Вы находитесь на странице: 1из 52

A methodology based on Galerkin method for solving the stochastic bending problem of a Euler Bernoulli.....

A methodology based on Galerkin method for solving the

stochastic bending problem of a Euler Bernoulli

beam supported by a foundation of Pasternak

4
5
6
7
8

Cludio Roberto vila da Silva Jnior


1

NuMAT/PPGEM, Federal University of Technology of Parana


Av. Seven of September, 3165, Curitiba, PR, Brazil.
2

PPGMNE/CESEC, Federal University of Parana

Polytechnic Center, Jardim das Americas

10

81531-980, Curitiba, PR, Brazil.

11

avila@utfpr.edu.br

12
13

Marcelo Borges dos Santos

14

NuMAT/PPGEM, Federal University of Technology of Parana

15

Av. Seven of September, 3165, Curitiba, PR, Brazil.

16

engmecanica.marcelo@hotmail.com

17
18

Andr Tefilo Beck

19

Department of Structural Engineering, EESC, USP

20

Av. Sancarlense, 400, So Carlos, SP

21

atbeck@sc.usp.br

22

Page 1 of 52

A methodology based on Galerkin method for solving the stochastic bending problem of a Euler Bernoulli.....

23

Ivan Moura Belo

24

NuMAT/PPGEM, Federal University of Technology of Parana

25

Av. Seven of September, 3165, Curitiba, PR, Brazil.

26

ivanbelo@utfpr.edu.br

Page 2 of 52

27
28
29
30
31

Abstract: In this paper is presented a methodology based on the Galerkin method, which is used to

32

determine the approximate solution of stochastic bending of a Euler-Bernoulli beam, resting on a

33

Pasternak foundation type. The uncertainty in beam and foundation stiffness coefficients is represented

34

by means of parameterized stochastic processes. In the first step of the proposed methodology is

35

realized a theoretical study on existence and uniqueness of solutions. In this step, it is emphasized by

36

hypothesis, the limitations on the parameters of uncertainty, and the choice of approximate solution

37

space. In the second step, the Galerkin method is used to obtain approximate numerical solutions. The

38

space of approximate solutions is built by tensor product in a finite-dimensional space, dense in the

39

space of theoretical solutions given by the theorem on existence and uniqueness. The Wiener-Askey

40

scheme of generalized chaos polynomials (gPC) is used to model the stochastic behavior of the

41

displacement responses. The performance of the approximate Galerkin solution scheme developed

42

herein is evaluated by comparing first and second order moments with the same moments evaluated

43

from Monte Carlo simulation.

44
45

Keywords: Lax-Milgram lemma, generalized chaos polynomials (gPC), Euler-Bernoulli beam,

46

Pasternak foundation, Galerkin method, Askey-Wiener scheme.

47

A methodology based on Galerkin method for solving the stochastic bending problem of a Euler Bernoulli.....

48

1.

Page 3 of 52

INTRODUCTION

49
50

The field of stochastic mechanics has been subject of extensive research and significant

51

developments in recent years. Stochastic mechanics incorporates the modeling of randomness or

52

uncertainty in the mathematical formulation of mechanics problems. This is in contrast to the more

53

established field of structural reliability, where uncertainty and randomness are also addressed, but

54

where problem solutions are obtained mainly based on deterministic mechanics models.

55

The analysis of stochastic engineering systems has received new impulse with use of finite element

56

methods to obtain response statistics. Initially, finite element solutions where combined with the Monte

57

Carlo method, and samples of random system response where obtained. Perturbation and Galerkin

58

methods where used in this context [1]. Such methods allowed representation of uncertainty in system

59

parameters or in loads by means of stochastic processes. At the end of the 80s, Spanos and Ghanem

60

[2] used the Galerkin finite element method to solve a stochastic beam bending problem, where

61

Youngs modulus was modeled as a Gaussian stochastic process. The space of approximate solutions

62

was built using the finite element method and chaos polynomials. These polynomials form a complete

63

system in L2 ( , , P ) =

64

polynomials and ( , , P ) is a probability space. The ideas presented in this study where innovative

65

and represented a new method to solve stochastic problems.

L2 ( , , P )

, where = span { i }i =0

is the space generated by the chaos

66

Babuska et al. [3] presented a stochastic version of the Lax-Milgram lemma. The authors showed

67

that, for certain problems of mechanics, use of Gaussian processes can lead to loss of coercivity of the

68

bi-linear form associated to the stochastic problem. This difficulty was indeed encountered in the study

69

of Silva Jr. [4], and resulted in non-convergence of the solution for the bending of plates with random

70

parameters. This non-convergence was due to the choice of a Gaussian process to represent the

71

uncertainty in some parameters of the system. This failure to converge also affects solutions based on

A methodology based on Galerkin method for solving the stochastic bending problem of a Euler Bernoulli.....

Page 4 of 52

72

perturbation or simulation methods. Importantly, this failure makes the theoretical problem bad post,

73

[44]. In this way the numerical solutions obtained with any numerical method, did not have any sense,

74

and are inconsistent with the fundamental studies of existence and uniqueness of solutions. The

75

methodology proposed in this paper, special attention is given to this question, section 2.1. Although

76

the message of Babuska et al. was assimilated by some researchers [24, 27-29], one finds very recent

77

papers where Gaussian processes are still used to represent the uncertainty in strictly positive

78

mechanical properties [31, 33, 34].

79

In recent years, much effort is being addressed at representing uncertainty in stochastic engineering

80

systems via non-Gaussian processes. In the paper by Xiu and Karniadakis [5] the Askey-Wiener

81

scheme was presented. This scheme represents a family of polynomials which generate dense

82

probability spaces with limited and unlimited support. This increases the possibilities for uncertain

83

system parameter modeling. The stochastic beam bending problem has been studied by several authors.

84

Vanmarcke and Grigoriu [6] studied the bending of Timoshenko beams with random shear modulus.

85

Elishakoff et al. [7] employed the theory of mean square calculus to construct a solution to the

86

boundary value problem of beam bending with stochastic bending modulus. Ghanem and Spanos [8]

87

used the Galerkin method and the Karhunem-Loeve series to represent uncertainty in the bending

88

modulus by means of a Gaussian process. Chakraborty and Sarkar [9] used the Neumann series and

89

Monte Carlo simulation to obtain statistical moments of the displacements of curved beams, with

90

uncertainty in the elasticity modulus of the foundation. Although they present numerical solutions for

91

stochastic beam problems, none of the papers referenced above address the matter of existence and

92

uniqueness of the solutions.

93

In the present paper, is shown a methodology, based on the Galerkin method to be applied to the

94

problem of bending of a stochastic Euler-Bernoulli beam resting on a foundation of Pasternak, in order

95

to obtain approximate numerical solutions for the transverse displacement of stochastic process.

96

Uncertainty is present in the stiffness coefficients of the beam and the foundation. The methodology

A methodology based on Galerkin method for solving the stochastic bending problem of a Euler Bernoulli.....

Page 5 of 52

97

developed in two steps: Existence and Uniqueness, and Galerkin method. The first step is performed a

98

study of existence and uniqueness of solutions of the theoretical problem at hand. Emphasis is placed

99

on limiting probability on the parameters that present uncertainty. Such assumptions are necessary for

100

implementing a stochastic version of the Lax-Milgram lemma [3]. As will be presented in sections 2.1

101

and 3 the approximation space is generated by a finite set of functions, which are elements of a

102

complete set in the space of theoretical solutions, defined in the first step. This study subsidizes and

103

gives consistency to the Galerkin solution scheme developed herein, avoiding the flaws of some

104

formulations mentioned earlier [31, 33, 34]. The uncertainties in Youngs modulus and in foundation

105

stiffness are represented by parameterized random processes [10]. In the second stage of labor, it is

106

used Galerkin's method to obtain numerical solutions for the approximate transverse displacement of

107

stochastic. The approximated solution space is constructed using isomorphism properties between

108

Sobolev and product spaces, using density between continuous functions and Sobolev spaces and using

109

spaces generated by L2 polynomials of the Askey-Wiener scheme [5, 11]. The Askey-Wiener scheme

110

is used to construct the approximate solution space. In section (6) the performance of the numerical

111

solutions obtained by the Galerkin method is evaluated by comparing the statistician moments of first

112

and second order stochastic process shift Performance of the Galerkin solution developed herein is

113

evaluated by comparing first and second order moments of random beam displacement responses with

114

the same statistics evaluated via Monte Carlo simulation. Feature histograms are still random variables

115

generated from the stochastic numerical approximation of displacement. These histograms are

116

compared with histograms obtained by Monte Carlo simulation and observed a good agreement. The

117

author has used successfully the methodology proposed in this work, and other engineering problems,

118

which had uncertainties in its parameters, such as, bending the Euler-Bernoulli and Timoshenko beams

119

[22],[24], Kirchhoff's plate [25], and heat conduction problems [45].

120

A methodology based on Galerkin method for solving the stochastic bending problem of a Euler Bernoulli.....

121

Page 6 of 52

2. BENDING OF EULER-BERNOULLI BEAMS ON PASTERNAK FOUNDATION

122
123

In this section, the strong and weak formulations of the problem of stochastic bending of Euler-

124

Bernoulli beams on Pasternak foundation are presented. At the end of this section, the Lax-Milgram

125

lemma is used to present a proof of existence and uniqueness of the solution. Let ( , , P ) be a

126

probability space, where is a sample space, is an -algebra and P is a probability measure. The

127

strong formulation of the stochastic problem of bending, of a Euler-Bernoulli beam, resting on a

128

foundation of Pasternak is defined as:

129

130

Find u L2 ( ( , , P ) ; C 4 ( 0, l ) ) , such as,

d 2 EI d 2u d du + u = q, x, 0, l ;
( ) ( )
W
dx2
dx P dx
dx2

u ( 0, ) = u ( l , ) = 0;
2
2
;
= d u2
= 0,
d u2
dx ( 0 ,) dx (l ,)

) (

(1)

131
132

where EI is the bending stiffness, P and W are foundation stiffness coefficients, u is the transversal

133

beam displacement and q is a load term. Making P 0 in Eq. (1), one obtains the problem of beam

134

bending on a Winkler-type foundation [23].

135
136

For the qualitative analysis regarding existence and uniqueness of the response, the following
hypotheses are considered:

137

138

(
(
(

)
)
)

a, a + : P : EI x, a,a , x 0, l = 1;
( )

H1: b, b + : P : P ( x, ) b, b , x 0, l = 1;

+

c, c : P : W ( x, ) c, c , x 0, l = 1.

(2)

A methodology based on Galerkin method for solving the stochastic bending problem of a Euler Bernoulli.....

139

Page 7 of 52

H2: q L2 , , P; L2 ( 0, l ) .

140
141

Hypothesis H1 ensures that the beam and foundation stiffness coefficients are positive-defined and

142

uniformly limited in probability [3]. Hypothesis H2 ensures that the stochastic load process has finite

143

variance. These hypotheses are necessary for the application of the Lax-Milgram Lemma, which is

144

used in the sequence to demonstrate the existence and uniqueness of the solution.

145
146

2.1. Existence and uniqueness of the solution

147
148

In this section, a brief theoretical study of existence and uniqueness of the solution of the

149

stochastic Euler-Bernoulli beam bending problem on Pasternack foundation, with uncertainty in beam

150

and foundation stiffness, is presented. Results presented in this section are based on classical results

151

from functional analysis and from theory of Sobolev spaces [3, 12, 13]. The study requires definition of

152

stochastic Sobolev spaces, tensor product and density between distribution spaces and Lp spaces. In

153

order to study existence and uniqueness, the abstract variational problem associated to the strong form

154

(Eq. 1) needs to be defined. The stochastic Sobolev space where the solution to the stochastic beam

155

bending problem is constructed is V = L2 ( , , P; Q ) . For fixed , one has,

156
157

d 2u
Q = u ( , ) H 2 ( 0, l ) u ( 0, ) = u ( l , ) = 0 2
dx

=
( 0,)

d 2u
dx 2

( l ,)

= 0 ,

(3)

158
159

160

hence V = u : ( 0, l ) u is mensurable and

u ( )

2
H 2 ( 0,l )

dP ( ) < + , where

H 2 ( 0,l )

is

the H 2 ( 0, l ) norm. For fixed , u ( , ) Q . On the other hand, for fixed x ( 0,l ) ,

A methodology based on Galerkin method for solving the stochastic bending problem of a Euler Bernoulli.....

Page 8 of 52

161

u ( x , ) L 2 ( , , P ) is a random variable. Defining the tensor product between g L 2 ( , , P ) and

162

w Q as u := w.g [26, 38], one notes that, for fixed ,

163
164

u ( , ) = w ( ) .g ( ) Q ,

165

whereas for a fixed x ( 0, l ) ,

(4)

166
167

u ( x, ) = w ( x ) .g () L2 ( , , P ) .

(5)

168
169

In order to obtain the isomorphism between V and L2 ( , , P ) Q , it is necessary to redefine the

170

differential operator D : L2 ( , , P ) Q L2 ( , , P ) L2 ( 0, l ) . This operator acts over an

171

element u V in the following way [14],

172

173

( )

D u : d w
x .g ,
dx ( ) ( )

(6)

174
175

where and 2 . From the definitions presented in Eq. (4) to Eq. (6), the isomorphism between

176

V and L2 ( , , P ) Q is obtained:

177

V = L ( , , P; Q )

2
V L ( , , P ) Q .
2
2
L ( , , P; Q ) L ( , , P ) Q
2

178

179
180
181

V is a Hilbert space, with inner product, (, )V : V V , defined as,

(7)

A methodology based on Galerkin method for solving the stochastic bending problem of a Euler Bernoulli.....

(u, v )V = ( u.v + Du.Dv + D2u.D2v ) ( x,)dxdP ( ) .

Page 9 of 52

182

(8)

183
184

The inner product defined in Eq. (8) induces the V norm u V = ( u,u )V2 , following Kinderlehrer and

185

Stampacchia [15]. The bilinear form a :V V is defined as:

186
l

187

a ( u, v ) = W .u.v + P .Du.Dv + EI .D2u.D2 v ( x,) dxdP ( ) .


0

(9)

188
189

The bilinear form a (, ) is obtained by multiplying Eq. (1) by v V and integrating in ( 0, l ) , using

190

the product rule for differentiation and the definition of space V given in Eq. (3b). In order to preserve

191

the equality, the right-hand side of Eq. (1) becomes,

192
l

193

f ( v ) = ( q.v )( x,) dxdP ( ) .

(10)

194
195

It is easy to see that Eq. (10) defines a linear functional f : V . Using the Cauchy-Schwartz

196

inequality, it can be shown that this functional is well-defined, hence f V' , where V' is the dual of V.

197

Hence, the abstract variational problem associated to Eq. (1) can be written as:

198
199

Find u V such that,

a ( u,v ) = f ( v ) , v V .

(11)

200

Theorem (Existence and Uniqueness): Let EI , P , W and q, be such that H1 and H2 (Eq. 2) are

201

satisfied. Then, a solution to the problem defined in Eq. (11) exists, and it is unique in V.

202

A methodology based on Galerkin method for solving the stochastic bending problem of a Euler Bernoulli.....

Page 10 of 52

203

Proof: The proof of existence and uniqueness uses the Lax-Milgram lemma [3, 16]. It is necessary to

204

show that the bi-linear form, Eq. (9), of the problem defined in Eq. (11), is continuous and coercive. In

205

order to do that, one makes use of the hypotheses of limited probability (H1), of the Cauchy-Schwartz

206

inequality [12] and of simple size comparisons:

207
208

a.

continuity

209

l 2

a ( u,v ) c u ( x,) dxdP ( )


0

+b

l 2

v ( x,) dxdP ( )
0

( Du ) ( x,) dxdP ( )

0
l

210

+a

( Dv ) ( x,) dxdP ( )

0
l

( D u ) ( x,) dxdP ( )

0
l

2
x
,
dxdP

D
v
( ) C u
( )( )

0
l

vV,

211
212

where C = max {a ,b,c} ;

213

b. coercivity

214
215

a ( u,u ) c u 2 ( x,) dxdP ( ) + b ( Du ) ( x,) dxdP ( ) + a ( D2u ) ( x,) dxdP ( ) c u V2 ,


2

216
217

where c = min {a,b,c} .

218
219

Following Eq. (10), the functional f : V , is linear and well-defined, hence in view of the Lax-

220

Milgram lemma, it is guaranteed that the problem defined in Eq. (11) has unique solution, and

221

continuous dependency on the data [3, 16]. Theoretical solutions to the abstract variational problem in

A methodology based on Galerkin method for solving the stochastic bending problem of a Euler Bernoulli.....

Page 11 of 52

222

Eq. (1), associated to the problem in Eq. (11), are found in V L2 ( , , P ) Q . Numerical solutions

223

in less abstract spaces are obtained herein: continuous functions of class C 2 , sequentially dense in Q ,

224

and a family of generalized chaos polynomials, belonging to the Askey-Wiener scheme, are used.

225
226

3.

UNCERTAINTY REPRESENTATION

227
228

In most engineering problems, complete statistical information about uncertainties is not available.

229

Sometimes, the first and second order moments are the only information available. The probability

230

distribution function is defined based on experience or heuristically. In order to apply Galerkins

231

method, an explicit representation of the uncertainty is necessary. Given the incomplete information

232

about the probability distribution of a given parameter, an hypothesis of finite dimensional noise is

233

assumed, following, for example, [41 - 43]. This implies that the uncertainty over a given input

234

parameter : ( 0,l ) will be represented in terms of a finite set of random variables,

235

( x,) = ( x, ( ) ) = ( x,1 ( ) ,, N ( ) ) .

(12)

236
237

From this hypothesis, the uncertainty in beam and foundation stiffness coefficients are modeled via

238

parameterized stochastic processes. These are defined from a linear combination of deterministic

239

functions and random variables [10],

240

241

( x, ) = ( x ) + i ( x ) i ( ) = ( x ) + t ( x ) . ( ) ,
i =1

242

(13)

A methodology based on Galerkin method for solving the stochastic bending problem of a Euler Bernoulli.....

Page 12 of 52

243

where ( ) is the expected value of random process ( , ) , : ( 0,l ) N is a vector-valued

244

function with terms i C0 ( 0, l ) C

245

independent random variables, such that:

( 0, l ) ,

i {1, , N} . ( ) = {i ( )}i =1 is a vector of


N

246
247

[ i ] = 0, i {1, , N} ,
.

P ( : i ( ) i ) = 1, i {1, , N} ,

(14)

248
249

where [] is the expected value operator. In Eq. (14), i is the image of random variable i , that is,

250

i = i ( ) , with i = [ ai , bi ] , i = bi ai < , i {1, , N} , limited. In this form, the image of

251

random vector : , with N , and in terms of {i }i=1 , is given by = i . Since the


N

i=1

252

random variables are independent, the joint probability density is given by,

253
N

254

( ( ) ) = i ( i ) ,

(15)

i =1

255

where i ( ) is the marginal probability density function of random variable i . Hence, the probability

256

measure dP ( ) is defined as,

257
N

258

dP ( ( ) ) = i ( i )d i .
i =1

259

(16)

A methodology based on Galerkin method for solving the stochastic bending problem of a Euler Bernoulli.....

Page 13 of 52

260

From the measure and integration theory [21], one knows that the probability measure defined in Eq.

261

(16) is the measure obtained from the product between probability measure spaces associated to the

262

random variables ( ) = {i ( )}i =1 , with i : i .

263
264

From the Doob-Dynkin lemma [40], the transversal displacement random process will be a function
of random variables ( ) = {i ( )}i =1 , hence,
N

265
266

u ( x, ) = u ( x, ( ) ) = u ( x, 1 ( ) , , N ( ) ) .

(17)

267
268

In this paper, polynomials of the Askey-Wiener scheme are used construct the problems solution space.

269
270

3.1. The Askey-Wiener scheme

271
272

The Askey-Wiener scheme is a generalization of chaos polynomials, also known as Wiener-chaos.

273

Chaos polynomials were proposed by Wiener [17] to study statistical mechanics of gases. Xiu and

274

Karniadakis [5] have shown the close relationship between results presented by Wiener [17] and Askey

275

and Wilson [18] for the representation of stochastic processes by orthogonal polynomials. Xiu and

276

Karniadakis [5] extended the studies of Ghanem and Spanos [8] and Ogura [19] for polynomials

277

belonging to the Askey-Wiener scheme.

278

The Cameron-Martin theorem [11] shows that Wiener-Askey polynomials form a base for a dense

279

subspace of second order random variables L2 ( , , P ) . Let L2 ( , , P ) be a separable Gaussian

280

Hilbert space and


= span {i } i =1

be an orthonormal base of Gaussian random variables. The

281

Wiener-Askey

representation

scheme

allows

of

any

second

order

random

variable

A methodology based on Galerkin method for solving the stochastic bending problem of a Euler Bernoulli.....

Page 14 of 52

282

, [35]. Let n ( ) be the vector


u L2 ( , ( ) , P ) , where ( ) is the -algebra generated by

283

space spanned by all polynomials of order less than "n",

284

{(

285

n ( ) = {i } i =1 : is the polynomial of degree n; i , i = 1,...,N;N < ,

(18)

:0: = 0 ( ) ,

(19)

286
287

with

288
289

:n: =
n ( )
n-1 ( ) ,

290
291

2
2
where
n is the closure of
n in L ( , , P ) . As shown by Jason [20], for L ( , ( ) , P ) , the

292

following orthogonal decomposition is admitted,

293
294

L ( , ( ) , P ) =
2

:n:

(20)

n=0

295
296

Equation (20) is an orthogonal decomposition of L2 ( , , P ) , known as Wiener-chaos decomposition

297

or simply chaos decomposition. One application of this decomposition is the representation of an

298

element X L2 ( , , P ) in terms of elements X n H :n: ,

299
300

X = Xn .
n =0

301

(21)

A methodology based on Galerkin method for solving the stochastic bending problem of a Euler Bernoulli.....

Page 15 of 52

302

Equation (21) represents an important result for the approximation theory applied to stochastic systems.

303

Solution of a stochastic system is expressed as a non-linear function in terms Gaussian random

304

variables. This function is expanded in terms of chaos polynomials as,

305
306

u i ( ) =

p 0 n1 ++ nr = p i1 ,,ir

j1 jr
i1ir

p i1 ( ) , , ir ( ) ,

(22)

307
308

1 jr
where p is the chaos polynomial of order p and ui1j
are the polynomial coefficients. The superir

309

index refers to the number of occurrences of ik ( ) . Chaos polynomials of order p are formed by an

310

Hermite polynomial, in standard Gaussian variables ( ) = ik ( )

311

Introducing a mapping in the sets of indexes {ik }k =1 and { jk }k =1 , Eq. (14) can be rewritten as,

r
k =1

of order less than p.

312
313

u ( ) = u ( ( ) ) ,

(23)

314
315

where is a multi-index, is a set of natural numbers with compact support, { }are chaos

316

polynomials, {u }are coefficients of the linear combination and : N is a vector of random

317

variables. In this paper, chaos polynomials are multi-dimensional Hermite polynomials,

318
319

( ( ) ) = hm ( m ( ) ) ,

(24)

m =1

320
321

where hm ( ) is a Hermite polynomial defined in random variable m . The inner product between

322

polynomials i and j in L2 ( , , P ) is defined as,

A methodology based on Galerkin method for solving the stochastic bending problem of a Euler Bernoulli.....

323

( , )
i

j L2 , , P
(
)

Page 16 of 52

= ( i . j ) ( ( ) )dP ( ) ,

(25)

324
325

where dP is a probability measure. These polynomials form a complete ortho-normal system with

326

respect to the probability measure, with the following properties:

327
328

0 = 1,

( , )
i

j L2 ( , , P )

= ij , i, j .

(26)

329
330

It is important to observe that in Eq. (21) the polynomials are orthogonal with respect to the standard

331

Gaussian density function of vector . The convergence rate is exponential for the case where the

332

random variable is Gaussian. For other random variables the convergence rate is smaller.

333

The proposal of the Wiener-Askey scheme is to extend the result presented in Eq. (23) to other

334

types of polynomials. In analogy to Eq. (18), taking


n ( H ) = span { i }i =1 , with H a separable

335

Hilbert space of finite variance random variables, one has that =

(H )

is a family of

336

polynomials of the Wiener-Askey scheme, generating a complete orthogonal system in L2 ( , , P ) .

337
338

Table 1: Correspondence between some random variables

339

and polynomials of the Askey-Wiener scheme.

340
341

The Askey-Wiener scheme represents a family of sub-spaces generated by orthogonal polynomials

342

obtained from ordinary differential equations [5]. Among them, the Hermite, Laguerre, Jacobi and

343

Legendre polynomials can be cited. Every sub-space generated by these polynomials is a complete

344

system in L2 ( , , P ) . The orthogonality between the polynomials is defined with respect to a weight

A methodology based on Galerkin method for solving the stochastic bending problem of a Euler Bernoulli.....

Page 17 of 52

345

function, which is identical to the probability density function of a certain random variable. For

346

example, the Gaussian density function is used as weight function to obtain the orthogonality between

347

Hermite polynomials. Table (1) shows the correspondence between subsets of polynomials of the

348

Askey-Wiener scheme and the corresponding probability density functions.

349
350

4.

GALERKIN METHOD

351
352

The Galerkin method is used in this paper to solve the stochastic beam bending problem with

353

uncertainty in the beam and foundation stiffness coefficients. In order to develop numerical solutions

354

which are compatible with the conditions for existence and uniqueness of the theoretical solution,

355

results presented by Besold [26] and Ryan [39] are used. An element from a space isomorph to the

356

space obtained via tensor product, V L2 ( , , P ) Q , can be represented from elements of

357

separable spaces, dense in spaces L2 ( , , P ) and Q . Hence, the strategy to construct approximated,

358

numerical solutions and to use Galerkins method is to use bases of finite dimensions, but dense in

359

L2 ( , , P ) and Q . Due to the simplicity of the spatial domain for the present problem, no spatial

360

discretizations are employed. Hence, functions employed in construction of the responses are defined

361

in the whole problem domain. For more complex spatial domains, special techniques like finite

362

element, boundary element or finite differences would have to be employed.

363
364

It is proposed that approximated solutions to the stochastic displacement response of the beam
have the following form,

365
366

u ( x, ( ) ) = ui i ( x, ( ) ) ,
i =1

367

(27)

A methodology based on Galerkin method for solving the stochastic bending problem of a Euler Bernoulli.....

Page 18 of 52

368

where ui , i are coefficients to be determined and i V are the test functions. Numerical

369

solutions to the variational problem defined in Eq. (11) will be obtained. Hence, it becomes necessary

370

to define spaces less abstract than those defined earlier, but without compromising the existence and

371

uniqueness

of

372

L2 ( , ,P )

373

2
d 2
d 2
= span i C ( 0, l , ) i ( 0 ) = i ( l ) = 0 2i ( 0 ) = 2i ( l ) = 0 , i

374

such that = Q ,

the

solution.

= L2 , ,P .

From

the

theorem

Consider

two

dx

L2 ( , , P )

of

Cameron-Martin

complete

dx

[11]

we

orthogonal

and

have

systems

= span {i }i =1 ,

= L2 ( , , P ) and define the tensor product between and as,

375

376

( )i ( x, ( )) = j ( x ) .k ( ( )) , with

j, k .

(28)

377
378

To simplify the notation, we will use i = ( ) . Since approximated numerical solutions are

379

derived in this paper, the solution space has finite dimensions. This implies truncation of the complete

380

orthogonal systems and . Hence one has,

381
382

2
d 2
d 2
m = span i C ( 0, l , ) i ( 0 ) = i ( l ) = 0 2i ( 0 ) = 2i ( l ) = 0 , i {1, ..., m} ,

dx

dx

(29)

383
384

and n
= span {i }i =1 , which results in VM = m n . The dimension of n , ( n = dim n

385

depends on the dimension of the random variable vector ( ) and on the order of chaos polynomials.

386

Let "s" be the dimension of random vector ( ) and p the order of chaos polynomials , then the

387

dimension of n is given by,

),

A methodology based on Galerkin method for solving the stochastic bending problem of a Euler Bernoulli.....

388

n=

( s + p )!
s!.p!

Page 19 of 52

(30)

389
390

Since dim n
< and dim n
< (following [39]), one has that the dimension of the

391

approximation space, VM , is given by,

392
393

M = dim (VM ) = dim ( m n ) = dim ( m ) .dim ( n ) = m.n

(31)

394
395
396

With the above definitions and results, it is proposed that numerical solutions are obtained from
truncation of the series expressed in Eq. (27) at the M th term,

397
M

398

uM ( x, ( ) ) = ui i ( x, ( ) ) .

(32)

i =1

399
400

Substituting Eq. (32) in Eq. (11), one arrives at the approximated variational problem:

401

402

403

Find {u }M M such that,


i i =1

M
a i , j ui = f ( j ) , j VM ;
i =1

(33)

where,

404
l

2
2
a i , j = W .i . j + P .Di .D j + EI .Di .D j

l
f =
q. x, )dxdP ( ) .
( j ) ( j )(
0

405

) ( x, ) dxdP ( ) ,
(34)

A methodology based on Galerkin method for solving the stochastic bending problem of a Euler Bernoulli.....

Page 20 of 52

406

The approximated variational problem (Eq. 33) consists in finding the coefficients of the linear

407

combination expressed in Eq. (32). Using a vector-matrix representation, the system of linear algebraic

408

equations defined in Eq. (33) can be written as,

409
KU = F ,

410

(35)

411
412

where K M ( ) is the stiffness matrix, U = {ui }i =1 is the displacement vector and F = { f i }i =1 is the

413

loading vector. Elements of the stiffness matrix are defined as,

414
K = kij

M M

, kij = ( W .i . j ) ( x, ) dxdP ( )
0

+ ( P .Di .D j ) ( x, ) dxdP ( )

415

(36)

+ EI .D2i .D2 j
0

) ( x, )dxdP ( ).

416
417

The load vector is given by

418

F = { f j}

419

M
j =1

f j = ( q. j ) ( x, )dxdP ( ) .

(37)

420
421

The sparseness of the stiffness matrixes for example 1, case (a) (to be presented), is shown in Figure

422

(1).

423
424
425

A methodology based on Galerkin method for solving the stochastic bending problem of a Euler Bernoulli.....

Page 21 of 52

426
427
428
429

0
1

10

2
3

20

430
431

30

5
6

40

432
433
434

50

8
9

60

10
11
0

6
nz = 42

10

70
0

20

40

60

nz = 1344

435

Figure 1: Sparcity pattern of stiffness matrix of example 1, case (a).

436

a) for m = 2, n = 5, p = 1; b) for m = 2, n = 35, p = 3.

437
438

Remember that p is the degree of the chaos polynomial. The stiffness matrix corresponding to

439

Figure (1a) has dimension 10 and nz = 42 (number of nonzero elements), whereas the matrix

440

corresponding to Figure (1b) has dimensions 70 and nz = 1344. The conditioning number (nC)

441

corresponding to these two matrixes is nC = 19.70 and nC = 25.79, respectively. These numbers show

442

that the conditioning number increases with the dimension of the approximation space,

443

(VM = m n ) . This can lead to ill-conditioning of the stiffness matrix and hence to loss of accuracy

444

of the approximated solution. For example 2, cases (a) and (b) - to be presented - the sparcity pattern is

445

similar. However, for example 1, case b) presents a different pattern, with nz = 100 and nC = 32.82 (for

446

p=2) and nz = 4592 and nC = 71.74, for p=3. In this paper, uncertainties in beam and foundation

447

, defined in the
stiffness coefficients are modeled using Legendre polynomials to construct space n

448

variables ( ) = {1 ( ) ,2 ( ) ,3 ( ) ,4 ( )} 1 2 3 4 = 11
, .

449

A methodology based on Galerkin method for solving the stochastic bending problem of a Euler Bernoulli.....

450

5.

Page 22 of 52

STATISTICAL MOMENTS, HISTOGRAMS AND ESTIMATES

451
452

In this section, the evaluation of first and second order moments, from the approximated Galerkin

453

solution, is presented. In an evaluation of the performance of the developed Galerkin scheme, these

454

moments are compared with the same moments computed through Monte Carlo simulation. In order to

455

reduce spurious correlations between the samples, Latin Hypercube Sampling (LHS) is used in the

456

simulations [36].

457

The statistical moment of kth order of a random variable u ( x, ) L2 ( , , P ) is obtained, for a

458

fixed x [ 0 ,l ] , taking the kth power of this random variable and integrating with respect to its

459

probability measure:

460

uk M ( x ) = u Mk ( x , ( ) ) dP ( ( ) )

461

times
k

i
i
1

i1

u ik i 1 i k

) ( x ) (

i1

i k

) ( ( ) ) dP ( ( ) ),

(38)

462
463

The integration term dP () is a probability measure, defined in Eq. (16). From Eq. (38), uk M ( ) is

464

given by,

465
k times

466

467
468
469

with

( x ) = ( u
i 1, j 1

i k,j k

i1

i 1 uik ik

) ( x )

i 1 , , i k

(39)

A methodology based on Galerkin method for solving the stochastic bending problem of a Euler Bernoulli.....

470

b1

bN

a1

aN

i 1 ,, i k = j 1 j k

) ( ( ) ) ( )
1

Page 23 of 52

( ) d 1 ( ) d N ( ) .

(40)

471
472

The integrals in Eq. (40) are called iterated integrals. The first order statistical moment, or expected

473

value, of the stochastic displacement process evaluated at a point x 0, l is,

474
m

475

u( x ) = u(i 1).n+1i ( x ) .

(41)

i =1

476
477

The second-order statistical moment (the variance) of this displacement is,

478
479

u2 ( x ) =
M

u
i =1 j =1 k = 2

ik

u jk ( i . j ) ( x ) .

(42)

480
481

The Monte Carlo estimates for expected value and variance are,

482

483

u ( x ) =

2 ( x ) =
u

1
N

u ( x, ( ) ) ;

i =1
j

(43)

2
1 N

;
u
x
x

(
)
(
)

u
j

N 1 j =1

484

485

where u x, ( j ) is the jth realization of random variable u = u ( x, ( ) ) , obtained from the solution

486

of Eq. (35), whose coefficients EI = EI x, ( ) , P = P x, ( )

487

evaluated for the jth sample of random vector ( j ) = {i ( j )} . Each realization u x, ( j ) ,

488

corresponding to the jth sample of random vector : , is given by:

i =1

and W = W x, ( )

are

A methodology based on Galerkin method for solving the stochastic bending problem of a Euler Bernoulli.....

489

Page 24 of 52

u x, ( j ) = ukj ( j ) k ( x ) = U t ( j ) ( x ) ,
k =1

(44)

490
491

where : [ 0, l ]

492

m = span {i }i =1 . U ( j ) = ukj ( j )

493

Eq. (39), obtained as solutions to the following linear system:

is a vector-valued function whose components {k }k =1 are functions, such that


m

) { (

)}

is an m vector, whose entries are the coefficients of

k =1

494
495

( KU ) ( ( j ) ) = F ( ( j ) ) .

(45)

496
497

In order to obtain realizations of vector U : m , corresponding to the jth sample of vector

498

: , it is necessary to evaluate stiffness matrix K : m ( ) for the sampling functions

499

EI = EI x, j

( ( )) ,

( ( )) and

= P x, j

( ( )) , evaluated in ( ) = { ( )}

= W x, j

i=1

500

In the paper, qualitative evaluations of the numerical solutions are presented, by comparing

501

histograms of the displacement random processes obtained via approximated Galerkin solution and via

502

Monte Carlo simulation. Histograms for random variable u ( x* , ) are obtained by evaluating the jth

503

realization

504

( x*, ( )) = ( x*, ( ) , ( ) , ( ) , ( )) ( 0, l ) ,

of

the
1

displacement
2

random

process,

Eq.

(44),

in

505
506
507

uM x*, ( j ) = ui i x*, ( j ) = ui ( )i x*, ( j ) .


M

i =1

i =1

(46)

A methodology based on Galerkin method for solving the stochastic bending problem of a Euler Bernoulli.....

Page 25 of 52

508

Samples of random vector { ( j )} j =1 = {1 ( j ) , 2 ( j ) , 3 ( j ) , 4 ( j )}

509

Carlo simulation, using Latin Hypercube Sampling, [36]. Twenty thousand samples (N=20,000) are

510

used in the solution of each problem.

N
j =1

are obtained from direct Monte

511
512

6.

NUMERICAL EXAMPLES

513
514

In this section, two numerical examples are presented, for bending of random Euler-Bernoulli

515

beams supported on random Pasternak foundations. In the first example, uncertainty on the beam

516

stiffness is considered. In the second example, uncertainty is assumed on the foundation stiffness. For

517

both examples, uncertainties in stiffness are represented using parameterized, weakly stationary,

518

random processes. In both examples, the beam is simply supported at both ends, the span equals one

519

meter

520

distributed load of q ( x ) = 1 kPa.m, x [ 0, l ] , following Figure (2).

521
522
523

(l = 1 m) ,

the cross-section has width b =

1
100

m and the beam is subject to an uniform

q ( x ) = 1 kPa.m

524

525
526
527

528
529

Figure 2: a) Simply supported beam on Pasternack foundation;

530

b) beam cross-section.

531

A methodology based on Galerkin method for solving the stochastic bending problem of a Euler Bernoulli.....

Page 26 of 52

532

In the examples, the following mean-value functions are considered for Youngs modulus, beam height

533

and for foundation stiffness parameters:

534

535

E ( x ) = 210 GPa, x [ 0, l] ;

h ( x ) = 1 50 m, x [ 0, l] ;

P ( x ) = 1000 N, x [ 0, l] ;

W ( x ) = 1000 Pa, x [ 0, l] .

(47)

536
537

The first and second order statistical moments of the numerical solution obtained via Galerkin

538

method are compared with the same moments evaluated via Monte Carlo simulation. In order to

539

evaluate the accuracy of the Galerkin solutions, the functions relative error in mean, u : [ 0, l ] + , and

540

relative error in variance, u : [ 0, l ] + , are defined as,


2

541

542

u
100% 1 u ( x ) ,
u ( x ) =
0,

u2
100% 1 u2 ( x ) ,
2 ( x ) =
u
0,

x ( 0, l ) ;
x {0, l} ;
x ( 0, l ) ;

(48)

x {0, l} ;

543
544

where u and u2 , are the Galerkin-based expected value and variance, respectively, and u and u2 are

545

the Monte Carlo estimates of the same moments. The accuracy of developed Galerkin solutions, in

546

approximating the expected value and variance of the displacement response, is evaluated for

547

numerical solutions using chaos polynomials of order p {1,2,3} . Monte Carlo estimates of expected

548


value and variance, ( u , u2 ) , are evaluated from Eqs. (41) and (42), using 20,000 samples. In order to

549

show that not only first and second order moments are accurately approximated, the histograms and

A methodology based on Galerkin method for solving the stochastic bending problem of a Euler Bernoulli.....

Page 27 of 52

550

cumulative distribution functions are also compared. Numerical results presented in this paper were

551

obtained in a HP Pavilion personal computer, running a MATLAB computational code.

552
553

6.1.Example 1: random beam stiffness

554
555

In this example, propagation of the uncertainty through the beam response, for two cases of

556

random beam stiffness, is studied. In case (a), uncertainty is assumed on the Youngs modulus of beam

557

material, E : [ 0, l ] + , which is modeled as a parameterized random process,

558
559

E ( x, ( ) ) = E + 3 . E 2.n 1 ( ) cos

n =1

( n.lx ) +

2.n

( ) sin ( n.lx ) ,

(49)

560
561

where E is the expected value and E is the standard deviation of Youngs modulus. In case (b),

562

uncertainty is on the beams cross-section height, h, which propagates into the beams stiffness

563

through the moment of inertia of the beams cross-section. The random process, h : [ 0, l ] + , is

564

given by,

565
566

h ( x, ( ) ) = h + 3 .h 2.n 1 ( ) cos
n =1

567

( n.lx ) +

2.n

( ) sin ( n.lx ) ,

(50)

A methodology based on Galerkin method for solving the stochastic bending problem of a Euler Bernoulli.....

Page 28 of 52

568

where h is the expected value and h is the standard deviation of beam height. For both cases,

569

( ) = {n ( )}n=1 ,

570

N = 2 , ( 2.N = dim ( ) ) . Numerical solutions are obtained for E = ( 101 ) . E and h = ( 101 ) .h . The

571

covariance function for random process E : [ 0, l ] + , is presented in Figure 3.

2.N

is

vector

of

four

uniform,

independent

random

variables,

hence,

Figure 3: Covariance function of Youngs


modulus, Example 1, case (a).
572
1
.C

( x, y )

573

For all examples presented herein, the covariance function

for the random processes is

574

similar to that shown in Figure 3. These functions are obtained in exact form (Eq. 49) from the

575

independency property of random variables {n }n =1 .


2.N

576

Results obtained via Monte Carlo simulation, and used as reference, are shown first. Figure 4

577

shows all the twenty-thousand realizations (NS =20,000) of the random transversal displacement of the

578

beam, for cases (a) and (b). The dispersion in transversal displacements is larger for the case when the

579

uncertainty is on beam cross-section height h, in comparison to the case with random Youngs

580

modulus. This is typical behavior for Euler-Bernoulli beams. Interestingly, it was observed that for

581

beams modeled using the Timoshenko beam theory [24], the behavior is opposite: for the same

A methodology based on Galerkin method for solving the stochastic bending problem of a Euler Bernoulli.....

Page 29 of 52

582

coefficient of variation, propagation of the uncertainty is larger for the case of random Youngs

583

modulus, in comparison to random cross-section height [37].

584

Figures 4 and 5 show the evolution of the Monte Carlo estimates for expected value and variance,

585

( ( 12 ) , ( 12 ) ) ,

586

observed that the Monte Carlo solution is stationary, showing that the solution is independent of N, for

587

N>10,000.

2
u

of random variable u ( 12 , ) , as the number of samples is increased. It can be

588
3

8.4

x 10

589

x 10

8.2

8.5

590

8.4
8.6

8.6

8.8

591

(0,5) 8.7

592

8.8

u(0,5) 9

9.2
9.4
9.6

8.9

593

9.8

9
0

594

0.2

0.4

0.6

0.8

1
N

1.2

1.4

1.6

1.8

10
0

0.2

0.4

0.6

0.8

x 10

1
N

1.2

1.4

1.6

Figure 4: Convergence of Monte Carlo simulation results for the mean of mid-spam

596

displacement ( x = 12 m ) for cases (a) - left - and (b) - right.


6

1.5

x 10

x 10

1.8

598

1.6
1.4

599

1.2

2(0,5)

600

2
4

x 10

595

597

1.8

2(0,5) 1
u

0.8

0.5

0.6

601

0.4
0.2

602

0
0

0.2

0.4

0.6

0.8

1
N

1.2

1.4

1.6

1.8

2
4

x 10

0
0

0.2

0.4

0.6

0.8

1
N

1.2

1.4

1.6

1.8

603

Figure 5: Convergence of Monte Carlo simulation results for standard deviation of mid-spam

604

displacement ( x = 12 m ) for cases (a) - left - and (b) - right.

605

2
4

x 10

A methodology based on Galerkin method for solving the stochastic bending problem of a Euler Bernoulli.....

Page 30 of 52

606

Figure 6 presents results for the expected value of mid-spam displacements, obtained trough Monte

607

Carlo simulation, and by means of the Galerkin solutions of order p {1,2,3} . The difference between

608

the curves is imperceptible, showing that even a Galerkin solution of order p=1 is already accurate.

609

Similar results are obtained for case (b), not shown here.

610
3

x 10

1
2
3
4
(x)
u
5
6
Monte Carlo
p=1
p=2
p=3

7
8
9
0

0.1

0.2

0.3

0.4

611

0.5
x

0.6

0.7

612

Figure 6: Expected value of mid-spam

613

displacements, example 1, case (a).

0.8

0.9

614
615

Figures 7a and 7b show the relative error of the Galerkin solutions, for cases (a) and (b),

616

respectively, for the expected value and using polynomials of order p {1,2,3} . Here it can be observed

617

that, although the error is already small for p=1, the Galerkin solutions converge monotonically to the

618

exact solution with increasing polynomial orders.

619
620
621
622

A methodology based on Galerkin method for solving the stochastic bending problem of a Euler Bernoulli.....

Page 31 of 52

623
1.4

0.045

624

0.04
0.035

625
626

p=1
p=2
p=3

1.2

p=1
p=2
p=3

0.03
(x)

627

0.8

0.025

(x)

0.02

0.6

0.015
0.4

0.01

628

0.2

0.005
0
0

629
630

0.1

0.2

0.3

0.4

0.5
x

0.6

0.7

0.8

0.9

0
0

0.1

0.2

0.3

0.4

0.5
x

0.6

0.7

0.8

0.9

Figure 7: Relative errors in expected value for example 1, cases (a) - left - and (b) - right.

631
632

Figures 8a and 8b show the variances, for cases (a) and (b), respectively, obtained via Monte Carlo

633

simulation and via Galerkin method for p {1,2,3} . Other than what was observed for the expected

634

value, approximations for the variance are only accurate for p=2. It is also observed that the variance is

635

larger for case (b), in comparison to case (a), as already observed in Figs. 4a and 4b.

636

1.5

x 10

1.8
Monte Carlo
p=1
p=2
p=3

637

1.4

640

Monte Carlo
p=1
p=2
p=3

1.6

638
639

x 10

1.2

u(x)

2(x)

0.8

0.5

0.6

641

0.4
0.2

642
643
644

0
0

0.1

0.2

0.3

0.4

0.5
x

0.6

0.7

0.8

0.9

0
0

0.1

0.2

0.3

0.4

0.5
x

0.6

0.7

0.8

0.9

Figure 8: Variance of mid-spam displacements for example 1, case (a) - left - and case (b) - right.

A methodology based on Galerkin method for solving the stochastic bending problem of a Euler Bernoulli.....

Page 32 of 52

645

Figures 9a and 9b show the relative error in variance, for cases (a) and (b), respectively, for the

646

Galerkin solutions using p {1,2,3} . It can be observed that the convergence in terms of variance is also

647

monotonic for increasing order p, as expected.

648
8

649
650

35
p=1
p=2
p=3

7
6

30

p=1
p=2
p=3

25

5
20

651

2(x) 4

2(x)

15

652
653

655

1
0
0

654

10

0.1

0.2

0.3

0.4

0.5
x

0.6

0.7

0.8

0.9

0
0

0.1

0.2

0.3

0.4

0.5
x

0.6

0.7

0.8

0.9

Figure 9: Relative errors in variance for example 1, cases (a) - left - and (b) - right.

656
657

Figures 10a and 10b present the histograms, for random variable u ( 12 , ) , for cases (a) and (b).

658

These histograms were obtained from Monte Carlo simulation and from Galerkin solutions with gPC of

659

order p = 3. An excellent agreement between the histograms can be observed, showing that the

660

developed Galerkin solution is accurate not only for the first and second moments, but for the whole

661

probability distribution of the response. The agreement between the histograms is best for case (a), but

662

it is still very good for case (b) of Example 1.

663
664
665
666
667
668

A methodology based on Galerkin method for solving the stochastic bending problem of a Euler Bernoulli.....

Page 33 of 52

669
8000

6000

670

Monte Carlo
p=3

Monte Carlo
p=3

7000

5000

671
672

6000

4000

5000
nu(0,5)4000

3000
nu(0,5)

673

2000

674

1000

3000
2000

675

1000

0
0.015

0.014

0.013

0.012

0.011

0.01
u(0,5)

0.009

0.008

0.007

0.006

0
0.035

0.03

0.025

0.02

0.015

0.01

0.005

u(0,5)

676

Figure 10: Histograms of mid-spam displacement response

677

for example 1: case (a) left and case (b) right.

678
679

Figures 11a and 11b show the corresponding cumulative probability distributions, for random

680

variable u ( 12 , ) , obtained from Monte Carlo simulation and from the approximated Galerkin solution

681

with p=3. Again, an excellent agreement can be observed.

682
683

684

0.9

0.9

0.8

0.8

685

0.7

0.7

0.6

0.6

CDFu(0,5)
0.5

CDF 0.5

0.4

0.4

0.3

0.3

686
687

u(0,5)

0.2

688
689

0.2
Monte Carlo
p=3

0.1
0
0.015

0.014

0.013

0.012

0.011

0.01
u(0,5)

0.009

0.008

0.007

0.006

Monte Carlo
p=3

0.1
0
0.035

0.03

0.025

0.02

0.015

0.01

u(0,5)

690

Figure 11: Cumulative probability distributions of mid-spam displacement response

691

for example 1: case (a) left and case (b) right.

692

0.005

A methodology based on Galerkin method for solving the stochastic bending problem of a Euler Bernoulli.....

693

Page 34 of 52

Summary of results for example 1:

694
695

Table 2 summarizes results of expected value, variance and corresponding relative errors for the

696

random variable obtained by fixing x = 12 m in the random process displacement, for cases (a) and (b)

697

of example 1. Results are presented for approximated solutions with p {1,2,3} . Monte Carlo estimates

698

of expected value and variance:

699
700

1
=0.00880204234854855 m;
case (a): u2 ( 21)
u ( )=1.4117530417803410 6 m2 ;
2

1
=0.00957574170796273 m;
case (b): u2 ( 21)
u ( )=1.6216703821786310 5 m2 .
2

(51)

701
702

Comparing the expected value and variance obtained via Galerkin solutions with simulation results,

703

one notes that the approximated solutions are always smaller than the corresponding Monte Carlo

704

estimates.

705
706

Table 2: Summary of numerical results of example 1, cases (a) and (b):

707

expected value, variance, relative errors in expected value

708

and variance evaluated at x = 12 m

709
710

It is observed in Table 2 that the error in variance, for case (a), is non-monotonic with the order of

711

chaos polynomials: the error is larger for p=3 than it is for p=2. This only happens for the displacement

712

at x = 12 m . In Figure 10 it can be observed that the maximum error in variance, for any x, is still

713

monotonically decreasing with increasing order p, as desired. Small deviations from monoticity can

714

occur due to small oscillations of the simulation results used as reference.

715

A methodology based on Galerkin method for solving the stochastic bending problem of a Euler Bernoulli.....

716

Page 35 of 52

6.2 Example 2: random foundation stiffness

717
718
719

In this example, uncertainty is considered on the Pasternak foundation stiffness. Two cases are
studied. In case (a), uncertainty on coefficient P : [ 0, l ] b , b is modeled as,

720
721

N P

( )

( )

x +
x
P ( x, ( ) ) = P + 3 . P 2.n1 ( ) cos n.l
2.n ( ) sin n.l ,

n =1

(52)

722
723

where P is the expected value and P is the standard deviation. In case (b), uncertainty is on

724

coefficient W : [ 0, l ] [ c , c ] , modeled as,

725
726

N W

( )

( )

x +
x
W ( x, ( ) ) = W + 3 . W 2.n 1 ( ) cos n.l
2.n ( ) sin n.l ,

n =1

(53)

727
728

where W is the mean value and W is the standard deviation. In both cases, ( ) = {n ( )} is a

729

vector of uniform, independent random variables. In this example, four random variables are used, with

730

N P = 2 and N W = 2 . Numerical solutions are obtained, for cases (a) and (b), with P ,

731

P = ( 101 ) . P and W , W = ( 101 ) . W , respectively. As in example 1, the uncertain foundation

732

stiffness coefficients (Eqs. 52 and 53) are parameterized, weakly stationary random processes.

733

The convergence of Monte Carlo simulation is similar to example 1 (shown in Figures 4 and 5).

734

Results show that the Monte Carlo estimate of beam displacement response is stable, and independent

735

of sample size for N>5000 samples. It is observed that the mean displacement is roughly the same, as

736

for example 1.

A methodology based on Galerkin method for solving the stochastic bending problem of a Euler Bernoulli.....

Page 36 of 52

737

Figure 12 presents results for the expected value of mid-spam displacements, obtained trough

738

Monte Carlo simulation, and by means of the Galerkin solutions of order p {1,2,3} . It can be observed

739

that the Galerkin solution is acceptable even for chaos polynomials of low order ( p = 1 ). Similar results

740

are obtained for case (b), not shown here.

741

742
743
744
745
746

x 10

1
2
3
4
(x)
u
5
6
7

Monte Carlo
p=1
p=2
p=3

747

9
0

748

0.1

0.2

0.3

0.4

0.5
x

0.6

0.7

0.8

0.9

Figure 12: Expected value of the displacement.

749
750

Figures 13a and 13b confirm these results, by showing the relative errors in expected value, for

751

cases (a) and (b), respectively. It can be observed that, although small for both cases, the error is

752

smaller for case (b). The order of magnitude of both errors is smaller than the errors (in expected value)

753

for example 1.

754
755
756
757
758
759
760

A methodology based on Galerkin method for solving the stochastic bending problem of a Euler Bernoulli.....

761

762

x 10

x 10

763
764

Page 37 of 52

5
6

4
(x)

(x)

765

5
4
3

2
p=1
p=2
p=3

766

767

0
0

0.1

0.2

0.3

0.4

0.5
x

0.6

0.7

0.8

0.9

p=1
p=2
p=3

1
0
0

0.1

0.2

0.3

0.4

0.5
x

0.6

0.7

0.8

0.9

Figure 13: Relative errors in expected value for example 2, cases (a) - left - and (b) - right.

768
769
770

The variance of random beam displacements for example 2, cases (a) and (b), obtained via Monte

771

Carlo simulation and via Galerkin solutions for p {1,2,3} is shown in Figures 14a and 14b. It can be

772

observed that the dispersion is larger for case (a). Since for both cases mean values and variances are

773

similar, one observes the influence of the differentiation order, for each foundation stiffness term, over

774

the variance of the displacement response.

775
9

776

x 10

777

Monte Carlo
p=1
p=2
p=3

4
2

2(x) 4
u

3
3

781

Monte Carlo
p=1
p=2
p=3

u(x)

780

11

778
779

x 10

0
0

0.1

0.2

0.3

0.4

0.5
x

0.6

0.7

0.8

0.9

0
0

0.1

0.2

0.3

0.4

0.5
x

0.6

0.7

0.8

0.9

782
783
784

Figure 14: Variance of mid-spam displacements for example 2, case (a) - left - and case (b) - right.

A methodology based on Galerkin method for solving the stochastic bending problem of a Euler Bernoulli.....

Page 38 of 52

785

Figures 15a and 15b show the corresponding relative errors in variance, for cases (a) and (b),

786

respectively. Although case (a) presented higher dispersion (Figure 14a), one observes that the

787

Galerkin solutions are better in terms of the variance for case (a). As for example 1, it is observed that

788

the relative error in variance is larger than relative error in expected values, for both cases. Interestingly

789

it is observed in Figure 15a that the error in variance is non-monotonic: the error is larger for p=3 or

790

p=2 than for p=1.

791
1

0.9

792

p=1
p=2
p=3

0.8

0.9
0.8

0.7

793

0.7

0.6

0.6

794
795
796
797
798

2(x)

0.5

2(x) 0.5

0.4

0.4

0.3

0.3

0.2

0.2

0.1

0.1

0
0

0.1

0.2

0.3

0.4

0.5
x

0.6

0.7

0.8

0.9

0
0

p=1
p=2
p=3

0.1

0.2

0.3

0.4

0.5
x

0.6

0.7

0.8

0.9

Figure 15: Relative errors in variance for example 2, cases (a) - left - and (b) - right.

799
800

Figures 16a and 16b present the histograms, for random variable u ( 12 , ) , for cases (a) and (b).

801

These histograms were obtained from Monte Carlo simulation and from Galerkin solutions with gPC of

802

order p = 3. An excellent agreement between the histograms can be observed, showing that the

803

developed Galerkin solution is accurate not only for the first and second moments, but for the whole

804

probability distribution of the response. The agreement between the histograms is best for case (b),

805

although it is excellent for both cases of Example 2.

806
807
808

A methodology based on Galerkin method for solving the stochastic bending problem of a Euler Bernoulli.....

Page 39 of 52

809
810
811
812

4500

4500

4000

4000

3500

3500

3000

3000

2500
nu(0,5)

813
814

815

2500
u(0,5)

2000

2000

1500

1500

1000

1000

500

500

0
8.9

8.85

8.8

8.75

8.7

8.65
u(0,5)

8.6

8.55

8.5

8.45

8.4
3
x 10

Monte Carlo
p=3

0
8.675

8.67

8.665

8.66

8.655

8.65 8.645
u(0,5)

816

Figure 16: Histograms of mid-spam displacement response

817

for example 2: case (a) left and case (b) right.

8.64

8.635

8.63

8.625
3

x 10

818
819

Figures 17a and 17b show the corresponding cumulative probability distributions, for random

820

variable u ( 12 , ) , obtained from Monte Carlo simulation and from the approximated Galerkin solution

821

with p=3. Again, an excellent agreement can be observed.

822

.
1

823
824
825

1
Monte Carlo
p=3

0.9
0.8

0.8

0.7

0.7

0.6

0.6

CDF 0.5

CDF 0.5

0.4

0.4

0.3

0.3

0.2

0.2

0.1

0.1

u(0,5)

u(0,5)

826
827
828

0.9

0
8.9

8.8

8.7

8.6
u(0,5)

8.5

8.4

8.3
3
x 10

Monte Carlo
p=3

8.675 8.67 8.665 8.66 8.655 8.65 8.645 8.64 8.635 8.63 8.625
u(0,5)

829

Figure 17: Cumulative probability distributions of mid-spam displacement response

830

for example 2: case (a) left and case (b) right.

831
832

x 10

A methodology based on Galerkin method for solving the stochastic bending problem of a Euler Bernoulli.....

833

Page 40 of 52

Summary of results for example 2:

834
835

Table 3 summarizes results of expected value, variance and corresponding relative errors for the

836

random variable obtained by fixing x = 12 m in the random process displacement, for cases (a) and (b)

837

of example 2. Results are presented for approximated solutions with p {1,2,3} . Monte Carlo estimates

838

of expected value and variance:

839
840

1
=0.00864795342664219 m;
case (a): u2 ( 21)
u ( )=6.1330965223349410 9 m2 ;
2

1
=0.00864724629346642 m;
case (b): u2 ( 21)
(54)
u ( )=6.8221090222025610 11 m2 .
2

841
842

Comparing the estimates for variance (Eq. 54), it can be observed that dispersion is greater for case (a).

843

This had already been observed in Figure 8, and indicates that when uncertainty is on a coefficient

844

associated to higher order derivatives, propagation of the uncertainty is larger.

845
846

Table 3: Summary of numerical results of example 2, for cases (a) and (b):

847

expected value, variance, relative errors in expected value

848

and variance evaluated at x = 12 m

849
850

Comparing the results in Table 3, for example 2, with the results in Table 2, for example 1, one

851

notes that the variances are much smaller for the cases where uncertainty is on the foundation stiffness

852

coefficients. Since for both problems the standard deviation is 10% of the mean, it can be concluded

853

that the propagation of the uncertainty to the beams displacement response is much larger when

854

uncertainty is on the beam stiffness. In a similar way, comparing the relative errors in expected value,

A methodology based on Galerkin method for solving the stochastic bending problem of a Euler Bernoulli.....

Page 41 of 52
-4

855

for examples 1 and 2, it is noted that expected errors are much smaller (less than 10 % ) for the cases

856

of random foundation stiffness, in comparison to random beam stiffness.

857

For case (b), example 2, the convergence in both expected value and variance is monotonic with

858

polynomial order, that is, the relative errors decrease as the polynomial order increases. The same is not

859

true for case (a), where some oscillation can be observed in terms of the convergence of relative errors

860

in expected value and variance. This had been observed in Figures 14a and 16a.

861

Another important aspect to be noted is the influence of the order of differentiation over the

862

variance of the displacement response. The variance of the displacement response is larger for case (a)

863

than for case (b). For case (a), the uncertain coefficient P is associated to a derivative of second order,

864

whereas for case (b) the uncertain coefficient w is associated to a derivative of zero-th order.

865
866

6.3 Comparison of Pasternak and Winkler foundations

867
868

The examples presented herein allowed a comparison of the random displacement responses for

869

uncertainties considered in the two coefficients of the Pasternak foundation (example 2, cases (a) and

870

(b)). In Silva Jr. et al. [23], similar results were obtained for the random displacements of Euler-

871

Bernoulli beams on Winkler foundations. For the beam on Winkler foundation, with uncertainty on

872

beam stiffness, the following values were obtained for the mean, variance, relative error in mean and

873

variance, for the displacement at x = 12 m , and using polynomials of order p = 3,

874

875

876

u ( 12 ) = 0.00942464045330815;

u2 ( 12 ) = 1.88638407354279 10-6 ;

1
u ( 2 ) = 0.0523945917209326;

1
u2 ( 2 ) = 1.34851682686812.

(55)

A methodology based on Galerkin method for solving the stochastic bending problem of a Euler Bernoulli.....

Page 42 of 52

877

It can be observed that the variance, and the error in variance, is much larger for the beam on Winkler

878

foundation.
For the beam on Winkler foundation with uncertain foundation coefficient W , the following

879
880

numerical results were obtained [23],

881
u ( 12 ) = 0.00923685601509934;

u2 ( 12 ) = 8.85995242558453 10-11;

-7
1
u ( 2 ) = 6.7389573238594710 ;

1
u2 ( 2 ) = 0.550566200510562.

882

(56)

883
884

By comparing Eq. (56) with results presented in Table 3, for case (b), it can be observed that

885

uncertainty propagation to the displacement response is slightly larger for the Winkler foundation, in

886

comparison to the Pasternak foundation. The relative errors in expected value and variance, however,

887

are slightly smaller for the Winkler foundation problem.

888
889

7.

CONCLUSIONS

890
891

In this paper, theoretical and practical results for bending of stochastic Euler-Bernoulli beams on

892

Pasternak-type foundations have been presented. The uncertainty in beam and foundation stiffness was

893

modeled as parameterized random processes. For that we present a methodology that is developed in

894

two stages. In the first stage presents a study of existence and uniqueness of solutions of the theoretical

895

problem. For this, the Lax-Milgram lemma was use to establish a proof for existence and uniqueness of

896

the theoretical solutions. This study This study evidences the constraints to the choice of the stochastic

897

processes for the representation of uncertainty, and indicates the set of functions used to generate the

898

space of approximate numerical solutions. The Galerkin method was employed in the construction of

A methodology based on Galerkin method for solving the stochastic bending problem of a Euler Bernoulli.....

Page 43 of 52

899

approximate, numerical solutions to Euler-Bernoulli beam bending on Pasternak foundation. Tensor

900

product of finite-dimensional spaces was used to construct the approximate solutions. In order to

901

represent the random displacement response, a family of Legendre polynomials, originated from the

902

Askey-Wiener scheme, were employed. The developed scheme was employed in the solution of four

903

example problems, where uncertainty was considered on the materials Youngs modulus, on cross-

904

section height, and on foundation stiffness parameters P and w . The performance of the developed

905

scheme was evaluated by comparing first and second order moments of the approximated, Galerkin

906

responses, with the same moments evaluated via Monte Carlo simulation, using 20,000 samples of the

907

displacement response. In general, it was observed that even low-order Galerkin solutions (p=1)

908

provide accurate approximations of the expected value of the response. For the variance of the

909

displacement responses, higher order polynomials were needed, but accurate results were already

910

obtained for p3, for all cases studied. Moreover, it was also shown that the histograms and the

911

cumulative probability distribution of the displacement responses are accurately represented with

912

Galerkin solutions or order p=3. If the whole probability distribution function is represented, than any

913

higher-order moment can also be represented with similar accuracy. Comparing the examples with

914

uncertainty in beam stiffness and in foundation stiffness, it was observed that, for the same standard

915

deviation (10% of the respective mean), propagation of the uncertainty to the displacement response is

916

much larger for the case of random beam stiffness. For the case of random cross-section height,

917

propagation of the uncertainty was higher than for random Youngs modulus. Comparing the two cases

918

with uncertain foundation stiffness, it was observed that for uncertain coefficient P , propagation of

919

the uncertainty to the displacement response was larger than for uncertain coefficient w . It was also

920

observed that propagation of the uncertainty to the displacement response is directly related to the order

921

of derivative associated to the random coefficient. Uncertainty propagation was larger for the case of

922

random beam stiffness (EI), which is associated to derivatives of fourth order. Second largest was the

A methodology based on Galerkin method for solving the stochastic bending problem of a Euler Bernoulli.....

Page 44 of 52

923

uncertainty propagation in case of random coefficient P , which is associated to a second-order

924

derivative. The smallest uncertainty propagation was observed for uncertain coefficient w , which is

925

associated to a zero-order derivative. Results obtained herein for the Euler-Bernoulli beam on

926

Pasternak foundation were also compared with results obtained elsewhere for a Winkler foundation. It

927

was observed that uncertainty propagation under a Winkler-type foundation is larger than for the

928

Pasternak foundation, for uncertain Youngs modulus and for uncertain foundation stiffness. The

929

developed Galerkin solution developed herein, using chaos polynomials of the Askey-Wiener scheme,

930

was shown to be an accurate and efficient solution for problems of stochastic Euler-Bernoulli beams on

931

Pasternak foundations. If a proper family of chaos-polynomials is selected, and if the conditions for

932

existence and uniqueness of the solution are respected, excellent convergence rates are obtained in the

933

approximation of random displacement responses.

934
935

REFERENCES

936
937
938
939
940

[1] Arajo JM, Awruch AM. (1994) On stochastic finite elements for structural analysis. Computers

and Structures; v. 52, n. 3, p. 461-469.


[2] Spanos PD, Ghanem R. (1989) Stochastic finite element expansion for media random. Journal

Engineering Mechanics ; v. 125, n. 1, p. 26-40.

941

[3] Babuska I; Tempone R, Zouraris GE. (2005) Solving elliptic boundary value problems with

942

uncertain coefficients by the finite element method: the stochastic formulation. Computer

943

Methods in Applied Mechanics and Engineering, v. 194, n. 12-16, p. 1251-1294.

A methodology based on Galerkin method for solving the stochastic bending problem of a Euler Bernoulli.....

Page 45 of 52

944

[4] Silva Jr. CRA.(2004). Application of the Galerkin method to stochastic bending of Kirchhoff

945

plates. Doctoral Thesis, Department of Mechanical Engineering, Federal University of Santa

946

Catarina, Florianpolis, SC, Brazil (in Portuguese),

947
948
949
950

[5] Xiu D, Karniadakis GE. (2002) The Wiener-Askey Polynomial Chaos for Stochastic Differential

Equations. SIAM Journal on Science Computing., v. 24, n. 2, p. 619-644.


[6] Vanmarcke EH, Grigoriu M. (1983) Stochastic finite element analysis of simple beams. Journal

of Engineering Mechanics; v. 109, n. 5, p.12031214.

951

[7] Elishakoff I, Ren YJ, Shinozuka M. (1995) Some exact solutions for the bending of beams with

952

spatially stochastic stiffness. International Journal of Solids and Structures; v. 32, n. 16, p. 2315-

953

2327.

954

[8] Ghanem R, Spanos PD. (1991) Stochastic Finite Elements: A Spectral Approach, Dover, NY.

955

[9] Chakraborty S, Sarkar SK.(2000) Analysis of a curved beam on uncertain elastic foundation.

956
957
958
959
960

Finite Elements in Analysis and Design; v. 36, n. 1, p. 73-82.


[10] Grigoriu M. (1995) Applied Non-Gaussian Processes: Examples, Theory, Simulation, Linear

Random Vibration, and Matlab Solutions. Prentice Hall,.


[11] Cameron RH, Martin WT. (1947) The orthogonal development of nonlinear functionals in series

of Fourier-Hermite functional. Annals Mathematics, n. 48, p. 385-392.

961

[12] Yoshida K. (1978) Functional analysis Springer Berlin.

962

[13] Adams RA. (1975) Sobolev spaces. Academic Press. New York.

963

[14] Matthies HG, Keese A. (2005) Galerkin methods for linear and nonlinear elliptic stochastic

964

partial differential equations. Computer Methods in Applied Mechanics and Engineering; v. 194,

965

n. 12-16, p. 1295-1331.

A methodology based on Galerkin method for solving the stochastic bending problem of a Euler Bernoulli.....

966
967
968
969

Page 46 of 52

[15] Kinderlehrer D, Stampacchia G. (1987) An Introduction to Variational Inequalities and Their

Applications, Society for Industrial Mathematics, p. 313,.


[16] Brenner SC, Scott LR. (1994). The Mathematical Theory of Finite Element Methods. Springer-

Verlag

970

[17] Wiener N. (1938) The homogeneous chaos American. Journal Mathematics, n. 60, p. 897-936.

971

[18] Askey R, Wilson J. (1985) Some Basic Hypergeometric Polynomials that Generalize Jacobi

972
973
974

Polynomials.Mem. Amer. Math. Soc., 319, AMS, Providence, RI.


[19] Ogura H.(1972) Orthogonal functionals of the Poisson process. IEEE Trans. Inform. Theory, v.

18, n. 4, p. 473481.

975

[20] Jason S. (1997).Gaussian Hilbert Spaces. Cambridge University Press.

976

[21] Fernandez PJ. (2002) Measure and Integration. IMPA, p. 198 (in Portuguese).

977

[22] vila S Jr. CR, Beck AT, Rosa E. (2009) Solution of the stochastic beam bending problem by

978

Galerkin method and the Askey-Wiener scheme. Latin American Journal of Solids and

979

Structures; v. 6, p. 51-72.

980

[23] vila S Jr. CR, Beck AT, Mantovani G, Azikri H. (2010) Galerkin Solution of Stochastic Beam

981

Bending on Winkler Foundations. Computer Modeling in Engineering & Science; v. 1729, p. 1-

982

31.

983
984

[24] vila S. Jr CR, Beck AT. (2011) Chaos-Galerkin solution of stochastic Timoshenko bending

problems. Computers & Structures; v. 89, 599-611.

985

[25] vila S Jr.CR, Beck AT. (2010) Bending of stochastic Kirchhoff plates on Winkler foundations

986

via the Galerkin method and the Askey Wiener scheme. Probabilistic Engineering Mechanics; v.

987

25, p. 172-182.

A methodology based on Galerkin method for solving the stochastic bending problem of a Euler Bernoulli.....

988
989
990
991

Page 47 of 52

[26] Besold P. (2000). Solutions to stochastic partial differential equations as elements of tensor

product spaces. PhD Thesis, Universitt Gottingen


[27] Babuska I, Chatzipantelidis P. (2002) On solving elliptic stochastic partial differential equations,

Computer Methods in Applied Mechanics and Engineering; v. 191, n. 37-38, p. 4093-4122.

992

[28] Soize C, Ghanem RG. (2009) Reduced Chaos decomposition with random coefficients of vector-

993

valued random variables and random fields. Computer Methods in Applied Mechanics and

994

Engineering; v. 198, n. 21-26, p. 1926-1934.

995

[29] Frauenfelder P, Schwab C, Todor RA. (2005) Finite elements for elliptic problems with stochastic

996

coefficients. Computer Methods in Applied Mechanics and Engineering; v. 194, n. 2-5, p. 205-

997

228.

998
999

[30] Stefanou G. (2009) The stochastic finite element method: Past, present and future. Computer

Methods in Applied Mechanics and Engineering; v. 198, n. 9-12, p. 1031-1051.

1000

[31] Chen NZ, Guedes Soares CG. (2008)Spectral stochastic finite element analysis for laminated

1001

composite plates. Computer Methods in Applied Mechanics and Engineering; v. 197, n. 51-52, p.

1002

4830-4839.

1003
1004
1005
1006

[32] Elman H, Furnival D. (2007) Solving the stochastic steady state diffusion problem using

multigrid. IMA J Numer Anal, n. 27, p. 675688.


[33] Sett K, Jeremi B, Kavvas ML. (2011) Stochastic elasticplastic finite elements. Computer

Methods in Applied Mechanics and Engineering;v. 200, n. 9-12, p. 997-1007.

1007

[34] Sarsri D, Azrar L, Jebbouri A, El Hami A. (2011) Component mode synthesis and polynomial

1008

chaos expansions for stochastic frequency functions of large linear FE models. Computers &

1009

Structures; v. 89, n. 3-4, p. 346-356.

A methodology based on Galerkin method for solving the stochastic bending problem of a Euler Bernoulli.....

1010
1011
1012
1013
1014
1015

Page 48 of 52

[35] ksendal BK. (2003) Stochastic Differential Equations: An Introduction with Applications.

Springer, Berlin.
[36] Olsson AMJ, Sandberg GE. (2002) Latin Hypercube Sampling for Stochastic Finite Element

Analysis. Journal of Engineering Mechanics; v. 128, n. 1, p. 121-125.


[37] Beck AT, Silva Jr. CRA. (2011) Timoshenko versus Euler beam theory: Pitfalls of a deterministic

approach. Structural Safety;v. 33, n. 1, p. 19-25.

1016

[38] Treves F. (1967) Topological Vector Spaces, Distributions and Kernels. Academic Press.

1017

[39] Ryan RA.(2002) Introduction to Tensor Products of Banach Spaces. Springer.

1018

[40] Rao MM, Swift JR. (2010)Probability Theory with Applications. Springer; 2nd ed.

1019

[41] Lin G, Tartakovsky AM, Tartakovsky DM. (2010) Uncertainty quantification via random domain

1020

decomposition and probabilistic collocation on sparse grids. Journal of Computational Physics; v.

1021

229, n. 19, p. 6995-7012.

1022

[42] Foo J, Yosibash Z, Karniadakis GE. (2007) Stochastic simulation of riser-sections with uncertain

1023

measured pressure loads and/or uncertain material. Computer Methods in Applied Mechanics and

1024

Engineering; v. 196, n. 41-44, p. 4250-4271.

1025

[43] Boyaval S., Le Bris C., Maday Y., Nguyen N. C. and Patera A. T. (2009) A reduced basis

1026

approach for variational problems with stochastic parameters: Application to heat conduction with

1027

variable Robin coefficient. Computer Methods in Applied Mechanics and Engineering; v. 198, n.

1028

41-44, p. 3187-3206.

1029
1030

[44] Salsa S. (2010) Partial Differential Equations in Action: From Modelling to Theory. Spring, 2nd

ed..

A methodology based on Galerkin method for solving the stochastic bending problem of a Euler Bernoulli.....

1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045

Page 49 of 52

[45] Silva Jr CRA, Beck AT, Suarez OAG. (2013) Galerkin Solution of Stochastic Reaction-Diffusion

Problems. Journal of Heat Transfer.

A methodology based on Galerkin method for solving the stochastic bending problem of a Euler Bernoulli.....

Page 50 of 52

1046
1047

Table 1: Correspondence between some random variables

1048

and polynomials of the Askey-Wiener scheme

1049
1050
1051
1052
1053
1054
1055
1056
1057
1058

Random variable

Polynomial

Gaussian

Hermite

Gama

Laguerre

Beta

Jacobi

Uniform

Legendre

Weight function

1
( +1)
++1)
( ++ 2 )
2 (
( +1)( +1)

Support

[ 0, + )

(1 ) (1 + )

1
b a

[ a, b ]

[ a, b ]

A methodology based on Galerkin method for solving the stochastic bending problem of a Euler Bernoulli.....

Page 51 of 52

1059
1060

Table 2: Summary of numerical results of example 1, cases (a) and (b):

1061

expected value, variance, relative errors in expected value

1062

and variance evaluated at x = 12 m


Case (a)
p
u ( 12 )

u2 ( 12 )

-0.008798316389903

1.3012802314298 10-6

0.0423306148505477

7.82522205237562

-0.008802491241301

1.40956270473782 10-6

0.0050998704043519

0.15515015570709

-0.008802626580317

1.41541143047253 10-6

0.0066374569206077

0.25913800671375

( 12 )

( 12 )

Case (b)
p

1063
1064
1065
1066
1067
1068
1069

u ( 12 )

u2 ( 12 )

-0.0094452704405395

1.07214504174748 10-5

1.36251865810766

33.8863770634378

-0.0095690939781480

1.5348684018844110-5

0.06942260993890

5.35262783658891

-0.0095778536745995

1.60881259466828 10-5

0.02205538433661

0.79287305556321

( 12 )

( 12 )

A methodology based on Galerkin method for solving the stochastic bending problem of a Euler Bernoulli.....

Page 52 of 52

1070
1071

Table 3: Summary of numerical results of example 2, for cases (a) and (b):

1072

expected value, variance, relative errors in expected value

1073

and variance evaluated at x = 12 m


Case (a)
p

u ( 12 )

u2 ( 12 )

-0.008647959156096

6.1797762049941310-9

6.6252143571266510-5

0.761111169361138

-0.008647959247087

6.18213009612544 10-9

6.73043078299650 10-5

0.799491310986781

-0.008647959247100

6.18213009612544 10-9

6.73044610031819 10-5

0.799500645487117

( 12 )

( 12 )

Case (b)
p

1074
1075

u ( 12 )

u2 ( 12 )

-0.00864724621762256

6.7588417616534 10-11

8.77086896812982 10-7

0.92738565659469

-0.00864724621763344

6.7588699713209 10-11

8.76961054203722 10-7

0.92697215297764

-0.00864724621763344

6.7588699735112 10-11

8.76961054203722 10-7

0.92697212087265

( 12 )

( 12 )

Вам также может понравиться