Anda di halaman 1dari 7

ARTICLE IN PRESS

Structural Safety xxx (2010) xxx–xxx

Contents lists available at ScienceDirect

Structural Safety
journal homepage: www.elsevier.com/locate/strusafe

Timoshenko versus Euler beam theory: Pitfalls of a deterministic approach


André Teófilo Beck a,*, Cláudio R.A. da Silva Jr. b
a
Department of Structural Engineering, EESC, University of São Paulo, Brazil
b
Department of Mechanical Engineering, Federal University of Technology of Paraná, Brazil

a r t i c l e i n f o a b s t r a c t

Article history: The selection criteria for Euler–Bernoulli or Timoshenko beam theories are generally given by means of
Received 24 July 2009 some deterministic rule involving beam dimensions. The Euler–Bernoulli beam theory is used to model
Received in revised form 20 April 2010 the behavior of flexure-dominated (or ‘‘long”) beams. The Timoshenko theory applies for shear-domi-
Accepted 26 April 2010
nated (or ‘‘short”) beams. In the mid-length range, both theories should be equivalent, and some agree-
Available online xxxx
ment between them would be expected. Indeed, it is shown in the paper that, for some mid-length
beams, the deterministic displacement responses for the two theories agrees very well. However, the arti-
Keywords:
cle points out that the behavior of the two beam models is radically different in terms of uncertainty
Euler–Bernoulli beam
Timoshenko beam
propagation. In the paper, some beam parameters are modeled as parameterized stochastic processes.
Uncertainty propagation The two formulations are implemented and solved via a Monte Carlo–Galerkin scheme. It is shown that,
Parameterized stochastic processes for uncertain elasticity modulus, propagation of uncertainty to the displacement response is much larger
Monte Carlo simulation for Timoshenko beams than for Euler–Bernoulli beams. On the other hand, propagation of the uncertainty
Galerkin method for random beam height is much larger for Euler beam displacements. Hence, any reliability or risk anal-
ysis becomes completely dependent on the beam theory employed. The authors believe this is not widely
acknowledged by the structural safety or stochastic mechanics communities.
Ó 2010 Elsevier Ltd. All rights reserved.

1. Introduction In this paper, it is shown that, for some mid-length beams,


deterministic displacement responses for the two beam theories
This paper presents a comparison of the Euler–Bernoulli and agree very well. In this case, the theories are generally accepted
Timoshenko beam theories, taking into account parameter uncer- as equivalent. However, it is shown in the paper that, although
tainties and uncertainty propagation. It is widely known that the the theories are equivalent when compared deterministically, their
Euler–Bernoulli beam theory properly models the behavior of flex- behavior is radically different in terms of uncertainty propagation.
ure-dominated (or ‘‘long”) beams. The Timoshenko theory is This is shown by means of some illustrative example problems.
known to apply for shear-dominated (or ‘‘short”) beams. In the In Section 2, formulation of the two beam theories is presented.
mid-length range, both theories should be equivalent, and some Representation of the uncertainty in beam parameters, via param-
agreement between them would be expected. eterized stochastic processes, is presented in Section 3. In the
The stochastic beam bending problem has been studied by sev- numerical examples, a Galerkin–Monte Carlo scheme is used to ob-
eral authors. Vanmarcke and Grigoriu [1] studied the bending of tain the random displacement fields. The Galerkin solutions are
Timoshenko beams with random shear modulus. Elishakoff et al. presented in Section 4. Section 5 shows the evaluation of first
[2] employed the theory of mean square calculus to construct a and second order moments of the Monte Carlo solution. Two
solution to the boundary value problem of bending with stochastic example problems are shown in Section 6, illustrating the large dif-
bending modulus. Ghanem and Spanos [3] used the Galerkin meth- ferences between the two formulations in terms of uncertainty
od and the Karhunem-Loeve series to represent uncertainty in the propagation. Section 7 discusses the effects of these differences
bending modulus by means of a Gaussian stochastic process. Cha- on reliability and risk analysis. Section 8 finishes the paper with
kraborty and Sarkar [4] used the Neumann series and Monte Carlo some conclusions.
simulation to obtain statistical moments of the displacements of
curved beams, with uncertainty in the elasticity modulus of the
2. Euler and Timoshenko beam formulations
foundation.
In this section, the strong and weak formulations of the prob-
lems of stochastic bending of Euler–Bernoulli and Timoshenko
* Corresponding author. Tel.: +55 16 3373 9460; fax: +55 16 3373 9482. beams are presented. The strong form of the Euler–Bernoulli beam
E-mail address: atbeck@sc.usp.br (A.T. Beck). bending problem is given by:

0167-4730/$ - see front matter Ó 2010 Elsevier Ltd. All rights reserved.
doi:10.1016/j.strusafe.2010.04.006

Please cite this article in press as: Beck AT, da Silva Jr. CRA. Timoshenko versus Euler beam theory: Pitfalls of a deterministic approach. Struct Saf (2010),
doi:10.1016/j.strusafe.2010.04.006
ARTICLE IN PRESS

2 A.T. Beck, C.R.A. da Silva Jr. / Structural Safety xxx (2010) xxx–xxx

8 2 
> d d2 w 3. Uncertainty representation
>
> 2 EIðx; x Þ  2 ¼ f ; 8ðx; xÞ 2 ð0; lÞ  X;
> dx
> dx
<
wð0; xÞ ¼ 0;
ð1Þ In most engineering problems, complete statistical information
>
> wðl; xÞ ¼ 0;
>
>   about uncertainties is not available. Sometimes, the first and sec-
>
: dw 
dx ð0;xÞ
¼ dw
dx ðl;xÞ
¼ 0; 8x 2 X; ond moments are the only information available. The probability
distribution function is defined based on experience or
where w is the transverse displacement field, EI is the bending stiff- heuristically.
ness, X is a sample space and f is a load term. The strong form of the In order to apply Galerkin’s method, an explicit representation
Timoshenko beam bending problem is given by: of the uncertainty is necessary. In this paper, uncertain parameters
8     are modeled as parameterized stochastic processes. These are de-
>
>
d
EIðx; xÞ  d/ þ GAðx; xÞ  dw  / ¼ 0; fined as a linear combination of deterministic functions and ran-
>
> dx dx dx
<d  
dom variables [9]:
dx
GAðx; xÞ  dw dx
 / ¼ f ; 8ðx; xÞ 2 ð0; lÞ  X; ð2Þ
>
>
>
> wð0; xÞ ¼ wðl; xÞ ¼ 0; X
N
: jðx; xÞ ¼ g i ðxÞni ðxÞ; ð6Þ
/ð0; xÞ ¼ /ðl; xÞ ¼ 0; 8x 2 X;
i¼1

where / is the angular displacement field, GA is the shear stiffness, where fg i gNi¼1 are deterministic functions and fni gNi¼1 are random
and the remaining symbols follow Eq. (1). The angular displace- variables.
ments stochastic process in Euler–Bernoulli theory is given by the
space derivative of the transverse displacement field. Both formula-
4. Galerkin method
tions are given for clamped–clamped boundary conditions.
In the sequence, elasticity modulus E and beam height h will be
The Galerkin method and direct Monte Carlo simulation are
assumed as stochastic processes. Hence, the displacement re-
used in this paper to obtain sample realizations of the beams ran-
sponses w and / will also be stochastic processes. In order to en-
dom displacements, from samples of the beams random
sure existence and uniqueness of the solutions, the following
parameters.
hypotheses are required:
Approximated solutions for the qth realization of the transverse
8
>  2 Rþ n f0g; j½a; a
9a; a  j < þ1; Pðfx 2 X : EIðx; xÞ 2 ½a; a
 ; displacement random process, for the Euler–Bernoulli beam, are
>
>
< 8x 2 ½0; lgÞ ¼ 1; given by:
H1 :
>  2 Rþ n f0g; j½s; s
> 9s; s j < þ1; Pðfx 2 X : GAðx; xÞ 2 ½s; s; X
m
>
:
8x 2 ½0; lgÞ ¼ 1; wqm ðx; xq Þ ¼ wiq ui ðxÞ; ð7Þ
i¼1
H2 : f 2 L2 ðX; F; P; L2 ð0; lÞÞ:
ð3Þ where fwiq gm m
i¼1 are coefficients to be determined and fui gi¼1 are
interpolating functions for the qth realization. Observing that
Hypothesis H1 ensures that the elasticity modulus and beam H20 ð0;lÞ
height are strictly positive and uniformly limited in probability C 20 ð0; lÞ ¼ H20 ð0; lÞ, and considering a complete orthonormal set
 U

[5]. Hypothesis H2 ensures that the stochastic load process has fi- 1
U ¼ fu i gi¼1 of U [10], such that span½U ¼ U . Since approxi-
nite variance. These hypotheses are necessary for application of the
mated numerical solutions are derived in this paper, the solution
Lax–Milgram Lemma, which ensures existence and uniqueness of
space has finite dimensions. This implies truncation of the complete
the solution, as well as continuous dependency on the data [5,6].
orthonormal set U, which results in Um ¼ fui gm i¼1 and
The abstract variational problem associated to the strong form
Um ¼ span½Um . The approximated variational problem associated
(Eq. (1)) of the stochastic Euler–Bernoulli beam bending problem
to the Euler–Bernoulli beam is obtained by inserting Eq. (7) in Eq.
is obtained as:
(4):
8
< Find
> w 2 Vsuch that :
R Rl  R Rl 8
For the qth realization; find fwiq gm 2 Rm such that :
d2 w d2 v >
X 0 EI  dx2  dx2 ðx; xÞdx dPðxÞ ¼ X 0 ðf  v Þðx; xÞdx dPðxÞ;
>
>  
i¼1
> <P m R Rl
: l 2 2
d u
8v 2 V: EIðx; xq Þ  ddxu2 i  dx2 j ðxÞdx wiq ¼ 0 f ðx; xq Þ  uj ðxÞdx;
> i¼1 0
>
ð4Þ >
:
8wj 2 W:
where V ¼ L2 ðX; F; P; UÞ with U ¼ H20 ð0; lÞ. ð8Þ
The abstract variational problem associated to the strong form
This problem can also be written in matrix form:
(Eq. (2)) of the stochastic Timoshenko beam bending problem is

obtained as: Find uq 2 Rn such that
8 ð9Þ
> Find ðw;/Þ 2 W such that : Kq uq ¼ Fq ;
>
>
>R R l 
> dw   R Rl
<
X 0 GA  dx  /  u ðx; xÞdxdPðxÞ ¼ X 0 ðf  uÞðx; xÞdxdPðxÞ; where Kq 2 Mm ðRÞ, with elements given by:
R R l  d/ dt R Rl  dw  
>
X 0 EI  dx  dx ðx; xÞdxdPðxÞ ¼ X 0 GA  dx  /  t ðx; xÞdxdPðxÞ;
>
> Z !
>
> l 2 2
d ui d uj
: q q q
8ðu; tÞ 2 W; K ¼ ½kij mm ; kij ¼ EIðx; xq Þ  2
 2
ðxÞdx: ð10Þ
0 dx dx
ð5Þ

where W ¼ L2 ðX; F; P; QÞ with Q ¼ H10 ð0; lÞ  H10 ð0; lÞ. Eq. (5) repre- The loading term is given by,
sents a system of variational equations for the coupled fields Z l
w ¼ wðx; xÞ and / ¼ /ðx; xÞ . Fq ¼ ffiq gm
i¼1 ; fiq ¼ f ðx; xq Þ  ui ðxÞdx: ð11Þ
0
Details of the formulation of stochastic Euler–Bernoulli beams
are given in [7]. For stochastic Timoshenko beams, details can be For the qth realization of Timoshenko beam displacements, approx-
found in Ref. [8]. imated Galerkin solutions are obtained as:

Please cite this article in press as: Beck AT, da Silva Jr. CRA. Timoshenko versus Euler beam theory: Pitfalls of a deterministic approach. Struct Saf (2010),
doi:10.1016/j.strusafe.2010.04.006
ARTICLE IN PRESS

A.T. Beck, C.R.A. da Silva Jr. / Structural Safety xxx (2010) xxx–xxx 3

8 8 8
>
> q P
m
>
>
 PN
>
>
 PN

< wm ðx; xÞ ¼ wiq wi ðxÞ;


> <l
> ^ wðxÞ ¼ N1 wðx; xi Þ; <l
> ^ /ðxÞ ¼ N1 /ðx; xi Þ;
i¼1 i¼1 i¼1
ð12Þ ^
> N h
 1 P i2 > N h
 1 P i2
>
> q P
m >
>
:r^ 2w ¼ N1 wðx; xi Þ  l
^ wðxÞ ; >
>
:r^ 2/ ¼ N1 /ðx; xi Þ  l
^ /ðxÞ :
>
: /m ðx; xÞ ¼ /iq wi ðxÞ; ðxÞ ðxÞ
i¼1 i¼1
i¼1
ð17Þ

where fðwiq ; /iq Þgm i¼1 are coefficients to be determined and fwi gi¼1
m
In order to study the effects of differences in uncertainty prop-
are interpolating functions. Let Q ¼ spanfwi gm i¼1 be a set generated agation in reliability or risk analysis, a simple reliability problem is
by truncation of a complete orthonormal set W ¼ fwi g1 i¼1 in Q, with
defined. An admissible displacement, at mid-spam, is defined as
l
wi 2 C 0 ð0; lÞ \ C 1 ð0; lÞ; 8i 2 N. Replacing Eq. (12) in Eq. (5), one ar- wADM ¼  200 , where ‘‘l” is the beam length. The associated proba-
rives at the approximated variational problem for the Timoshenko bility of failure is given by:
beam:
Pf ¼ PðBÞ; ð18Þ
8
> For the qth realization; find fðwiq ; /iq Þgm 2 R2m such that; where P stands for probability and B ¼ x 2 X l
xÞ P  200 . jwð2l ;
>
> 9 i¼1
> 8 hR l
>   i This can be estimated from the same set of simulated displace-
>
> > EIðx; xq Þ  dw  wj ðxÞdx wiq >
m < = R
i
>
> P 0 dx ments, by:
>
> h i
l
¼ 0 f ðx; xq Þ  wj ðxÞdx;
> > Rl
> >  X
< i¼1 :  GAðx; x q Þ  ðw  w ÞðxÞdx / iq
; 1 N
0 i j
bf ¼
P 1B ðxi Þ; ð19Þ
>
> Pm nR h   i o N i¼1
>
> l dwi dwj
> i¼1 0 EIðx; xq Þ  dx  dx ðxÞ þ GAðx; xq Þ  ðwi  wj ÞðxÞ dx /iq
>
>
>
> where 1B : X ! f0; 1g with:
>
> Pm hR   i
>
> l
: ¼ 0
GAðx; xq Þ  dw dx
i
 wj ðxÞdx wiq ; 8wj 2 Qm : 1; x 2 B;
i¼1 1B ð xÞ ¼ ð20Þ
ð13Þ 0; x R B;
The approximated variational problem consists in finding the is the characteristic function of set B.
coefficients of the linear combination expressed in Eq. (13). Using
a vector–matrix representation, the system of linear algebraic 6. Numerical examples
equations defined in Eq. (13) is written as:
In this section, two numerical examples are presented. In the
8
q q 2m first example, the elasticity modulus is considered a random field.
< For the qth realization; find ðw ; / Þ 2 R
> such that :
q q q q q
ð14Þ In the second example, the height of the beam’s cross-section is
A w þB / ¼F ;
>
: q q random. In both cases, uncertainty is modeled by parameterized
C w ¼ Dq /q ; stochastic processes. In both examples, the beam is clamped at
both ends, the span (l) equals one meter, the cross-section is rect-
where Aq ; Bq ; Cq ; Dq 2 Mm ðRÞ. Elements of these matrices are given angular with b ¼ 30 1 1
m and h ¼ 25 m and the beam is subject to an
by: uniform distributed load of f ðxÞ ¼ 100 kPa=m; 8x 2 ½0; l.
Fig. 1 shows the exact, deterministic transverse (left) and angu-
8 q Rl  
>
> A ¼ ½aqij mm ; aqij ¼ 0 EIðx; xq Þ  dw i
 wj ðxÞdx; lar (right) displacement responses, obtained via Euler–Bernoulli
>
> dx
>
>
> R and Timoshenko beam theories. These results are obtained for
< Bq ¼ ½bqij mm ; bqij ¼  0l GAðx; xq Þ  ðwi  wj ÞðxÞdx;
  the mean values of the parameters to be considered random in
R l
>
> Cq ¼ ½cqij mm ; cqij ¼ 0 GAðx; xq Þ  dw i
 wj ðxÞdx; the following. It is observed that the two theories yield very close
>
> dx
>
>
>
: Dq ¼ ½dq  q Rl h 
dwi dwj
 i results, with transverse mid-spam displacements agreeing within
ij mm ; dij ¼ 0 EIðx; xq Þ  dx  dx ðxÞ þ GAðx; xq Þ  ðwi  wj ÞðxÞ dx:
97%. From a deterministic point of view, the two theories could
ð15Þ be considered equivalent, for this beam.
The loading term is given by Eq. (11). Solution of the linear sys-
tem in Eq. (14) is obtained as: 6.1. Random elasticity modulus

( 1 In this example, the elasticity modulus is modeled as a param-


/q ¼ ðAq Cq Dq þ Bq Þ1 Fq ; eterized stochastic process:
1 1
ð16Þ
wq ¼ Cq Dq ðAq Cq Dq þ Bq Þ1 Fq : pffiffiffi h x xi
Eðx; xÞ ¼ lE þ 3  rE n1 ðxÞ cos þ n2 ðxÞ sin ; ð21Þ
It is important to note that conversion of the continuous prob- l l
lem (Eq. (5)) to the discretized form (Eq. (13)) results in de-cou- where lE is the mean value, rE is the standard deviation and fn1 ; n2 g
pling of the displacement fields w and /, following Eq. (16). are uniform orthogonal random variables. Numerical solutions are
obtained for rE ¼ ð10
1
Þ  lE .
Results obtained via Monte Carlo simulation are shown in
5. Statistical moments and reliability problem Figs. 2–5. Fig. 2 shows the envelope (largest and smallest values)
among the 15,000 samples obtained, for transverse (left) and angu-
In the following, Monte Carlo simulation is used to study the lar (right) beam displacements. Fig. 3 shows the mean values, and
propagation of uncertainty through the Timoshenko and Euler– Fig. 4 shows the variance of both displacement fields, obtained for
Bernoulli bending models. In order to compare the solutions, it is the two beam theories. Fig. 5 shows the cumulative distribution
interesting to focus on some statistics of the results. function, obtained via simulation, of the displacement fields.
Estimates for expected value and variance of random variables Results presented in Fig. 1 suggest that the Euler–Bernoulli and
wðxÞ ¼ wðx; xÞ and /ðxÞ ¼ /ðx; xÞ, for a fixed point x 2 ½0; l, are ob- Timoshenko beam theories are equivalent for this problem. Now,
tained from the set of displacement fields samples fwðx; xi ÞgNi¼1 Figs. 2–5 make very clear that the two theories are completely dif-
and f/ðx; xi ÞgNi¼1 : ferent in terms of uncertainty propagation. It is observed that the

Please cite this article in press as: Beck AT, da Silva Jr. CRA. Timoshenko versus Euler beam theory: Pitfalls of a deterministic approach. Struct Saf (2010),
doi:10.1016/j.strusafe.2010.04.006
ARTICLE IN PRESS

4 A.T. Beck, C.R.A. da Silva Jr. / Structural Safety xxx (2010) xxx–xxx

Fig. 1. Exact deterministic solutions for transverse displacements (left) and angular displacements (right).

Fig. 2. Envelope of samples for transverse (left) and angular (right) beam displacements.

Fig. 3. Mean value of transverse (left) and angular (right) beam displacements.

uncertainty in elasticity modulus propagates much more through The two sets of Monte Carlo realizations, obtained for the Euler
the Timoshenko model than through the Euler–Bernoulli beam and Timoshenko beam displacements, can be written as:
model. The explanation for this behavior can be drawn from a com- 8 n
parison of Eqs. (1) and (2). The uncertainty in elasticity modulus >
> Ew ¼ wxi ðxÞ 2 Rjwxi ðxÞ ¼ wðx; xi Þ; ðx; xi Þ 2 ½0; l  fxi gNi¼1 ;
>
>
also represents uncertainty in the stiffness modulus G, through >
>
< 00 00
w solution of Eq:ð1Þ:g;
the relation: n
>
>
>
> Tw ¼ wxi ðxÞ 2 Rjwxi ðxÞ ¼ wðx; xi Þ; ðx; xi Þ 2 ½0; l  fxi gNi¼1 ;
E ¼ 2Gð1 þ tÞ: ð22Þ >
>
: 00 00
w solution of Eq:ð2Þ:g:
where t is the Poisson coefficient. The two uncertainty terms affect
the coupled system of Timoshenko beam equations. ð23Þ

Please cite this article in press as: Beck AT, da Silva Jr. CRA. Timoshenko versus Euler beam theory: Pitfalls of a deterministic approach. Struct Saf (2010),
doi:10.1016/j.strusafe.2010.04.006
ARTICLE IN PRESS

A.T. Beck, C.R.A. da Silva Jr. / Structural Safety xxx (2010) xxx–xxx 5

Fig. 4. Variance of transverse (left) and angular (right) beam displacements.

Fig. 5. Cumulative distribution functions of transverse beam displacements.

It is observed in Fig. 2 (left) that Ew  Tw . Hence, there are real- height is larger for the Euler–Bernoulli response than for the Tim-
izations of the Timoshenko beam displacements which are not oshenko displacements. Hence, for this example, Ew  Tw .
contained in the set of realizations of Euler displacements. Results To understand this result, the first term of Eq. (1) can be written
presented in Fig. 1 show no hint of this behavior. in the following form:

2 
d d/
EI 
2
¼ f: ð25Þ
6.2. Random cross-section height dx dx
When this equation is solved for /, and the result is used in Eq.
In this example, the beam cross-section height is modeled as a
(2) to find the transverse displacement w, one notes that the solu-
parameterized random process: 2
tion is proportional to h . For the Euler–Bernoulli beam, this dis-
3
pffiffiffi h x xi placement is proportional to h . This explains the differences in
hðx; xÞ ¼ lh þ 3  rh n1 ðxÞ cos þ n2 ðxÞ sin ; ð24Þ beam height uncertainty propagation for the two beam models,
l l
and why the propagation is larger for the Euler beam.
1 Comparing Figs. 7 and 3, it is observed that the agreement be-
where lh is the mean value, rh ¼ 10  lh is the standard deviation
and fn1 ; n2 g are uniform, independent random variables. tween the two theories is better, for this example, in comparison
Results obtained via Monte Carlo simulation are shown in to the random elasticity modulus. Comparing Figs. 8 and 4, it is ob-
Figs. 6–9. Fig. 6 shows the envelope (largest and smallest values) served that the variance is smaller for the random beam height
among the 15,000 samples obtained, for transverse (left) and angu- example.
lar (right) beam displacements. Fig. 7 shows the mean values, and
Fig. 8 shows the variance of both displacement fields, obtained for 7. Effect on reliability and risk analysis
the two beam theories. Fig. 9 shows the cumulative distribution
function, obtained via simulation, of the displacement fields. From the results presented in Section 6, it is clear that differ-
It is first observed that the agreement between the two theories ences in uncertainty propagation will affect any reliability or risk
is better for this problem, although far from ideal. However, it is analysis based on the Euler or Timoshenko beam theories. This is
noted that results have opposite trends in terms of uncertainty confirmed in this section, and quantified for the example problems
propagation: the propagation of uncertainty in random beam considered in the study.

Please cite this article in press as: Beck AT, da Silva Jr. CRA. Timoshenko versus Euler beam theory: Pitfalls of a deterministic approach. Struct Saf (2010),
doi:10.1016/j.strusafe.2010.04.006
ARTICLE IN PRESS

6 A.T. Beck, C.R.A. da Silva Jr. / Structural Safety xxx (2010) xxx–xxx

Fig. 6. Envelope of samples for transverse (left) and angular (right) beam displacements.

Fig. 7. Mean value of transverse (left) and angular (right) beam displacements.

Fig. 8. Variance of transverse (left) and angular (right) beam displacements.

Table 1 shows failure probability results obtained for the two for the random elasticity modulus example, Ew \ B ¼ Ø. This im-
beam theories, and for an admissible mid-spam displacement of plies that, for the Euler beam model, the probability of event B is
l
wADM ¼  200 (Eq. (18)). These results were obtained via simple zero, that is, the probability of failure is zero. On the other hand,
Monte Carlo simulation. It is clear that the results are completely for the Timoshenko beam theory, there is some probability associ-
dependent on the beam theory used in the analysis. ated to this event. This probability can be drawn from Fig. 5, and is
A qualitative assessment of failure probability results can be given in Table 1. For the case of random beam height, it can be ob-
drawn from Figs. 2, 5, 6 and 9. In Fig. 2, it can be observed that, served in Fig. 6 that Ew \ Tw \ B–Ø. Hence, the failure probabilities

Please cite this article in press as: Beck AT, da Silva Jr. CRA. Timoshenko versus Euler beam theory: Pitfalls of a deterministic approach. Struct Saf (2010),
doi:10.1016/j.strusafe.2010.04.006
ARTICLE IN PRESS

A.T. Beck, C.R.A. da Silva Jr. / Structural Safety xxx (2010) xxx–xxx 7

Fig. 9. Cumulative distribution functions of transverse beam displacements.

Table 1 beam than for the Timoshenko beam. Hence, although the Timo-
Effect of beam theory on failure probability results. shenko and Euler–Bernoulli beam theories appear to be equivalent
Problem ^f
P kl ¼ lwADM for the mid-length beam considered, the propagation of uncer-
wðL=2Þ
tainty to the beams displacement response is radically different.
Euler–Bernoulli Timoshenko Euler Timoshenko
As a consequence, any reliability or risk analysis becomes com-
Random E 0.0000 0.2310 2.78 2.33 pletely dependent on the theory employed.
Random h 0.1007 0.0208 2.70 2.70
There are no pitfalls in the Timoshenko or Euler–Bernoulli beam
theories presented herein. What the title of the manuscript sug-
gests is that there are pitfalls in using pure deterministic judgment
are nonzero for both beam models. These failure probabilities can when comparing these formulations, in order to choose one of
be drawn from Fig. 9, and are given in Table 1. them for a reliability or risk analysis.
Apart from the minor (3%) difference between the deterministic
Euler and Timoshenko solutions of this problem, the safety coeffi- Acknowledgements
cient for the deterministic problem is given by:
Sponsorship of this research project by the São Paulo State
wADM 0:005
k¼   ¼  l  ¼ 2:78: ð26Þ Foundation for Research – FAPESP (Grant No. 2008/10366-4) and
w 2l w 2
by the National Council for Research and Development – CNPq
This coefficient is the same for both Euler and Timoshenko (Grant No. 305120/2006-9) is greatly acknowledged.
beam formulations: hence, it clearly does not take into account
the differences in uncertainty propagation and in failure probabil- References
ities. The central safety coefficients, which are given in Table 1, are
[1] Vanmarcke EH, Grigoriu M. Stochastic finite element analysis of simple beams.
also not sufficient to provide uniform reliability for this problem. J Eng Mech 1983;109(5):1203–14.
[2] Elishakoff I, Ren YJ, Shinozuka M. Some exact solutions for the bending of
beams with spatially stochastic stiffness. Int J Solids Struct 1995;32(16):
8. Conclusions
2315–27.
[3] Ghanem R, Spanos PD. Stochastic finite elements: a spectral approach. NY:
In this paper, it was shown that two beam theories, which Dover; 1991.
seemed perfectly equivalent when compared in terms of determin- [4] Chakraborty S, Sarkar SK. Analysis of a curved beam on uncertain elastic
foundation. Finite Elem Anal Des 2000;36(1):73–82.
istic response, behave radically different in terms of uncertainty [5] Babuska I, Tempone R, Zouraris GE. Solving elliptic boundary value problems
propagation. Hence, the very notion that the theories are equiva- with uncertain coefficients by the finite element method: the stochastic
lent is limited to the realm of determinacy, and is unfounded when formulation. Comput Methods Appl Mech Eng 2005;194(12–16):1251–94.
[6] Brenner SC, Scott LR. The mathematical theory of finite element
uncertainty propagation is considered. methods. Springer-Verlag; 1994.
Two very simple examples were presented to illustrate the is- [7] Silva Jr CRA, Beck AT, Rosa E. Solution of the stochastic beam bending problem
sue, involving the Timoshenko and Euler–Bernoulli beam theories. by Galerkin method and the Askey–Wiener scheme. Lat Am J Solids Struct
2009;6:51–72.
A mid-length beam was considered, and it was shown that deter- [8] Silva Jr CRA, Beck AT. Chaos-Galerkin solution of stochastic Timoshenko
ministic displacement responses obtained by the two theories bending problems. Comput Struct, submitted for publication.
agreed within 97%. However, uncertainty in the elasticity modulus [9] Grigoriu M. Applied non-gaussian processes: examples, theory, simulation,
linear random vibration, and matlab solutions. Prentice Hall; 1995.
propagates much largely for the Timoshenko beam, in comparison
[10] Kreyszig EO. Introductory functional analysis with applications. Wiley; 1989.
to the Euler beam. When uncertainty in beam height is considered,
propagation to the displacement response is larger for the Euler

Please cite this article in press as: Beck AT, da Silva Jr. CRA. Timoshenko versus Euler beam theory: Pitfalls of a deterministic approach. Struct Saf (2010),
doi:10.1016/j.strusafe.2010.04.006

Anda mungkin juga menyukai