Structural Safety, 1 (1982) 3-13

Elsevier Scientific Publishing Company, Amsterdam - Printed in The Netherlands

SYSTEM RELIABILITY DEVELOPMENTS IN STRUCTURAL

ENGINEERING
F. Moses
Case Institute of Technology, Case Western Reserve University, Cleveland, OH (U.S.A.)

(Received February 25, 1982; accepted March 28, 1982)

ABSTRACT
Two major limitations occur in present
structural design code developments utilizing
reliability theory. The notional system reliabilities may differ significantly from calibrated
component reliabilities. Secondly, actual failures
are often due to gross errors not reflected in
most present code formats. A review is presented of system reliability methods and further
new concepts are developed. The incremental
load approach for identifying and expressing
collapse modes is expanded by employing a
strategy to identify and enumerate the significant structural collapse modes. It further isolates the importance of critical components in

the system performance. Ductile and brittle

component behavior and strength correlation is
reflected in the system model and illustrated in
several examples. Modal combinations for the
system reliability are also reviewed. From these
developments a system factor can be addended
to component safety checking equations. Values
may be derived from system behavior by substituting in a damage model which accounts for
the response range from component failure to
collapse. Other strategies are discussed which
emphasize quality assurance during design and
in-service inspection for components whose behavior is critical to the system reliability.

INTRODUCTION

were based on second-moment approximations (means and coefficients of variation) but

recent studies have utilized more sophisticated methods incorporating frequency distributions to assess component reliabilities with
safety indices [1,5].
Two major limitations exist in present design code development: (a) The overall structural system reliability can significantly differ
from the calibrated component reliabilities for
beams, columns, connections, piles, etc.; (b)

Structural reliability theory has had considerable impact in recent years on structural
design. For example, partial safety factors
based on load and strength uncertainties have
been proposed in the United States, Canada
and Great Britain for building, bridge and
offshore platform design codes [1-4]. A criterion has evolved to achieve uniform component reliabilities. Initially, reliability analyses

Actual structural failures are often not related

to load and strength parameters checked by
design codes, but are due to accidents, fabrication mistakes, changes and gross human
errors not under the control of either the
reliability formulation or the design codification.
This paper reviews present system reliability methods and develops new concepts. The
aim is both the solution of notional or calculated system reliabilities and to identify critical structural components for quality assurance. The latter identification can help to
reduce gross error failures. Some examples
may clarify the need for such comprehensive
system reliability methods.
(1) A recent series of long-span roof failures
has led to investigation and review of present
codes for such application. One report suggests that safety factors be modified to emphasize critical components [6]. This is a clear
recognition of system analysis but no methodology is currently available to balance such
safety factors.
(2) Pile foundations--due to large uncertainties in pile-soil behavior the calculated
safety indices for piles are often lower than
for other members. An increase in their safety
indices, however, may be unnecessary since
individual pile failure is rarely serious. In
most instances, the ultimate foundation
capacity has considerably larger margins than
the single most heavily loaded pile Controlling
design.
(3) Fatigue failure--several instances of reported cracks in highway and offshore platform members may not always have serious
consequences because of system behavior. A
code specification of redundancy, however,
can be confusing as in bridge codes [7], since
redundancy may only be satisfied by a rational system analysis.
(4) In-service assessment--certain events
may require a risk analysis of existing structures. For example, in-service inspection which
detects flaws, the occurrence of an accident

(collision, fire, damaged element, etc.) or

changes in performance data. An appropriate
strategy can often be devised by combining
estimated changes in strength capacity with
any experienced load history and an appropriate system analysis modet. For example, comparing: (a) immediate closure, (b)
temporary operation and repairs based on
load models with shorter exposures or (c)
unconstrained operation supplemented by
more frequent inspections.
The considerable research in system reliability is due to such situations arising in
nuclear plants, offshore platforms, highway
bridges and other structures where ultimate
capacity rather than component: response is
the important factor. In addition, there are
field data in a Bayesian sense to be utilized
when system performance is introduced [8].

SYSTEM RELIABILITY BACKGROUND

In recent years there have been ~mproved
solutions to system reliability problems [9-12].
To appreciate the complexity and limitations
of such analyses several basic definitions are
reviewed [9]. Structural topology may be classified as series (or weakest-link) in which any
component failure causes system collapse, and
parallel (or fail-safe) in which alternate load
paths exist. Ductile and brittle component
models define the extremes of post-failure
behavior after an element reaches its maxim u m (random) capacity. Statistical correlations between collapse failure modes arise
through common loading variables or correlated strengths due to similar fabrication.
material and even modelling uncertainties.
Although any structure may combine all
these topological, post-failure behavior and
correlation characteristics these models help
identify the range of system problems. For
example, a parallel (redundant) system loses
its advantage over a series if the load uncertainty greatly exceeds strength uncertainty/.

Similarly, even a redundant brittle system may

possess advantages if strength uncertainty is
sufficiently high to minimize the likelihood of
a cascading sequence of failures. There have
been several studies which investigated these
idealized models [9,10,12,13].
For application to large scale realistic
structures the system analysis should be divided into two parts: (a) Engineering modelling, which means identification, description
and enumeration of the various collapse
modes, and (b) Probabilistic calculations, to
determine individual mode failure probabilities (or safety indices) and then combine them
into an overall system assessment. For simple
structures such as one story frames (plastic
analysis) the identification and enumeration
of collapse modes in terms of loads and component resistances is straightforward. For
large structural frameworks the search for
critical modes is difficult and it is not possible
to assure that all significant failure modes
have been found [10]. The modelling must
usually seek collapse mode expressions in the
form:

gj-- ZCjkR ~ -- Y.biS,,

(1)

where the failure function g j < 0 means the

j t h collapse mode has occurred. R~, is the k t h
component strength and S i is the i th load
term. Nonlinear system failure functions have
also been studied, but are more difficult to
formulate [12]. The system reliability R s is
expressed as:
R s = Probability [all g j > 0].

(2)

Collapse mode events are correlated

through loading and resistances so an exact
solution to Eqn. 2 is usually impossible.
Several investigations considered this combination problem by either finding approximate solutions or bounds for R s [10,11,13-15].
Monte Carlo simulation is the only precise
method to combine the statistics of each collapse mode to estimate the system reliability
R s [16]. Estimates of R s using Clark's method

of deriving statistical moments of the minim u m of correlated random variables has been
reported with reasonable accuracy [16].
Another approach is to use quadrature techniques [17]. Recently, there have been considerable improvements in system reliability
bounds. These upper and lower bounds on
system reliabilities include correlation [13]. A
detailed comparison is underway by Schueller
and Co-workers [23] to assess and compare
the computational efficiency and accuracy of
collapse mode probabilities and their combination into R s for simple frames analyzed by
plastic design. The remaining part of this
paper therefore concerns the modal identification with special emphasis on large-scale analysis.

A PRACTICAL EXPRESSION OF SYSTEM

RELIABILITY
To investigate a general system reliability
approach consider the following incremental
analysis techniques [16,18]. It identifies a
failure mode by following a load path from
initial component failure to system collapse
and leads to a linear failure expression such
as eqn. 1. This approach was also applied to a
platform jacket structure [19]. The steps may
be briefly outlined as follows:
(1) Analyze the intact structural system.
(2) Find the critical component based on
mean resistances and the load factor to reach
component failure.
(3) Remove this component from carrying
additional loading and reanalyze the system
to find the next component failure.
(4) Repeat the reanalyses, removing at each
step a component that has failed and continue
until the system collapses.
(5) The outcome is a resistance expression
which can be combined with the load variable.
In general, the failure path is guided by
mean component strengths or some similar

criteria. The advantage is that it can incorporate ductile, brittle or intermediate component behavior and statistical correlation. In
this method loading is incremented proportionally with a factor r. Thus, the total load at
collapse comprises load increments r~, r2, r>
etc. corresponding to successive component
failures. Each component is expressed by its
utilization strength R~. Let a,:r~ be the utilization of component i during the j th load increment. The relationship between component
strength and load increments can be expressed by a utilization matrix A with terms
a,j. To illustrate for ductile components let
the sequence of component failures be R]', R';.
... The s,,stem behavior may be expressed as:

Re

a~

a21

a22

...

rl
r2

0
aml

am2

r,~7
(3)

System failure occurs when member m fails.

A zero term in the A matrix indicates a member is no longer loaded after exhausting its
utilization ratio. [Brittle or semi-brittle behavior can also be included in the modelling [16]
and is illustrated below.]
The next step expresses the load increments
in terms of component strength by solving
eqn. 3. The sum of load increments gives the
system resistance at collapse:
R s = Y~ri= ?ECiR e,

LOAC

LOA[

I PEAK A

\ /

i/

i/

i:

RESPONSE
(a)

:L}

RESPONSE

Fig. 1. Load-response curves: (a) Ductile system; (b)

Brittle system.

lapse. In the instance of brittle behavior or

unloading after component failure the incremental load matrix is modified for loads
transferred back into the system. In this situanon there may be several peaks occurring
prior to collapse as shown in Fig lib). The
m a x i m u m load corresponding to system collapse capacity is the largest of the individual
peak loads. For illustration, the two parallel
m e m b e r system is shown m Figure2 along
with a Venn diagram showing the failure region for strength of members 1 and 2 under
fixed load L. Failure occurs if R~ and R 2 are
within the shaded region. The dotted curve
bounds the failure region for ductile behavior.
The discontinuties in brittle failure regions
make it harder to formulate than ductile reliability models (see example 3 below).

{4)

where C i is the sum of terms in the i th colu m n of the inverse of matrix A [18]. Equation
4 expresses the collapse mode in terms of
component strengths. It thus provides the
modal mean and variance including strength
correlation and subsequently combines with
the loading variables to give the g function.
A ductile system may be expected to follow
a load deflection curve such as in Fig. l(a)
with load monotonically increasing until col-

k,,

L /"V'

R~

/
1

/// ~.

////////~

2,

L/2

Fig. 2. Two-member parallel example and associated

Venn diagram (Solid line--britile, dotted line--ductile).

two had failed we would obtain:

EXAMPLES
Several simple examples will describe the
incremental technique. Also, the results with
brittle components and partial failure sequences will be described.
Example 1: Consider the three parallel
load-sharing members shown in Fig. 3 which
by simple inspection give a system capacity
equal to the sum of member strengths 1, 2
and 3. Assume the failure sequence is 1, 2 and
then 3. The sequence is unimportant if behavior is linear between load increments. With an
intact system the load is shared equally or
all=a21 =a31 = 1/3. After 1 fails 2 and 3
share equally and after 2 fails member 3
carries the entire load. Equation 3 becomes:

R'i

R;

1/3

1/2

r2 .

1/3

1/2

r3

(S)

Solving for r and summing gives:

R s = r~ +

r2

+ r 3 = R{ + R~ + R ; ,

(6)

which is the expected answer. Note that if the

analyses were terminated after only member

// "////

"///// //

R s = r 1+ r 2 = R

1 + 2 R 2.

(7)

The accuracy of truncated solutions depends

on the additional remaining capacity and the
variance and correlation of deleted component strengths.
Example 2: If we look at the example in
Fig. 4 with a failure sequence determined from
incremental loading of 1, 2 and 3 we obtain
for eqn. 4:
R~ ]
Ri

1/2
=

O lrl 1
1/2

R~

1/2

r3

which gives for system capacity:

R s = r I + r 2 + r 3 = R~ + R~.

(9)

Equation 9 corresponds to one of the two

collapse modes of the system, the other being
components 2 and 4 failing. The failure of
c o m p o n e n t 2 before 3 does not affect the
system expression capacity. However, a
strategy is needed to find additional collapse
/ z

' I

II

Ii11/

[ ,.

1
Fig. 3. Three member parallel example.

(8)

Fig. 4. Four member-two tier example.

modes since the incremental loading technique converges to a single collapse mode
defined by mean component strengths. This
limitation is removed in the next section.
Example 3: A brittle behavior is illustrated
with the model in Fig. l(b) with failure sequence 1 and 2. In this case, loading in member 2 includes load increment I plus load
transfer caused by member 1 failing, plus load
increment 2. or:
R'~ = r ] / 2 .

(IO)

R~ = r ] / 2 - - R~ + r2.

(11)

LOAD

._d

/
/
/
/

/
LATERAL TOP DEFLECTION

Fig. 5. Load response curve for X-braced frame of offshore platform [ 19].

Two peaks must be checked in this instance.

Solution of eqns. 10 and 11 gives:

r I - 2R~.
r 2 - R~

(12)

2R~.

(13)

As illustrated in Fig. l(b), the system capacity

found from sequence 1-2. R~-2
S ,IS
R~- 2 = M a x [ r l , r 1+r2] = M a x [ 2 R ~ , R ~ ] .

(14}

If the other failure sequence, namely 2-1. ~s

evaluated in the same way we have:
R 2'

= Max[R~, 2R~].

(15)

Since for multiple failure modes the system

assessment requires the smallest of any capacity we have
RSystem ~--- Min(Max(2R~, R~ ), Max(R~, 2R~)).

(16)
Equation (16) provides the exact representation of the shaded portions of the failure
region in the Venn diagram in Fig. 2.
Applications of the incremental loading approach to more complex truss and frame
structures have been reported [16,18,19]. Figure 5 illustrates the response for an X-braced
frame of an offshore platform which had seven
component failures before collapse [19]. The
shape of the response curve is quite typical
with a rapid slope decrease after a few element failures. Such behavior suggests that the
load incrementation may be stopped to give a

pseudo-mode expression which may be sufficiently accurate for most system analyses.

EXTENSION
TO M U L T I - C O L L A P S E
MODE IDENTIFICATION
The incremental load technique described
is useful for finding only a single failure sequence. A search for other modes is reported
with heuristic techniques or even Monte Carlo
simulation [16]. The latter leads to trial structures which may have different collapse
modes. By using artificially high strength
coefficients of variation there is reduction in
the number of trials needed to find the significant additional collapse modes [ 16]. Nonlinear
programming has also been used for finding
ductile collapse modes for frames first
analyzed for basic plastic mechanisms [10].
For large systems a more efficient search
technique is still needed.
As a general solution to modal identification the following technique has been employed. It extends the concept of utilization
ratios in a logical strategy. The procedure
begins with critical components and finds the
influence of their ductility, and the possible
existence of redundant load paths. For design
and quality assurance it also provides the

overall importance of a member to the load

carrying capacity of the system under a particular loading condition. The steps are outlined as follows:
(1) Analyze the structural system intact with
no members failed.
(2) Identify critical components. This may
be based on (a) mean strength as done above,
(b) a safety index or (c) other measures of
importance based on professional judgement.
(3) Remove a critical component recognizing its ductile, brittle or semi-brittle behavior.
(4) Examine the changes in utilization ratios
a u which occur following the reanalysis. Determine components which have large changes
in a u as candidates for a failure sequence.
(5) For each component identified in step 4
extrapolate the load factor to find its failure
value. Ignore sequences which have high load
factors. This is illustrated below.
(6) Remove in succession components identified in steps 4 and 5 as remaining important.
(7) The procedure is continued until either
(a) collapse occurs, (b) a large load factor is
needed to cause additional failures and hence
the path can be ignored or (c) several components have failed with little change in load
factor suggesting that the load is flattening
out. (Also, the modelling assumptions become
more uncertain, e.g. linearity of material,
equilibrium based on undeformed geometry,
etc.).
(8) Continue the analysis (steps 3-7) with
other critical members.
This strategy may be viewed with a failure
tree (See Fig. 7). Each branch leads to a linear
failure mode expression as in eqn. 1. Some
paths may be similar so failure mode correlation in assessing system reliability is important. These procedures are further aided
by several considerations.
(a) The structural reanalysis following component failures may be extrapolated using the
initial intact analysis and a reanalysis algorithm.
(b) Only two or three failed components

may provide a sufficiently accurate reliability.

(c) Changes in utilization ratios for typical
frameworks suggest that a component failure
significantly affects only members located adjacent to it or in a parallel load path arrangement.
The identification strategy can be automated for analyzing large frameworks although this has not yet been carried out.
Naturally, some of these assumptions may fail
in certain situations and there can be a very
large number of reanalyses to perform. Nevertheless, it offers a systematic procedure using
engineering behavior and existing analysis
programs to identify critical modes.
Truss Example:
The truss (pinned ends)
shown in Fig. 6 was studied by Gorman who
identified all its failure modes [16]. The member dimensions and resistances are shown.
Table I shows the intact utilization ratios with
member 8 as the most critical (highest a u
value). Table l also shows the respective
changes in utilization ratios if this member is
removed. The possible resulting failure paths

21

",z"

13

Fig. 6. Ten member truss example [16].

10
TABLE 1

Data for example truss in Fig. 6

Component

Strength

Intact analysis

Element 8

Brittle e x a m p l e
E l e m e n t 8 failed

Tension

Comp.

Force ~

a, ,. b

Force a

change
ill a , 1 b

20

10

2
3
4
5
6
7
8
9
10

20
20
20
20
20
5
5
5
5

10
l0
10
l0
10
2.5
2.5
2.5
2.5

2,25
0,76
-- 0.74
-- 2.25
- 0.49
0.01
0.89
-0.92
0.90
-0.90

0.11
0.04
0.07
0.23
0.05
0
0,18
0.37
0.18
0.36

2,26
0
1.5
0.24
- 1.0
0.49
1.80

1.0
0
2.1
0
2.0
~
2.0

0,0o6
0,35
0.35
0,t)O(~
023
0.23
042

0.88
- 0.92

1.0
1.0

0.O07
- 0007

a at load factors equal to 1.0

expressed as percent of total strength

due to unit unloading of member 8

are therefore 8-3, 8-5, 8-6. and 8-7 as illustrated in Fig. 7. However. the extrapolated
(mean) load factors at which these components fail, namely 8.1, 11.3. 24.7 and 4.1.
respectively suggests that only sequence 8 - 7
is important. Removal of member 7 causes
collapse of the system. The incremental load
matrix for this path is:

[R871

[0.920

l 01[rr 1 ]

than ductile we would follow a different analysis. From Table 1. we see that member 7 is
most influenced by unloading in 8. To express
sequence 8 - 7 we have:
R~ -- 0.92rl
R7 =0.89rl + 0.42R s + 1.8r 2

The system capacity is the maximum of r~ and

r~ - r2 which gives:

,,7,
Mean failure loads

Solution for the collapse mode expression

gives:
R s = r 1+ r2 =

0.55R s

+ 0.56R 7.

(19)

-~

{ 18}

Substituting the respective strengths (compression and tension) for members 7 and 8
gives one of the identical important modes
given by Gorman (about 5 out of 27 modes
are important) [16]. In a similar manner starting with another critical member, say 10.
would give another important mode. Repeating the process leads to all of the significant
modes.
If member 8 were assumed brittle rather

2.7+

5.4

8.1

Critical
Member

2.7 + 8.6

= 11.3

2.7+22.0

=24.7

2.7+

1.42 = 4 . t 2

Members with High Change

in Utilization Ratio

Fig. 7. Failure tree following member 8 failure.

11
of the form:

R s = Max[rl, rl + r2]
-- Max[1.09R s, 0.555Rv]

(20)

The procedure can be repeated with other

paths leading to each collapse mode. For brittle component behavior, the overall system
reliability becomes a problem of finding the
distribution of the minimum of maximum
peaks for each sequence.

DISCUSSION AND CONCLUSIONS

(1) A method has been outlined for identifying collapse mode expressions for large
structural systems which incorporate both
ductile and brittle components. It utilizes existing structural analysis programs with incremental loadings and reanalyses following
successive component failures. The array of
different potential collapse modes are
determined in a systematic strategy looking at
changes in component utilization ratios following element failures. This approach is intended to simplify the difficult problem of
modal identification.
(2) The identification of collapse modes
leads logically to a combination problem to
determine overall system reliability. This
problem is summarized in the paper and references should be consulted. The solution may
be used to formulate a system optimization
for proportioning components. It requires
mathematical programming methods and has
not yet been attempted on large examples
[15].
(3) The strategy for identifying collapse
modes examines changes in component utilization ratio and has both design code and
quality assurance applications. At present,
design codes are based on element safety
checks with partial safety factors calibrated to
achieve uniform component reliabilities. This
approach can be modified to incorporate system capability by using a checking equation

~sdPiRi~ ~.ykQk

(21)

where R i is the nominal strength, Qk the

nominal loading and their respective strength
factor ~i and load factor ~'k. A system factor
q's is added to the partial factor equation to
represent the system consequences of a member failure. By examining collapse modes
which follow a component failure, a composite ratio of collapse failure load divided by
member failure load can be estimated. A recent study [20] incorporated this ratio into an
expected damage calculation. Such damage
models represent the expected consequences
to a structural system following a single comp o n e n t failure which progresses from
serviceable losses to collapse.
In a further study [21,22], the damage model
included the probability distributions of loading, component strength and collapse strength
and the correlation between the latter
strengths. A damage function is taken as zero
if the component does not fail and 1.0 if
collapse occurs. A damage index is defined by
integrating the damage function over load
and strength variables. To date [22], only normal distributions have been considered, single
variate loadings and either zero or complete
correlation between component and collapse
strength.
Further it was reported [21,22] how the
system factor q's could be used to obtain
uniform designs with the same expected
damage level. The important influences on q's
were found to be the mean ratio of collapse to
component failure load and the shape of the
damage function.
For example, in the truss example this mean
ratio equals 1.52 for member 8, i.e., the mean
collapse load (4.12) divided by the mean component load (2.70). In the instance where
several collapse modes may occur after a component fails, a composite bound must be developed.
Additional applications included highway

12

bridge structures [22] which emphasize the

importance of system reserve capactty because overloads routinely occur due to rising
truck weights.
(4) The study of collapse modes from component failure to collapse also provides
strategies for quality assurance. For example.
components whose failure precipitates collapse deserve further inspection and quality
control attention during design, fabrication
and construction. Similarly, the impact of
members prone to fabrication defects or possible accidental damage during structural lives
can be examined from a system realiability
viewpoint. A balanced risk strategy can be
developed including optimal total investment
for material and quality assurance. Reliability
is seen as a dynamic status possibly changing
during the lifetime. For example, different
approaches are needed for components which
show early signs of distress and hence could
be repaired.
Reliability includes the probability of
damage detection, distribution of residual
strength and the load magnitude assessed over
the repair period. Numerous applications to
marine and bridge structures are readily apparent.

ACKNOWLEDGEMENTS
This paper was written while the author
was on sabbatical at The Marine Technology
Center of Imperial College, London. supported by the Science and Engineering Research Council (U.K.). The work draws on
earlier studies under a National Science
Foundation (U.S.) grant at Case Institute of
Technology in which the contributions of
Gorman and Ghosn were important. The
calculations for the two-tier example were
performed by Mohammed R. Rashedi.
graduate student at Case.

REFERENCES
1 B. Ellingwood, T.V. Ga~ambos. J.C. McGregor and
C.A. Cornell. Development of a Probability Based
Load Criteria for American National Standard A58.
National Bureau of Standards. June. 1980.
2 P.F. Csagoly and R.A. Dorton. The Development of
Ontario Highway Bridge Design Code. Transportation Research Record No. 665. Vol 2. 1978.
3 A.R. Flint. B.W. Smith. M.J. Baker and W. Manners.
The derivation of Safety Factors for Design of Highway Bridges, Proc. Conf. on the New Code for the
Design of Steel Bridges. Cardiff. March. 1980.
Granada Pub. Co.. London. 1981,
4 F. Moses. Guidelines for Calibrating API RP2A for
Reliability-Based Design, Final Report. API PRAC
80-22. American Petroleum Institute. Dallas. Texas.
1980.
5 CIRIA. Rationalization of Safety and Serviceability
Factors in Structural Codes, Construction lndustry
Research and Information Association. Report 63.
1977.
6 Towards Safer Long-Span Buildings, The American
Institute of Architects. Long Span Building Panel.
AIA Catalog Number: 6-N101. 198t.
7 AASHTO. Standard Specification for Highway
Bridges, American Association of State Highway
and Transportation Officials. Washington, D.C..
1977.
8 F. Moses. Bayesian Calibration of Platform Reliability. ASCE Spec. Conf. on Probabitistic Mechanics
and Structural Reliability, Jan., 1979, Tucson.
Arizona.
9 F. Moses. Reliability of structural .systems. J. Struct.
Div. ASCE. ST-9. 100 (1974).
10 A.H.-S. Ang and H-F. Ma, On the reliability of
structural systems, in: T. Moan and M. Shinozuka
(Eds.), Proc. 3rd Intl. Conf. on Structural Safety and
Reliability, Trondheim. Norway, 23-25 June. 1981.
Elsevier Scientific Pub. Comp., 198t.
II E.H. Vanmarcke. Matrix formulation of reliability
analysis and reliability based design, Comp. Struct..
13 [1971 ~.
12 G Edwards. P.J. Tudhope and M.A. Halsen. Offshore jacket platform Failure mode analysis. KIVI
Symp. on Safety and Reliability of Structures, T.H.
Delft. Sept., 1980.
13 O. Ditlevsen, Narrow reliability bounds for structural systems, J. Struct. Mech.. 1 (4) {1979}.
14 C.A. Cornell, Bounds on the reliability of structural
systems. J. Struct. Div. ASCE. STI, 93 ~1967).

13
15 J. Stevenson and F. Moses, Reliability analysis of
frame structures, J. Struc. Div. ASCE, 96 (1970).
16 M.R. Gorman, Reliability of Structural Systems,
Report CE 79-2, Dept. of Civil Engineering, Case
Western Reserve University, Cleveland, Ohio, May,
1979.
17 M.R. Gorman and F. Moses, Direct estimates of
structural system reliability, Proc. ASCE 7th Conf.
on Elec. Comp., St. Louis, Aug., 1979.
18 F. Moses, Structural system reliability and optimization, Comp. Struct., 7 (1977).
19 F. Moses and B. Stahl, Reliability analysis format
for offshore structures, Paper OTC 3046, Offshore
Technology Conference, Houston, Texas, May, 1978.
20 M.R. Gorman and F. Moses, Partial factors for

structural damage, ASCE Proceedings on Probabilistic Methods in Structural Engineering, ASCE, St.
Louis, Oct., 1981, ASCE, New York.
21 F. Moses and M.J. Ghosn, Requirements for a reliability-based bridge code, ASCE Proceedings on
Probabilistic Methods in Structural Engineering,
ASCE, St. Louis, Oct., 1981, ASCE, New York.
22 M.J. Ghosn, Reliability Formulation of Highway
Bridge Loadings, Report 81-1, Dept. of Civil Engineering, Case Western Reserve University, Cleveland, Ohio.
23 M,J. Grimmelt and G.I. Schueller, Benchmark study
on methods to determine collapse failure probabilities of redundant structures, Structural Safety, in
press.