acta mechanica solida sinica 30 (2017) 638–646

Available online at www.sciencedirect.com

journal homepage: www.elsevier.com/locate/CAMSS

Theoretical analysis of non-probabilistic reliability

based on interval model

Xu-Yong Chen a,b, Jian-Ping Fan b,∗, Xiao-Ya Bian a

a School of Resource and Civil Engineering, Wuhan Institute of Technology, Wuhan 430073, China
b School of Civil Engineering & Mechanics, Huazhong University of Science and Technology, Wuhan 430074, China

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

Article history: The aim of this paper is to propose a theoretical approach for performing the non-
Received 21 March 2017 probabilistic reliability analysis of structure. Due to a great deal of uncertainties and limited
Revised 31 October 2017 measured data in engineering practice, the structural uncertain parameters were described
Accepted 7 November 2017 as interval variables. The theoretical analysis model was developed by starting from the 2-D
Available online 1 December 2017 plane and 3-D space. In order to avoid the loss of probable failure points, the 2-D plane and
3-D space were respectively divided into two parts and three parts for further analysis. The
Keywords: study pointed out that the probable failure points only existed among extreme points and
Non-probabilistic root points of the limit state function. Furthermore, the low-dimensional analytical scheme
Reliability was extended to the high-dimensional case. Using the proposed approach, it is easy to find
Interval model the most probable failure point and to acquire the reliability index through simple compari-
Theoretical analysis son directly. A number of equations used for calculating the extreme points and root points
Probable failure point were also evaluated. This result was useful to avoid the loss of probable failure points and
meaningful for optimizing searches in the research field. Finally, two kinds of examples were
presented and compared with the existing computation. The good agreements show that
the proposed theoretical analysis approach in the paper is correct. The efforts were con-
ducted to improve the optimization method, to indicate the search direction and path, and
to avoid only searching the local optimal solution which would result in missed probable
failure points.
© 2017 Published by Elsevier Ltd on behalf of Chinese Society of Theoretical and Applied
Mechanics.

Euclidean norm by searching the most probable failure point

1. Introduction
The reliability of probability events could be measured using
reliability or reliability index [12]. The evaluation of reliabil-
ity depends strictly on the probability density function (PDF).
Unfortunately, it is hard to obtain the PDF for most engineer-
ing structures. However, due to its apparent geometric signif-
icance, the reliability index can be determined based on the
∗
Corresponding author.
E-mail address: jpfan@hust.edu.cn (J.-P. Fan).
(MPP) at the failure surface when the explicit PDF is hard to
find [18].
In modern civil engineering field, the probabilistic reliabil-
ity analysis has been applied extensively because it can con-
veniently deal with uncertainties widely existing in structural
and geotechnical engineering. However, this method has two
disadvantages. First, the probabilistic reliability analysis de-
pends strongly on the PDF. The achievement of PDF relies on
a large amount of statistical data [6]. For some important and
complicated structures, there exist quite amount of uncertain
parameters but very few or insufficient statistical data. There

https://doi.org/10.1016/j.camss.2017.11.003
0894-9166/© 2017 Published by Elsevier Ltd on behalf of Chinese Society of Theoretical and Applied Mechanics.
 acta mechanica solida sinica 30 (2017) 638–646
are many restrictions for the application of the probabilistic
reliability analysis method in engineering practice. Second,
the probabilistic reliability is very sensitive to the variation of
model parameters (mean and variance). The small errors of
statistical data will lead to considerable errors for the struc-
ture, i.e., the unreality and unreliability of the calculated re-
sults [1,6].
In the 1990s, Ben-Haim [2,3] proposed the concept of non-
probabilistic reliability and suggested using the convex model
to describe the uncertain—but—bounded parameters for en-
gineering structures. The non-probabilistic reliability is an ef-
fective method to deal with the reliability problems that only
few or insufficient statistical data can be achieved. Elishakoff
[5] further defined a non-probabilistic safety factor to mea-
sure the non-probabilistic reliability index by using the inter-
val theory. The above-presented non-probabilistic reliability
theory by Ben-Haim [2,3] and Elishakoff [5] was not involved
in probability at all and could overcome the inextricable dif-
ficulty that the traditional probability model faced. So the
non-probabilistic reliability is an appropriate choice when the
available data of uncertainties are insufficient or limited. This
was illustrated by Guo et al. [9,10], who contrasted the prob-
abilistic and non-probabilistic reliability methods through
modeling concept, model construction, formulations for com-
putation, etc. Therefore, the non-probabilistic structural reli-
ability has become a hot research direction, and the corre-
sponding research approach has also aroused wide concern
from the theoretical and engineering fields. By virtue of the
theory of non-probabilistic interval process, Wang et al. [24] in-
vestigated an interval process model of fatigue crack propaga-
tion, and evaluated the reliability of fatigue crack growth pre-
diction via limited measured data.
The non-probabilistic reliability index based on the in-
terval model is actually the minimum norm of the coordi-
nate vector in the standardized space, and solving the reli-
ability index is actually an optimum problem with equality
constraint. For linear performance functions, one can easily
obtain the analytical expression of the non-probabilistic re-
liability index. However, the performance functions are gen-
erally non-linear in practical engineering. For simple non-
linear performance functions, Guo et al. [9,10] suggested
using the definition approach, the transfer approach and the
optimization approach to solve the non-probabilistic reliabil-
ity index. For complex and strongly non-linear performance
functions, researchers frequently use the optimized iterative
algorithm to search the solution. For the reliability index de-
fined by the Euclidean norm, the MPP can be obtained along
the normal direction of the limit state surface. For the non-
probabilistic reliability index based on the interval model de-
fined by the infinite norm, the MPP may not be along the nor-
mal direction of the limit state surface. In order to simplify the
search process, some other researchers suggested using the 1-
dimensional optimization method [4,13] and the space search
algorithm [7]. The search directions for these two approaches
are fixed along the 45° diagonal. This search direction is cor-
rect only for the linear performance function, because only
part of the probable failure points is searched for the non-
linear performance function. In addition, if the performance
functions are in the normalized quadratic expression, the se-
quence quadratic programming (SQP) method can be used to
639
solve the non-probabilistic reliability index based on the inter-
val model [15]. Recently, the gradient projection method (GPM)
was proposed to solve the non-probabilistic reliability index
[19]. In fact, the GPM is a general method suitable for the non-
probabilistic convex model, although the convergence process
during the iteration needs special treatment.
The non-probabilistic reliability method mentioned above
has been further developed and commonly used in reliabil-
ity analysis and design of engineering structures. Jiang et al.
[14] presented a semi-analytical method for calculating the
non-probabilistic reliability index based on interval models.
Based on the probabilistic reliability model and interval arith-
metic, Qiu and Wang [21] built an interval estimation model
for the reliability of probabilistic and non-probabilistic hybrid
structural system. Ni and Qiu [20] constructed a hybrid relia-
bility model which contains randomness, fuzziness and non-
probabilistic uncertainty based on the structural fuzzy ran-
dom reliability and non-probabilistic set-based models. Wang
et al. [22] further developed a new hybrid reliability analysis
technique based on the convex modeling theory for structures
with multi-source uncertainties. This hybrid reliability analy-
sis technique was conducted to solve the convex modeling re-
liability problem and to further analyze the correlation within
uncertainties. Jiang et al. [16] proposed one non-probabilistic
reliability model with convex model uncertainty for the struc-
tural reliability analysis. Jiang et al. [17] further proposed a
non-probabilistic structural reliability analysis method based
on a multidimensional parallelepiped convex model. Besides,
Wang et al. [23] investigated a new formulation and numeri-
cal solution of reliability-based design optimization (RBDO) of
structures exhibiting random and uncertain—but—bounded
(interval and convex) mixed uncertainties.
The literature indicates that, researchers mainly paid at-
tention to the optimization search algorithm and conver-
gence of iteration process, but ignored the distribution of non-
probabilistic reliability index and numbers of probable failure
points. This results in that the final searched design point is
not the optimal solution. This paper will carry out the theoret-
ical study to solve the non-probabilistic reliability index, and
provide the distribution and the theoretical solving method
for the non-probabilistic reliability index. The example anal-
ysis will also be conducted to verify the effectiveness and ac-
curacy of the presented method.
2. Definition of non-probabilistic reliability
index
In engineering structures, resistance (R) is generally required
larger than the effect of actions (S). R and S are the functions
of geometry, material properties, load, and so on. When there
exist uncertainties for R and S, the parameters (such as geom-
etry, material properties, load, and so on) may vary in a certain
interval or follow a certain probabilistic distribution. For some
parameters, R > S may be satisfied; while for others, R > S may
not be satisfied. What may be concerned with is the probabil-
ity of R > S.
The non-probabilistic reliability analysis is used to study
whether or not the structure rigorously meets the require-
ment: M = R-S > 0. All parameters are regarded as belonging
640 acta mechanica solida sinica 30 (2017) 638–646

to one certain deterministic set. The smaller is the fluctuation

range of the system performance caused by these parameters,
the more reliable is the structure system.
Considering interval variables (x1 ,x2 ,…,xn ) and estab-
lishing performance function M = G(x1 ,x2 ,…,xn ), the vari-
ation range of xi is denoted by xi = [xi L ,xi U ] (i = 1,2,…,n).
The surface G(x1 ,x2 ,…,xn ) = 0 (turns to a curve for the
case of n = 2) will divide the design variable space into
two parts: G(x1 ,x2 ,…,xn ) > 0 that stands for the safety re-
gion, and G(x1 ,x2 ,…,xn ) < 0 that stands for the failure re-
gion. G(x1 ,x2 ,…,xn ) = 0 stands for the failure surface. Let
Fig. 1 – Linear performance function.
ML = minG(x1 ,x2 ,…,xn ) and MU = maxG(x1 ,x2 ,…,xn ), then the
non-probabilistic reliability index η was defined as [10]:

Mc MU + ML infinite norm. η > 1 demonstrates that the actual fluctuation

η= = U (1)
M r M − ML range of structure performance does not intersect the failure
region, and the structure is reliable. A larger value of reliabil-
(1) If η > 1, ML > 0 and G(x) > 0 can be obtained based on Eq. (1).
ity index η means a longer distance between the actual fluc-
This means that the safety region and the failure region
tuation region of the structure performance and the failure
of the structure do not intersect, and the structure is in a
region, and therefore, higher reliability of the structure.
reliable state.
Actually, to solve the non-probabilistic reliability index is to
(2) If η < −1, MU < 0 and G(x) < 0 can be obtained based on
search one set of basic variables x or u at the limit state surface
Eq. (1). This means that the structure is in the failure re-
G(x) = 0, for which the infinite norm of x or u is minimum. Tak-
gion and will inevitably fail.
ing the 2-dimensional plane problem for example, G(u1 ,u2 ) = 0
(3) If −1 < η < 1, the safety region and the failure region of the
is one plane curve. The formula of max(|u1 |,|u2 |) is to search
structure intersects and there may be G(x) > 0 or G(x) < 0
for the maximum coordinate value of the special point at the
for the structure. The structure may either be safe or fail,
plane curve, and the formula of min(u∞ ) is to search for one
which is not reliable. For this state, the structure reliability
point with the minimum value measured by the infinite norm.
needs to be further analyzed with the non-probabilistic set
The point, at which the minimum distance from the origin of
reliability [8].
coordinates to the failure surface is obtained for this linear
limit state function, is the intersection of the 45° line through
For general linear performance function:
the origin and the linear limit state function (seen in Fig. 1).

m 
n Based on the definition of non-probabilistic reliability men-
G= ai Ri − b jS j (2) tioned above, one can obtain: η = |ur | = |us |, and the non-
i=1 j=m+1 probabilistic reliability index based on the interval model
can be expressed as: η = (Rc –Sc )/(Rr + Sr ). This is the non-
the non-probabilistic reliability index was estimated as [10]:
probabilistic reliability index defined by Eq. (1) when the per-
⎧ m a Rc − n c m n formance function is linear. As stated in the previous section,
⎨ mi=1 i ir ni=m+1 bi Si r ; ai Rci − c >0
|ai |Ri + i=m+1 |bi |Si i=1 i=m+1 bi Si in practical engineering, the performance functions always
η= i =1 (3)
⎩ m n
0; c − c ≤0 demonstrate strongly non-linear, which increases the com-
i=1 ai Ri i=m+1 bi Si
plexity in solving the non-probabilistic reliability index. The
where Ri (resistance) and Sj (effect of actions) are the internal theoretical study for solving the non-probabilistic reliability
variables; Rci and Scj are centers of Ri and Sj , respectively; Rri and index with non-linear cases will be discussed in the next sec-
Srj are radii of Ri and Sj , respectively; and ai and bj are linear tion.
coefficients.
In practical engineering, the performance functions al-
ways exhibit in different forms and are almost non-linear. 3. Theoretical study of non-probabilistic
Normalizing the internal variable x, one has xi = xi c + xi r ui , reliability index
where ui is the normalized internal variable, and its vari-
ation interval is [−1,1]. Then the x space can be trans- The reliability index, which is the minimum distance of the
formed into the u space, and the continuous performance probable failure points from the origin of coordinates with
function M = G(x) = G(x1 ,x2 ,…,xn ) can be transformed into the infinity-norm measurement, is thoroughly determined by
M = G(u) = G(u1 ,u2 ,…,un ). After that, the value range of ui the most probable failure points. Then through geometric
should be relaxed to [-∞,+∞]. The non-probabilistic reliabil- analysis, all probable failure points can be solved for the limit
ity index η based on the interval model was defined as [10]: state functions of 2-, 3-, and n-dimensional cases. These prob-
  able failure points may be the extreme points or the root
η = min (u∞ ) = min max (|u1 |, |u2 |, . . . , |un | ) points. The most probable failure points can be obtained by
(4)
s.t.G(u1 , u2 . . . un ) = 0 comparing the values of all probable failure points. Finally, the
According to Eq. (4), η is the minimum distance from the reliability index can be computed using the minimum of the
origin of coordinates to the failure surface measured by the infinity-norm for the most probable failure points.
 acta mechanica solida sinica 30 (2017) 638–646
Fig. 2 – MPP in the region above the 45° line.
Fig. 3 – MPP in the region below the 45° line.
3.1. 2-dimensional case
The limit state function G(u1 ,u2 ) = 0 represents a plane curve.
The 45° line divides the first quadrant of the coordinate plane
into two parts. The MPPs may be in the region above the 45°
line, in the region below the 45° line, or on the 45° line (the
boundary between these two regions).
(1) If the MPP is in the region above the 45° line, as shown in
Fig. 2, there is: η = min(u∞) = min|u2| due to |u2| > |u1|.
For this case, the implicit limit state function G(u1 ,u2 ) = 0
can be transformed into the explicit function u2 = F(u1 ), then
the minimum point of u2 = F(u1 ), or (u1 ,u2 ) is the MPP.
(2) If the MPP is in the region below the 45° line, as shown in
Fig. 3, there is: η = min(u∞ ) = min|u1 | due to |u1 | > |u2 |.
For this case, the implicit limit state function G(u1 ,u2 ) = 0
can be transformed into the explicit function u1 = F(u2 ), then
the minimum point of u1 = F(u2 ), or (u1 ,u2 ) is the MPP.
(3) If the MPP is on the 45° line, as shown in Fig. 4, the reliability
index η can be solved using the function G(u1 = ± u2 ) = 0,
and there is: η = min(u∞ ) = min|u1 | = min|u2 |.
When the limit state function G(u1 ,u2 ) = 0 is known, the
non-probabilistic reliability index can be computed using the
theoretical method described below:
1) Solving all the extreme points for u1 and u2 based on the
limit state function G(u1,u2) = 0;
2) Solving all the root points based on the limit state function
G(u1 = ±u2) = 0;
641
Fig. 4 – Case of MPP on the 45° line.
3) Obtaining the non-probabilistic reliability index η, which
is the minimum value of the extreme points and the root
points mentioned above.
3.2. 3-dimensional case
Supposing G(u1 ,u2 ,u3 ) = 0 is the limit state function for the 3-
dimensional problem, the entire space can be divided into
three zones (respectively denoted by Part Ⅰ, Part Ⅱ, and Part
Ⅲ) by using three planes: u1 = u2 , u1 = u3 , u2 = u3 , as shown in
Fig. 5. These three planes (u1 = u2 , u1 = u3 , u2 = u3 ) are the in-
terfaces of the three zones. In Part Ⅰ represented by the green
zone, the limit state function G(u1 ,u2 ,u3 ) = 0 can be trans-
formed into u1 = F(u2 ,u3 ), so to search for the MPPs is to find
the minimum of u1 . In Part Ⅱ represented by the blue zone,
the limit state function G(u1 ,u2 ,u3 ) = 0 can be transformed into
u2 = F(u3 ,u1 ), so to search for the MPPs is to find the minimum
of u2 . In Part Ⅲ represented by the yellow zone, the limit state
function G(u1 ,u2 ,u3 ) = 0 can be transformed into u3 = F(u1 ,u2 ),
so to search for the MPPs is to find the minimum of u3 . Be-
sides the extreme points in space, the extreme points at the
interface and the boundary should be carefully considered,
and compared with the extreme points in space.
As a summary, when the limit state function G(u1 ,u2 ,u3 ) = 0
is given, the non-probabilistic reliability index can be com-
puted using the theoretical method described below:
1) Solving all the extreme points for u1, u2, and u3 based on
the limit state function G(u1,u2,u3) = 0;
2) Reducing dimensionality, then solving all the extreme
points of G(u1 = ± u2,u3) = 0, G(u1 = ± u3,u2) = 0, and
G(u1,u2 = ± u3) = 0 (computation is conducted as the
2-dimensional problem);
3) Reducing dimensionality again, then solving all
the root points based on the limit state function
G(u1 = ±u2 = ±u3) = 0;
4) Obtaining the non-probabilistic reliability index η, which
is the minimum value of the extreme points and the root
points obtained above.
3.3. n-dimensional case
According to the study on the 2- and 3-dimensional cases,
the MPPs appear only in the extreme points and the root
points of the limit state function G(u1 ,u2 ,…,un ) = 0. The non-
642 acta mechanica solida sinica 30 (2017) 638–646

Fig. 5 – Division of 3-dimensional space domain. (For interpretation of the references to colour in this figure legend, the
reader is referred to the web version of this article.)

probabilistic reliability index for the n-dimensional case can ber of limit state functions for solving the root points is always
be computed using the theoretical method described below: 2n -1 , which cannot be reduced.
According to the above analysis, in order to obtain the
non-probabilistic reliability index, the extreme points and root
1) Solving all the extreme points for u1, u2, …, and un based
points of the limit state functions should be found. The MPPs
on the limit state function G(u1,u2,…,un) = 0;
must be among these extreme points and root points. The
2) Reducing dimensionality, then solving all the extreme
non-probabilistic reliability index η is the minimum value of
points of G(ui = ± uj,uk) = 0 for the total n(n-1) limit
these extreme points and root points. The computation of the
state functions (computation is conducted as the (n-1)-
extreme points and the root points of the limit state functions
dimensional problem);
can be easily conducted using the existing mathematical tool,
3) Reducing dimensionality again, then solving all the ex-
such as MATLAB software, by analyzing the function behav-
treme points of the limit state functions (computation is
iors.
conducted as the (n-2)-, (n-3)-, …, and 2-dimensional prob-
lems in sequence);
4) Reducing dimensionality again, then solving all
the root points based on the limit state functions
G(u1 = ±u2 = … = ±un) = 0 (2n-1 functions in total); 4. Application examples
5) Achieving the non-probabilistic reliability index η, which
is the minimum value of the extreme points and the root 4.1. Performance function with analytical expression
points obtained above.
Now consider the performance functions, which can be ex-
pressed in the analytical form and are normalized.
For the n-dimensional problem, there are (1 + n!(n-1)!/2)
limit state functions for solving the extreme points and 2n -1
limit state functions for solving the root points. The non-
probabilistic reliability index η is the minimum value of 4.1.1. Example 1
these extreme points and root points. In summary, for the 3- Performance function: G = (u1 -1)2 + (u2 -4/5)2 –4u3 + 5.5.
dimensional problem, there are 11 limit state functions; for The extreme point: (1, 4/5, 1.375);
the 4-dimensional problem, there are 81 limit state functions; The root points: 1.7, 2.1.
for the 5-dimensional problem, there are 1457 limit state func- These results are shown in Fig. 6, which indicate that the
tions; and for the 6-dimensional problem, there are 43,233 MPP is (1, 4/5, 1.375). This MPP is the extreme point. The non-
limit state functions. With consideration of the forms of per- probabilistic reliability index η is 1.375, which is in accor-
formance functions, the number of limit state functions for dance with the result obtained by using the gradient projec-
solving the extreme points can be reduced. However, the num- tion method [19].
 acta mechanica solida sinica 30 (2017) 638–646
Fig. 6 – MPP of example 1.
4.1.2. Example 2
Performance function: G = 567(0.6 + 0.072u1 )(2.18 + 0.109u2 )
−0.5(32.8 + 1.64u3 )2 .
The extreme points: (−8.33, −20, −20);
The root points: 5.7695, 20, −1.7578, −6.0935, −1.1556,
−56.7733.
These results are shown in Fig. 7, which indicates that the
MPP is (−1.1556, −1.1556, −1.1556). This MPP is a root point.
The non-probabilistic reliability index η is 1.1556, which is in
accordance with the result obtained in literature [11].
4.1.3. Example 3
Performance function: G = (u1 -7/4)2+ (u2-4/5)2 –4u
3 + 5.5.
The extreme points: (1.75, 0.8, 1.375), (1.275, 1.275, 1.4878),
(1.4048, 1.4048, 0.8);
The root points: 3.0329, 1.5171.
These results are shown in Fig. 8, which indicates that the
MPP is (1.4048, 1.4048, 0.8). This MPP is the extreme point. The
non-probabilistic reliability index η is 1.4048, which is in ac-
cordance with the result obtained by using the gradient pro-
jection method [19].
4.2. Structure in practical engineering
4.2.1. Cantilever girder under two concentrated loads
As shown in Fig. 9, two concentrated loads are applied on
the cantilever girder at the points with distances b1 and b2
from the fixed end, respectively. Supposing that the cantilever
girder fails under the condition of mmax  ≥ mcr , where mcr is
the critical moment and mmax is the maximum moment ap-
plied on the girder, the marginal intervals of the uncertain
643
variables are: p1 ∈[4.4,5.6] kN, p2 ∈[1.7,2.3] kN, b1 ∈[1.8,2.2] m,
b2 ∈[4.5,5.5] m, and mcr ∈[32,40] kN m.
Based on the failure criteria for the cantilever girder, the
performance function can be established as follows:
M = mcr − p1 b1 − p2 b2 = 0 (5)
Taking the normalized transformation for Eq. (5), one has:
G = 36 + 4u1 − (5 + 0.6u2 )(2 + 0.2u3 ) − (2 + 0.3u4 )(5 + 0.5u5 ) = 0
(6)
The extreme points: none;
The root points: −225.6964, −220.9192, −212.5097,
−68.9293, −33.9668, −19.0385, −12.7651, −10.0000, −9.1027,
−8.2195, −7.7374, −5.9259, −4.6423, −3.4538, −2.4142, 1.7446,
2.3631, 2.5097, 2.8498, 3.1126, 6.5101, 17.1575, 53.3333, 64.8862,
187.1502.
Comparing the extreme points and the root points ob-
tained above, the MPP, (1.7446, −1.7446, −1.7446, −1.7446,
−1.7446), can be obtained. This MPP is a root point. The non-
probabilistic reliability index η is 1.7446, which is in accor-
dance with the result obtained by using the gradient projec-
tion method [19].
4.2.2. 10-bar truss structure
As shown in Fig. 10, the cross-sectional areas of the 10-bar
truss structure are denoted as A1 –A10 , respectively. The ver-
tical loads p1 and p2 are applied on nodes 2 and 3 of the
644 acta mechanica solida sinica 30 (2017) 638–646

Fig. 7 – MPP of example 2.

Fig. 8 – MPP of example 3.

truss structure. The vertical loads p1 and p2 are both in- and 3 must be smaller

than 0.5 × 10−3 m. So the performance
terval variables, the normal means of which are 10 × 103 N function is g = (δy2 + δy23 )/2 ≤ 0.5, where δ y 2 and δ y 3 are the
2

and the deviations of which are 10% and 25%, respectively. vertical displacements of nodes 2 and 3.
The elastic moduli of bars  1, 2, 3, 
4,  5 , and 6 are all In order to avoid the lengthy mathematical (analytical)
E1 = 2.0 × 1011 Pa, and the elastic moduli of bars  7 ,
8 ,
9 , and derivation, the displacements of the appointed nodes under

10 are all E2 = 2.2 × 1011 Pa. For this 10-bar truss structure, the the loads are solved using the numerical method described as
root-mean-square values of vertical displacements of nodes 2 follows:
 acta mechanica solida sinica 30 (2017) 638–646
Fig. 9 – Cantilever girder subjected to two concentrated
loads.
Fig. 10 – Truss structure.
Table 1 – Non-probabilistic reliability index for 10-bar
truss structure.
Member number Cross-sectional area (10−2 m2 )
Case 1 Case 2
1 2.0 1.6152
2 2.0 1.0486
3 2.0 1.6312
4 2.0 0.0065
5 2.0 0.0065
6 2.0 0.0065
7 2.0 0.6297
8 2.0 0.7284
9 2.0 0.0065
10 2.0 1.2497
η 6.394 1.197
η (Luo and Kang 2008) 6.134 1.148
(1) Let p1 = p2 = 10 × 103 N. The displacements of nodes 2 and
3 along the y-axis caused by vertical load p1 alone are re-
spectively denoted as δ 21 and δ 31 , and the displacements
of nodes 2 and 3 along the y-axis caused by vertical load p2
alone are respectively denoted as δ 22 and δ 32 . δ 21 , δ 31 , δ 22 ,
and δ 32 can be obtained by using the finite element method.
(2) Under the combined effects of p1 and p2 , the displace-
ments of nodes 2 and 3 along the y-axis are respec-
tively expressed as δ y 2 = (1 + 0.1η)δ 21 + (1 + 0.25η)δ 22 and
δ y 3 = (1 + 0.1η)δ 31 + (1 + 0.25η)δ 32 .
(3) The
critical condition can be determined as g =
(δy22 + δy23 )/2 = 0.5 or δy22 + δy23 = 0.5, based on which
the non-probabilistic reliability index η can be solved, as
shown in Table 1.
From Table 1, the non-probabilistic reliability indices η are
6.394, 1.197, and 1.047 for Case 1, Case 2, and Case 3, re-
645
spectively. They are generally in accordance with the results
obtained by Luo and Kang (2008). Of course, it should also
be pointed out that the results obtained using the proposed
method are slightly larger than the results from Luo and Kang
(2008). This is due to the use of hyper ellipsoidal model in Luo
and Kang (2008)’s study.
5. Summary and conclusions
The main contribution of this paper is that one rigorous ana-
lytical approach based on the interval model was proposed to
search for the optimal solution. In this process, the following
highlights can be drawn:
(1) The basic thought of developing this analytical approach
initiated from the 2- and 3-dimensional spaces, and was
extended to the n-dimensional space.
(2) The study first pointed out that the probable failure points
could only exist in the extreme points and root points of
the limit state function.
(3) The dimensionality reduction idea was used to search for
the probable failure points in the presented analytical ap-
proach.
(4) The number of functions for calculating the extreme points
and root points were precisely evaluated in the process of
solving the probable failure points.
(5) Several examples are analyzed and compared with the
Case 3
existing studies. The good agreements between them
1.5750 demonstrate the correctness and effectiveness of the pre-
1.0694 sented analytical approach.
1.5454
0.0065
The values of the presented analytical approach are mainly
0.0065
0.0065
embodied in two aspects:
0.5988
0.7069 (1) When the performance function can be analytically ex-
0.0065 pressed and has interval variables less than 5, using the
1.2297 presented analytical approach can expediently compute
1.047
the non-probabilistic reliability index with small amount
1.001
of computational task and precise results.
(2) For complex engineering problems, the presented analyti-
cal approach can be used to estimate the number and re-
gion of the failure points. When an optimization search
method is used, the analytical approach appears to be an
important reference for avoiding the loss of failure points.
Acknowledgment
The authors would like to express their gratitude to the
National Natural Science Foundation of China (51408444,
51708428).
references
[1] Y. Ben-Haim, Convex models of uncertainty in radial pulse
buckling of shells, J. Appl. Mech. 60 (3) (1993) 683–688.
[2] Y. Ben-Haim, A non-probabilistic concept of reliability,
Struct. Saf. 14 (4) (1994) 227–245.
646 acta mechanica solida sinica 30 (2017) 638–646
[3] Y. Ben-Haim, A non-probabilistic measure of reliability of
linear systems based on expansion of convex models, Struct.
Saf. 17 (2) (1995) 91–109.
[4] X.Y. Chen, C.Y. Tang, C.P. Tsui, J.P. Fan, Modified scheme based
on semi-analytic approach for computing non-probabilistic
reliability index, Acta Mech. Solida Sin. 23 (2) (2010) 115–123.
[5] I. Elishakoff, Discussion on: a non-probabilistic concept of
reliability, Struct. Saf. 17 (3) (1995) 195–199.
[6] I. Elishakoff, Essay on uncertainties in elastic and
viscoelastic structures: from A. M. Freudenthal’s criticisms to
modern convex modeling, Comput. Struct. 56 (6) (1995)
871–895.
[7] J.P. Fan, S.J. Li, X.Y. Chen, Optimal searching algorithm for
non-probabilistic reliability, Chin. J. Comput. Mech. 29 (6)
(2012) 831–834 (in Chinese).
[8] J.P. Fan, S.J. Li, W. Qi, X.Y. Chen, Safety evaluation of
non-probabilistic reliability model of structures, Chin. J. Solid
Mech. 33 (3) (2012) 325–330 (in Chinese).
[9] S.X. Guo, Z.Z. Lv, Comparison between the non-probabilistic
and probabilistic reliability methods for uncertain structure
design, Chin. J. Appl. Mech. 20 (3) (2003) 107–110.
[10] S.X. Guo, Z.Z. Lv, Y.S. Feng, A non-probabilistic model of
structural reliability based on interval analysis, Chin. J.
Comput. Mech. 18 (1) (2001) 56–62 (in Chinese).
[11] S.X. Guo, L. Zhang, Y. Li, Procedures for computing the
non-probabilistic reliability index of uncertain structures,
Chin. J. Comput. Mech. 22 (2) (2005) 227–231 (in Chinese).
[12] J.E. Hurtado, D.A. Alvarez, The encounter of interval and
probabilistic approaches to structural reliability at the design
point, Comput. Methods Appl. Mech. Eng. 225–228 (2012)
74–94.
[13] T. Jiang, J.J. Chen, P.G. Jiang, Y.F. Tuo, A one-dimensional
optimization algorithm for non-probabilistic reliability
index, Eng. Mech. 24 (7) (2007) 23–27.
[14] T. Jiang, J.J. Chen, Y.L. Xu, A semi-analytic method for
calculating non-probabilistic reliability index based on
interval models, Appl. Math. Model. 31 (7) (2007) 1362–1370.
[15] C. Jiang, Z. Zhang, X. Han, Y.C. Bai, An evidence-theory-based
reliability analysis method for uncertain structures, Chin. J.
Theor. Appl. Mech. 45 (1) (2013) 103–115.
[16] C. Jiang, R.G. Bi, G.Y. Lu, X. Han, Structural reliability analysis
using non-probabilistic convex model, Comput. Methods
Appl. Mech. Eng. 254 (2013) 83–98.
[17] C. Jiang, Q.F. Zhang, X. Han, Y.H. Qian, A non-probabilistic
structural reliability analysis method based on a
multidimensional parallelepiped convex model, Acta Mech.
225 (2014) 383–395.
[18] J.O. Lee, Y.S. Yang, W.S. Ruy, A comparative study on
reliability-index and target-performance-based probabilistic
structural design optimization, Comput. Struct. 80 (3–4)
(2002) 257–269.
[19] S.J. Li, J.P. Fan, W. Qi, X.Y. Chen, The gradient projection
method for non-probabilistic reliability index based on
interval model, Chin. J. Comput. Mech. 30 (2) (2013) 192–197
(in Chinese).
[20] Z. Ni, Z.P. Qiu, Hybrid probabilistic fuzzy and
non-probabilistic model of structural reliability, Comput. Ind.
Eng. 58 (3) (2010) 463–467.
[21] Z.P. Qiu, J. Wang, The interval estimation of reliability for
probabilistic and non-probabilistic hybrid structural system,
Eng. Fail. Anal. 17 (5) (2010) 1142–1154.
[22] L. Wang, X.J. Wang, Y. Xia, Hybrid reliability analysis of
structures with multi-source uncertainties, Acta Mech. 225
(2) (2014) 413–430.
[23] L. Wang, X.J. Wang, R.X. Wang, X. Chen, Reliability-based
design optimization under mixture of random, interval and
convex uncertainties, Arch. Appl. Mech. 86 (7) (2016)
1341–1367.
[24] L. Wang, X.J. Wang, H. Su, G.P. Lin, Reliability estimation of
fatigue crack growth prediction via limited measured data,
Int. J. Mech. Sci. 121 (2017) 44–57.