Vous êtes sur la page 1sur 9

Mathl. Comput. Modelling Vol. 18, No. 11, pp.

59-67, 1993
Printed in Great Britain. All rights reserved
0895-7177/93 $6.00 + 0.00
Copyright@ 1993 Pergamon Press Ltd
Nonprobabilistic Modelling of
Dynamically Loaded Beams
under Uncertain Excitations
I. S. SADEK
Department of Mathematical Sciences
University of North Carolina at Wilmington, Wilmington, NC 28403, U.S.A.
J . M. SLOSS
Department of Mathematics
University of California, Santa Barbara, CA 93106, U.S.A.
S. ADALI* AND J. C. BRUCH, JR.
Department of Mechanical and Environmental Engineering
University of California, Santa Barbara, CA 93106, U.S.A.
(Received and accepted J une 1993)
Abstract-A nonprobabilistic model for a dynamically loaded beam is developed, when the forcing
function and the initial conditions may contain uncertain components. Uncertainty in these functions
is required to be bounded in the Lz norm and expressed using a finite number of eigenmodes. The
specific expressions are determined for the dynamic loads which yield the least favorable deflection.
Numerical examples are given to illustrate the solution procedure and to assess the effect of various
problem parameters.
1. INTRODUCTION
In vibration problems, it is often the case that the exact nature of the dynamic loads is not known
in a precise manner. Such problems are usually treated by using the techniques of probabilistic
modelling, which requires a knowledge of the distribution of the probability function for the
loading or a knowledge of the random variables. These techniques have been extensively treated
in the literature (see, for example, [l-4]). H owever, complete probabilistic information may
not be available under certain circumstances and it becomes important to develop models for
handling uncertain input, data. In the present study, a nonprobabilistic approach to the modelling
of uncertainty is described by requiring the uncertain components of the dynamic loads to be
arbitrary and bounded. In this approach, the uncertain phenomena are represented by convex
sets in a suitable finite-dimensional space, and the method is referred to as convex modelling [5].
Earlier works on nonprobabilistic modelling of uncertainty include [6,7]. The convex modelling
has been applied to the study of shell imperfections [8,9], scatter in material properties [lo],
stress concentration factors (111, and optimal design of laminates under uncertain loading [12].
Uncertain vibrations of beams were studied by stochastic methods in [13].
Herein, the forced vibrations of a beam subject to uncertain excitations are investigated. The
partial information needed for convex analysis involves bounds on the L2 norms of the uncertain
*On leave from the Department of Mechanical Engineering, University of Natal, Durban, South Africa.
Downloaded from http://www.elearnica.ir
60
I. S. SADEK et al.
components of the forcing function and the initial conditions. Subject to these bounds, the
specific excitation producing the least favorable dynamic response is determined by analyzing
the convex sets associated with the uncertainties in the input. A proof if given, showing that
the extreme points of the intersection of the types of convex sets arising in this study lie in the
intersection of the boundaries of the convex sets considered. The method of Lagrange multipliers
is used to compute the optimal expressions for the uncertain functions.
Numerical results are given for beams subject to an uncertain forcing function and to uncertain
initial conditions. The maximum deflections produced by the least favorable excitations are
determined. The effects of the uncertainty bounds, the location, and time on the least favorable
deflection are investigated.
2. PROBLEM STATEMENT
The equation governing the forced vibrations of a uniform beam is given by
EI y + m g = p(z) G(t),
(I)
where EI is the flexural stiffness, y(z, t) the transverse displacement, m the mass per unit length,
and F(z) G(t) indicates the uncertain forcing function. In (l), a prime denotes differentiation
with respect to the space variable x and a dot
conditions on the motion are expressed as
Y(Z, 0) = %x)7
with respect to the time variable t. The initial
(2)
where the functions 6(x) and G(x) also contain uncertain information. The boundary conditions
on the beam are taken to be simply supported.
The forcing function and the initial conditions are assumed to be the sum of deterministic and
uncertain parts, viz.
F(z) = Fe(z) + f(z),
W = 40(x) + 9(z), %4 = $0(z) + 1z(z),
(3)
where the subscript 0 denotes the known deterministic part and a tilde, the unknown part,
which contains uncertain information. The uncertainty in these functions is required to be
bounded in the Lz norm which imposes the conditions
where ~1, ~2, and ss are prescribed measures of maximum allowable uncertainty. It is noted that
for a well-defined problem of this type, the Lz norm of the derivative of the uncertain displacement
initial condition, i.e., 11~!511~, has to be bounded. The problem investigated in this study involves
determining the least favorable response of the beam to excitations with uncertain components
of the type previously described.
3. METHOD OF SOLUTION
The eigenmodes of the freely vibrating beam are given by Y,(z) = sin(n?rz/L), which satisfy
the boundary condition and the orthogonality condition
J
L
Y,(x) Y,(z) dx =
0, for m # 12,
for m = n, J
LY~W34dz=
0 formfn,
?;, form=n (5)
0
y,
0 7
Nonprobabilistic Modelling
61
where 7 = L/2 and T,., = (n?r)2/(2L). The forcing functions and the initial conditions can be
expanded in terms of the eigenmodes of the beam, viz.,
n=l n=l
n=l n=l
where the coefficients a,, b,.,, and c,, are determined from
b,, = 1
r (II. 40(X) K(X) dz7
I
@o(x) Y,(x) dx.
(10)
The coefficients ii,, &,,,, and &, of the uncertain excitations are not known a priori, however, it is
assumed that zi, = &, = & = 0, for n > N. By virtue of condition (4), they satisfy
Expanding the solution of (1) in terms of the eigenmodes, we obtain
y(z,t) = 2 Z9z.Q) K(x).
n=l
(12)
The equations satisfied by Zn(t) are derived by inserting Y(z,t) from (12) and P(z) from (3)
into (1) and using (6), viz.,
&(t) + 0; Zn(t) = ; (a, + &) G(t),
(13)
where 0, = (EI/m)/2 (nr/L)2 is the nth eigenfrequency of the freely vibrating beam. The
solution of (13) is given by
1 t
Z,(t) = (Y, cos R,t + 0, sin R,t + -
I mQn 0
(a, + &) G(t) sin % (t - 0 dE,
(14)
where the coefficients CX,, and P,, are determined from the initial conditions (2). From (3), (7),
(8)) and (14)) LY, and Pn are computed as
cn + E,
an=bn+ii,, &=-.
Qn
(15)
Now, the solution of the forced vibration problem follows from (12), (14), and (15) as
Y(z, t) = 2 [(b, + L) COS Rnt + & (C, + En) sin R,t
n=l
1 t
+-
s m% 0
(a, + iin) G(E) sin fL(t - t) 4
1
Y,(x). (16)
62
I. S. SADEK et al.
4. UNCERTAINTY ANALYSIS
The uncertain coefficients ii,, &,, and C, are undetermined, but are required to satisfy the
inequalities (11) which represent three convex sets
in the coordinates (&, &, &). The intersection of these three convex sets is a non-empty compact
convex set S. It is known that every affine functional whose domain is a compact convex set
takes on its maximum value on the set of extreme points of its domain (see [14, Theorem 15,
p. 2141). Since y(z,t) in (16), for a given point z = 20 and time t = to is a linear function of the
3N variables
U
%z,&&
>>
N
ll=l
when the initial conditions are prescribed, then y(zo,ts) takes
on its maximum value at an extreme point of the boundary of S. However, at every extreme
point of S, equality for all three constraints in (11) must hold. (See Appendix.) Therefore, to
obtain the least favorable response, {&}, {&}, and {&} must satisfy the equality constraints.
5
N N
rig=& C#=&;, c yc; = E$
(17)
?Z=l n=l n=l
This observation enables us to employ the method of Lagrange multipliers to compute the
optimal values of the uncertain coefficients causing the least favorable response of the beam. The
Lagrangian for this problem is given by
L = Y(zo,to) + A,
(18)
The extremum condition yields
dC
--0
dti*
dL o dC o
I z=, ac,=.
from (W, (1%
and (19), it follows that
in which
&.(zo,to) = -&
s
to
G (t) sin fL (to - E) dt Y, (~0) ,
n 0
, 1
(19)
(20)
(21)
B,(zo,to) = - cos (ado) K (x0),
2
&(x0, to) = -& sin (W0) Yn (20).
n
The values of the Lagrange multipliers can be computed by inserting ii,, i,, and & from (20)
and (21) into the equalities (17). This calculation gives
112
,
(24
1
112
,
and
(23)
(24)
Nonprobabilistic Modelling 63
5. NUMERICAL ANALYSIS
The solution procedure and the quantitative behavior of beams subject to uncertain excitations
are illustrated with two numerical examples. The first example involves an uncertain distributed
forcing function and the second one uncertain initial disturbances.
Example 1. Distributed Forcing Function
We consider a simply supported beam of length L, vibrating under the transient force
p(x) G(t) = p; z H(t) f j(z) H(t),
(25)
where p is the magnitude of the deterministic force at x = L, and H(t) is the Heaviside function
given by H(t) = 0, for t < 0, and H(t) = 1, for t 1 0. The initial conditions are taken as
6(x) = 0, @t(z) = 0.
(26)
From equations (9), (lo), and (26), it follows that
a, = (-l)n+lZ, b,=&=c,=&=O.
(27)
Inserting the values of the coefficients from (27) into the solution y(z, t) in (16), we obtain
y(s,t) = g -J- [(-I)+% + &] (1 - cosSZ,t) sin 7.
n=l mflz
Using (20) and (22), the uncertainty coefficients are computed as
Gi, = F
Jzel (1- COS(~~T~TO)) sin(n?rXo)
L1/2 n4 [C,=, $ (1 - cos (k2 *Ir2 To))~ sin2 (k7rXo)] 12
(28)
where To = 77 to, 7 = (EI/m) j2 Lm2, and X0 = x0/L. Substituting zi, from (29) into (28) and
introducing the dimensionless quantities
T=vt,
El
e1 = ~312 m 772
the solution is obtained as
Y( X, T) =2 (-l)n+l -& f
fiel (1 - cos (n2 7r2 TO)) sin (nTX0)
n=l
n8 7r4
[
CF==, & (1 - cos (k 2 ?r2 To))~ sin2 (k R x0)] 12
X (1 -cos(n~To)) sin(n7rX),
(30)
(31)
where the plus sign for 5, is taken for the least favorable deflection response. To compare the
relative norms of the deterministic and uncertain loadings, we compute
IIFol12 = (6 (y) dx) I2 = pm.
(32)
Let Il.fll, = ~1 = RI(Fo(12 = RP&%
w
h
ere R is a proportionality constant. From (30), it
follows that el = R P/a.
The least favorable deflection of the beam at a given point X = X0 and time T = TO is
obtained by evaluating Y(X, T), g lven by (31), at X = X0, T = TO. Figure 1 shows the curves
of Y(O.5,To) plotted against TO, for R = 0.1 and 0.5, with P = 1.0 and N = 15. The curves
Y(X0,l.O) versus X0 for the same problem parameters are shown in Figure 2. Both figures show
that the least favorable deflection of the beam depends on the ratio R = ~[$~~,/~lFo~l,. Moreover,
the difference between the deflections of the beam under the deterministic load only (R = 0) and
under the combination of deterministic and uncertain load (R > 0) is a function of X0 and To.
64 I. S. SADEK et al.
I
- UNCERTAINIXX
DETERMINISllC CASE
0.0 0.2 0.4 0.6 0.8 1 .o
T.
Figure 1. Least favorable and deterministic deflections, plotted against time
x0 = 0.5 for Example 1
0.02
s
s
Y
0.01
I
- UNCERWNCASE
TO at
0.00 v \
0.0 0.2 0.4 0.6 0.8 1 .o
x0
Figure 2. Least favorable and deterministic deflections, plotted against Xo at
To = 1.0 for Example 1.
DElERMlNlSllC CASE
Example 2. Initial Conditions
We consider a simply supported beam of length L subject to initial conditions
y(z,O) = dc sin 7 + q(x), g(2,O) = w0 + &CC),
(33)
where do and ve are the magnitudes of the deterministic deflection and velocity at t = 0. The
forcing function P(z) G(t) is taken as zero. From (lo), it follows that
bl = do, b, = 0, n > 2; c, = - 2vo (1 - cosna) 7 n 2 1.
n7r
(34)
Noting that C& = n2 rr2 17, for the simply supported beam, we compute (Y, and pn from (15), viz.,
CYI =dc+&, on = E,,
72 > 2, P?l =
2vo (1 - cos n7r) + 2,
qn3x3 n27r27)
n > 1.
(35)
From (16) and (30), it follows that
Y(X,T) = 2 (%
n=l
co6 (n 7r2 T) + p, sin (n2 x2 T)) sin (nnX) ,
(36)
Nonprobabilistic Modelling 65
where oyn and ,f3, are given by (35). The uncertainty coefficients i, and En are computed from
(21), (23), and (24) as
6, = F
(2L)/2 .s2 cos (n2 ~~ TO) sin (n 7r X0)
(37)
7rn2
[
Cr_=, $
cos2 ( k2 r2 TO) sin2 (k r Xc)] 12
2
E,=F -
0
i/2
e3 sin (n 7r2 To) sin (n 7r X0)
L
n2
[
CL, i$
sin2 (k2 79 TO) sin2 (k * X0)] 12
(38)
To compare the deterministic and uncertain initial conditions, we choose
where Rs and Rs are constants independent of ~$6 and $0. By inserting ~2 and ~3 from (39)
into (37) and (38), respectively, &, and En can be computed. The deflection is now determined,
from (35)-(38), as
co
Y(X, T) = &I cos n2T sin nX +
x[(
do R2 cos (n x2 To) sin (n?r Xo)
n=l
712 [c;==, $
cos2 (k2 r2 To) sin2 (k A Xo)] 12
cos (n2 7r2 T)
+uo
2 (1 - cosnn)
fi Rs sin (n r2 TO) sin (n w Xo)
n3 7r3
r
n4 x2 [Cc==, $
sin (I? 7r2 TO) sin2 (k P XO)]
1,2) sin (n2n2T)] sin(n?rX)140j
where & = do/L and GO = vo/(Lq) are the dimensionless initial deflection and velocity. It is
observed that there are four possible plus and minus combinations in equation (40) for Y(X,T).
The least favorable deflection is determined by computing these four possibilities and choosing
the largest deflection for a given Xc and TO.
Figure 3 shows the curves of Y(O.5,Ta) plotted against TO, for C& = Cc = 1.0 and R2 = R3 =
R = 0 (deterministic case), Rs = RS = R = 0.1, and Rs = Rs = R = 0.5, and N = 15. It is
observed that the least favorable deflection for R = 0.5 is given by a nonsmooth curve.
- UNCERTAIN CASE
DETERMlNiSTlC CASE
-2 1 1
0.0 0.2 0.4 0.6 0.8 1 .o
TO
Figure 3. Least favorable and deterministic deflections, plotted against time To at
Xo = 0.5 for Example 2.
6. CONCLUSIONS
The problem of finding the least favorable response of a vibrating beam with incomplete in-
formation on dynamic loads is studied. The maximum deflection under dynamic loads is deter-
mined, subject to bounds on the L2 norms of the uncertain components of the forcing functions
66
I. S. SADEK et al.
and the initial conditions. The approach outlined in the present study allows the computation of
a worst-case dynamic loading without having full probabilistic information on the nature of the
uncertainty.
Numerical results are given, to illustrate the solution procedure and investigate the effect
of various problem parameters on the least favorable deflection. The difference between the
deterministic deflection and the deflection due to uncertain loading is a measure of the influence
of the uncertainty. The present approach allows the computation of the worst case for this
deviation from the deterministic case.
APPENDIX
Let A= (Zil,...,&),B = (&,...,&), and C = (El,...,&);
)\(A, B, C)112 = IAl2 + lB12 + ICI;
Sl={(A,B,C) 1 /A&1}, S2={(A,B,C) 1 IBlp5~2} 7 S3=((A,W ) ICj75e};
W={(A,B,C) 1 \Ala=~l} ,G={(AIB,c) ( IBlp=~2} ,ag={(A,B,C) 1 IC\7=~3).
LEMMA. If P = (A", B", Co) is an extreme point of the compact convex set S = S1 fl S2 n S3,
then
PO E asI n as, n as, ,
i.e., lAolo = ~1, IBlp = ~2, and lCl_, = ~3.
NOTE 1. S1 n 52 n 273 is indeed compact since d, > 0, ,& > 0, and ;Yn > 0, for 1 5 n 5 N.
NOTE 2. It is not true in general if T = TI n TZ is a compact convex set with Tl and TZ
convex, that an extreme point of T lies on aT1 n dT2, e.g., TI = {(z, y, z) ) x2 + y2 _< 1, z2 5 9},
T2={(~,y,z)(2~+y~+z~I2}.
PROOF. By contradiction. Assume P $ dS1 n a& n a&. Without loss of generality, we may
sssume
lAoI, < ~1 - 61,
lB"jp = ~2, (C"jy = ~3, 0 < 61 <&I.
Consider the set of points P = {(A, B", Co)} which lie in the sphere SJ where
S4={PIIIP-P0JI<S1}.
(1) Note that S4 c S. This follows since, clearly S4 C SZ and S4 C S3. Also, since
1~1, = IA-A' +A~/, I IA-AOJ, + )AOI, = J\P - poll + 1~~1~ < hl+el -& = Ed,
S, c &, and hence S4 c S.
(2) Consider the two points on the surface of the sphere aS4
Q+ = (A', B", Co)+ (&,O,... ,O), Q- = (A', B", C')+(-61, O,...,O),
and the line joining them
l(X) = XQ- + (1 - X) Q+, OIXIl.
Clearly every point of a(x) belongs to 54 c S. However,
pO=e l
0
2
is also an interior of a(x), i.e., P is not an extreme point of S and the lemma is proved.
I
NOTE. The lemma can be generalized, together with the proof to an arbitrary finite number of
given cylinders, provided the convex sets are described by disjoint sets of coordinates.
Nonprobabilistic Modelling
67
REFERENCES
1. Y.K. Lin, Probabilistic Theory of Structural Dynamics, McGraw-Hill, New York, NY, (1967).
2. I. Eliihakoff, Probabilistic Methods in the Theory of Structures, Wiley-Interscience, New York, NY, (1983).
3. C.Y. Yang, Random Vibration of Structures, John Wiley and Sons, New York, NY, (1986).
4. G.I. Schueller and M. Shinozuka, Stochastic Methods in Structural Dynamics, Mortinus Nijhoff Publishers,
Dordrecht, The Netherlands, (1987).
5. Y. Ben-Haim and I. Elishakoff, Convex Models of Uncertainty in Applied Mechanics, Elsevier Science
Publishers, Amsterdam, The Netherlands, (1990).
6. F.C. Schweppe, Uncertain Dynamic Systems, Prentice-Hall, Englewood Cliffs, NJ, (1973).
7. Y. Ben-Haim, The Assay of Spatially Random Material, Kluwer Academic Publishers, Dordrecht, The
Netherlands, (1985).
8. Y. Ben-Haim and I. Elishakoff, Nonprobablistic models of uncertainty in the non-linear buckling of shells with
general imperfections: Theoretical estimates of the knockdown factor, ASME Journal of Applied Mechanics
56 (2), 403-410 (1989).
9. I. Elishakoff and Y. Ben-Haim, Dynamics of a thin cylindrical shell under impact with limited deterministic
information on its initial imperfections, Journal of Structural Safety 8 (l-4), 103-112 (1990).
10. I. Elishakoff, P. Elisseeff and S.A.L. Glegg, Non-probabilistic modelling of scatter in material properties for
vibration and analysis of viscoelastic structures, AIAA Journal (to appear).
11. D. Givoli and I. Elishakoff, Stress concentration at a nearly circular hole with uncertain irregularities, ASME
Journal of Applied Mechanics 59 (2), S65-S71 (1992).
12. S. Adali, Convex and fuzzy modelling of uncertainties in the optimal design of composite structures, In
Proceedings of IUTAM Symposium on Optimal Design with Advanced Materials, pp. 18-20, August 1992,
Lungby, Denmark, Elsevier (to be published).
13. C.C. Chang and H.T.Y. Yang, Random vibration of a flexible, uncertain beam element, ASCE Journal of
Engineering Mechanics 117 (lo), 2329-2350 (1991).
14. P.J. Kelly and M.L. Weiss, Geometry and Convedty: A Study in Mathematical Methods, John Wiley, New
York, NY, (1979).

Vous aimerez peut-être aussi