You are on page 1of 35

Optimal Reliability-Based Aseismic Design

of High-Rise Buildings
Alfredo H-S. Ang
University of California, Irvine, California, USA

David De Leon
Autonomous University of Mexico State, Toluca, Mexico
and
Wenliang Fan
Chongqing University, Chongqing, China

Presentation at National Congress on Earthquake Engineering, Guadalajara, Mexico


21 September 2017
Outline
• Introduction – main theme
• On uncertainties – two broad types: aleatory and epistemic
• Reliability from aleatory uncertainty
• Effects of epistemic uncertainty
• Reliability with statistical confidence levels
• On optimality of structural design
• Life-cycle cost design
• Example of a complex system –
• 3D FEM of a 15-story building in Mexico City subjected to 1985 earthquake
• Confidence level appropriate for system design
• Conclusions
• References cited
Introduction
Proposal -- presenting a practical probability basis for optimal design of
complex structural systems under uncertainty.
Thus far, reliability-based design of structural components has been
established (based on calibration) but not for a complete structure as a
system.
Related question -- For a complex structural system, how do we determine
the target reliability (i.e., safety index) for the optimal design of the
complete structure, as a total system?
A complex system – for sufficient accuracy, requires a digitized model
(e.g.,FEM) with 105 – 106 finite elements or dof’s.
Current practice, invariably requires Monte Carlo simulations, in spite
of its limitations (computation expensive; e.g., for failure probability <10-4).
On Uncertainties
In engineering, uncertainties are expected, and are unavoidable. Main purpose of
reliability engineering is to systematically handle uncertainties.

Practical Approach -- By separating uncertainties into two broad types; namely,


aleatory and epistemic types,
the safety index of a system becomes a random variable, associated with
specified “statistical” confidence.

The aleatory type being the variability or randomness in nature, whereas the
epistemic type is our inability to model reality, referred to, respectively, as “data-
based” and “knowledge-based” uncertainties.

Aleatory type -- measured through statistical observations.


Epistemic type -- seldom measurable; often requires subjective judgments.
In practice, it is sufficient to limit the epistemic type to the errors or inaccuracies
in estimating the mean state of nature. The mean state is of first importance,
whereas errors in its standard deviation (and higher moments) may be secondary.

In developing a design, or in formulating criteria for design, both types of


uncertainty are relevant and must be explicitly included.

Proposed:
a rational basis for including the two types of uncertainty in formulating
reliability-based design of structural systems;
logical process for this formulation -- systematically described and
illustrated.
Being part of nature, the aleatory type cannot be reduced; whereas the epistemic type
may be reduced with improved knowledge of the state of nature.

Real uncertainty is the epistemic type; reducing this uncertainty is seldom practical.
However, minimizing its effects is practically feasible and important in engineering
design.
Aleatory type is information with variability, but is really not an uncertainty.

For consistency and to permit quantitative analysis, the two types of uncertainty must
be expressed in the same terms – i.e., in terms of probability.
The aleatory type -- based on the statistical variability of observed information.
Epitemic type requires subjective modeling -- as a range of possible errors in
predicting the state of the real world.
On Optimality of Structural Design

Optimality in the design of structures depends on the field of application.

In aero-space engineering, optimal design aims to achieve minimum weight.

For land-based or earth-bound structures, cost is a more important design objective.


In this latter regard, although the initial cost is important, minimizing the
total life-cycle cost is more realistic as a design objective.
Expected Life-Cycle Cost, E(LCC), Design of Structures

The expected life-cycle cost of a structure should include all cost items over the
operational life of the structure (50-100 yr). That is, the total E(LCC) will consist of,

CT = CI + CM + CS

where, CI = the initial cost, including design and construction cost;


CM = the maintenance cost, including costs of inspection and repair,
damage and failure costs, retrofit and demolition costs, etc;
CS = societal cost, including possible life-loss cost, cost of injuries, cost
associated with revenue loss, etc.

All cost items must be expressed in “present value”, taking into account the discount
rate of the region.
Total life-cycle cost, CT
E(LCC)

Maintenance cost, Initial cost, CI


CM+CS

Mean safety index for


minimum E(LCC) design

Mean safety index, E()

Figure 1: E(LCC) versus mean safety index for respective cost items
Optimal design
E(LCC)

A specific design

Mean safety index of optimal design

Mean safety index, E(β)

Figure 2: Plot of various designs with varying mean safety indices E(β) and associated E(LCC)
Performance of design due to aleatory uncertainty

Effects of variability in information - must necessarily be expressed in terms of


probability.

For a complex system, the PDEM (Li and Chen, 2009) is effective to obtain
the PDF of the ultimate system performance, Zmax , i.e., fZmax (z)
. . From which the
mean reliability, R, can be assessed through its one-dimensional integration; that is

R= ò fZmax (z)dz (1)


W
where W is the safe domain of the system.
For design, however, the effects of the epistemic type of uncertainty must also be
included. Question – How?

For this purpose. the result from PDEM is useful as follows:

Obtaining the PDF fZmax (z) is of special rrelevance – this provides a rational
and practical basis for including the effects of the epistemic uncertainty in the
reliability-based design of a complex system.
Effects of epistemic uncertainty
Eq.1, namely R= ò fZ (z)dzgives the mean reliability (a single value by PDEM)
max
W
due to the aleatory uncertainty. To include the effects of the epistemic uncertainty,
it is reasonable to limit the errors in the estimation of the mean-value of Zmax. That
is, the range (or variation) of possible mean-values of Zmax represents the effects
of the epistemic uncertainty.

In other words, the mean-value of Zmax , m Z , becomes a random variable; its


PDF may be assumed, judgmentally, to be a lognormal distribution with a
mean of 1.0 and a specified c.o.v. (or equivalent range of values).
Therefore, we have fZmax (z) -- the PDF of Zmax
and -- the PDF of m Z
max

m
Because Z (the mean of Zmax ) is a random variable, the reliability of the
max
system will also be a random variable.
To obtain the PDF of the system reliability requires the convolution
integration of these two PDFs indicated above. That is, by theorem of total
probability, ¥

ò
W
ò
R= fZ m (z) fmZ (m )dzd m
0
(2)

where W is the safe domain of the system.


By simple Monte Carlo simulation, obtains histogram of all possible reliabilities
(or safety indices) of the system -- contains effects of both types of uncertainty
in the estimated system reliability.

Through this histogram, safety index is associated with statistical “confidence”.


A high-confidence level of the safety index -- appropriate and may be specified
for the safe design of a complete system.

A high confidence level in design, therefore, serves also to minimize the effects
of the epistemic uncertainty.
Confidence level for engineering design

Question -- What confidence level is appropriate for the design of a


complex critical system?
To answer this important question, let us examine the confidence level(s) that
underlies good professional practice; particularly in the successful design of existing
major critical systems.

1. Consider the design of offshore oil production platforms -- The industry


standards for design (American Petroleum Institute, 1993; Mexican PEMEX, 2000)
require the safety index for the design of production platforms to be b = 3.3 – 3.5.
In the study of life-cycle cost (LCC) design of offshore platforms, De Leon and Ang (2008)
shows the histogram of b for the minimum LCC design in Figure 6. Whereas Figure 7
summarizes the 90% b for different designs with an optimal 90% b = 3.45. Clearly, in the case of
offshore production platforms, a confidence of 90% is consistent with good professional practice,
i.e., within b = 3.3 – 3.5.
1.8
54
1.6

E(LCC) million USD


52
1.4
75%  90% 
50
1.2 3.21
Frequency

1 48

0.8
46
0.6
44
0.4
42
0.2

0 40
2.3
2.35
2.4
2.45
2.5
2.55
2.6
2.65
2.7
2.75
2.8
2.85
2.9
2.95
3
3.05
3.1
3.15
3.2
3.25
3.3
3.35
3.4
3.45
3.5
3.55
3.6
3.65
3.7
3.75
3 3.1 3.2 3.3 3.4 3.5 3.6 3.7

 90% 
Figure 6: Histogram of b for minimum life-cycle cost design Figure 7: 90% b versus life-cycle cost
2. Design of Cable-Stayed Bridges – In the case of the design of cable-stayed bridges,
consider the particular bridge in Jindo, Korea (Han and Ang, 2008).
For the minimum life-cycle cost design of the bridge, the histogram of the safety index,
β, is shown in Figure 8 , indicating that the mean β is 2.28, whereas the 90% β is 3.23.
1400

1200 Mean 
75% 
90% 

Frequency
1000

800

600

400

200

2.06
0.01
0.22
0.45
0.68
0.91
1.14
1.37
1.6
1.83

2.29
2.52
2.75
2.98
3.21
3.44
3.67
3.9
4.13
4.36
4.59
4.82
Value of safety index, β
Figure 8: Histogram of β for minimum LCC design of Jindo bridge
The safety index in the actual design of the bridge was inferred to have slightly
higher than the 90% β of 3.23; perhaps within 90%-95%. See Figure 9.
1500 1468.14
1434.37

1400

E(LCC) in million USD


1341.26

1300

1223.96
Minimum
1200
Actual Design
1100 1078.29
1031.8
1016.16
985.56 989.76
1000

900
0.8 1.85 2.28 2.67 3.01 3.32 3.85 4.29 4.65

Value of safety index, β

Figure 9: E(LLC) vs β for different designs of the Jindo bridge


Therefore, the confidence level for the minimum E(LCC) design of cable-stayed
bridges appears to be in the range of 90% - 95%.

In summary, it appears that for the optimal design of critical


engineering systems, a confidence level within 90%-95% is
consistent with good professional practice, and may be considered
acceptable to ensure safety for the design of a complete system.
Example of Optimal Design of a Complex System

A 15-Story R/C Building in Mexico City

15-Story R/C building in Mexico City 3D FEM of 15-story building


Building was subjected to two directions of ground motions, EW and NS.

EW Ground Motion NS Ground Motion

Ground Motions of 1985 Mexico Earthquake


2D Elevation Sections of the 15-story Building
Table 1: Details of Column Sizes and Rebars at Various Story Levels
Level Section Stirrup A Stirrup B Details Rebars A Rebars B
(cm) ( below)
Basement to 150X60 # 4 @8” #3 @10” 1 -1 2#10+1#8 1#8
level 3
Level 4 to 125X60 #3 @8” #3 @10” 1-2 2#10 1#8
level 6
Level 7 to 100X60 #3 @8” #3 @10” 1-3 1#10+1#8 1#6
level 8
Level 9 to 85X60 #3 @8” - 2-1 2#8+1#6 -
level 10
Level 11 to 60X60 #3 @10” - 2-2 2#8 -
level 14
Level 15 45X45 and #3 @10” - 2-3 1#8+1#6 -
40X40
Table 2: Concrete initial Young’s modulus (×104MPa)

Distribution Normal Normal Normal


Mean value 3.250 3.150 3.000
Standard deviation 0.325 0.315 0.300
DETAIL 1-1
Rebar DETAIL 2-1
sA
2#10+
1#8 Rebars A
Stirrups A
2#8+
Rebar #3@8”
Stirrups Stirrups B Rebars B 1#6
A sA
#3@10” 1#8
#4@8” 2#10+
1#8
DETAIL2-2
DETAIL 1-2
Rebar
sA Rebars A
2#10 2#8
Stirrups A
#3@10”
Stirrups Stirrups Rebars B Reba
A B 1#8 rs A DETAIL 2-3
#3@8” 2#10
#3@10
DETAIL 1-3 Rebars A
Rebars 1#8+1#6 Stirrups A
A Reba #3@10”
1#10+ rs A
1#8 1#10
Stirrups Stirrups Rebars B +1#8
A B 1#6
#3@8” #3@10

Graphic Details of Columns and Reinforcement Re-Bars
Response Analysis and Optimal Design of Building

3D FEM model of the building is quite involved.


Number of finite elements is over 72,000,
Total number of nodes and dofs, over 63,000 and 200,000, respectively.
3D FE Model of the 15-story Building
Response analysis by PDEM – Used a small sample size, of 135
deterministic response solutions. Accuracy of results? (see below)

For the original, as-built building, the PDF of the ultimate performance
function, Zmax of the building response -- shown in Fig. 13 with twin modes;
and the mean safety index  = 2.73. Fitted Lognormal PDF is also shown
in Fig. 13 with β = 2.77.

Assume the epistemic uncertainty (inaccuracy of the mean-value of Zmaxx )


can be modeled also with a lognormal PDF with mean of 1.0 and a c.o.v.
of 0.10. Convolution integration (by MCS) of this lognormal PDF and the
lognormal of Zmax with mean = 0.9 and standard deviation = 0.035 yields the
histogram of the safety index for the 15-story building, shown in Fig. 14.
Results for Original As-Built Case

15 120

Fitted Lognormal PDF: 100


mean = 0.9
std. dev. = 0.035
10 Computed PDF 80

60

5 40

20

0 0
0.6 0.7 0.8 0.9 1 1.1 1.2 1.5 2 2.5 3 3.5 4

Zmax Safety index β

Figure 13: Computed PDF of Zmax of the building and Figure 14: Histogram of  of the original 15-story building
corresponding fitted lognormal PDF based on lognormal PDF of Zmax
On Accuracy and Effectiveness of PDEM

Only 135 sample points, with corresponding associated probabilities, of the building
responses were used in the PDEM reliability calculations for each case of the building
design.

To verify the accuracy with this small sample size of 135, calculations with 226 sample
points were also performed for the case of the original as-built building design –
the results: with 135 sample points, the range of safety index with 90% to 95%
confidence is  = 3.05-3.18,
whereas with 226 sample points, the corresponding range is  = 3.07-3.20.

The difference between the results for the two sample sizes is around 0.6%, indicating that
with 135 sample points for the original case, give reasonably accurate safety indices for the
15-story building. May assume that the other cases of the same building will have the same
accuracy.
PDEM calculations for different percentages of the original as-built structure:
80%, 90%, 95%, 100%, 105%, 110%, 120%, 130%.

Results (Table 3): safety indices  and expected life-cycle costs, E(LCC)
Table 3: Safety Index  and E(LCC) for all cases
Case (% of original)* 80% 90% 95% 100% 105% 110% 120% 130%

Mean β 0.023 0.465 1.71 2.72 2.85 2.99 3.31 3.74


E(LCC) ** 876 576 82.86 11.24 9.59 8.46 7.34 7.21
90% conf. β 0.025 0.524 1.91 3.05 3.23 3.39 3.71 4.22
E(LCC)** 875 539 55.14 7.46 6.80 6.59 6.69 7.07
95% conf. β 0.026 0.545 2 3.18 3.37 3.55 3.87 4.39
E(LCC)** 874 526 45.67 6.73 6.36 6.30 6.60 7.05
*in percentage of the original as-built structure
**in million U.S. dollars
7.5
7.6
Minimum  = 3.55
7.5
7
7.4
Expected life-cycle cost (Million USD)

Expected life-cycle cost (Million USD)


7.3

7.2 6.5

7.1
Minimum
7  = 3.39 6
6.9

6.8
5.5
6.7

6.6
5
6.5
3 3.2 3.4 3.6 3.8 4 4.2 4.4
3 3.2 3.4 3.6 3.8 4 4.2 4.4
95% confidence 
90% confidence 

Plots of alternative designs with 90% confidence  Plots of alternative designs with 95% confidence 
versus respective E(LCC) versus respective E(LCC)

With 90-95% confidence, the optimal design of the 15-story building would need a safety index of  = 3.39-3.55
which is 110% of the original design; i.e., to obtain the minimum E(LCC) design would require 110%, of the original
as-built building. The original as-built structure was designed with a safety index of  = 3.05-3.18 with the same
90-95% confidence.
Conclusions
The main conclusions of this study may be summarized as follows:
1. The required safety in the design of a structural system cannot be prescribed as that of a
structural component. Each structural system is unique; thus, the required safety for its
design must be determined independently for each system.
2. A procedure for the optimal design of a structural system is proposed; it is based on
achieving the minimum expected life-cycle cost design of the structure.
3. For reliability assessment and design of complex systems, the PDEM is a most effective
method.
4. As illustrated in the example of the 15-story R/C building, the proposed procedure shows
that the building in Mexico City could have been designed for earthquake resistance with
higher safety and at lower cost (minimum saving of around $0.5million).

In addition to the above conclusions, it is well to emphasize that for the accuracy of results,
calculations with 3D models (such as 3D finite element models) of engineering systems may
be necessary, especially for complex systems; results obtained with corresponding simplified
2D models could be misleading.
References Cited
American Petroleum Institute (API), 1993, “Recommended Practice for Planning, Designing and Constructing Fixed
Offshore Platforms – Load and Resistance Factor Design”, API RP 2ª LRFD, First Edition, July, Washington, D.C.

Ang, A.H-S. and Ma, H-F., 1981, “On the reliability of structural systems”, Proc. Third ICOSSAR, Trondheim, Norway.

Chen, JB., and Li, J. 2007, Development-process-of-nonlinearity-based reliability evaluation of structures, Probabilistic
Engineering Mechanics, Elsevier, 22, 267-27J

Chen, JB., Ghanem, and Li, J., 2009, “Partition of the probability-assigned space in probability density evolution
analysis of nonlinear stochastic structures”, Probabilistic Engineering Mechanics, Elsevier, 24, 27-42.

De Leon, D. and A.H-S. Ang, 2008, Confidence bounds on structural reliability estimations for offshore platforms,
Journal of Marine Science and Technology, Vol. 13, No 3

Ellingwood, B. and Galambos, T.V., 1982, “Probability-based Criteria for Structural Design”. Structural Safety,
Elsevier Scientific Publishing

Han, S.H. and Ang, A. H-S., 2008, Optimal design of cable-stayed bridges based on minimum life-cycle cost
Proceedings. IABMAS’08, Seoul, Korea

Li, J. and Chen, JB. 2009, Stochastic Dynamics of Structures, John Wiley & Sons, 103(GT11), 1227-1246.

Petróleos Mexicanos (PEMEX), 2000. “Diseño y Evaluación de Plataformas Marinas Fijas en la Sonda de Campeche”,
NRF-003-PEMEX-2000, Rev. 0, Committee of Normalization for PEMEX and Subsidiaries Organisms.
Thank you for your attention!

You might also like