2jt) J J

— ec —oo

(2jz) 2 J J H^m!«! J

_0C —00 *■ ■*

From Eqs. 12.104 and 12.109, )fc m ,„ and fi mjn are related as follows:



(12.110)



356



*02= i "02-Aoi

k03=[M)3- 3/*OL"02 + 2^01

^10 = /"10

hi =Pii-f*wPoi

hi = ^12 -PicPca - 2uoiMu + 2aio"6i

&20 = /*20 ~/%

*2i = fill - 2^10^11 -fioiM20 + lu-iduoi



^30=^30-3^10^20 + 2^



10



(order in e)


(0)


(1 + e 2 )


(e+e 3 )


(6)


(0)


(e+e 3 )


d+e 2 )


(0)



(6+e 3 )



(12.111)



Following these general formulations by Vinje, M. Hineno 67 calculated the proba-
bility distribution function of the wave height, treating the wave process as nonlinear, as
was mentioned in Section 9.2.1, and will be summarized as follows.

From these relations with p, mn the order of i m ine's was obtained, which is a
smallness parameter that appeared in Eq. 12.100 as is listed in Eq. 12.111. Expanding
Eq. 12.110 around e = into a Taylor series and truncating at 0(e ) gives



pOi, ^2) =



W



exp



-|*2o(O)0l-!*te(O)0i



1 + *io(O)(i0i) + - *3o(O)(z0i) 3 + - *d3(O)(i02)-





- ^'i2(0)(^i)(/6> 2 ) 2



^^*® d&xdd^



(12.112)



where fc,y(0) = *y| e=0 , *y(0) = — £<;| e=0 -
After manipulation,



357



1



(_2 _2

2k 2 o 2*02



hU kw



Jho



**> H3|^.] + l_*S8_ ff3



y/^20 / 6 £ 2 o v^20 \ V&20 I 6 koi Jkm ' \ v*02



2 *02 v/^20 \ V ^20 / \ V&02



. (12.113)



Here Z? m (x) is a Hermite polynomial,

00

H n (x) = -i f r^Oo + i /2 f) n df



ff oW=l

B 2 (x) = x 2 -1

H 3 (x) = x 3 -3x



H n+ i (x) = xH n (x) - nH„^i (x)



(12.114)



convention of (0) is omitted.

Under the assumption that z\{t) is narrow banded, the number of zero-crossings of

an output z(t) is equal to the number of maxima.

Therefore, the expected number of z-up crossings per unit time is given as already
shown in Section 12.5, Eq. 12.44,



E[N + (2i)]=| J |z :



2I p(zi,z 2 )dz 2 .



(12.115)



358



Also, the expected number of mean level up-crossings is



(12.116)



The probability distribution function of a maxima then lies above Zj is
Substituting Eq. 12.113 into Eq. 12.115 and carrying on the integration gives



(12.117)



E[N + (z)] = — / -^ exp (

2n -/ £ 2 o \ 2£ 2 o



1+4^^!




*12



+ - — 1± r=H 1




(12.118)



As already shown in Eq. 12.111, the cumulants are calculated from the m,nth
moment of response, and the m,nth moment can be expressed by using the frequency
functions and the input spectrum as some of the examples that were shown in Section
11.4. With the order of each term up to (e) taken into account, and with the spectrum
function, which is a real function, the cumulants can be written as follows:



359



£ 10 = e/z 10 =e s(co) G2(co,-(o)d(o

— 00
00

■/

— 00

00

=1

— 00

(
#12 - €hi2 = € (<Di(D2 + 2(02) S((0i)s(Q)2)



k 20 = h 2 o= I s(a>) \Gi((ofd(o



^02 = ^02 = (O S((0) \Gi((0)\ da>



— 00

00 00

,2n



x [Gi(- co i)Gi(-eo 2 )G 2 (£o i,(0 2 ) (12.119)

+ Gi(-C0i)Gi(-C02)G2(£0i,Ct>2)] dO)\d(l)2



00 00



^30 = e^30= 3e s(<ui)(,sa>2)

— 00 —00

x [G 1 (-w 1 )G 1 (-(o 2 )G2((Oi,(02)

+ G*i(-o)i)G*i(-(02)G2((0 1,(02)] d(0ida)2,

where * indicates a complex conjugate,

s(q)) is the two-sided spectrum of input x(t),
G\{a>) is the linear frequency response function



00

-I



Gi(a>) = giirfc^dd,



360



and G2(co 1,0)2) is the quadratic response function

00 00

G 2 (co U (o 2 )= J { g2(n,r2)e- n ^- h ^dr 1 dT 2 .



Through these manipulations, Hineno obtained his expression for the probability
distribution function as



/>(z) = exp -



2h 2 o



h 2 o 6h 20 \ h 20 I 2aiq2«20



(12.120)



The probability distribution density function p(z) and the expected \/n highest value
z\ u are obtained from the relations



p(z) = -—P(z),
dz



(12.121)



J z p(z)dz



'■\/n



Zl/„



/ P(z)dz



00

-I



z p(z)dz.



(12.122)



Zl/n



If we use Eq. 12.120 for P(z) and consider that in real calculations, the smallness
parameter c is absorbed in the computation of G 2 (a> i,o>2) and does not appear explicitly,



p(z) = exp -



z 2
2/j 2 o



^10 J ^30 h\2 1

h 5 + z

/Z20 2/l 2 2/220^02 ^20



] 1 / feig /l 30 | /in \ /?30 | _2 [ ^30 j

yh 2 o\h 20 2h 2 20 2h 20 ho 2 ) 2^ J " 6^



(12.123)



This is the final expression of the probability distribution density function of the maxima.
Using this expression, M. Hineno calculated the probability characteristics of the maxima
and minima of the waves that were treated as nonlinear with the quadratic response
function, as shown in Eq. 9.10 in Section 9.2.1,



361



G 2 {coi,(o 2 )=— (aj + coij
G 2 {a)i,-(D 2 ) =-—(col-(
with the linear response assumed as G\(w) = 1 .



10- 1



P^



10"V



10-^ -




Fig. 12.13. Probability density function of the maxima of waves.
(From Hineno. 67 )

Figure 12.13 shows the probability distribution density function for maxima of the
waves compared with the experimental data by a certain research worker for model
waves with a wave spectrum that is almost of the Pierson-Moskowitz type. In Fig. 12.13,
the + marks indicate the experimental data analyzed as nonlinear waves and the other
marks (A, O) are experimental data analyzed on the assumption of linear waves, the solid
line being the theoretical relations as linear for this model waves. Results computed by
Eq. 12.123 are shown by a dotted line, and the agreement with + signs is quite good,
especially in the larger amplitude range, which is important practically.

Figures 12.14, 12.15, and 12.16 illustrate the calculated results for nonlinear waves,
using a modified Pierson-Moskowitz wave spectrum with //1/3 = 11.6 m, Toi = 16.1

sec as the input. From these figures, we can fmd the extent of nonlinearity of the waves
and also the effects of nonlinearity or the effect of the distortion of the wave forms on the
difference of the probability distribution function for maxima and minima.



362



Hineno 88 also calculated the relative motion of a semi-submersible in this kind of
nonlinear wave, ass umin g a linear response to the excitation by the waves.



1.0

o.s

O.G
0.4
0.2
0.0





H 1/3 - 11.6m


T<> -16.1 sec


■\


MAXIMA




MINIMA "


LINEAR '




-







2



Fig. 12.14. Cumulative probability distribution function of wave amplitude.
(From Hineno. 67 )



H 1/3 - 11.6m

T a - 16.1 sec

MAXIMA

MINIMA ~
LINEAR




Fig. 12.15. Probability density distribution function of wave amplitude.
(From Hineno. 67 )

12.72 Wide-Banded Case

J. F. Dalzell 56 (1984) extended this technique further. He did not assume narrow
handedness of the response and did not truncate the functional polynomials at n=2 but
continued to n=3. He thus formulated the technique for calculating the probability distri-
bution function of extremes of the nonlinear responses to Gaussian inputs. Here the
characters of the nonlinear frequency response functions up to degree 3 are assumed to be
known from the analysis as discussed in Section 11.5. The nonlinear response Y(t) to
Gaussian input X(t) is expressed by



363



4.0




Fig. 12.16. Expected 1/n highest values of wave amplitude.
(From Hineno. 67 )



Y(t) = Jgi(Ti) x(t-T{) dci

+ 82(Ti,T2) x(t-n) x(t-z 2 ) dXidX2

+ g3(Ti,T 2 ,r 3 ) x{t-ti) x(t-r 2 ) x(t-r 3 ) dr 1 dr 2 dx i . (12.124)

(Limits of integrals - » — i- oo are omitted throughout this section.)

As was assumed in Section 11.5, here the kernels or the «th degree impulse
response functions g n (X\,X2 . . . r n ) are real, time invariant, completely symmetrical in
the variables g„(Ti,T 2 . . . t„) = g„(r 2 t 3 . . . t„,ti)= . . . for any rearrangement of the
variables r ; , and sufficiently smooth and integrable so that there exist n-fold Fourier
transforms,

g„(Tl,t 2 . . . T ")=7^r I I • • • I G n(a>l,0>2 ■ ■ • 0)n)



exp i^ojyr,-ptt)ida)2 . • . dco„ (12.125)



364



G n ((l)i,(D2 ■ ■ ■ 0) n )= J ... I g„(Ti,T 2 , . • - r„)



exp



. 7=1



<fridr 2 . . . dr„. (12.126)



The function G„(a>i, a>2 . . . (o„) is the nth degree frequency response function and
is also symmetric in its arguments G„(a>\,(02 . . . co„) = G n ((02,(0\, . . . (O n )= ... for
any rearrangement of (Oj because the impulse response functions are real.

G n (-(O h -Q>2, ■ ■ ■ -(0„) = G* n (Q)i,0) 2 , ■ ■ ■ (O n ) (12.127)

here the * denotes the complex conjugate.

In Dalzell's paper, 56 the spectrum 5^(0)) was defined a little differently from those
used by this author in Parts I, II, and in. He used 2n times our s(co) ,

S^co) = 2jcs((0), (12.128)

and also took the one-sided spectrum

tfxxtM] = 2s(co) = -S^w), forcu > 0. (12.128')

71

The initially assumed functional polynomial process was reformulated as the
response to a white noise excitation, and a new set of frequency response functions was
defined which contains both characteristics of the original frequency response to the
excitation and that of the excitation spectrum.

The two-sided spectrum of white noise is

S w (a>) = 1, (12.129)

and the autocorrelation of white noise is a delta function

Rww(r) = — exp[i(OT]dco = 6(t). (12.130)

2jc j

We think of the spectrum filter L{w) that expresses the spectrum of input 5^(0))
with the white noise as

S^ico) = al \L(a))\ 2 (12.131)



365



Ua» = lS ^ )] " 2 . (12.131-,



The linear, quadratic, and cubic frequency response functions connecting the white
noise W(t) with the output Y(f) will be

o x H\{a>) = o x L{o))G{(o)
olH 2 ((D h (D2) = a 1 x L{0} l )L{0} 1 )G 1 {(Oi,(O2) (12.132)

O x H 3 ((Di,(D 2 (D3) = olL((0 1 )L(Q) 2 )L((03)G3((Oi,(02,C03,).

Then Y(t) can be related to the white noise input W{t) as
Y(t) = o x \h 1 (x l )W(t-x 1 )dx 1



+ o\ J J h 2 (? h r 2 )W(t - n)W(t - x 2 )dx 1 dx 2

+ o\\ \h 2l {x l ,x 2 ,X3)W{t-x l )W{t-x 2 )W{t-r ?> )dx l dx 2 dr3, (12.133)

where

h\{x\) =— — Hi(o))exp[ia)Xi]da)

2ji j

h 2 (t u x 2 ) = ——j H 2 (coi,co 2 ) exp[z'(<yiri + (o 2 x 2 )]da)idw 2 (12.134)

hs(Ti,X 2 ,Tl)= 3 // 3 (<yi,G) 2 ,G)3)exp[/(a>iT 1 + a)2r2+CU3T3)]da>i^(U2rf(y;.

The derivatives of the output are then
Y(t) = a x \h' 1 (x 1 )W(t-x 1 )dx 1

+ (j2 A\ fori, r 2 W(t - x x )W{t - x 2 )dx x dx 2

+ ol\ | J hzixux^xzWit-XxWit-xJWit-xiWxrfxrfxz, (12.135)
where



366



*iw - y



H\{a>) exp(i(OT\)da)



l_



h2(ri,T 2 )= 2 I I ^2(tt>1.0»2)«q>[/(tt»lTl+fi>2T2)]^Wl^»2 (12.136)



^3(^1,^2,^3) = 3 I I H3((Ouo>2,o)3)exp[i(a)iTi+ 0)2*2 + Q>3 r 3)]dQ>ido>2d<o?,.



The terms //„(<y) are defined as

H\(o)) = ia)Hi(co)

H2(0)\,0)2) = i(Q)i + 0)2)H2(0)i,Q)2) (12.137)

H ?,(0) 1,0)2,0)?,) = i(0)\ + 0)2 + 0)-i)H-s{0) 1,0)2,0)3)

and



-a x \h

ol\ h2(Xi,



r 2 )W(t - Ti)W(t - T 2 )dr 1 dT 2

k(Ti,T2,T3)W(t-Ti)W{t-X2)W(t-t3)dTidt2dr3, (12.138)



where



^1 ( r i) = — U\ (o))exp(io)Ti)do)
2ji j

^2(Tl,T 2 ) = ——J ^/2(^l J O> 2 )exp[j(C0iTi + 0)2T2)]d(DidO)2 (12.139)

^3(Tl,T 2 ,r 3 )=— — 3 //3((U 1 ,a)2,W 3 )exp[/(a>iTi+G)2T2+O' 3 T3)]rfcyidC02rf0>3,



and //„ are defined as



367



Hi(a)) = -o) 2 Hi(o))
H 2 (o) 1,(02) = -(o)i + a)2) 2 H 2 (co 1,0)2)

H 3 (0)i,0) 2, 0)3) = -(0)1+0)2 + 0)3) 2 Hi(0) 1,0)2,0)3).



(12.140)



In addition, the characteristics of the products of white noise were fully utilized as

follows:

\

Wi(t-Ti)W(t-T 2 ) ■ . . W(t -T N ) = for N odd,



W(t-Xi)W(t-X2) = <5(ti - r 2 ) = 6 12,

W(t - Ti)W(t - T 2 )W(t - T?,)W(t - T 4 ) = 6 12634 + 613624 + <5 14^23,

W(t - n)W(t - z 2 )W(t - r 3 )W(t - u)W(t - x 5 )W(t - t 6 )

= 612634656 + 612635646 + 612636645
+ 613624656 + 613625646 + 613626645
+ 614623656 + 614625636 + 614626635
+ 615634626 + 615632646 + 615636624
+ 616634652 + 616635642 + 616632645,



) (12.141)



where (5^ = d(r, - xj).

Eq. 12.141 is the special case of Eqs. 11.68, 11.69, and 11.70 in Section 11.5.

If we do not assume narrow handedness of the output spectrum, the expected

number of maxima of the response Y greater than Y = | per unit time is approximated as



N.



00
! + = J J \Y\ p(Y,0,Y) dY dY.

I-00



(12.142)



Similarly, the expected number of minima of Y less than Y = | per unit time is
approximated as



368



tff= I \Y\ p(Y,0,Y) dY dY. (12.143)

-oc

Then, the expected number of maxima regardless of magnitude per unit time will be

oc o

Nl = [ \Y\ p(Y,0,Y) dY dY. (12.144)

— ce —so

."s imilar ly the expected number of minima per unit time will be



N a = \Y\ p(Y,Q,Y) dY dY. (12.145)

-oo

Because maxima and m inim a are paired in the same record of response A^ = N& ,
from Eqs. 12.142 and 12.144, the probability that a maximum will be less than £ is ap-
proximated



Prob[Maximum < £] = 1 - I — %■ J . (12.146)



Similarly,



Prob[Minimum < |] = —I- ). (12.147)

\ No* /



Then, the probability densities of maxima and minima are obtained by differentiat-
ing Eqs. 12.146 and 12.147 with respect to |, as



p + ® = -X- f 171 p£, 0,Y)dY (12.148)

N<x> J



369



00

p~(M) = -£=- f lii pd, o, f) dr. (12.149)







In the same way as we did for the joint probability density p(Y, Y) in Section 12.7.1,
relations between the joint moment generating function <f>{it, it, it ) and the joint cumulant
generating function K(it, it, it) were used to give



(it, it, it) =\



<f>(it, it, it) =||| p(Y, Y, Y) exp [itY + itY + itY] dY dY dY

jkm J' '

Here n ]km is the joint moment

fijkm = Y j Y k Y m =fff YJY k Y m p(Y,Y,Y) dY dY dY (12.151)

K(it, it, it) = log <f>(it, it, it) = X -TTT7 <& & )* #' ) m - (12-152)

Therefore

<p(it,ii,H') = exp pT(ir,if,ir)] (12.153)

where y, fc, and m are positive integers whose sum is greater than zero. Therefore, the
inverse transform of Eq. 12.150, from Eqs. 12.150 and 12.152, is



370



p(Y, Y,Y) = — i-y exp f- [iYt + iYt - iYt}) <p(it, ii, it ) dt, di, d'i

= —!-j exp [- [iYt + &i - zYr'jl exp [K{it, it, it)] dt, di, d'i



= (2*) 3 JJ J



exp \-iYt-iYi-iYi+K(it,ii,ii)\ dt,dt,dt



^■| i ! | exp \(k m -Y) it+(koio-Y) it + (kooi-Y)it



(2jt)



-t-{*200? 2 + ko 2 tf 2 + W^* 2k m ti+ 2k m t'i + 2ko U ii'



(12.155)



+ Y -^- (ity' (iiftfr] dt,di,dt. (12.154)

^- ilklml J

jkm J

Here j, k, and m are now positive integers whose sum is greater than 2.
The relations between joint moment and joint cumulants are

£100=^100

*oio = ,"010

&001 =/*ooi

^200 = variance of Y

&020 = variance of Y

kooz - variance of Y.

Nondimensionalizing the variables as follows

z = (Y-k l0 o)klg>

z = &-hw)lt& (12.156)

2 = (r-%)i)^5

and



371



5 — I A, 2 Q0



= t k l {l (12.157)



and



S = t *ot>2



jkm = WmHk j k k k™ ) ' (12 - 158)

J.K.m.{K 20Q K Q2Q KQQ2)



p(z,z,z) = ——j exp J- izs - z'z i- zz i j

-—\s z + s 2 +s 2 + 2Xuoss'+ 2A m ss + 2Aons's\



+ X -W^y (")* (") m ^ ds ds - (12.159)

y/fcm

Expanding the characteristics of the moments as shown in Eq. 12.159 into joint
moments and modifying them into joint cumulants, Dalzell derived the expression for
joint cumulants until the fourth degree from the modified frequency response character-
istics H\(w), #2(0 1,02), S3(o>\Q)2fi>3), and functions of w by the same type of style as
Eq. 12.119 for the case of p{z, z). Since these manipulations require many transactions
involving the higher order terms in the expansions, he checked the order of magnitude of
the functions, suggested the order to truncate the approximations and, by laborious
manipulation utilizing Hermite polynomial, he obtained his approximation.

From these expressions, arbitrarily denoting the standardized maxima or minima of
response by v he finally derived the expression for p(v, 0,'i) using the cumulants up to
the fourth order and then the probability distribution function for maxima and minima by


p + (v) = -i- f Bp(y,0,z)dz

Ar 00 J

— 00



= - Z ^r [ zp(v,0,z)dz' (12.160)

iVoo J



372



p-(






\z\ p(v,0,z)d'z



= -U| zp(y,0,z)dz.

Woo J





(12.161)



Finally he derived the expressions for his first and second approximations p|(v),
pl(v) and P2(v), piiv) as follows. First, by neglecting the term higher (than 3) order
joint cumulants, the first approximation is,



pftv) =



(2.T) 1 / 2



exp



V
2?



+ (l_ e 2)l/2 v exp



vd-e 2 ) 1 / 2



(12.162)



Pl(v) =



Qxi 11



exp



2e 2



-(l-e 2 ) 1/2 v exp



2



vd-e 2 ) 1 / 2



(12.163)



where 0(a) is the Gaussian cumulative distribution function and e is the spectrum band
width parameter.

Thee was introduced in Eq. 12.162 and Eq. 12.163 from the relations as follows that
comes from the characters of joint cumulants,



-•^•101 =



&020



(*2<x>*002) 1/2



= (l-e 2 ) 1 / 2


1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 17 19 20

Online LibraryYasufumi YamanouchiA review of statistical studies of seakeeping qualities → online text (page 17 of 20)