Latest Trends on Engineering Mechanics, Structures, Engineering Geology

Advanced Modeling Approaches for Reliability Analysis of Steel Structures

ABAYOMI OMISHORE
Department of Structural Mechanics
Brno University of Technology, Faculty of Civil Engineering
Veveří Str. 95, Brno
CZECH REPUBLIC
[email protected]; http://www.vutbr.cz

Abstract: One important approach in the identification of uncertainty is sensitivity analysis. The sensitivity analysis is
crucial for building, understanding and applying complex mathematical models to reliability problems of building
structures. Attainment of limit state is generally a random event, which is studied in the reliability theory by means of
the probability theory and mathematical statistics. Methodology based on ANOVA variance-based techniques is used
for the sensitivity analysis. The sensitivity indices were evaluated applying the Monte Carlo method. Sensitivity
analysis is used to identify the dominant input random quantities and their higher order interaction effects. The
imperfections which interact and may thus generate extreme values of resistance capacity have been identified.
The results obtained may be utilized in standards for design.

Key-Words: Reliability, Frame, Column, Structure, Sensitivity, Safety.

1 Introduction The system response is understood as the ultimate

The identification and representation of uncertainty is resistance capacity and maximal deformation of
recognized as an essential component in model structure. The present paper is aimed at the sensitivity
applications [1]. The sensitivity analysis is a analysis of the ultimate resistance capacity of a steel
fundamental tool in the building and understanding of plane frame, see Figure 1. It is focused on the application
mathematical models as it provides information on the of such sensitivities to obtain new modeling approaches
behavior of the underlying simulated system [1]. The for problems of steel structures. This information can be
sensitivity analysis evaluates how the variations in the used to support the development of optimal design and
model output can be apportioned to variations in model the reduction or simplification of existing ones.
parameters [2]. Nowadays, many sensitivity analysis Sensitivity analysis is often carried out to identify the
techniques are available [3]. In the case of simple most relevant material and geometrical characteristics,
mathematical models, Taylor series expansion can be on which of them it is therefore reasonable to
used to approximate the model, and the analytical concentrate the efforts during the calibration activity.
differential sensitivity index can be derived [1].
However, it would be difficult to use the Taylor series
approximation for complex models; there are needed 2 Sensitivity and Statistical Analysis
more advanced techniques [1]. The sensitivity analysis The sensitivity analysis evaluates how the variations in
of steel frame structures is aimed at safety and the model output can be apportioned to variations in
serviceability assessment in the limit state methods. model parameters [3]. In the presented article, the
sensitivity analysis is applied to the study of the
influence of initial imperfections on the resistance
capacity. The sensitivity analysis can be generally
divided into two groups: (i) deterministic sensitivity
analysis and (ii) stochastic sensitivity analysis [4-13].
The variance-based techniques are sometimes called
ANOVA techniques for ANalysis Of VAriance [14].
With regard to the limit states of structures, the ultimate
or fatigue resistance and deflection are frequently
considered to be the output quantities [15-21]. Stochastic
methods are based upon various assumptions, and it is
difficult to compare the results. However, each of the
Fig.1 Steel plane frame geometry methods has its informative capability of a different
type.

ISSN: 1792-4294 172 ISBN: 978-960-474-203-5

 Latest Trends on Engineering Mechanics, Structures, Engineering Geology

2.1 Input Random Quantities

Experimentally obtained material and geometrical Sensitivity index Sij expresses the effect of doubles
characteristics of steel products made by a dominant on the monitored output. Other Sobol’ sensitivity indices
Czech producer, see [22, 23] were applied to the enabling the quantification of higher order interactions
problem solved. Input quantities Xi of left column are may be expressed similarly.
yield point fy1, cross-sectional height h1, cross-sectional
width b1, web thickness tw1, flange thickness tf1 and ∑S +∑∑S +∑∑∑S
i ij ijk + ... + S 123...M = 1 (3)
Young modulus E1. Input quantities of right column are i i j >i i j >i k > j

fy2, h2, b2, tw2, tf2, E2. Input quantities of cross beam are
fy0, h0, b0, tw0, tf0, E0. All the input parameters Xi, given The number of members in (13) is 2M-1, i.e., for M=3,
synoptically in Table 1, are statistically independent of we obtain 7 sensitivity indices S1, S2, S3, S12, S23, S13,
one another. S123; for M=18, we obtain 262143 sensitivity indices; it is
excessively large for practical usage. The computational
Table 1: Statistic characteristics of the input quantities demand represents the main limitation in the
No. Member Symbol Mean value Std. deviation determination of all members of (3).
1. h1 * 100.09 mm 0.44 mm
b1 * 2.3 Results of Sensitivity Analysis
2. 49.6 mm 0.49 mm
Column

The Monte Carlo method was employed for the

Left

*
3. tw1 4.74 mm 0.18 mm calculation of sensitivity indices. The model output Y is
*
4. tf1 6.75 mm 0.31 mm the resistance of the column under tension calculated in
6. fy1 *
297.3 MPa 16.8 MPa each simulation run of the Monte Carlo method.
7. h0 *
120.11 mm 0.53 mm Resistance of column is yield point multiplied by cross
* section area. Resistance of frame is minimum of
8. Cross b0 58.81 mm 0.57 mm
*
resistance of left and right columns. Nine thousand
9. Beam tw0 5.37 mm 0.2 mm simulation runs were applied in our study. Results of
*
10. tf0 7.64 mm 0.35 mm sensitivity analysis depicted in Figure 2 illustrate that the
12. fy0 *
297.3 MPa 16.8 MPa variance of yield point of both columns has the greatest
13. h2 *
100.09 mm 0.44 mm effect on the variance of ultimate resistance. A further
* dominant quantity is the flange thickness of both
14. b2 49.6 mm 0.49 mm
columns of the I profile. The sum of higher-order
Column
Right

*
15. tw2 4.74 mm 0.18 mm sensitivity indices is approximately 0.25. It means that
*
16. tf2 6.75 mm 0.31 mm the interactions of higher orders between input random
18. fy2 *
297.3 MPa 16.8 MPa quantities also have a significant effect.
* Histogram

2.2 Sensitivity Indices

The description of Sobol’ decomposition theory is listed,
e.g., in [24]. The Sobol’ first order sensitivity indices
may be written in the form:

V (E (Y X i ))
Si = (1)
V (Y )

Sobol’ proposed an alternate definition

S i = corr (Y , E (Y X i )) based on the evaluation of the
correlation between output random value Y and the
conditional random arithmetical mean E (Y X i ) .
Analogously, we can write the second order sensitivity
indices:

V (E (Y X i , X j ))
S ij = − Si − S j (2)
V (Y ) Fig.2 Sensitivity analysis results

ISSN: 1792-4294 173 ISBN: 978-960-474-203-5

 Latest Trends on Engineering Mechanics, Structures, Engineering Geology

2.4 Statistical Analysis

In Fig. 3, there are presented simulations runs of the These conclusions are of great importance for the
Monte Carlo method. All the input quantities from Table calculation of the design value of resistance capacity.
1 were considered to be random characteristics. The The design value can be calculated, according to the
arithmetic mean of resistance capacity is 313.85 kN; the standard EN1990, as 0.1 percentile by the value
standard deviation is 17.18 kN; standard skewness is - 260.8 kN, see Fig. 4. The design value calculated
0.097; standard kurtosis is 3.08. Taking low values of according to the EUROCODE 3 is 249.1 kN. The design
standard skewness and of standard kurtosis into value 260.8 kN is higher than 249.1 kN; it means that the
consideration, it can be supposed that the runs of Monte design according to the EUROCODE 3 is safe. From the
Carlo method can be approximated by the Gauss practical point of view, the differences between both
probability density function. This assumption was values are, however, very small.
confirmed by three distribution tests carried out
independently from each other, namely by Chi-square 2.5 Limit States Methods
test, Kolmogorov-Smirnov test and Anderson-Darling
test. The findings that the hypothesis of Gauss stochastic What is the connection among the results of sensitivity
model of resistance capacity should not be rejected were analysis, statistical analysis and reliability of design
the conclusions of every test. like? The reliability of steel structures can be increased
by decreasing the variance of input geometrical and
material parameters which are of random character in
general. An objection can be made saying that it is not
sufficient to focus the attention only to the variance,
nevertheless, with exception of yield strength, the
majority of mean values are approximately equal to
characteristics values. The sensitivity analysis supplies
us the information for which of parameters the decrease
of variance will have the largest effect on the decrease of
the random ultimate resistance. The presented analyses
can be applied analogously also to the serviceability
limit state problems, e.g., when evaluating the maximum
deflections of bridges. Also the deformation of struts in
structural system can be significant from the point of
view of the serviceability limit state; it is one of the
problems which is paid relatively low attention in the
field of the reliability of steel structures.

Fig.3 Monte Carlo runs of resistance capacity

3 Conclusion
The past 33 years have witnessed the advent of
numerous computer models in different application areas
including physics, chemistry, environmental sciences,
and ecology [1]. At the beginning, the structures of
models were relatively simple and the outputs of models
were assumed to be error free. Today, more and more
complex models have been developed, and at the same
time, the uncertainty influence analysis methods
enabling the identification of sophisticated model
dependences [3] are getting improved [25]. One
important distinction between Sobol' and classical
sensitivity is that the Sobol' sensitivity analysis detects
interactions of input quantities through the second and
higher order terms, while classical sensitivity methods
give only derivatives with respect to single quantities.
The imperfections which interact and may thus
Fig.4 Histogram and design value of resistance capacity generate extreme values of resistance capacity have been

ISSN: 1792-4294 174 ISBN: 978-960-474-203-5

 Latest Trends on Engineering Mechanics, Structures, Engineering Geology

identified. In the frame with columns under tension, the J. Cariboni, D. Gatelli, M. Saisana, S. Tarantola,
yield point of the left column, yield point of the right Global Sensitivity Analysis Guiding the Worth of
column and the mutual interaction of these quantities Scientific Models, New York: John Wiley and
had a significant effect on the ultimate resistance as Sons, 2007.
corroborated by the second order sensitivity index (2). [4] C. Szymczak, Sensitivity Analysis of Thin-Walled
the flange thickness of the left and the right columns is Members, Problems and Applications, Thin-
next in order. The other input random quantities Walled Structures, Vol.41, No.2-3, 2003, pp.271-
influence the resistance capacity only little. 290.
It is evident that the implementation of probabilistic [5] Z. Kala, Sensitivity Analysis of the Stability
methods in reliability assessment presents a number of Problems of Thin-Walled Structures, Journal of
problems. With the development of the algorithms Constructional Steel Research, Vol.61, No.3,
optimizing problems of structures, these procedures can 2005, pp. 415-422.
contribute to a qualitative improvement of the safety and [6] Z. Kala, Sensitivity Assessment of Steel Members
reliability analysis of structures [26-32]. In order for the under Compression, Engineering Structures,
mathematical models employed for reliability analysis to Vol.31, No.6, 2009, pp. 1344-1348.
provide realistic information on the reliability of real [7] Z. Kala, J. Kala, Variance-Based Sensitivity
steel structures, it is necessary that input random Analysis of Stability Problems of Steel Structures
quantities are obtained from experimental research on using Shell Finite Elements and Nonlinear
ample samples. Many new measurements of Computation Methods, In Proc. of the 2nd WSEAS
noncommercially aimed research are available for model Int. Conf. on Engineering Mechanics, Structures
parameters, and therefore the probability functions or at and Engineering Geology (EMESEG ’09), Rodos
least the variances of the parameters are known. Island (Greece), 2009, pp.89-94.
The input random parameters may be generally [8] Z. Kala, J. Kala, J. Sensitivity Analysis of
divided into two basic groups [3]. The first group Stability Problems of Steel Structures Using Shell
includes those variables whose statistical characteristics Finite Elements and Nonlinear Computation
can be positively influenced during production (yield Methods, WSEAS Transactions on Applied and
strength, geometric characteristics) and those that are not Theoretical Mechanics, Vol.4, No.3, 2009, pp.
sufficiently sensitive to changes in production 105-114.
technology (e.g. variability of Young’s modulus E). The [9] L. Puklicky, Imperfection Sensitivity and
first group of quantities may be further subdivided into Insensitivity of Building Structures, In Proc. of the
two subgroups: (i) quantities for which mean value and 14th WSEAS International Conference on Applied
standard deviation can be changed by improving Mathematics (MATH '09), Puerto de la Cruz,
production quality [3]. Examples of such variables Canary Islands (Spain), pp.89-92, 2009, ISBN:
include Young’s modulus; (ii) variables, the mean value 978-960-474-138-0, ISSN: 1790-2769.
of which cannot be significantly changed, because it [10] J. Melcher, M. Škaloud, Z. Kala,
should approximately correspond to the nominal value, M. Karmazínová, Sensitivity and Statistical
for e.g. geometric characteristics of profile dimensions. Analysis Within the Elaboration of Steel Plated
Girder Resistance, Advanced Steel Construction,
Vol.5, No.2, 2009, pp. 120-126.
Acknowledgement [11] H.A. Mang, X. Jia, G. Hoefinger, Hilltop Buckling
The article was elaborated within the framework of as the Alfa and Omega in Sensitivity Analysis of
projects of GAČR 103/08/0275, AVČR IAA201720901 the Initial Postbuckling Behavior of Elastic
and CIDEAS 1M68407700001(1M0579). Structures, Journal of Civil Engineering and
Management, Vol.15, No.1, 2009, pp. 35-46.
References: [12] A. Iosif, I. Sarbu, Computing of Cavitation
[1] C. Xu, G. Gertner, Extending a Global Sensitivity Characteristic and Sensitivity Curves for Francis
Analysis Technique to Models with Correlated Pump-Turbine, WSEAS Transactions on Fluid
Parameters, Computational Statistics & Data Mechanics, Vol.5, No.2, 2010, pp. 55-65.
Analysis, Vol.51, No.12, 2007, pp.5579-5590. [13] H. Shareef, A. Mohamed, K. Mohamed,
[2] R. Confalonieri, Monte Carlo Based Sensitivity Sensitivity of Compact Fluorescent Lamps during
Analysis of Two Crop Simulators and Voltage Sags: An Experimental Investigation,
Considerations on Model Balance, European WSEAS Transactions on Power Systems, Vol.5,
Journal of Agronomy, Vol.33, No.2, 2010, pp.89- No.1, 2010, pp. 22-31.
93. [14] SR. Arwade, M. Moradi, A. Louhghalam,
[3] A. Saltelli, M. Ratto, T. Andress, F. Campolongo, Variance Decomposition and Global Sensitivity

ISSN: 1792-4294 175 ISBN: 978-960-474-203-5

 Latest Trends on Engineering Mechanics, Structures, Engineering Geology

for Structural Systems, Engineering Structures, D. Žilionienė, Evaluation of Soil Shear Strength
Vol. 32, No.1, 2010, pp. 1-10. Parameters Via Triaxial Testing by Height Versus
[15] M. Kotelko, Load-capacity Estimation and Diameter Ratio of Sample, The Baltic Journal of
Collapse Analysis of Thin-Walled Beams and Road and Bridge Engineering, Vol.IV, No.2,
Columns - Recent Advances, Thin-Walled 2009, pp. 55-60.
Structures, Vol.42, No.2, 2004, pp.153-175. [27] E.R. Vaidogas, V. Juocevičius, Assessment of
[16] M. Kotelko, Load-carrying Capacity and Energy Structures Subjected to Accidental Actions using
Absorption of Thin-Walled Profiles with Edge Crisp and Uncertain Fragility Functions, Journal
Stiffeners, Thin-Walled Structures, Vol.45, No.10- of Civil Engineering and Management, Vol.15,
11, 2007, pp.872-876. No.1, 2009, pp.95-104.
[17] Z. Kala, Stability Problems of Steel Structures in [28] V. Gribniak, G. Kaklauskas, S. Idnurm, D.
the Presence of Stochastic and Fuzzy Uncertainty, Bačinskas, Finite Element Mesh Size Effect on
Journal of Thin–Walled Structures, Vol.45, No. Deformation Predictions of Reinforced Concrete
10-11, 2007, pp. 861-865. Bridge Girder, The Baltic Journal of Road and
[18] Z. Kala, Fuzzy Probability Analysis of the Fatigue Bridge Engineering, Vol.V, No.1, 2010, pp.19-27.
Resistance of Steel Structural Members under [29] M. Kotelko, K. Kowal–Michalska, J. Rhodes,
Bending, Journal of Civil Engineering and Stability of Structures, Thin-Walled Structures,
Management, Vol.14, No.1, 2008, pp. 67-72. Vol.45, No.10-11, 2007, pp. 809-809.
[19] L. Puklicky, Z. Kala, Application of Fuzzy [30] M. Kotelko, K. Kowal-Michalska, T. Kubiak,
Analysis to Stability Problems of Steel-Concrete Z. Kolakowski, R. Gradzki, Estimation of Load-
Structural Members, In Proc. of the 2nd WSEAS Carrying Capacity of Multi-layered Plated
Int. Conf. on Engineering Mechanics, Structures Structures, Thin-Walled Structures, Vol.46, No.7-
and Engineering Geology (EMESEG ’09), Rodos 9, 2008, pp. 1003-1010.
Island (Greece), 2009, pp.111–113. [31] V. Jankovski, J. Atkociunas, SAOSYS Toolbox as
[20] M. Karmazínová, J. Melcher, Z. Kala, Design of MATLAB Implementation in the Elastic-Plastic
Expansion Anchors to Concrete Based on Results Analysis and Optimal Design of Steel Frame
of Experimental Verification, Advanced Steel Structures, Journal of Civil Engineering and
Construction, Vol.5, No.4, 2009, pp. 390–405. Management, Vol.16, No.1, 2010, pp. 103-121.
[21] Z. Kala, L. Puklicky, A. Omishore, M. [32] V. Ungureanu, M. Kotełko, R.J. Mania, D.
Karmazínová, J. Melcher, Stability Problems of Dubina, Plastic Mechanisms Database for Thin-
Steel-Concrete Members Composed of High Walled Cold-Formed Steel Members in
Strength Materials, Journal of Civil Engineering Compression and Bending, Thin-Walled
and Management, Vol.16, No.3, 2010. (in print) Structures.(in print) doi:10.1016/j.tws.2010.01.004
ISSN 1392-3730 print / ISSN 1822-3605 online
[22] J. Melcher, Z. Kala, M. Holický, M. Fajkus,
L. Rozlívka, Design Characteristics of Structural
Steels Based on Statistical Analysis of
Metallurgical Products, Journal of Constructional
Steel Research, Vol.60, No.3-5, 2004, pp.795-808.
[23] Z. Kala, J. Melcher, L. Puklicky, Material and
Geometrical Characteristics of Structural Steels
Based on Statistical Analysis of Metallurgical
Products, Journal of Civil Engineering and
Management, Vol.15, No.3, 2009, pp. 299–307.
[24] I.M. Sobol', Sensitivity Estimates for Nonlinear
Mathematical Models, Matematicheskoe
Modelirovanie 2, pp.112–118, 1990. (in Russian,
translated into English in Sobol' 1993)
[25] M. Krolak, Z. Kolakowski, M. Kotelko, Influence
of Load-Non-Uniformity and Eccentricity on the
Stability and Load-Carrying Capacity of
Orthotropic Tubular Columns of Regular
Hexagonal Cross-Section, Thin-Walled Structures,
Vol.39, No.6, 2001, pp. 483-498.
[26] J. Amšiejus, N. Dirgėlienė, A. Norkus,

ISSN: 1792-4294 176 ISBN: 978-960-474-203-5