See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/285833516

Fuzzy finite element analysis of structures with uncertainty in load and material
properties

Article in Journal of Structural Engineering (Madras) · June 2006

CITATIONS READS

6 54

2 authors, including:

Venkata Rama Rao Mallela

Vasavi College of Engineering
21 PUBLICATIONS 109 CITATIONS

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Interval FEA of thin plates View project

All content following this page was uploaded by Venkata Rama Rao Mallela on 08 December 2015.

The user has requested enhancement of the downloaded file.

Journal of Structural Engineering No. 33–13
Vol. 33, No.2, June–July 2006 pp. 129–137

Fuzzy finite element analysis of structures with uncertainty in load and

material properties

M.V. Rama Rao* and R. Ramesh Reddy**

In this paper, the cumulative effect of load and material uncertainties is explored using fuzzy-finite element analysis.
Example problems are taken up, calculations are performed and results are discussed. The need for introducing
uncertainty in load and material property (Young’s modulus) in the analysis of solid mechanics problems is
underlined. Uncertainty is introduced in load and material properties by choosing triangular membership functions
with equal spread about the crisp value. Uncertain response of the structures to the fuzzy interval values of load and
Young’s modulus is obtained in terms of fuzzy interval displacements and the results are post-processed to obtain
fuzzy interval values of bending moment, shear force and axial forces. Sensitivity analysis is performed to evaluate
the relative sensitivity of displacements and forces to multiple uncertainties. The result demonstrates the need for
introducing simultaneous uncertainties in the analysis of solid mechanics problems.

Till date, classical non-fuzzy finite element analysis is the obtained by stochastic and interval versions of finite ele-
most popular mathematical tool to model the static and dy- ments.
namic response of structures to a high degree of computa- Rao and Sawyer6, Rao and Berke7, Rao and Chen Li8
tional precision. Classical finite element analysis makes use have developed different versions of interval-based finite
of floating point numbers to describe the physical quantities elements to account for uncertainties in engineering prob-
and the output generated is crisp and non-fuzzy in nature. lems. But these works were primarily developed to suit nar-
Classical finite element method, despite its advantages, is row intervals and approximate numerical results. A new
not suited to handle uncertainty in the design variables of the search-based algorithm to solve a system of linear interval
structural system i.e. displacements, loads, bending moments equations was developed by Rao and Chen Li8. The algo-
and shear forces. A realistic or natural way of representing rithm performs search operations with an accelerated step
uncertainty in engineering problems might be to consider the size in order to locate the optimal setting of the hull of the
values of unknown variables, which are defined within in- solution.
tervals that possess known bounds1. Muhanna and Mullen2 developed a fuzzy-based matrix
Extensive research helped understanding the behaviour method of structural analysis for the calculation of extreme
of imprecisely defined systems using fuzzy logic. Uncer- values of structural response for all possible loading combi-
tainty in the input data as well as the behaviour of systems nations.
was explained by introducing vagueness in qualitative terms Muhanna and Mullen3 dealt with the formulation by
in the definition of design variables of the problem. Use of fuzzy-finite elements for solid mechanics problems. The
fuzzy logic to understand and model the behaviour of struc- fuzzy approach to treating uncertainties in continuum me-
tural systems is of recent origin2. Concerted efforts were chanics is applied to individual instances of load, geometric
made since then to handle uncertainty in engineering prob- and material uncertainties thus obtaining sharp inclusion of
lems realistically by introducing fuzziness in material and fuzzy solution in comparison with the exact solution.
geometric properties of structural systems and also service Muhanna and Mullen1 handled uncertainty in mechanics
loads to which the structures are exposed to during their de- problems using an interval-based approach. Element-by-
sign life period3. element (EBE) technique was employed to obtain a very
An interval based finite element method was developed sharp enclosure for the fuzzy solution by eliminating the
by Koyluoglu et al4 to deal with pattern loading and struc- sources of overestimation.
tural uncertainty. In addition, linear programming and trian- A practical approach for analyzing the structures with
gle inequalities were used for the solution of simultaneous fuzzy parameters was developed by Akpan et al9. The uncer-
linear interval equations. tainties in material, loading and structural properties were
Koyluoglu et al5 demonstrated the problem of shear represented by convex normal fuzzy sets. Vertex solution
frames with uncertain properties and compared the results methodology that was based on !-cut representation was

* Associate Professor in Civil Engineering, Vasavi College of Engineering, Ibrahimbagh, Hyderabad-31, ** Professor in Civil Engineering, University
College of Engineering (Autonomous), Osmania University, Hyderabad-7

JOURNAL OF STRUCTURAL ENGINEERING VOL. 33, NO.2, JUNE–JULY 2006 129

used for the fuzzy analysis. Response surface methodology Using Rayleigh- Ritz approach and invoking the stationarity
and combinatorial optimization were used to determine the of !* leads to
binary combinations of the fuzzy variables that resulted in
[K] {U} + [C#]T{!} = {P} (4)
fuzzy responses at an !- cut level. These binary combina-
tions of the fuzzy variables were then used to obtain extreme Expressing [K] = [D] [S#] , multiplying Eq.3 on both sides by
responses to the finite element model. [D] [C#]T and adding to Eq.(4) leads to
A cable-stayed bridge was modelled and analysed using
[D] [S#]{U} +[D] [C#] T [C#]{U} ={P} -[C#]T{!} (5)
fuzzy-finite element analysis by Rama Rao and Ramesh
Reddy10. The Centre Canal Bridge at Obourg, Belgium was Taking
chosen as an example problem. Load uncertainty was intro- [R#] = [S#]+[Q] (6)
duced by a triangular membership function. Static response
of the bridge to the fuzzy loading is obtained in terms of Eq. (5) can be simplified as
fuzzy displacements. The results demonstrate the effective- [D][R#]{U} = {P}-[C#]T{!} (7)
ness of introducing fuzziness in the analysis of Cable-stayed
bridges. Equation (7) can be recast as
{U} =[R#]-1[D] -1 {{P}-[C#] T{!}} (8a)
SCOPE OF THE PRESENT STUDY
and
In the present study, the concept of handling multiple uncer- {U} =[R#]-1[M]{#} (8b)
tainties in load and material properties using fuzzy-finite
1
element methodology is applied to the case of analysis of Equations 8(a) and 8(b) were solved by Muhanna by ap-
simple structures. The current work makes a study of the proximating the vector of internal forces {!} by its mid-
effect of fuzziness in service loads and material properties of point (crisp) vector {!c} to solve the equations when the load
linear elastic structures in terms of displacement, rotation, vector {P} is crisp and Young’s modulus E is uncertain.
shear force and bending moment, by solving example prob- Attempts by the authors to apply the same approach in
lems. The sensitivity of structural response to the simultane- the presence of load and material uncertainties could not
ous variation of load and material uncertainties is explored. yield acceptable results. Thus, a modified Muhanna’s
A brief description of the properties of fuzzy interval method is introduced in order to consider the simultaneous
numbers and the concept of !-cut are given in the works of effect of load and material uncertainties.
several authors2, 3.
Formulation of linear interval equations- present study
ELEMENT-BY-ELEMENT MODEL FOR HANDLING Using extension principle, the set of Eqs. (3) and (4) can be
UNCERTAINTIES rewritten at a given level ($- cut values) of material uncer-
tainty ! and load uncertainty % (0# $, %#1) as
The element-by-element (EBE) model1 has the following
characteristics: [C#]{U}$%={0} (9)
a) The set of elements of the model is de-assembled. Each [K]${U}$% + [C#]T{!}$% = {P}% (10)
element has its own set of nodes so that a given node be-
longs to only one element. where
b) A set of additional constraints (in addition to the con- [K]$ = interval stiffness matrix at a specified level of
straints that are imposed at the boundary) is introduced to material uncertainty $
force unknowns associated with coincident nodes to have {P}% = interval load vector at a specified level of material
identical values. uncertainty %
{!}$% = interval internal force vector at a specified level
Formulation of linear interval equations- Muhanna’s of material uncertainty ! and load uncertainty %
approach respectively.
{U}$% = interval displacement vector at a specified level of
The variational formulation for an interval case of a discrete
material uncertainty ! and load uncertainty % re-
element-by-element structural model is given as
spectively.
!* = ½ {U} T [K]{U}–{U} T {P} +{!} T {[C"]{U}-{V}} (1) It may be here noted that [K]$ is a function of ! alone,
{P}% is a function of % alone, while {!}$% and {U}$% are func-
with the conditions tions are both $ and %.
""/"{Ui} = {0} for all i (2) The interval stiffness matrix [K]$ is a symmetric indefi-
where !, [K], [C"], {U} and {P} are potential energy, stiff- nite square matrix owing to the de-assembled state of ele-
ness matrix, constraint matrix, displacement vector and load ments in the EBE model1. All the elements of the stiffness
vector respectively. matrix have the Young’s modulus as the multiplier. There-
Here, {!}, the vector of Lagrange multipliers, represents fore, stiffness uncertainty for an element is equal to the ma-
terial uncertainty $ (of the Young’s modulus).
the internal forces.
Therefore, If the unknowns of each element are numbered consecu-
tively, then the stiffness coefficients of [K]! will be clus-
[C#]{U}={0} (3) tered in the form of square diagonal blocks each of size

130 JOURNAL OF STRUCTURAL ENGINEERING VOL. 33, NO.2, JUNE–JULY 2006

equal to the number of unknowns per element. The size of {!}$ % $ [%l, %u] {!}cc (19)
[K]$ is n × n where n is the product of number of degrees of
Substituting the approximate value of {!}$% in Eq. (13), the
freedom per each element and total number of elements in
resulting matrix interval equation is
the structure. The vector of Lagrange multipliers {!}$% is an
interval vector. It represents the vector of internal forces that [D]$[R"]{U}$% = {{P}% -[%l, %u] [C#]T{!}cc} (20)
are exposed because elements in the EBE model are kept
separate till the end. {!}$% depends on applied loads and im- The same equation can be expressed as
posed constraints on the boundary and also on structural {U}$% = [R"]-1[M]{#}$ (21)
stiffness (in the case of statically indeterminate structures).
Equation (10) can be introduced as where [M] is a matrix of size n × m where m is the number
of elements and {#}$ is an interval vector of size m × 1 con-
[D]$[S&]{U}$% = {P}% -[C#]T{!}$% (11) taining interval material properties of m elements taken from
the diagonal entries of [D]$.
where [D]$ is an diagonal matrix of size n × n containing
In order to solve Eq. (21) it is necessary to compute the
interval terms corresponding to material uncertainty at level
! and [S#] is a deterministic singular matrix of size n × n. value of the mid-point internal force vector {!}cc. This is
If Eq. (9) is multiplied on either side by [D]![C#]T and the done by considering the assembled Finite Element model of
result is added to Eq.(11), the structure with deterministic value of structural stiffness
subjected to fuzzy interval loading at a level of load uncer-
[D]$[[S#]+[C#]T[C#]] {U}$% = {{P}% -[C#]T{!}$%} (12) tainty %.
The following matrix equation is solved in order to ob-
[D]$[R"]{U}$% = {{P}% -[C#]T{!}$%} (13) tain the fuzzy displacement vector {Uc%}3, 10.
[K"]{Uc%} = {P%} (22)
{U}$% = [R"]-1 [D]$-1 {{P}% -[C#]T{!}$%} (14)
where [K"] is the fixed-point (crisp) stiffness matrix.
If the interval vector {!}$% can be determined exactly, then
the solution of Eq. (14) represents the exact hull of interval
system of equations SOLUTION OF LINEAR INVERVAL MATRIX
EQUATIONS
[K]{U} = {P} (15)
Making use of the inclusion theory developed by several
This is true in the case of statically determinate structures,
interval researchers11, the interval matrix equations given by
where the vector of internal forces depends only on applied
Eq. (21) are tackled.
loads and boundary conditions and does not depend on Considering [A] =[D$][R"], {x} = {U}$% and {b} = {P}%-
structural stiffness. Accordingly, the internal force vector is [%l,%u][C"]T {!}cc, Eq. (21) can be recast as
independent of material uncertainty ! in this case. This leads
to the condition {!}$% = {!}% in the case of statically deter- [A] {x}={b} (23)
minate structures, resulting in an exact solution to the prob-
Equation (23) represents a set of interval matrix equations.
lem [K]{U} = {P}.
Solution of these equations using the inclusion theory results
However, in the case of statically indeterminate struc-
in an optimal enclosure AHb known as the hull of the solu-
tures, {!}$% depends on structural stiffness (with an uncer-
tion (Fig. 1) . Making use of Jansson’s algorithm1, 11, the
tainty level ! of material property) in addition to loads (with
lower bound vector {x} and the upper bound vector {y} en-
an uncertainty level %) and boundary conditions (crisp).
closing hull of the solution are obtained. The solution vector
Therefore evaluation of the exact value of {!}$% is uncertain.
{U}$% is taken as the average of {x} and {y}.
Effect of multiple uncertainties ! and " on structural
response
At a given level of uncertainty $, the interval load vector
{P}% and vector of internal forces can be expressed as
{P}% = [%l, %u] {P}c (16)
and
FIG. 1 BOUNDS ON HULL SOLUTION
{!}$ % = [%l, %u] {!}$c (17)
where %l and %u are the lower and upper bounds of the nor-
malised interval at a given %-cut level of load uncertainty. CASE STUDIES
Here, {!}$c is the vector of internal forces for crisp load
To illustrate the applicability of the present approach,
(%=1) for a material uncertainty $. The exact value of {!}$c
three example problems are considered viz. a fixed beam
is approximated with {!}cc
(adapted from the work of Akpan et al 9), a plane truss and a
Thus,
plane frame given by Fig. 2, Fig. 3 and Fig. 4 respectively.
{!}cc $ {!}$c (18) The material and geometric properties of these structures are
given in Tables 1 - 3 respectively. Linear elastic analysis is
On substituting Eq. (18) in Eq. (17) results in performed in all the cases. Uncertain values of Young’s

JOURNAL OF STRUCTURAL ENGINEERING VOL. 33, NO.2, JUNE–JULY 2006 131

modulus E$ and load P% are given as E$ = Ec [1-', 1+'] and TABLE 2
P% = Pc [1-(, 1+(] where Ec and Pc are the nominal values MATERIAL AND GEOMETRIC PROPERTIES OF PLANE
of Young’s modulus and load respectively. Here ' = TRUSS
($max/100)*(1-$) and ( = (%max/100)*(1-%). The cumulative Area of Cross section 0.01 m2 Span of each 10.0m
effect of multiple uncertainties is investigated in all the of each member bay
cases. All the loads on each structure are uncertain and pos- Young’s Modulus (E) 200 GPa Material Uncer- ±5% about
sess a common level of load uncertainty %. For example, the tainty $max mean value
uncertain loads acting on the plane frame (Fig. 4) of E
Load Uncertainty %max ±100% about mean value of load

TABLE 3
MATERIAL AND GEOMETRIC PROPERTIES OF PLANE
FRAME
Columns Cross sec- 6.0e-2 m2 Moment of 2.0e-4 m4
tional Area Inertia
Beam Cross sec- 7.5e-2 m2 Moment of 3.91e-2 m4
tional Area Inertia
FIG. 2 FIXED BEAM Material Uncer- ±5% Load Un- ±100% Young’s 200
tainty certainty Modulus GPa

at % = 0.8 are [20,30] kN/m on element 1, [80,120] kN at

node 2 and [40,60] kNm at node 3 in the specified direc-
tions.

Membership Functions for material and load

uncertainties
Membership functions of material and load uncertainties
adopted for the fixed beam are indicated in Fig. 5 and 6. The
membership functions of material and load uncertainties
adopted for the case of plane truss and plane frame are given
FIG. 3 PLANE TRUSS in Fig. 7 and 8 respectively. The membership functions for
material uncertainty indicate the normalised values of uncer-
tain Young’s modulus, taking the crisp values as 1.0. Fur-
ther, the membership function for load uncertainty indicates
the normalized values of uncertain load, taking the crisp
values as 1.0.

1 1.000
Memebership Value

0.75 0.995 1.005

(alpha)

0.5 0.990 1.010

0.25 0.985 1.015

FIG. 4 PLANE FRAME 0 0.980 1.020

0.980 0.985 0.990 0.995 1.000 1.005 1.010 1.015 1.020
Design Quanity
TABLE 1
FIG. 5 FIXED BEAM-MEMBERSHIP FUNCTION FOR
MATERIAL AND GEOMETRIC PROPERTIES OF FIXED
MATERIAL UNCERTAINTY
BEAM
Span 10.16 metres (400 inches)
Effect of simultaneous variation of ! and " on displace-
Area of Cross section 0.07165 m2 (111.06 inch2)
ments and rotations
Moment of Inertia 1.2486×10-3 m4 (3000 inch4)
Uniformly distributed load 70.076 kN/m (400lb/inch) Table 4 shows a comparison of the results obtained for the
Location of load Entire span fixed beam using the present approach with the results ob-
Young’s Modulus (E) 206.9 GPa (30×10-6 psi)
tained by Akpan et al9 using the response surface methodol-
ogy. Figure 9 represents the membership function for mid-
Material Uncertainty of E ($max) ±2% about mean value of E
span displacement (mm) for three cases of uncertainty. It is
Load Uncertainty (%max) ±10% about mean value of observed that the present results compare very well with the
load results of Akpan et al9 for all the three identical cases.

132 JOURNAL OF STRUCTURAL ENGINEERING VOL. 33, NO.2, JUNE–JULY 2006

 1 1 tion of vertical displacement at node 2 of the plane truss.
Table 7 represents the simultaneous variation of rotation at
Memebership Value(beta)
0.75 0.975 1.025 node 3 of the plane frame. It is observed that the width of the
interval increases along and across each table.
0.5 0.95 1.05 Figure 10 depicts the variation of mid-span displacement
of the fixed beam at various levels of material uncertainty $
0.25 0.925 1.075
while load is uncertain with ±2.5% variation about mean
0 0.9 1.1
value. Figure 11 depicts the variation of horizontal dis-
0.9 0.925 0.95 0.975 1 1.025 1.05 1.075 1.1
placement of the plane truss at node 4 at various levels of !
Design Quanity while load is uncertain with ±20% variation about mean
value. Figure 12 depicts the variation of vertical displace-
FIG. 6 FIXED BEAM - MEMBERSHIP FUNCTION FOR LOAD ment of the plane truss at node 2 at various levels of ! while
UNCERTAINTY load is crisp. Figures 13 and 14 represent the variation of
vertical displacement at node 2 and rotation at node 3 of
1 1 plane frame at various levels of % while E is uncertain with
±2% variation about mean value.
Membership Value (alpha)

0.8 0.99 1.01

1.00 -7.715E-03 -7.339E-03
0.6 0.98 1.02
0.80
0.4 0.97 1.03 -7.754E-03 -7.302E-03
0.60

alpha
0.2 0.96 1.04
-7.793E-03 -7.266E-03
0.40
0 0.95 1.05
0.95 0.97 0.99 1.01 1.03 1.05 -7.833E-03 -7.230E-03
0.20
Normalised value of Young's Modulus
0.00 -7.873E-03 -7.195E-03
FIG. 7 MEMBERSHIP FUNCTION FOR MATERIAL -7.9E-03 -7.8E-03 -7.7E-03 -7.6E-03 -7.5E-03 -7.4E-03 -7.3E-03 -7.2E-03 -7.1E-03
UNCERTAINTY Mid-span displacement (mm)

FIG. 10 FIXED BEAM - VARIATION OF MID-SPAN

1 1 DISPLACEMENT AT BETA=0.75
Membership Value

0.8 0.8 1.2

1 3.161E-05 4.741E-05
0.6 0.6 1.4
0.8 3.076E-05 4.870E-05
0.4 0.4 1.6
0.6 2.991E-05 4.999E-05
alpha

0.2 0.2 1.8

0.4 2.907E-05 5.130E-05
0 0 2
0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8 2.0 0.2 2.824E-05 5.262E-05
Normalised Load
0 2.741E-05 5.395E-05
FIG. 8 MEMBERSHIP FUNCTION FOR LOAD UNCERTAINTY 2.7E-05 3.2E-05 3.7E-05 4.2E-05 4.7E-05 5.2E-05
Horizontal displacement (m)
Fi 11 Pl T V i ti f H i t l di l t f d 4 t

-7.527 FIG. 11 PLANE TRUSS-VARIATION OF HORIZONTAL

Degree of belief (alpha or beta)

1.00 DISPLACEMENT OF NODE 4 AT BETA = 0.8

E and Load
-7.754 -7.302 Load only
0.75 -7.565 -7.490
E only
-7.715 -7.339
-7.984 -7.080 1 -1.993E-04
0.50 -7.603 -7.453
-7.904 -7.151
0.8 -2.013E-04 -1.974E-04
-8.215 -6.859
0.25 -7.642 -7.416
-8.092 -6.963 0.6 -2.034E-04 -1.954E-04
alpha

-8.280 -6.774
-6.641
0.00 -7.681 -7.379
0.4 -2.055E-04 -1.935E-04
-8.450 -8.200 -7.950 -7.700 -7.450 -7.200 -6.950 -6.700
Displacement (mm)
0.2 -2.076E-04 -1.917E-04

FIG. 9 FIXED BEAM - MID-SPAN DISPLACEMENT UNDER 0 -2.098E-04 -1.898E-04

FUZZY-STATIC LOAD -2.100E-04 -2.060E-04 -2.020E-04 -1.980E-04 -1.940E-04 -1.900E-04
Vertical displacement(m)

Table 5 represents the simultaneous variation of mid- FIG. 12 PLANE TRUSS-VARIATION OF VERTICAL
span displacement for the fixed beam for various combina- DISPLACEMENT OF NODE 2 AT BETA=1.0
tions of ! and $. Table 6 represents the simultaneous varia-

JOURNAL OF STRUCTURAL ENGINEERING VOL. 33, NO.2, JUNE–JULY 2006 133

 TABLE 4
FIXED BEAM –MID SPAN DISPLACEMENT - COMPARISON OF PRESENT RESULTS WITH RESPONSE SURFACE
APPROACH AT DEGREE OF BELIEF (! OR ") = 0.0
Akpan (mm) Present (mm)
Uncertainty Akpan (inch)
Interval Normalised interval Interval Normalised interval
Load and E [-0.332, -0.2615] [-8.433,-6.642] [-1.119, -0.881] [-8.450, -6.641] [-1.120,0.880]
Load only [-0.326,0.267] [-8.280,-6.782] [-1.099,-0.901] [-8.280, -6.774] [-1.100,-0.900]
E only [-0.303, -0.290] [-7.671,-7.366] [-1.02,0.980] [-7.681, -7.379] [-1.02,0.9799]

TABLE 5
FIXED BEAM- SIMULTANEOUS VARIATION OF MID-SPAN DISPLACEMENT (×10-3 METRES)
wrt ! AND "
%%
1.0 0.75 0.50 0.25 0.0
$&
1.0 [-7.527,-7.527] [-7.715,-7.339] [-7.904,-7.151] [-8.092,-6.963] [-8.280,-6.774]
0.75 [-7.565,-7.490] [-7.754,-7.302] [-7.943,-7.115] [-8.132,-6.928] [-8.322,-6.741]
0.50 [-7.603,-7.453] [-7.793,-7.266] [-7.984,-7.080] [-8.174,-6.893] [-8.364,-6.707]
0.25 [-7.642,-7.416] [-7.833,-7.230] [-8.024,-7.045] [-8.215,-6.859] [-8.406,-6.674]
0.0 [-7.681,-7.379] [-7.873,-7.195] [-8.065,-7.010] [-8.257,-6.826] [-8.450,-6.641]

TABLE 6
PLANE TRUSS-SIMULTANEOUS VARIATION OF VERTICAL DISPLACEMENT OF NODE 2
(×10-5 METRES) wrt ! AND "
%%
1.0 0.8 0.6 0.4 0.2 0.0
$&
1.0 [-19.93,-19.93] [-23.92,-15.95] [-27.91-11.96] [-31.89,-7.97] [-35.88,-3.99] [-39.87,0.00]
0.8 [-20.13,-19.74] [-24.16,-15.79] [-28.19,-11.84] [-32.22,-7.89] [-36.25,-3.94] [-40.28,0.01]
0.6 [-20.34,-19.54] [-24.41,-15.63] [-28.48,-11.72] [-32.55,-7.81] [-36.62,-3.89] [-40.69,0.02]
0.4 [-20.55,-19.35] [-24.66,-15.48] [-28.78,-11.60] [-32.89,-7.73] [-37.01,-3.85] [-41.12,0.02]
0.2 [-20.76,-19.17] [-24.92,-15.33] [-29.08,-11.49] [-33.24-7.65] [-37.39,-3.81] [-41.55,0.02]
0.0 [-20.98,-18.98] [-25.18,-15.18] [-29.39,-11.38] [-33.59-7.58] [-37.79,-3.78] [-41.99,0.02]

TABLE 7
PLANE FRAME-SIMULTANEOUS VARIATION OF ROTATION AT NODE 3(×10-3 RADIANS) wrt ! AND "
%%
1.0 0.8 0.6 0.4 0.2 0.0
$&
1.0 [0.653,0.653] [0.369,0.938] [0.084,1.223] [-0.201,1.508] [-0.486,1.793] [-0.771,2.078]
0.8 [0.625,0.682] [0.343,0.969] [0.060,1.256] [-0.222,1.544] [-0.505,1.831] [-0.788,2.119]
0.6 [0.598,0.710] [0.317,0.999] [0.037,1.290] [-0.244,1.580] [-0.524,1.870] [-0.804,2.160]
0.4 [0.570,0.738] [0.292,1.031] [0.013,1.324] [-0.265,1.616] [-0.543,1.909] [-0.821,2.201]
0.2 [0.542,0.767] [0.266,1.062] [-0.010,1.358] [-0.286,1.653] [-0.562,1.948] [-0.838,2.243]
0.0 [0.514,0.796] [0.241,1.094] [-0.033,1.392] [-0.307,1.690] [-0.581,1.988] [-0.855,2.286]

1 1.838 1.987 1 0.598 0.710

0.8 1.185 2.657 0.8 0.317 0.999

0.6 0.533 3.327 0.6 0.037 1.290

beta
beta

0.4 -0.119 3.996 0.4 -0.244 1.580

0.2 -0.772 4.666 0.2 -0.524 1.870

0 -1.424 5.336 0 -0.804 2.160

-1.5 -0.5 0.5 1.5 2.5 3.5 4.5 5.5 -1.0 -0.5 0.0 0.5 1.0 1.5 2.0 2.5
rotation(X0.001 radians)
Vertical Displacement(X0.01m m)

FIG. 14 PLANE FRAME-VARIATION OF ROTATION AT

FIG. 13 PLANE FRAME-VARIATION OF VERTICAL
NODE 3 AT ALPHA = 0.6
DISPLACEMENT AT NODE 2 AT ALPHA = 0.6

134 JOURNAL OF STRUCTURAL ENGINEERING VOL. 33, NO.2, JUNE–JULY 2006

 TABLE 8
PLANE TRUSS-SIMULTANEOUS VARIATION OF AXIAL FORCE (kN)
IN MEMBER 4 wrt ! AND "
%%
1.0 0.8 0.6 0.4 0.2 0.0
$&
1.0 [-7.902,-7.902] [-9.483,-6.322] [-11.063,-4.741] [-12.64,-3.161] [-14.22,-1.580] [-15.805,0.0]
0.8 [-8.124,-7.687] [-9.748,-6.150] [-11.373,-4.612] [-12.99,-3.075] [-14.62,-1.537] [-16.247,0.0]
0.6 [-8.351,-7.477] [-10.02,-5.982] [-11.69,-4.486] [-13.36,-2.991] [-15.03,-1.495] [-16.702,0.0]
0.4 [-8.585,-7.273] [-10.30,-5.818] [-12.019,-4.364] [-13.74,-2.909] [-15.45,-1.455] [-17.170,0.0]
0.2 [-8.826,-7.074] [-10.59,-5.659] [-12.36,-4.244] [-14.12,-2.830] [-15.89,-1.415] [-17.652,0.0]
0.0 [-9.074,-6.880] [-10.89,-5.504] [-12.70,-4.128] [-14.52,-2.752] [-16.33,-1.376] [-18.148,0.0]

TABLE 9
PLANE FRAME-SIMULTANEOUS VARIATION OF BENDING MOMENT (kNm) AT NODE 2 wrt ! AND "
%% 1.0 0.8 0.6 0.4 0.2 0.0
$&
1.0 [-81.33,-81.33] [-106.34,-56.31] [-131.36,-31.30] [-156.37,-6.29] [-181.4,18.73] [-206.4,43.74]
0.8 [-77.97,-84.74] [-110.14,-53.33] [-135.54,-28.68] [-160.94,-4.04] [-186.3,20.61] [-211.7,45.26]
0.6 [-74.67,-88.20] [-114.01,-50.37] [-139.81,-26.08] [-165.62,-1.78] [-191.4,22.52] [-217.2,46.81]
0.4 [-71.41,-91.73] [-117.97,-47.45] [-144.19,-23.48] [-170.40,0.48] [-196.7,24.45] [-222.9,48.41]
0.2 [-68.20,-95.33] [-122.00,-44.55] [-148.68,-20.90] [-175.40,2.75] [-202.0,26.40] [-228.7,50.05]
0.0 [-65.02,-98.99] [-126.13,-41.67] [-153.28,-18.32] [-180.40,5.04] [-207.6,28.39] [-234.7,51.74]

Effect of simultaneous variation of ! and " on shear 1 8.834

forces and bending moments
0.8 7.067 10.601
Table 8 represents the simultaneous variation of axial force
in member 4 in plane truss. Similarly Table 9 represents the 0.6 5.301 12.368
beta

simultaneous variation of bending moment at node 2 of

0.4 3.534 14.135
plane frame. Here also, it may be observed that the width of
the interval increases along and across the tables. 0.2 1.767 15.902
Figure 15 depicts the variation of axial force in member
17.669
4 of the plane truss at various levels of material uncertainty 0 0

! when load is uncertain (with ±20% variation about mean 0.00 2.00 4.00 6.00 8.00 10.00 12.00 14.00 16.00 18.00
Axial Force (kN)
value). Figure 16 depicts the variation of axial force in
member 10 of the plane truss at various levels of load uncer- FIG. 16 PLANE TRUSS-VARIATION OF AXIAL FORCE IN
tainty % when E is crisp. Figure 17 depicts the variation of MEMBER 10 AT ALPHA = 1.0
bending moment at node 2 of the plane frame at various lev-
els of material uncertainty ! when load is crisp. Figure 18
1 81.33
depicts the variation of axial force in member 1 of the plane
frame at various levels of load uncertainty % when E is un- 0.8 77.97 84.74
certain with ±2% variation about its mean value.
0.6 74.67 88.2
alpha

1 -9.483 -6.322 0.4 71.41 91.73

0.8 -9.748 -6.15 0.2 68.2 95.33

0.6 -10.02 -5.982 0 65.02 98.99

alpha

65.0 70.0 75.0 80.0 85.0 90.0 95.0 100.0

Bending Moment (kNm)
0.4 -10.3 -5.818
FIG. 17 PLANE FRAME-VARIATION OF BENDING
0.2 -10.59 -5.659 MOVEMENT AT NODE 2 AT BETA = 1.0

0 -10.89 -5.504
-11.00 -10.00 -9.00 -8.00 -7.00 -6.00 -5.00 In all the tables, width of interval of structural response
Axial Force (kN)
with reference to material uncertainty is found to be less
compared to the variation with reference to load uncertainty.
FIG. 15 PLANE TRUSS-VARIATION OF AXIAL FORCE IN
MEMBER 4 AT BETA = 0.8 This is because less uncertainty is associated with Young's
modulus compared to load.

JOURNAL OF STRUCTURAL ENGINEERING VOL. 33, NO.2, JUNE–JULY 2006 135

It can be observed in all the cases that the variation of re-
sponse is triangular in the presence of a single uncertainty
and is trapezoidal in the presence of multiple uncertainties.

1 53.2 61.6

0.8 33.9 81.9

0.6 14.5 102.3

beta

0.4 -4.8 122.6

0.2 -24.2 142.9

0 -43.5 163.2
-50.0 -25.0 0.0 25.0 50.0 75.0 100.0 125.0 150.0 175.0
Axial Force (kN)

FIG. 18 PLANE FRAME-VARIATION OF AXIAL FORCE IN FIG. 19 PLANE FRAME-SENSITIVITY ANALYSIS AT NODE
MEMBER 1 AT ALPHA = 0.6 2 AT BETA=1.0

Sensitivity analysis
Sensitivity is defined as the relative percentage variation
of structural response quantity (such as displacement, force
or moment) to a prescribed percentage variation of load or
material properties about the respective mean values. Sensi-
tivity Analysis is performed to evaluate the relative sensitiv-
ity of uncertain structural response to changes in the levels
of uncertainty of load and material property. Sensitivity
Analysis is performed for the plane frame in the present
study (Table 10 and 11). Percentage variation about the
mean value is computed after normalisation of the intervals.
An interval [a, b] can be normalised by dividing its lower
and upper bounds a and b by the mid-point µ = 0.5*(a+b).
Thus a Young’s modulus [199,201] GPa can be normalised
as [0.995,1.005], the variation about the mean being ±0.5%.
In general, a normalised interval [1-)1, 1+)2] indicates that
lower bound and upper bound variations of the given inter-
FIG. 20 PLANE FRAME - SENSITIVITY ANALYSIS AT NODE
val about its nominal (mean) value are )1 and )2 respectively. 2 AT ALPHA = 1.0

TABLE 10
VARIATION OF FORCES AND MOMENTS AT NODE 2 AT " = 1.0 (NORMALISED VALUES)
$ Young’s Modulus (E) Axial Force in Member 1 Shear Force Bending Moment
1.0 [1.00,1.00] [-1.0000,-1.0000] [1.0000,1.0000] [-1.0000,-1.0000]
0.8 [0.99,1.01] [-1.0370,-0.9637] [0.9793,1.0212] [-1.0369,-0.9636]
0.6 [0.98,1.02] [-1.0746,-0.9279] [0.9590,1.0430] [-1.0742,-0.9277]
0.4 [0.97,1.03] [-1.1129,-0.8928] [0.9392,1.0652] [-1.1122,-0.8921]
0.2 [0.96,1.04] [-1.1520,-0.8581] [0.9198,1.0881] [-1.1507,-0.8568]
0.0 [0.95,1.05] [-1.1919,-0.8238] [0.9008,1.1115] [-1.1899,-0.8219]

TABLE 11
VARIATION OF FORCES AND MOMENTS AT NODE 2 AT ! = 1.0 (NORMALISED VALUES)
% Load Axial Force in Member 1 Shear Force Bending Moment
1.0 [1.00,1.00] [-1.0000,-1.0000] [1.0000,1.0000] [-1.0000,-1.0000]
0.8 [0.80,1.20] [-1.0370,-0.9637] [0.9793,1.0212] [-1.0369,-0.9636]
0.6 [0.60,1.40] [-1.0746,-0.9279] [0.9590,1.0430] [-1.0742,-0.9277]
0.4 [0.40,1.60] [-1.1129,-0.8928] [0.9392,1.0652] [-1.1122,-0.8921]
0.2 [0.20,1.80] [-1.1520,-0.8581] [0.9198,1.0881] [-1.1507,-0.8568]
0.0 [0.00,2.00] [-1.1919,-0.8238] [0.9008,1.1115] [-1.1899,-0.8219]

136 JOURNAL OF STRUCTURAL ENGINEERING VOL. 33, NO.2, JUNE–JULY 2006

 Figure 19 represents relative percentage variation of %max Maximum Load uncertainty
structural response quantities viz. axial force, bending mo- "* Potential Energy Functional
ment and shear force at node 2 of the plane frame for a cor- {!} Vector of internal forces
responding percentage variation of Young’s modulus (E)
when the loads are crisp (%=1.0). The slopes of these plots REFERENCES
are 3.828, 3.562 and 2.23 respectively. Figure 20 represents
relative percentage variation of structural response quantities 1. Muhanna, R.L. and Mullen, R.L., “Uncertainty in Me-
viz. axial force, bending moment and shear force at node 2 chanics Problems-Interval-Based-Approach”, J.
of the plane frame for a corresponding percentage variation Engg. Mech., Vol. 127, No. 6, 2001, pp 557-566.
of load when the E is crisp ($=1.0). The slopes of these plots
2. Mullen, R.L. and Muhanna, R.L., “Bounds of Struc-
are 1.727, 1.538, 1.000 respectively. Thus it is observed that
sensitivity of response is more in the case of material uncer- tural Response for All Possible Loading Combina-
tainty compared to load uncertainty. It is also observed that tions”, J. of Struct. Engg., ASCE, Vol. 125, No. 1,
axial force has the greatest sensitivity and shear force has the 1999, pp 98-106
lowest sensitivity in both cases. 3. Muhanna, R.L. and Mullen, R.L., “Formulation of
Fuzzy Finite-Element Methods for Solid Mechanics
CONCLUSIONS Problems”, J. Comp. Aided Civil and Infrastructure
Engg., 14, 1999, pp. 107-117.
In this paper, the interval-based approach proposed by Mu- 4. Koyluoglu, H.U., Cakmak, A.S. and Nielsen, S.R.K.,
hanna and Mullen has been modified to take into account the
“Interval algebra to deal with pattern loading and
multiple uncertainties in load and material properties simul-
structural uncertainty”, J. Engg. Mech., ASCE,
taneously. The proposed method is illustrated by applying it
121(11), 1995, pp. 1149-1157.
to three simple problems viz. fixed beam, plane truss and
plane frame. A new approximation to the vector of internal 5. Köyluoglu, U. and Elishakoff, I., “A Comparison of
forces is found to yield a sharp solution in all the problems. Stochastic and Interval Finite Elements Applied to
In all the cases structural response is tabulated and is found Shear Frames with Uncertain Stiffness Properties”,
to vary for various combinations of load and material uncer- Comp. and Struc., Vol. 67, No. 1-3, 1998, pp.91-98.
tainties. Sensitivity analysis is found to be a useful tool to 6. Rao, S.S. and Sawyer, P., “Fuzzy Finite Element Ap-
evaluate the relative sensitivity of structural response in the proach for Analysis of Imprecisely Defined Systems”,
presence of multiple uncertainties. The present methodology AIAA J., Vol. 33, No. 12, 1995, pp. 2364-2370.
enables the designer to perform a detailed evaluation of the
effect of multiple uncertainties on structural behaviour. 7. Rao, S.S. and Berke, L., “Analysis of Uncertain Struc-
tural Systems using Interval Analysis”, AIAA J., Vol.
35, No. 4, 1997, pp. 727-735.
NOTATIONS
8. Rao, S.S. and Li Chen, “Numerical Solution of Fuzzy
E Young’s Modulus Linear Equations in Engineering Analysis”, Int. J.
[C#] Constraint Matrix of size n × 1 Numer. Meth. Engg. Vol. 43, 1998, pp. 391-408.
[D] A diagonal matrix containing interval values of
9. Akpan, U.O., Koko, T.S., Orisamolu, I.R. and Gallant,
Young’s Modulus B.K., “Practical fuzzy Finite element analysis of struc-
[K] Stiffness matrix of size n × n tures”, Finite Elements in Analysis and Design 38
[K"] Fixed-point (crisp) stiffness matrix (2001) 2001, pp. 93-111. (Elsevier Publishing).
[M] A matrix of size n × m
10. Rama Rao, M.V. and Ramesh Reddy, R., “Fuzzy Fi-
{P} Load Vector of size n × 1
nite Element Analysis of a Cable- Stayed Bridge”, 8th
[S#] A deterministic singular matrix of size n × n International Conference on Innovations in Planning,
{U} Vector of displacements of size n × 1 Design and Construction Techniques in Bridge En-
m No. of elements gineering, Hyderabad, India, 2003.
n No. of degrees of freedom for the structure
11. Jansson, C., “Interval Linear Systems with symmetric
$ Material uncertainty
matrices, skew-symmetric matrices, and dependencies
% Load uncertainty in the right hand side”, Computing, 46, 1991, pp. 265-
$max Maximum material uncertainty 274.

JOURNAL OF STRUCTURAL ENGINEERING VOL. 33, NO.2, JUNE–JULY 2006 137

View publication stats