safety factor and inverse reliability
TRANSCRIPT
-
8/3/2019 Safety Factor and Inverse Reliability
1/26
I. Introduction1Inverse reliability measures are becoming popular in reliability community. Various2
inverse reliability measures have been developed. This paper focuses on discussing the3
interconnection of various inverse reliability measures, examining methods to compute4
inverse reliability measures and the advantages of using them over direct reliability measures.5
Traditionally, structural safety was defined in terms of safety factors, which were used6
to compensate for uncertainties in loading and material properties, and for inaccuracies in7
geometry and theory. Safety factors permit design optimization using computationally8
inexpensive deterministic methods. In addition, it is relatively easy to estimate the change in9
structural weight of over or under designed structures needed to satisfy a target safety factor10
requirement (Qu and Haftka 2003, 2004).11
Probabilistic approaches in design optimization allow incorporation of available12
uncertainty data and thus provide more accurate measures of safety. Structural safety is13
measured in terms of probability of failure to satisfy some performance criterion. The14
probability of failure is often expressed in terms of a reliability index. This reliability index is15
the ratio of the mean to the standard deviation of the safety margin distribution, which is the16
difference between the capacity and the resistance of the system.17
Optimization using probabilistic approaches called reliability-based design18
optimization (RBDO) enables to gauge the structural safety better but is computationally19
significantly more expensive compared to deterministic approaches. In addition, the20
difference between the computed probability of failure or reliability index and their target21
values does not provide the designer with easy estimates of the change in the design cost22
needed to achieve these target values.23
Safety factors are defined as the ratio of the capacity to resistance of a system. In24
deterministic approaches, safety factors are determinate, typically calculated for mean values25
-
8/3/2019 Safety Factor and Inverse Reliability
2/26
-
8/3/2019 Safety Factor and Inverse Reliability
3/26
-
8/3/2019 Safety Factor and Inverse Reliability
4/26
4
inverse measure as percentile performance. Traditionally, design for robustness involves1
minimizing the mean and standard deviation of the performance. Here, Du et al. (2003)2
proposed to replace the standard deviation by percentile performance difference, which is the3
difference between the percentile performance corresponding to the left tail of a CDF and the4
right tail of that CDF. They demonstrated increased computational efficiency and more5
accurate evaluation of the variation of the objective performance. In an effort to address6
reliability-based designs when both random and interval variables are present, Du and7
Sudijianto (2003) proposed the use of percentile performance with worst-case combination of8
the interval variables for efficient RBDO solutions.9
Du and Chen (2004) developed the sequential optimization and reliability assessment10
(SORA) to improve the efficiency of the probabilistic optimization. The method is a serial11
single loop strategy, which employs percentile performance and the key is to establish12
equivalent deterministic constraints from probabilistic constraints. This method is based on13
evaluating the constraint at the most probable point of the inverse measure in Section IV14
below) based on the reliability information from the previous cycle. This is referred to as15
design shift (Chiralaksanakul and Mahadevan 2004; Youn et al. 2004). They show that the16
design quickly improves in each cycle and is computationally efficient. The sequential17
optimization and reliability assessment, however, is not guaranteed to lead to an optimal18
design. Single-level (or unilevel) techniques that are equivalent to the standard RBDO19
formulation are based on replacing the RIA or PMA inner loop by the corresponding Karush-20
Kuhn-Tucker conditions. Here again, Agarwal et al. (2004) showed that the PMA approach is21
more efficient than the unilevel RIA approach due to Kuschel and Rackwitz (2000).22
The several inverse measures discussed above are all based on the common idea of23
using the inverse of the cumulative distribution function. The numerous names for the inverse24
measures contemplate that they were developed by different researchers for different25
-
8/3/2019 Safety Factor and Inverse Reliability
5/26
5
applications. Since these inverse measures come under various names, it is easy to fail to1
notice the commonality among them. The main purpose of this paper is to highlight the2
relationship between the inverse measures. The objectives of this work include:3
(i) To discuss the relationship (or minor differences) between the various inverse4
measures5
(ii) To discuss methods available for calculating these inverse measures6
(iii) To explore the advantages of using inverse measures instead of direct measures.7
Section II describes inverse reliability measures. Calculation of inverse measures by MCS is8
discussed in Section III. Section IV describes calculation of inverse measures using moment-9
based techniques, followed by discussion of using inverse measures in RBDO in Section V.10
Section VI demonstrates the concepts with the help of a beam design example, and Section11
VII provides concluding remarks.12
13
II. Inverse Reliability Measures14Birger Safety Factor15
The safety factor, S is defined as the ratio of the capacity of the system cG (e.g.,16
allowable strength) to the response rG with a safe design satisfying r cG G . To account for17
uncertainties, the design safety factor is greater than one. For example, a load safety factor of18
1.5 is mandated by FAA in aircraft applications. To address the probabilistic interpretation of19
the safety factor, Birger (1970) proposed to consider its cumulative distribution function20
(CDF) SF :21
( ) Prob( )cSr
GF s s
G= (1)22
Note that unlike the deterministic safety factor, which is normally calculated for the23
-
8/3/2019 Safety Factor and Inverse Reliability
6/26
6
mean value of the random variables, c rG G in Eq. (1) is a random function. Given a target1
probability, ftargetP , Birger suggested a safety factor s* (which we call here the Birger safety2
factor) defined in the following equation3
Prob( ) Prob( )cS ftarget r
GF (s*) s* S s* P
G= = = (2)4
That is, the Birger safety factor is found by setting the cumulative distribution function5
(CDF) of the safety factor equal to the target probability. That is, we seek to find the value of6
the safety factor that makes the cumulative distribution function of the safety factor equal to7
the target failure probability. This requires the inverse of the CDF, hence the terminology of8
inverse measure.9
10
Probabilistic Sufficiency Factor11
Qu and Haftka (2003, 2004) developed a similar measure to the Birger safety factor,12
calling it first the probabilistic safety factor and then the probabilistic sufficiency factor13
(PSF). They obtained the PSF by Monte Carlo simulation and found that the response surface14
for PSF was more accurate than the response surface fitted to failure probability. Later, they15
found the reference to Birgers work in Elishakoffs review (2001) of safety factors and their16
relations to probabilities. It is desirable to avoid the term safety factor for this entity because17
the common use of the term is mostly deterministic and independent of the target safety level.18
Therefore, while noting the identity of the Birger safety factor and the probabilistic19
sufficiency factor, we will use the latter term in the following.20
Failure happens when the actual safety factor S is less than one. The basic design21
condition that the probability of failure should be smaller than the target probability for a safe22
design may then be written as:23
Prob( 1) (1)f S ftarget P S F P= = (3)24
-
8/3/2019 Safety Factor and Inverse Reliability
7/26
7
Using inverse transformation, Eq. (3) can be expressed as:1
11 ( )S ftarget F P s*
= (4)2
The PSF concept is illustrated in Fig. 1. The design requirement ftargetP is known and the3
corresponding area under the probability4
density function of the safety factor is the5
shaded region in Fig. 1. The upper bound of6
the abscissa s* is the value of the PSF. The7
region to the left of the vertical line 1S = 8
represents failure. To satisfy the basic9
design condition s* should be larger than or10
equal to one. In order to achieve this, it is11
possible to either increase cG or12
decrease rG . The PSF s* , represents the factor that has to multiply the response rG or divide13
the capacity cG , so that the safety factor be raised to 1.14
For example, a PSF of 0.8 means that rG has to be multiplied by 0.8 or cG be15
divided by 0.8 so that the safety factor ratio increases to one. In other words, this means that16
rG has to be decreased by 20 % (1-0.8) or cG has to be increased by 25% ((1/0.8)-1) in order17
to achieve the target failure probability. It can be observed that PSF is a safety factor with18
respect to the target failure probability and is automatically normalized in the course of its19
formulation.20
PSF is useful in estimating the resources needed to achieve the required target21
probability of failure. For example, in a stress-dominated linear problem, if the target22
probability of failure is 10-5
and a current design yields a probability of failure of 10-3
, one23
cannot easily estimate the change in the weight required to achieve the target failure24
Fig. 1 Schematic probability density of
the safety factor S . The PSF is the value
of the safety factor corresponding to the
target probability of failure
-
8/3/2019 Safety Factor and Inverse Reliability
8/26
-
8/3/2019 Safety Factor and Inverse Reliability
9/26
-
8/3/2019 Safety Factor and Inverse Reliability
10/26
-
8/3/2019 Safety Factor and Inverse Reliability
11/26
-
8/3/2019 Safety Factor and Inverse Reliability
12/26
-
8/3/2019 Safety Factor and Inverse Reliability
13/26
13
can also be used to approximate the structural response in order to reduce the computational1
cost. Qu and Haftka (2004b) employ PSF with MCS based on RSA to design stiffened panels2
under system reliability constraint.3
4
IV. Inverse Measure Calculation by Moment-Based Methods5Moment-based methods provide for less expensive calculation of the probability of6
failure compared to MCS, although they are limited to a single failure mode. These methods7
require a search for the most probable point (MPP) on the failure surface in the standard8
normal space. The First Order Reliability Method (FORM) is the most widely used moment-9
based technique. FORM is based on the idea of the linear approximation of the limit state10
function and is accurate as long as the curvature of the limit state function is not too high.11
When the limit state has a significant curvature, second order methods can be used. The12
Second Order Reliability Method (SORM) approximates the measure of reliability more13
accurately by considering the effect of the curvature of the limit state function (Melchers,14
1999, pp 127-130). All the random variables are to be transformed to the standard normal15
variables with zero mean and unit variance.16
Moment-based methods are employed to calculate the reliability index, which is17
denoted by and related to the probability of failure as:18
( )fP = (13)19
where is the standard normal cumulative distribution function. Respective target values of20
and failure probabilities are also related in the same manner. In FORM, can be21
calculated as = U , where U is the vector of standard normal variates (variables with22
normal distribution of zero mean and unit variance). The standard normal variates are23
obtained from a transformation of the basic random variables X, which could be non-normal24
-
8/3/2019 Safety Factor and Inverse Reliability
14/26
14
and dependent. In the standard normal space the point on the first order limit state function at1
which the distance from the origin is minimum is the MPP.2
Figure 4 illustrates the concept of the reliability index and MPP search for a two3
variable case in the standard normal space. In reliability analysis, concerns are first focused4
on the ( ) 0G =U curve. Next, among the various values possible (denoted by 1 , 2 , 3 ),5
the minimum is sought. The corresponding point is the MPP. This process can be6
mathematically expressed as:7
To find* = u ,8
MinimizeT
U U 9
Subject to: ( ) 0G =U (14)10
where u* is the MPP. The calculation of the failure probability is based on linearization of11
the limit function at the MPP.12
Inverse reliability measures can also be computed through moment-based methods.13
Figure 5 illustrates the concept of inverse reliability analysis and MPP search. The circles14
represent the curves with the target curve represented by a dashed circle. Here, among15
the different values of limit state functions that pass through the target curve, the one with16
minimum value is sought. The value of this minimal limit state function is the PPM as shown17
by Tu et al. 1999. The point on the target circle with the minimal limit state function is18
sought. This point is also an MPP and in order to avoid confusion between the usual MPP and19
MPP in inverse reliability analysis, Du et al. (2003) coined the term most probable point of20
inverse reliability (MPPIR) and Lee et al. (2002) called it the minimum performance target21
point (MPTP). Du et al. (2003) developed the sequential optimization and reliability analysis22
method in which they show that evaluating the probabilistic constraint at the design point is23
-
8/3/2019 Safety Factor and Inverse Reliability
15/26
15
1equivalent to evaluating the deterministic constraint at the most probable point of the inverse2
reliability. This facilitates in converting the probabilistic constraint to an equivalent3
deterministic constraint. That is, the deterministic optimization is performed using a4
constraint limit which is determined based on the inverse MPP obtained in the previous5
iteration. Kiureghian et al. (1994) proposed an extension of the Hasofer-Lind-Rackwitz-6
Fiessler algorithm that uses a merit function and search direction to find the MPTP. In Fig. 5,7
the value of the minimal limit state function or the PPM is -0.2. This process can be8
expressed as:9
Minimize: ( )G U 10
Subjected to: T target= =U U U (15)11
In reliability analysis the MPP is on the ( ) 0G =U failure surface. In inverse reliability12
analysis, the MPP search is on the targetcurve.13
14
15
16
Fig. 5 Inverse reliability analysis and MPP
for target probability of failure of 0.00135
( = 3)
Fig. 4 Reliability analysis and MPP
-
8/3/2019 Safety Factor and Inverse Reliability
16/26
-
8/3/2019 Safety Factor and Inverse Reliability
17/26
-
8/3/2019 Safety Factor and Inverse Reliability
18/26
18
Table 4 Comparison of inverse measures for
w=2.4526 and t=3.8884 (stress constraint)
p
1f fP ( P )
N
= (19)1
In this case, the failure probability of 0.0013 calculated from 100,000 samples has a standard2
3
Minimize objective functionA=wtsuch that 3 4
Inverse Reliability AnalysisMethod Reliability
Analysis
FORM
(RIA)
FORM
(PMA)
MCS (PSF)
(Qu and Haftka,2003)
Exact
Optimum
(Wu et al.
2001)
w 2.4460 2.4460 2.4526 2.4484Optima
t 3.8922 3.8920 3.8884 3.8884
Objective Function 9.5202 9.5202 9.5367 9.5204
Reliability Index 3.00 3.00 3.0162 3.00
Failure Probability 0.00135 0.00135 0.00128 0.00135
5
deviation of 1.1410-4
. It is seen from Table 3 that the designs obtained from RIA, PMA and6
PSF match well. Since the stress Eq. (17) is a linear function of random variables, the RIA7
and PMA are exact. The more conservative design from PSF is due to limited sampling of8
MCS.9
10
Comparison of Inverse Measures11
The relation between PSF and PPM in Eq. (10) is only approximate when PPM is12
calculated by FORM and PSF by MCS. PSF suffers from sampling error, and PPM from13
error due to linearization. For the linear stress constraint, and with a large Monte Carlo14
sample, the difference is small, as seen in Table 4. It may be expected that the MPTP16
(minimum performance target point)18
should also be close to the point used19
to calculate PSF. This result may be20
useful, because when a response21
surface is used for an approximation to the response, it is useful to center it near the MPTP.22
Method FORM MCS (1107samples)
Pf 0.001238 0.001241
Inverse
Measure PPM: 0.00258 PSF: 1.002619
Table 3 Comparison of optimum designs for the stress constraint
-
8/3/2019 Safety Factor and Inverse Reliability
19/26
-
8/3/2019 Safety Factor and Inverse Reliability
20/26
20
one. The structure can be made safer by scaling both w and tby the same factor c. This will1
change the stress and displacement expressed in Eq. (17) and Eq. (18) by a factor ofc3
and c4,
2
respectively, and the area by a factor ofc2. If stress is most critical, it will scale as c
3and the3
PSF will vary with the area, A as:4
1 5.
*0
0
As* s
A
=
(20)5
Equation (20) indicates that a one percent increase in area will increase the PSF by 1.56
percent. Since non-uniform scaling of width and thickness may be more efficient than7
uniform scaling, this is a conservative estimate. Thus, for example, considering a design with8
a PSF of 0.97, the safety factor deficiency is 3% and the structure can be made safer with a9
weight increase of less than two percent, as shown by Qu and Haftka (2003). For a critical10
displacement state, s* will be proportional to 2A and a 3% deficit in PSF can be corrected in11
under 1.5% weight increase. While for more complex structures we do not have analytical12
expressions for the dependence of the displacements or the stresses on the design variables,13
designers can usually estimate the weight needed to reduce stresses or displacements by a14
given amount.15
16
Design for System Reliability by MCS and PSF17
Monte Carlo simulation is a good method to use for system reliability analysis with18
multiple failure modes. The allowable deflection was chosen to be 2.25" in order to have19
competing constraints (Wu et al., 2001). The results are presented in Table 6. It can be20
observed that the contribution of the stress mode to failure probability dominates the21
contribution due to displacement and the interaction between the modes. The details of the22
design process are provided in Qu and Haftka (2004). They demonstrated the advantages of23
using PSF as an inverse safety measure over probability of failure or the safety index as a24
-
8/3/2019 Safety Factor and Inverse Reliability
21/26
21
normal safety measure. They showed that the design response surface (DRS) of the PSF was1
much more accurate than the DRS for the probability of failure. For a set of test points the2
error in the probability of failure was 39.11% from the DRS to the PSF, 96.49% for the DRS3
to the safety index, and 334.78% for the DRS to the probability of failure4
5
6
7
**100,000 samples, Pf1, Pf2, Pf1 Pf2 - Failure probabilities due to stress constraint,8
displacement constraint, and intersection between the modes, respectively.9
10
Design for System Reliability by Moment-Based Method and PPM11
Moment-based methods are restricted to address single mode failures. A common way12
of addressing multi-mode failures using moment-based methods is to prescribe reliability13
indices for each failure modes. Tu et al. (1999) adopted this procedure to solve system14
reliability problems using PMA. They showed that the PPM helps accelerate the15
convergence. The system reliability for the beam design example with separate reliability16
indices for the stress mode and displacement mode are performed by RIA and PMA and the17
results are presented in Table 7 for RIA and Table 8 for PMA. The convergence of the18
objective function is shown in Fig. 7. Comparing Tables 7 and 8, it is clear that the optimal19
values converge more quickly in PMA compared to RIA. Figure 7 illustrates that there is a20
smoother convergence in PMA compared to RIA.21
22
23
24
25
Optima Objective
FunctionPf Safety
Index
Pf1 Pf2 Pf1 Pf2 PSF
*s
w = 2.6041t= 3.6746
9.5691 0.001289
3.01379 0.001133 0.000208 0.000052 1.0006
Table 6 Design for System Reliability by PSF, Qu and Haftka (2004)**
-
8/3/2019 Safety Factor and Inverse Reliability
22/26
-
8/3/2019 Safety Factor and Inverse Reliability
23/26
23
Fig. 7 Convergence of objective function in (a) RIA and (b) PMA
(a) (b)
1
V. Concluding Remarks2The paper described various inverse measures and their usage for reliability based3
design optimization (RBDO). In particular, the relationship between two inverse safety4
measures, PPM and PSF was established. The computation of inverse measure by Monte5
Carlo simulation (MCS) and moment-based techniques was discussed. Several advantages of6
inverse measures were illustrated. They can accelerate convergence in RBDO, increase the7
accuracy of design response surfaces and maintain accuracy with MCS even when the failure8
probability is very low. Moreover, inverse measures can be employed to estimate the9
additional cost required to achieve the target reliability. These features of inverse measure10
make it a valuable resource in RBDO. A simple beam example was used to demonstrate some11
of the benefits.12
Acknowledgments13
This work has been funded in part through NASA Cooperative Agreement NCC3-994, the14
"Institute for Future Space Transport" University Research, Engineering and Technology15
Institute.16
-
8/3/2019 Safety Factor and Inverse Reliability
24/26
24
References1
1) Agarwal, H., Lee, J.C., Watson, L.T., and Renaud J.E, ``A unilevel method for reliability2based design optimization'', Proceedings of the 45th AIAA/ASME/ASCE/AHS/ASC3
Structures, Structural Dynamics, and Material Conference, Palm Springs, CA, April 19-4
22, 2004, AIAA Paper 2004-2029.5
2) Chiralaksanakul, A., and Mahadevan, S., Multidisciplinary Design Optimization Under6Uncertainty, in press, Optimization and Engineering, 2005.7
3) Du, X., Sudjianto, A., and Chen, W., "An Integrated Framework for Optimization under8Uncertainty Using Inverse Reliability Strategy," ASME Journal of Mechanical Design,9
Vol.126, No.4, 2004, pp. 561-764.10
4) Du, X., and Chen, W., Sequential Optimization and Reliability Assessment Method for11Efficient Probabilistic Design,Journal of Mechanical Design, Vol 126, No. 2, 2004, pp.12
225 233.13
5) Elishakoff. I., Interrelation between Safety Factors and Reliability, NASA report CR-142001-211309, 2001.15
6) Elishakoff I., Safety Factors and Reliability: Friends or Foes? Kluwer Academic16Publishers, Dordrecht, The Netherlands, 200417
7) Frangopol, D. M., Maute, K., Life-cycle Reliability-based Optimization of Civil and18Aerospace Structures, Computers and Structures, Vol. 81, No. 7, 2003, pp. 397-410.19
8) Kirjner-Neto, C., Polak, E., Kiureghian, A. D., An Outer Approximations Approach to20Reliability-based Optimal Design of Structures, Journal of Optimization Theory and21
Applications, Vol. 98, No. 1, pp 1-16.22
9) Kiureghian, A. D., Zhang, Y., and Li, C. C., Inverse Reliability Problem, Journal of23Engineering Mechanics, Vol. 120, No. 5, 1994, pp. 1154-1159.24
-
8/3/2019 Safety Factor and Inverse Reliability
25/26
25
10)Kuschel, N., and Rackwitz, R., A new approach for structural optimization of series1systems,Applications of Statisticsand Probability,Vol.2, No 8, 2000, pp987-9942
11)Lee, J. O., Yang, Y. S., Ruy, W. S., A Comparative Study on Reliability-index and3Target-performance-based Probabilistic Structural Design Optimization, Computers and4
Structures, Vol. 80, No. 3-4, 2002, pp. 257-269.5
12)Li, H., and Foschi, O., An Inverse Reliability Measure and Its Application, Structural6Safety, Vol. 20, No. 3, 1998, pp. 257-270.7
13)Melchers, R. E., Structural Reliability Analysis and Prediction, Wiley, New York, 1999.814) Madsen, H. O., Krenk, S., and Lind, N. C., Methods of Structural Safety, Prentice-Hall,9
NJ, 1986.10
15) Tu, J., Choi, K. K., and Park, Y. H., A New Study on Reliability Based Design11Optimization, Journal of Mechanical Design, ASME, Vol 121, No. 4, 1999, pp. 557-12
564.13
16) Lee, T. W and Kwak, B. M., A Reliability-based Optimal Design Using Advanced First14Order Second Moment Method, Mechanics of Structures and Machines, Vol. 15, No. 4,15
1987, pp. 523 542.16
17) Qu, X., and Haftka, R. T., Reliability-based Design Optimization Using Probabilistic17Safety Factor, Proceedings of the 44th AIAA/ASME/ASCE/AHS/ASC Structures,18
Structural Dynamics, and Material Conference, Norfolk, VA, April 7-10, 2003, AIAA19
Paper 2003-1657.20
18) Qu, X., and Haftka, R. T., Reliability-based Design Optimization Using Probabilistic21Sufficiency Factor, Journal of Structural and Multidisciplinary Optimization, Vol. 27,22
No.5, 2004, pp. 314-325.23
19) Qu, X., Singer, T., and Haftka, R. T., Reliability-based Global Optimization of24Stiffened Panels Using Probabilistic Sufficiency Factor, Proceedings of the 45th25
-
8/3/2019 Safety Factor and Inverse Reliability
26/26
26
AIAA/ASME/ASCE/AHS/ASC Structures, Structural Dynamics, and Material Conference,1
Palm Springs, CA, April 19-22, 2004, AIAA Paper 2004-1898.2
20)Royset, J.O., Kiureghian, A. D., Polak, E., Reliability-based Optimal Structural Design3by the Decoupling Approach, Reliability Engineering and System Safety,Vol 73, No. 3,4
pp.213-2215
21) Wu, Y T., Shin Y., Sues, R., and Cesare, M., Safety Factor Based Approach for6Probabilitybased Design Optimization, Proceedings of 42
ndAIAA/ ASME/7
ASCE/AHS/ASC Structures, Structural Dynamics and Materials Conference, Seattle, WA,8
2001, AIAA Paper 2001-1522.9
22) Youn, B. D., Choi, K. K., and Park, Y. H., Hybrid Analysis Method for Reliability-10based Design Optimization, Journal of Mechanical Design, ASME, Vol. 125, No. 2,11
2003, pp. 221-232.12
23)Youn B. D., Choi K. K., and Du. L., Enriched Performance Measure Approach (PMA +)13for Reliability-based Design Optimization, AIAA Journal, Vol. 43, No. 4, 2005, pp. 874-14
884.15
16
17
.18