Professional Documents
Culture Documents
www.elsevier.com/locate/compstruct
Abstract
In this paper a brief presentation of five reliability methods taken from literature is given. The methods are successively applied to
two examples, a beam made of isotropic material and a laminated composite plate. Loads, geometries and material properties of
both examples are defined as normal, uncorrelated stochastic variables. The aim of the paper is to compare and discuss the per-
formances of the methods. Furthermore, the utility of the so-called ‘‘Directional Cosines’’ is put in evidence, and its utility in Robust
design is suggested.
Ó 2002 Elsevier Science Ltd. All rights reserved.
Nomenclature
Table 1 considered safe when this ratio is higher than 1, i.e., the
Data for example 1 buckling load is higher than the applied one.
Stochastic variable Mean, l Standard deviation, r
9
J (mm) 1:067 10 100,000 3.3. Convergence criteria in the FORMs
E (N/mm2 ) 73,000 1000
L (mm) 5000 2
Q (N) 10 3 In all the FORMs the convergence criteria adopted is
the following:
ðkÞ ðk1Þ
the GSMM is calculated. Data for this example are gi- jbHL j jbHL j
ðkÞ
<e ð10Þ
ven in Table 1. jbHL j
i.e., the algorithm stops when the difference between the
3.2. Example 2
last bHL and the previous is smaller than a given e, which
in these calculations has been set to 105 .
A laminated composite rectangular plate, simply
supported on the four sides is loaded by compressive
distributed forces acting in its mid-plane (Fig. 1). The 3.4. Incomplete MCMs
quantity to be checked for defining the performance
function is the buckling load. The stochastic variables Along with the MCM above explained, a series of
are the material properties of the ply (E1 , E2 , G12 ), the ‘‘incomplete’’ MCMs have been launched. These in-
thicknesses and the fibre orientation angles of the plies, complete MCMs are runs during which only one (or a
the value of the loads and the plateÕs dimensions. The few) stochastic variable is sampled, while the others are
data defining this case are listed in Table 2. kept fixed at their mean values. The purpose of this was
The PoissonÕs coefficient is the only deterministic to verify the statement that the DC are indexes of the
variable, and its value is m12 ¼ 0:28. The mean value of degree of participation of the variables in determining
P x is allowed to range from 10,000 to 100,000 N/m, in the reliability level.
order to calculate pointwise the relation R ¼ RðP x Þ. The
performance function is written using the expression of
the buckling load given for the special orthotropic 4. Results and discussion
laminate [7]; though the laminates produced in this work
are not special orthotropic, they satisfy the relations Tables 3 and 4 report the results of MCMs and
D16 D26 FORMs for both the examples, and the corresponding
c¼p ffiffiffiffiffiffiffiffiffiffiffiffiffiffi < 0:2 d¼p ffiffiffiffiffiffiffiffiffiffiffiffiffiffi < 0:2 ð8Þ plots are represented in Figs. 3 and 4. Along with the
4
D311 D22 4
D322 D11
results of the FORMS the number of steps needed to
As a consequence, the buckling load (as well as the achieve convergence are also given. Figs. 5a and 6a
fundamental frequency and the displacement of the represent the DC for the two examples, calculated for
centre of the plate) can be calculated with great accuracy each of the considered displacement and loading con-
by means of the same expression used for orthotropic ditions, and Figs. 5b and 6b give the plots of the results
laminates. The PF is therefore [7]: of the incomplete MCMs. As one can observe, the
( )
p2 ðD11 ðm=aÞ4 þ 2ðD12 þ 2D66 Þðm=aÞ2 ðn=bÞ2 þ D22 ðn=bÞ4 Þ
gðs; h; Nx ; Ny Þ ¼ min 2 2
1 ð9Þ
m;n Px ðm=aÞ þ Py ðn=bÞ
The fraction represents the ratio between the buckling higher the absolute value of the directional cosine of a
load and the external load Px , i.e., Nx;cr =Px . The plate is variable, the nearer the partial MCMÕs result of that
Table 2
Data for example 2
Dimensions of the plate Mechanical properties of a lamina Loads Thickness of
a (m) b (m) E1 (N/m )2 2
E2 (N/m ) G12 (N/m ) 2
Nx (N/m) Ny (N/m) a lamina (m)
Table 3
Results for example 1
Dl (mm) MCM FORMs
FPIM HSM DCM SAM
Val. N° it. Val. N° it. Val. N° it. Val. N° it.
4 2.2528 2.2531 4 2.2531 3 2.2362 3 2.2531 5
5 4.7248 4.7288 4 4.7288 3 4.7111 4 4.7288 6
8 24.9862 24.9828 5 24.9828 3 24.979 3 24.9828 6
9 36.5750 36.5814 4 36.5814 3 36.5808 3 36.5814 6
10.01 49.6913 49.6931 5 49.6931 3 49.6931 3 49.6931 6
10.03 49.9573 49.9579 5 49.9579 3 49.9579 3 49.9579 6
12 74.2922 74.2946 5 74.2946 3 74.2946 3 74.2946 6
14 90.5783 90.5778 4 90.5776 3 90.5776 4 90.5776 6
20 99.9522 99.9512 6 99.9510 4 99.951 5 99.951 8
Table 4
Results for example 2
P x (N/m) MCM FORMs
FPIM HSM DCM SAM
Val. N° it. Val. N° it. Val. N° it. Val. N° it.
10,000 92.2833 100 15 100 6 99.896 2 100 10
20,000 99.3218 100 14 100 6 99.9996 4 100 11
30,000 99.7363 99.9999 12 99.9998 6 100 6 99.9998 11
40,000 99.5865 99.9809 13 99.975 5 99.991 5 99.9749 11
50,000 97.2109 99.2243 14 99.112 4 99.3352 5 99.1113 12
60,000 84.8663 90.1703 14 89.7035 4 90.1531 4 89.6945 14
70,000 55.355 56.5843 13 56.4686 3 56.466 4 – –
80,000 22.3649 16.9513 15 17.4802 4 17.7539 4 17.4754 9
90,000 4.9225 1.9026 18 2.1242 5 2.3529 4 2.1222 11
100,000 0.546 0 60 0.0911 5 0.1225 4 0.0909 11
variable to that of the complete MCM, for a given Dl or reliabilities that in the first example are almost coinci-
Px (dotted lines). A practical consequence of this infor- dent, and in the second are very near to one another.
mation is that the MCMs can be performed by sampling Typically, the variables with greater DC were the
only the most important variables, with a considerable external loads q and P x , in the two cases; this is also due
shortening of the computing time. All the MCMs, both to the fact that the loadÕs standard deviations (Tables 1
complete and incomplete, used 106 trials for each point. and 2) were taken much greater than those of the other
A glance to Tables 3 and 4 reveals that the quickest variables, therefore the importance of the load dis-
FORMs are DCM and HSM, because they took an persion in determining the reliability was considerable.
inferior number of iterations to reach convergence, with In example 2 the best incomplete MCM, i.e., that
160 M. Di Sciuva, D. Lomario / Composite Structures 59 (2003) 155–162
Table 5
Results of the application of the GSMM to example 1
Calc. numb. Dl (mm) lZ bHL RGSM (%)
1 4 )6.035 )2.003 2.2593
2 5 )5.035 )1.671 4.7357
3 8 )2.035 )0.675 24.9711
4 9 )1.035 )0.344 36.5604
5 10.01 )0.025 )0.0083 49.6681
6 10.03 )0.005 )0.0017 49.9329
7 12 1.965 0.652 74.2839
8 14 3.965 1.316 90.5893
9 20 9.965 3.307 99.9529
For all the points rz ¼ 3:013.
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
4 2 4 2 2 2 4 2
lL lq 2
lL lq 2
lL lq 2
lL
rD ¼ rg ¼ 2
rJ þ 2
rE þ rL þ r2q ¼ 3:01 mm ð12Þ
8lJ lE 8lJ lE 2lJ lE 8lJ lE
The ‘‘deterministic’’ displacement of the tip of the beam, Among the four FORMs tested, the faster have been
calculated by setting the four variables to the corre- the DCM and the HSM.
sponding mean values, is
lq l4L References
¼
D ¼ 10:03 mm ð13Þ
8lE lJ
[1] Verderaime V. Aerostructural safety factor criteria using determin-
istic reliability. J Spacecr Rockets 1993;30(2):244–7.
As one can see, this value does not correspond to the true
[2] Tichy M. Applied methods of structural reliability. Dordrecht:
mean value of the tip displacement lD , due to additional Kluwer Academic Publisher; 1993.
terms containing the standard deviations of some input [3] Townsend J, Meyers C, Ortega R, Peck J, Rheinfurth M,
data. Numerical results of GSMM are given in Table 5. Weinstock B. Review of the probabilistic failure analysis
162 M. Di Sciuva, D. Lomario / Composite Structures 59 (2003) 155–162
methodology and other probabilistic approaches for application [6] Hahn GJ, Shapiro SS. Statistical models in engineering. New York:
in aerospace Structural Design. NASA Technical Paper 3434, John Wiley & Sons; 1967.
November 1993. [7] Jones RM. Mechanics of composite materials, Scripta book.
[4] Sundararajan CR. Probabilistic structural mechanics handbook, Tokyo: McGraw-Hill; Washington, 1975.
Theory and industrial applications. New York: Chapman & Hall; [8] Taguchi G. System of experimental design engineering methods to
1995. optimize quality and minimize costs, UNIPUB, Kraus Interna-
[5] Kalos MH, Whitlock PA. Monte Carlo methods: vol 1st, basics. tional Publisher, New York: Unipub-Kraus; Dearborne: American
New York: Wiley; 1986. Supplier Institute, 1987.