Anda di halaman 1dari 14

r.,l~---r.,,-7~ ,7~,..-,.

r,lwl

Structural Safety Vol. 18, No. 2/3, pp. 67-80, 1996


ELSEVIER
PII: S0167-4730(96)00006-9

Copyright 1996 Elsevier Science Ltd. Printed in The Netherlands. All rights reserved 0167-4730/96 $15.00 + .00

Reliability-based condition assessment and LRFD for existing structures


Bruce R. Ellingwood
Department of Civil Engineering, The Johns Hopkins University, Baltimore, MD 21218, USA

Abstract

Safety and serviceability requirements for limit states design in North America have been developed and implemented in codes of practice during the past 20 years. The load and resistance criteria in limit states design are based on probabilistic methods. Statistical data on loads and on strengths are integrated by the reliability analysis to yield criteria that are consistent with a prescribed measure of reliability. The focus of code work to date has been on criteria for design of new buildings. However, professional practice frequently involves evaluation of existing structures. Research to support the development and implementation of probability-based condition assessment procedures for existing buildings is in progress. Copyright 1996 Elsevier Science Ltd.

Keywords: Buildings (codes); Design (buildings); Limit states; Loads (forces); Probability theory; Reliability; Safety; Standards; Structural
design; Structural engineering

1. Introduction
Safety and serviceability provisions in codes and standards used in the design of new buildings and other structures take the form: Required strength < Design strength. Deformation due to service loads < Deformation limit. (1) (2)

Such performance checks are intended to ensure that the structure is safe under extreme loads and remains functional under service loads. During the past 20 years, the advantages of limit states design (or LRFD, as it is termed in the USA) over allowable stress design methods have become apparent [ 1], and practically all recent advances in codified design have followed a limit states approach. The focus in probability-based codified design development has been almost entirely on new construction. However, a significant amount of modern structural engineering practice concerns existing buildings and other structures, especially in mature urban areas where space is limited, costs of new or replacement construction are high, and the desire to preserve the architectural heritage is strong. There have been only limited attempts to transfer the ideas of limit states design for new construction to existing facilities. Some of the issues that must be addressed in doing so are reviewed in this paper.
67

68

B.R. Ellingwood

2. Structural reliability summary


Uncertainties arise from variations in loads and material strength properties, dimensions, natural and man-made hazards, insufficient knowledge, and human error in design and construction. Advances in structural reliability theory and probabilistic methods for analyzing uncertainty and setting performance criteria for design have accompanied the development of limit states design as an engineering field of endeavor [2]. A limit state is defined by a mathematical model derived from principles of mechanics and experimental data which relates the resistance and load variables for the specific behavioral condition of interest; G(X) = 0 , (3)

in which X = (X l . . . . . X n) are resistance or load variables that, in general, are random. The function G(X) is defined, by convention, such that failure occurs when G ( X ) < 0. Thus, the limit state probability becomes e.g., [3],

Pf= ffx (x)dx,

(4)

in which f x 0 = joint density function of X and the integration is performed over the region of x which G(x) < 0. The integration in Eq. 4 usually is difficult for realistic structural problems. Firstand second-order reliability methods ( F O R M / S O R M ) provide an alternate measure of reliability in terms of a reliability index, /3 [4,5]. Provided that the limit state is well-behaved, the probability,

(5)
often is an excellent approximation for the true limit state probability, making FORM a tool for approximate numerical integration of Eq. 4 and /3 and Pf more or less interchangeable as reliability measures. Alternatively, with advances in computation have come highly efficient Monte Carlo methods for direct evaluation of Eq. 4, e.g., [6,7]. Statistical data on the mean, standard deviation or coefficient of variation and probability distribution of each resistance and load variable used in the reliability analysis should be representative of values that would be expected in situ and should reflect uncertainties due to inherent variability, load and resistance modeling and prediction when data are limited, and measurement. Such data have been described in detail elsewhere [8,9]. A brief summary of data that will be used in subsequent examples is presented in Table 1. The structural parameters and dead loads are modeled as

Table 1 Statistical data on structural loads Variable Dead load Live load (50 years) Live load (25 years) Yield strength, F r Section modulus, Z x Flexure model bias, B Mean 1.05D 1.00L 0.85L 1.05Fy,, 1.00Z+ n 1.02 c.o.v. 0.10 0.25 0.35 0.11 0.06 0.05 CDF Normal Type I Type I Lognormal Normal Normal

B.R. Ellingwood

69

random variables, while the live load is modeled as a pulse process, and the live load statistics therein pertain to the maximum load to occur in 50 or 25 years.

3. Probability-based codified designbcurrent status


In LRFD, the design strength in Eq. 1 is represented by ~bRn, in which Rn = nominal (or code-specified) strength for the limit state of interest computed using specified nominal strengths, Fni, and dimensions and ~b = a resistance or performance factor that depends on the variability in strength and nature of the limit state. In the limit states criteria being developed for the Eurocodes, design strength is determined by first dividing the individual nominal material strengths, Fn, by partial material factors, i.e. Fdi = Fni//Ymi, and then calculating strength directly using these Fdi. Both code formats are philosophically similar and lead to approximately the same structural designs for similar materials and load envelopes. The methods used to arrive at partial factors of safety and nominal strengths and loads now are relatively robust [1]. A partial list of the probability-based load requirements that now appear in the ASCE 7 Standard on Minimum Design loads [9,10] is: 1.2D + 1.6L + 0.5S, (0.9 or 1.2)D + 1.3W+ 0.5L, (0.9 or 1.2)D + 1.0E + 0.5L + 0.2S,

(6) (7) (8)

in which D, L, W and E = nominal dead, live and wind loads, and structural action due to earthquake, respectively. The structural action due to earthquake is based on a ground motion with an estimated probability of 0.1 in 50 years [11]. The commentaries to ASCE 7-95 also contain, for the first time, load combinations for serviceability and accidental limit states. Structural analysis using these load combinations defines the required strength in Eq. 1.

4. Evaluation of existing structures


LRFD, as described above, presently is applicable to new construction. ~Events during the service life of a completed building or other structure may require an assessment of structural adequacy and lead to possible modifications for further use. The condition assessment of an existing structure must be situation-specific and thus is a unique technical process that is difficult to codify. The concerns of safety, serviceability and durability are the same as with a new structure. The condition assessment must ensure that these performance requirements are met with acceptable reliability. However, current guidelines for evaluating existing structures e.g. [12,13], do not have a reliability basis.

4.1. Scope of application


Structural evaluations are conducted for many reasons: changes in tenancy; concern about faulty building materials or construction methods; discovery of a design/construction error after the building is occupied; concern about structural deterioration from normal operating or environmental

70

B.R. Ellingwood

conditions discovered during routine inspection/maintenance; structural damage following extreme environmental events such as windstorms or earthquakes, severe fires or other accidental/abnormal events; complaints from tenants regarding serviceability. The structural engineer must normally perform the structural condition assessment with minimally invasive techniques. Moreover, structural modifications resulting from this assessment, if required, generally must preserve the architectural integrity and heritage of the building. Changes in a structure are often brought on during its service life by operating and environmental conditions, including excessive service loading, an aggressive environment, improper maintenance/repair, or accidents. Cracks and other signs of surflcial distress; excessive permanent deformations; changes in natural frequency or damping; dimensional stability; pretensioning loss; all are evidence of structural deterioration. Among the most pervasive mechanisms giving rise to structural deterioration are corrosion of structural steel or steel reinforcement; damage to the concrete due to attack of sulfates, acids and other chemicals, expansive aggregate reactions, frost attack, salt crystallization, and microbiological attack; and damage due to fatigue/fracture from repetitive loads. Changes in strength brought on by these stressors are stochastic, and are difficult to model mathematically at the current state-of-the-art. Design documentation for buildings over 20 years old tends to be poor, and original design calculations and working drawings usually are unavailable. Design codes, materials and construction practices may have changed, and material specifications may be difficult to obtain. Not all members may be accessible for inspection. Information on past extreme structural loading may be unknown. Previous maintenance, if conducted, may have caused the strength and stiffness to change; in any event, maintenance records often are unavailable. On the other hand, many older buildings have performed well without indications of damage or failure. They generally were designed to higher factors of safety and, if well-maintained, may have reliability levels that are substantially in excess of those in modern construction. An illustration of the decrease in reliability index during the past century is presented in Table 2 for a compact flexural member subjected to dead plus live load and supporting a tributary area of 400 ft 2 (37 mE). An analysis for reinforced concrete shows a similar trend. Most of this decrease has arisen from reductions in design loads and factors of safety, reflecting increasing confidence placed in improved material quality, analysis, design and construction practice. The fact that a structure has survived the severities of construction and has performed successfully in service (perhaps for as much as 50 years or more) is primafacie evidence of at least minimally acceptable reliability. For certain types of construction presenting significant offsite hazards, it may be feasible to perform a full time-dependent reliability analysis, view the changes in the conditional failure rate and reliability function with time, and determine rational in-service condition assessment, inspection and maintenance policies that limit the limit state probability to a specified level, e.g. [ 14]. However, such an assessment requires a substantial investment in reliability analysis and acquisition of supporting data. The majority of buildings evaluated for possible retrofit are under five stories in height. Traditional LRFD methods will prove more suitable for the vast majority of condition assessments of ordinary buildings and other structures. Numerous sources of uncertainty cloud the condition assessment of existing structures. Some of the uncertainties reflected in the load and resistance factors (modeling approximations, deviations from specified dimensions and strengths) may not be present or may be less than in new construction, particularly when in-situ measurements have been taken during the course of the evaluation. In other

B.R. Ellingwood

71

Table 2 Evolution in reliability index for steel design Period Material 1873 Wrought iron 1889 Steel (buildings) 1923 1923 1936 1955 1963 1963 1972 1972 1978 1982 1986 Steel (A36) Steel (A36) Steel (A7) Steel (A7)

Allowable stress, psi Producer specifications 12,000 16,000 Ist edn, AISC ASD 18,000 20,000 A58.1-1955 7th edn, AISC ASD 22,000 A58.1 - 1972 22,000 8th edn, AISC ASD A58.1 - 1982 22,000 1st edn, AISC LRFD

(MPa) (83 MPa) (110 MPa) (124 MPa) (138 MPa)

/3 6.3 5.4 5.2 4.8

Steel (A36) Steel (A36)

(152 MPa) (152 MPa)

4.0 2.9

(152 MPa)

3.0 2.6

cases, the uncertainties can be greater, e.g. for inaccessible members and connections where construction details cannot be inspected and verified.

4.2. Probability-basedevaluation of existing buildings


Survival of a qualification test or a service load history provides additional information about reliability. When a condition assessment of an existing structure is conducted, the knowledge gained can be used to update the prior estimate of reliability, P} or /3'. The condition assessment might consist of deflection measurements, passage of a load test, crack mapping and so forth. Denoting the results of the inspection by event H, the updated limit state probability e.g. [15], is

U=P[G(X)<OIH]=P[G(X)<Oand

H]/P[H].

(9)

This conditional probability can be determined by any of several techniques for reliability analysis. The computations in the examples to follow all were performed by importance sampling with a FORM analysis as an independent check [6,7]. The condition assessment method(s) should be selected to make the correlation between the events {G(X) < 0} and H as strong as possible, within economic or physical constraints. Strong correlation (or dependence) sharpens the posterior estimate of reliability. For example, if strength and stiffness are positively correlated, as with concrete or wood, strength might be inferred from structural deformation. On the other hand, if the correlation is weak, the prior and posterior estimates of reliability will be nearly the same. While Eq. 9 provides a conceptual basis for updating the estimated reliability of an existing structure, criteria for evaluation likely will continue to be specified in an LRFD type of format and

72

B.R. Ellingwood

will be based on component behavior. Designers are familiar with this approach and the prospects for codifying the evaluation process are better. Criteria for assessment of existing buildings must be flexible enough to handle the additional updated measurements of dead loads, material strengths and dimensions, and other information. For the most part, this flexibility is not present in the LRFD Specification [16], ACI Standard 318 [17], and codes for new buildings. A condition assessment of an existing building or other structure can be performed by structural analysis, by test, or by a combination thereof. Verification of safety is most difficult because it requires extrapolation beyond the range of observations.
4.3. Evaluation by analysis

The evaluation by analysis should be consistent with design requirements for new buildings.
4.3.1. Factors on structural actions

Dead loads are better defined in existing buildings than in new construction, where the structural calculations may be essentially completed before interior finishes, attachments and other contributors to dead load are finalized. Other structural loads generally are stochastic processes with intensities that vary in time, and one needs to take the expected service period into account in setting required load combinations. A decision must be made as to whether the building is to be renewed following rehabilitation or the overall service period is to remain the same (presumed service life remains, for example, at 50 years). In the former case, the required loads should be essentially the same as for a new building. If the life expectancy is less, however, there is a lower probability of extreme loads in the (reduced) future service period, and less conservative load combination requirements could be used in the supporting structural calculations. This decision should involve the building owner, architect and structural engineer, as it has significant economic implications.
4.3.2. Factors on structural strength

Uncertainties that are used to derive LRFD resistance criteria for new construction are based on a presumption of modern methods of structural analysis and an examination of many material tests, measurements of dimensions, etc., and are typical of what one would obtain if a new building were selected at random. The situation is different with existing buildings, where strengths and dimensions may be determined from in-situ measurements. Strengths determined by ASTM specifications or other standardized procedures may not provide an accurate picture of the in-situ strength of the structure. For example, the static yield strength for steel in a W-shape is typically 0 . 9 5 ( F y - 4 ksi) or 0.95(Fy - 28 MPa) in which Fy is the reported test strength determined from a tensile coupon taken from the web and subjected to mill test [18]. Similarly, the specified compression strength of protected concrete under static rate of load early in the service life might be 0.7f~ + 1 ksi (0.7f" + 7 MPa) [ 19], increasing to about 150% of this at an age of 10 years or more. As a simple example, we consider the W24 76 girder in Fig. 1, taken from an interior floor of an existing building, for which the governing loads are dead and live load. The girder is A36 steel (nominal yield strength of 36 ksi or 248 MPa). The dead load is 0.75 k i p / f t (10.9 k N / m ) and the concentrated live load is 22 kip (98 kN). This girder, if new, would be designed for flexure according to [16],
0 . 9 F y . Z x . = 1 . 2 M o + 1.6M L,

(10)

B.R. Ellingwood

73

L = 22 k (98 kN)
wD

= 0.75 k/ft (10.9 kN/m)

ll!lll JlJlllllliill JllllZJJllIJ iJilJllJltllIJl Jill [IJll


W 24 x 76 [~L 20 ft (6.1 m, ~{_.F. 20 ft (6.1 m,

Fig. 1. Simply supported beam designed by LRFD.

in which Fy, and Zx, = nominal yield strength and plastic section modulus and M o, M c = moments due to dead and live load. The flexural limit state is,
G() = BFyZx - Q = O,

(11)

in which B, Fy and Zx were defined in Table 1 and Q represents the combined (random) structural action due to dead and live loads. Using the statistics of resistance and load terms summarized in Table 1 as a basis, this design equation results in a girder with a flexural limit state probability of 0.0051 and reliability index of approx. 2.6 on a 50-year basis [8]. Suppose that the service life of the W24 76 girder is to be extended by 50 years from the present; measurements of material strength, dimensions and weights in support of the assessment might indicate that measured yield strength Fy m, section modulus, Zxm, and dead load, D,,, differ from the nominal design values. Assume that the mill test report indicates a strength of 50 ksi (345 MPa), with an estimated coefficient of variation of 0.07. The (static) yield strength then can be revised upward to 0.95(50 - 4) or 44 ksi (303 MPa). No deterioration in beam section is observed. On the other hand, measurements of weights of construction materials and permanent attachments indicate that the original dead load used in design was underestimated by an average of 10%; however, as a result of measurement, the coefficient of variation now is 0.05 rather than 0.10. Based on the revised statistics and in the absence of further information on live load, the estimated /3 for the W24 X 76 girder increases from 2.6 to 3.4. The girder clearly is stronger than expected, and could withstand additional live load without violating the reliability requirements of the code. How much additional live load the beam could withstand can be seen by incrementing L until /3 decreases to 2.6; it is found in this instance that L can be increased by about 30%. Alternatively, a safety check for the existing girder that is consistent in reliability terms with that for a newly designed girder can be developed from the statistical data obtained in situ and presented in the previous paragraph. The fact that additional data were obtained on dead load and strength suggests that the partial safety factors on those design parameters might be different. On the other hand, no additional information was obtained on live load. Using the same reliability analysis and a target reliability index of 2.6 as before, we obtain (by trial and error),
0.88FymZxm = I.IMDm + 1 . 6 M L.

(12)

Note that the partial factors 0.88 and 1.1 are applied to the measured values. With some additional complexity in format, one could envision resistance and load criteria that vary depending on the quality of inspection, extent of in-situ measurements, potential failure modes and consequences. A simple illustration of how this might be done is provided in Allen [20].

74

B.R. Ellingwood

4.4. Evaluation by test (proof loading)


A load test of a structure may be required to demonstrate structural integrity and acceptable performance during a proposed service period. A load test can be used to validate and improve analytical modeling of structural behavior for design purposes; to verify design assumptions and member, connection and system capacities, especially of newly retrofitted components; to establish redundancy and the interaction between primary and secondary members; and to identify potential failure mechanisms. Occasionally, an on-site load test may be specified when the original design documents are unavailable and safety cannot be verified by other means [21]. Unfortunately, current codes and standards [10,16,17] provide only limited guidance on appropriate test methods, load levels and response measurements. A typical load test involves the following general steps: (i) Load the structural component slowly, in several approximately equal stages, to a maximum of 1.5-2 times the design live load, holding the load at each stage for one hour and measuring deformations; (ii) Hold the maximum test load for 24 hr, measuring deformations at the end of that period; (iii) Remove the test load and measure residual deformations 24 hr following removal of all load. If 75% of the deformation from superposed load is recovered, the component passes the load test. Such procedures do not have a reliability basis, and the extent to which they demonstrate compliance with code safety objectives is uncertain. Passage of a load test provides confirmation of structural resistance, and allows the previously estimated reliability of the structure to be updated [22-24]. As an example, we consider the same W24 76 steel beam as previously (Fig. 1), designed to withstand dead plus and live load according to Eq. 10 [16]. To assess its integrity, the beam now is subjected to a proof load of magnitude q *. Passage of the proof test implies that,

H= {FyZx- MD- Mq. > O },

(13)

in which Mq. = moment due to proof load (deterministic). If the dead load were measured accurately prior to test, M D would be essentially deterministic, but an accurate determination is not always feasible. The updated reliability index, fl", is determined from, /3"= ~ - 1 ( 1 - P[G(X) < 0 I H ] ) , (14)

in which the conditional probability is determined from Eq. 9. The variation in updated failure probability, P}', for a future service life of 50 yr with increasing proof load is illustrated in Fig. 2 (denoted P]' (50;0) in Fig. 2), in which q * = aL. The proof load factor, ot is applied only to the live load, the presumption being that the dead load is known. The probability of failure during the proof load test, P(H---) = 1 - P(H) is also shown. For proof loads less than 1.2 L, surviving the proof test is noninformative insofar as the estimate of reliability is concerned, and /3" =-/3'. Indeed, the impact of the load test on reliability apparently is negligible until the load reaches a level of about 140% of L, or approximately 75% of the nominal strength, Fy,,Zx, ,. To decrease P}' by a factor of 5 (or increase /3" from 2.6 to 3.1, one would have to proof load to approximately 2.0L. It is likely that nonstructural or minor structural damage would become evident at this level. The plot of P(H---) increases very rapidly when a exceeds about 1.3, and at a proof load of 1.6L, the probability of failure during the proof test has increased to the (perhaps unacceptable)

B.R. Ellingwood

75

0.006

0.005
I_

0.004

0.005
.p

.fl

O. 002 0 L ~L

0.001

/
0.8 1.0 1.2 1.4
proof

--

,.(,,

0.0oo

1.6
load,

1.8
a-q'/L

2.0

2.2

normalized

Fig. 2. Reliability following p r o o f load otL ( V R ~ O. 13).

value of 0.05. On the other hand, test loading to less than 75% of nominal strength does not provide sufficient evidence to update the prior reliability estimate o f / 3 ' = 2.6. Thus, determining a requires a tradeoff between additional information gained by increasing the load and the increased likelihood of costly damage or failure to the structure tested. If the objective is acceptable performance during a fixed service interval (for example, 50 years), a proof test may be more informative if performed during the service life rather than at the beginning. Fig. 2 compares (updated) probabilities of failure in 50 years, given survival of the proof test at the beginning and at 10 and 25 years into the service life [denoted P)'(50;10), and P~'(50;25)]. Passing a proof test half-way through the service life can decrease the overall limit state probability for the 50-year service period by a factor of more than 2. For example, the probability of failure in a 50-year period, given survival for 25 years and a confirming proof test to 1.5L at that time prior to continued service, is 0.002 (or/3" = 2.9). Fig. 3 illustrates how these probabilities change when the coefficient of variation in the strength increases from 0.13 to 0.18. Here, the proof load provides evidence of improvement in subsequent reliability at when q * exceeds approx. 1.2L, but again, one would have to load to approx. 1.8 L to increase reliability by a factor of 5. Results of similar analyses, not reported here in detail, suggest that proof loading at the customary levels provides a more informative measure of subsequent safety if the variability in resistance tends to be on the high side, as it would be for concrete and masonry structures. Conversely, proof loading is less informative if the strength variability is relatively small, as it would be for most steel structures. Despite the fact that load tests at customary load levels of 1.0-1.5L provide little insight into structural reliability, they nonetheless have considerable value when used to validate computerized structural modeling and nonlinear structural analysis, which does not require loading to near-failure levels. The role of load testing in structural quality assurance should be re-examined in this light. Surviving a service load history which is usually stochastic in nature may provide evidence of

76
0.018
0.016-

B.R. Ellingwood
I I I I I I I I I I I I I -1-

I} (.

0.014 0.0120.010"

..... .....

PIHI P",[50;O)
P " t { 5 0 ; 10 )

Jr j.
1

m
o

O. 00B"

O. 006o

0 . 004" O. 002 O. 000

0.8

1.0

1.2
normal ized

1.4 proof"

1.6 load.

1.8
,,=qi/L

2.0

2.2

Fig. 3. Reliability following proof load aL (VR = 0.18)

structural reliability that is comparable to what might be learned from a proof test. Fig. 4 compares the (cumulative) probability of failure during a service life of 50 years with the probability of failure, given survival through t years for the W24 76 beam subjected to dead and live considered previously (Fig. 1). Both curves are essentially linear because the combined load process is stationary in time and the resistance, while random variable, does not vary in time. Survival for 25 years reduces

without O.OOg .....


O. O O B
L -I

strength degredetion: P| T< t l P[ T < 5 O I T > t ] ,~ in 5 0 yrs:


f. /'/" /" `% ,/ / `% % .% o/./.o/ "

`% ~, 0.007,
0. 006.

`%

107. d e g r a d a t i o n ..... P I T < t l


`% `%-- .

1.1

P[T<5OIT>t] `% ,%

t...

`% ~. "-.

O. 005
O. 004
In IO

~.

/" w'-""

O. 0 0 3
L O.

./ / ,/ /"
,-,--,, .

O. 0 0 2

"". " - ,

`% `%
-. ,.
., ,

0.001
0.000

,-

,-

-,

-.

10

20

30

40

50

t {y e a r )

Fig. 4. Proof load required for reliability assurance.

B.R. Ellingwood

77

the limit state probability by a factor of approx. 2. One would have to proof load the beam to approx. 1.8L successfully to make the same statement about reliability during a 50-year service period (cf. Figs. 2 and 4). If the beam degrades, e.g. due to cbrrosion, the cumulative probability of failure during 50 years increases. This also is illustrated in Fig. 4, where it is assumed that uniform corrosion causes the plastic section modulus to decrease linearly to 90% of its initial value after 50 years. Such a loss is consistent with corrosion penetration of the flange and web of the W24 76 modeled by C(t) = At 8, in which A and B are experimentally determined constants that depend on the environment, and B typically is in the range 1 / 2 - 2 / 3 [25]. That survival of service load is more informative in the case of active corrosion can be seen from the more rapid decrease in P[T < 50IT> t] with elapsed time, t.

5. Current unresolved issues and research needs There are a number of issues that need to be addressed in research to develop and implement rational reliability-based condition assessment criteria. The following issues have arisen in the writer's own professional experience. Readers undoubtedly will be able to add to the list. In contrast to new construction, the economic impact of required changes to existing buildings and other structures to meet updated safety requirements may be very large [20]. This is a significant current issue in the area of seismic rehabilitation. Many jurisdictions require that if renovations exceed some percent (for example, 50%) of the value of the property, the building must be upgraded to conform to current standards (and, by inference, the level of reliability inherent in new construction). In other cases, a reduced level of performance may be approved by the authority having jurisdiction. What is the appropriate policy if the code loads have gone up and the building cannot meet the new code, even if it met the old? What risk is acceptable in this case, and what target reliability measures should be met by the evaluation criteria? What are the legal implications of using assessment criteria for old buildings that are different from those used for new? Validation of reliability models by comparing predicted and observed structural failure rates is not practical or informative for the majority of civil engineering structures because the failure rates are low. There are only limited failure data available, and seldom does a sufficient sample of a given type of structure accumulate before changes in technology or the economic climate cause changes in construction. Details of most structural failure investigations are held confidential. Because of these difficulties, the first generation of reliability-based codes was developed by calibration [8,9] to existing acceptable practice for new construction. This calibration option requires rethinking in applying it to existing construction, since what is existing acceptable practice is not clear. Many older structures were constructed with archaic materials. How does one determine strength criteria probabilistically for members such as cast-iron columns or wrought-iron beams? Moreover, when old members are reinforced with new materials, what are appropriate strength criteria for the resulting composite structural member? How does one determine dimensions and other structural properties of repaired sections, e.g. slabs repaired by adding concrete, concrete columns strengthened with external steel reinforcing jackets, etc? Methods for determining such engineering properties are not well defined. What are suitable limit states for evaluating existing structures? Current design of new construction

78

B.R. Ellingwood

focuses on individual members. Consideration of the behavior of the structural system as a whole may pay handsome economic dividends in structures with even a modest capacity for load sharing. What sort of consideration (if any) should be given to nonstructural components? In older structures, for example, partitions were not uncoupled from the frame as they are in new construction, giving rise to much stiffer structures. The beneficial effect of such stiffening might be taken into account in some way. How? For some limit states related to earthquake, strength is lesser in importance to ductility [26], and performance is determined by construction detailing. How should one verify an energy dissipation limit state? What sort of structural inspections should be required? Resistance criteria ought to depend on the level of NDE, and more comprehensive attempts at determining in situ strength should be rewarded with less conservative criteria. But how does one trade off the need to gather information on design details and material properties for inaccessible structural elements with the damage to and disruption of the building functions? What additional analyses should be required in support of a decision to continue service? How does one distinguish between structural evaluations performed prior to repair and after repair? Little information exists to describe the impact of repair on structural reliability. This is of particular concern when repair operations, such as those involving welding, have the potential to cause further damage to the structure. Finally, how does one go about codifying procedures for condition assessment of existing buildings, bridges and other structures, given the unique nature of each situation? For a codification effort to be successful, certain generally applicable principles and procedures must be identified. As a minimum, the following would be necessary: a statement of general principles and requirements for acceptable structural component and system behavior, based on structural reliability; requirements for data collection by inspection and nondestructive evaluation; recommended statistically-based methods for analysis and synthesis of data; and a reliability-based but deterministically appearing performance check, load test, or other action required to verify safety. The code should support the engineer in addressing the factors identified previously: changes in load-carrying requirements due to changes in use, potential structural deterioration due to aggressive environmental factors, accidents or abuse; and service life extensions.

6. Conclusions

Probability-based safety checking procedures can be developed for existing buildings and other structures. The reliability framework provides a rational basis for analyzing uncertainty, guides the acquisition of the additional data that are most significant for ensuring future acceptable performance, and organizes information into a form useful for engineering decision analysis. Moreover, probability-based limit states analysis provides a clear link between research, both analytical and experimental, and in-service experience. Only limited guidance for evaluating existing facilities quantitatively can be found in current design and regulatory documents. Research to support the development of such guidelines is both challenging and opportune.

B.R. Ellingwood

79

Acknowledgements
This w o r k w a s c o n d u c t e d as part o f a research p r o g r a m to e x a m i n e issues pertaining to t i m e - d e p e n d e n t reliability o f aging structures, supported, in part, b y the N u c l e a r R e g u l a t o r y C o m m i s sion and the US A r m y C o r p s o f Engineers. T h e assistance o f M r B a i d u r y a B h a t t a c h a r y a , g r a d u a t e student in the D e p a r t m e n t o f Civil E n g i n e e r i n g , in p e r f o r m i n g the c o m p u t a t i o n s is gratefully acknowledged.

References
[1] Ellingwood, B., Probability-based codified design: past accomplishments and future challenges. Structural Safety, 1994, 13, 159-176. [2] Ellingwood, B., Probabilistic risk assessment. In Engineering Safety, ed. D. Blockley, McGraw-Hill International Series in Civil Engineering, Berkshire, UK, 1992, pp. 89-116. [3] Ang, A. H.-S. and Cornell, C. A., Reliability bases of structural safety and design. Journal of the Structural Division, 1974, 100, 1755-1769. [4] Hohenbichler, M. and Rackwitz, R., New light on first- and second-order reliability methods. Structural Safety, 1987, 4, 267-284. [5] Bjerager, P., On computation methods for structural reliability analysis. Structural Safety, 1990, 9, 79-96. [6] Schueller, G. I. and Stix, R., A critical appraisal of methods to determine failure probabilities. Structural Safety, 1987, 4, 293-310. [7] Melchers, R. E., Importance sampling in structural systems. Structural Safety, 1989, 6, 3-10. [8] Galambos, T. V., Ellingwood, B., MacGregor, J. G. and Cornell, C. A., Probability-based load criteria: assessment of current design practice. Journal of the Structural Division, 1982, 108, 959-977. [9] Ellingwood, B., MacGregor, J. G., Galambos, T. V. and Comell, C. A., Probability-based load criteria: load factors and load combinations. Journal of the Structural Division, 1982, 108, 978-997. [10] Minimum design loads for buildings and other structures, ASCE 7-95, American Society of Civil Engineers, New York. [11] NEHRP recommended provisions for the development of seismic regulations for new buildings, 1992. FEMA Report 223, Federal Emergency Management Agency, Washington, DC. [12] Manual for maintenance inspection of bridges, 1983. American Association of State Highway and Transportation Officials, Washington, DC. [13] Strength evaluation of existing concrete buildings, ACI 437-1982. American Concrete Institute, Detroit, MI. [14] Mori, Y. and Ellingwood, B., Reliability-based service life assessment of aging concrete structures. Journal of Structural Engineering ASCE, 1993, 119, 1600-1621. [15] Madsen, H., Model updating in reliability theory. Proc. ICASP5, Vancouver, BC, 1987, pp. 564-577. [16] Load and resistance factor design specification for structural steel buildings (1986, revised 1993), American Institute of Steel Construction, Chicago, IL. [17] Building code requirements for reinforced concrete, AC1318-1989, revised 1995, American Concrete Institute, Detroit, MI. [18] Galambos, T. V. and Ravindra, M. K., Properties of steel for use in LRFD. Journal of the Structural Division, ASCE, 1978, 104, 1459-1468. [19] MacGregor, J. G., Mirza, S. A. and Ellingwood, B., Statistical analysis of resistance of reinforced and prestressed concrete members. AC1 Journal, 1983, 80, 167-176. [20] Allen, D. E., Limit states criteria for structural evaluation of existing buildings. Canadian Journal of Civil Engineering, 1991, 18, 995-1004. [21] Fu, G., Highway bridge rating by nondestructive proof-load testing for consistent safety. Report FHWA/NY/RR95/163, 1995, New York State Department of Transportation, Albany, NY. [22] Nowak, A. S. and Tharmabala, T., Bridge reliability evaluation using load tests. Journal of Structural Engineering ASCE, 1988, 114, 2268-2279.

80

B.R. Ellingwood

[23] Hall, W. B. and Tsai, M., Load testing, structural reliability and test evaluation. Structural Safety, 1989, 6, 285-302. [24] Moses, F., Lebet, J. and Bez, R., Applications of field testing to bridge evaluation. Journal of Structural Engineering ASCE, 1994, 120, 1745-1762. [25] Komp, M. E., Atmospheric corrosion ratings of weathering steels--calculation and significance. Materials Performance, 1987, 26, 42-44. [26] Ellingwood, B., Probability-based codified design for earthquakes. Engineering Structures, 1994, 16, 498-506.

Anda mungkin juga menyukai