You are on page 1of 12

BUILDING VULNERABILITY ASSESSMENT USING PUSHOVER METHODS - A TURKISH CASE STUDY Edmund BOOTH1, Juliet BIRD2 and Robin

SPENCE3
ABSTRACT

Non-linear static pushover analysis forms one key element of current methodologies of performance-based seismic design. It is also central to the estimates of building vulnerability used by the HAZUS earthquake loss methodology. However, despite the theoretical appeal of the methodology, recent comparisons of predictions using HAZUS with damaged recorded in the 1999 Kocaeli earthquake have shown poor agreement, not significantly better than those using intensity based methods. This paper examines further the reasons for this poor correlation, and uses a series of simple Monte Carlo simulations to examine the effect of varying degrees of brittleness and ductility on predicted damage distributions. These Monte Carlo simulations used the capacity spectrum method, and found that in allowing for the randomness of demand and capacity, the degree of brittleness markedly affected damage distributions in a way that appears plausible by comparison with observed data. This dependence on brittleness cannot be directly obtained from the original HAZUS methodology, nor from intensity based methods.
Keywords: Capacity spectrum, insurance loss, earthquake loss, HAZUS, Monte Carlo simulation

1.

INTRODUCTION

Predictions of the scale of future losses in earthquakes are vitally important to the insurance industry; inter alia, they are needed to set premiums and the size of financial reserves the industry should maintain. In the past, these predictions used empirical data, based on intensity. The methodology introduced by HAZUS (FEMA 1999), using the capacity spectrum method, offered a radical change in approach; the ground motion measure was a response spectrum, and so could be based directly on instrumentally recorded strong motion data, allowing a much more rational approach to the influence of factors such as frequency content and soil conditions on response.
1

edmund@booth-seismic.co.uk

Imperial College, London

CAR Ltd, Cambridge, UK

Moreover, structural vulnerability was expressed by a capacity curve, which could be directly related to fundamental structural parameters like yield strength, stiffness and ductility in a way that was impossible in an intensity based approach. However, an intensity based approach has its advantages, too. The data are very clearly based on real damage data, rather than theoretical predictions; structural calculation of damage has often been notoriously inaccurate. Moreover, when dealing with a very large population of buildings, an intensity based model with few parameters has some appeal over a HAZUS type model with many parameters which may be difficult to estimate on a country wide scale. Also, insurance personnel without a background in structural engineering or engineering seismology tend to prefer the simpler intensity approach to one producing answers from a complex black box. In 1998, two of the authors were involved in the preparation of seismic loss estimates for the whole of a large industrialized and highly seismic country (Turkey). Despite the reservations noted above, a HAZUS methodology, suitably adapted for local conditions, was adopted (Bommer et al., 2002). Early in the project, the devastating Kocaeli and Dzce earthquakes in Western Turkey occurred, creating a large amount of damage data for events which were rich in instrumental recordings, and which affected a large stock of rather uniform recent construction; these events provided an ideal (and rare) opportunity for validation. This paper discusses previous work using these data which suggested that the HAZUS methodology did not (in this case) provide significant predictive advantages over intensity based methods. The paper goes on to describe a Monte Carlo based modification to HAZUS which gives some insights into the reasons for these poor predictions of damage. 2. 2.1 PREVIOUS VALIDATIONS, USING TURKISH DAMAGE DATA

Zones dominated by ground shaking damage

A study by Spence et al. (2003) compared predicted damage with that reported by the AIJ (2001) and EEFIT (2003) at three locations in the epicentral area of the Kocaeli earthquake. The ground-shaking demand was defined using smoothed elastic spectra, selected to agree with the recordings from the Kocaeli earthquake, and modified to match local soil conditions; it was also defined in terms of intensity. The study focused on the predicted performance of reinforced concrete frame buildings, of either good or poor quality, and unreinforced masonry buildings. The estimation of damage using the capacity spectrum approach was based upon the HAZUS methodology (FEMA, 1999), with adaptations for Turkish conditions, while that using intensities employed vulnerability functions derived from damage data from previous Turkish earthquakes; further details are given in Spence et al. (2003). The study concluded that neither displacement-based (ie HAZUS-based) nor intensity-based approaches showed good agreement with the observed results (Fig. 1).

1 0.9 0.8 Proportion of total 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0
Undamaged Slight

0.7 0.6 0.5 0.4 0.3 0.2 0.1 0

Moderate

Extensive

Complete

MDR

Undamaged

Slight

Moderate

Extensive

Complete

MDR

Figure 1: Comparison of predictions and observed damage (Spence et al., 2002) 2.2 Zone including ground failure

a) Mid-rise RC frame buildings

b) Masonry buildings with RC slabs and roofs

Subsequently, the above study was extended to consider a region of extensive ground failure (Adapazari), in order to investigate the influence of ground failure on damage distributions (Bird et al., in press). Adapazari was damaged by a combination of ground failure, causing buildings to settle, rotate and slide on their foundations, and ground shaking. In this study, the intensity-based approach was not considered, partly due to the unsuccessful results obtained in the previous work, and additionally because of the inability of intensities to distinguish between the occurrence of ground failure and other damaging hazards such as local amplification due to soft soils. Again the damage estimation was based upon the HAZUS methodology, with some modifications where appropriate for the conditions in Adapazari. Liquefactioninduced damage was considered in two ways, either by increasing the site category to site class E, or by following the default methodology presented in HAZUS.
1 0.9 0.8 Proportion of total 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 None Slight Moderate Extensive Complete MDR Predicted - shaking Predicted - liquefaction Predicted - combined Observed 1 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 Mean Damage Ratio

Figure 2: Predicted vs observed damage to mid-rise RC buildings in Adapazari The results suggested that the additional work required to incorporate liquefaction into the damage methodology was not warranted in this case, since there was no obvious improvement in the results compared to those obtained ignoring liquefaction-induced damage. This is illustrated in Figure 2, where the third set of 3

Mean Damage Ratio

Displacement - poor Displacement - good Intensity - poor Intensity - good Observed

Displacement Intensity Observed

0.9 0.8

predictions shows the combined damage due to liquefaction and ground shaking, following the HAZUS methodology. Further conclusions from this study were as follows. Existing intensity scales were observed to be lacking in that they do not include ground-failure induced damage such as settlement or rotation of buildings. Without a unified scale for both ground shaking and ground-failure induced damage states, an accurate prediction of damage in liquefied areas will be almost impossible to develop. The importance of the foundation type on ground-failure induced damage was also noted. It was suggested that the modelling of liquefaction effects is less significant than the structural model used to determine the ground shaking-induced damage, which should therefore receive (initially at least) the greatest attention. 3 3.1 RMS beskat survey DAMAGE DISTRIBUTIONS

Figure 3 shows the locations of six out of the thirty or so sites where Risk Management Solutions (RMS) Inc conducted a survey of 4 7 storey (beskat) RC buildings (Johnson et al., 2000). The surveys were conducted within one week of the Kocaeli earthquake at sites within 10km of the fault. They were therefore all in the very near field of a large event and were conducted within days of its occurrence by a uniform team; moreover, the 6 sites considered in this study had apparently similar, stiff ground conditions. Figure 4 shows the damage distributions recorded; it can be seen that they are very different, and do not seem to be correlated strongly to features such as distance and direction from the source. Without a lot of further information, therefore, it would therefore be impossible to construct a single model that could get even fairly close to all these recorded data. However, at least the capacity spectrum method offers the chance to investigate further what parameters might be important in determining damage distribution, in a way not possible with intensity based measures. This realization informed the further studies reported below.

Figure 3: Locations of RMS beskat surveys shown in Figure 4 4

70% 60% Percentage in each category 50% 40% 30% 20% 10% 0% None Light Mod Heavy Part/total collapse 1: Adapazari 2: Izmit SE 3: Izmit W 4: Degirmendere 5: Hereke 6: Yalova W

Figure 4: Recorded damage from RMS beskat survey 3.2 Type I and Type II damage distributions

The term damage distribution refers to the distribution of the different damage states of none through to complete for a particular building class. Generally it is assumed that such distributions will have only one peak, with a roughly symmetrical spreads to either side of the peak. HAZUS (FEMA 1999) generally predicts such distributions (termed here Type I), found also in the damage probability matrices used in intensity-based methodologies such as ATC-13 (ATC, 1985). However, a number of damage datasets from the 1999 Kocaeli earthquake collated for this paper display damage distributions that did not follow this trend. Of the ten distributions presented in Figure 5, four have distributions defined here as Type II; they have two peaks, one at complete damage, and the other at a lower damage level. The salient features of the sites summarized in Figure 5 and the others considered in this paper, are presented in Table 1; no obvious reasons for the Type II distributions emerge. One solution for the prediction of these variable distributions, as well as the variability in recorded damage at similar sites noted in the previous section would be to consider the mean damage ratio (MDR) only, and to focus calibration efforts etc. onto this composite parameter, on the basis that the complexity of actual damage distributions cannot be reproduced using simplified methods. Whilst this may be appropriate for some cases, such as those where only the overall loss ratio is of interest, it is not conducive to improving damage estimation approaches and understanding the variables that influence the distribution of earthquake damage. This study therefore devotes itself to improving the understanding of why these different damage distributions occur and thus how to predict them in future earthquakes. Two causes for Type II damage distributions are now discussed. The first is that they represent damage caused by two independent, superimposed mechanisms.

80% Percentage in each category

Hereke

Yalova West

Izmit A

Izmit SE

Adapazari DegirA mendere

Izmit B

Glck Hill

Glck Coast

Adapazari B

60% 40%

20%

0%
N S M E C N S M E C N S M E C N S M E C N S M E C N S M E C N S M E C N S M E C N S M E C N S M E C

Figure 5. Selected damage distributions for mid-rise RC apartment buildings Table 1: Characteristics of sites shown in Figure 5. Dist.to Damage Source fault distribution Site Soils RMS 4 km Type I Stiff 1: Hereke RMS 4 km Type II Stiff 2: Yalova West RMS 7 km Type I Stiff 3: Izmit A RMS 3 km Type II Stiff 4: Izmit SE RMS Type I Stiff 8 km 5: Adapazari A RMS 1 km Type I Stiff 6: Degirmendere CAR 8 km? Type I Stiff 7: Izmit B AIJ 2 km Type I Stiff 8: Glck Hill AIJ 1 km Type II Soft 9: Glck Coast EERI 4 km Type II Soft 10: Adapazari B Ground failure Unknown Minor None None Major Moderate None None Minor None

Superimposition might for example arise in a survey zone which has buildings damaged by both ground shaking and liquefaction. Even within the relatively small zone of central Adapazari, two distinct types of damage were observed, either those with structural damage caused by ground shaking, or those on liquefied soil, which suffered foundation failures, manifested as settlement or tilting of the building, with little or no evidence of damage due to ground shaking in the form of cracks, deformation or partial or complete collapse. Although in many ways this is a plausible explanation for these damage distributions, there are a number of points that do not support it, namely: Table 1 shows no clear relationship between the type of damage distributions and the occurrence of extensive ground failure. As noted previously, intensity scales used in damage surveys such as these tend not to include liquefaction-related damage. Thus, by strict application of such scales, a building that had suffered major tilting as a result of ground failure, but otherwise remained undamaged structurally, should be classified as undamaged. Surveyors may make their own subjective decisions that a building that has

tilted is damaged beyond repair, but the application described above is the only one that can be interpreted without uncertainty. The study by Bird et al. (2004) discussed in Section 2.2 attempted to reproduce observed damage by superimposing ground shaking and ground failure damage, but without much success.

Therefore, the superposition of two separate damage distributions cannot entirely explain the Type II damage distributions. Another possible cause relates to structural ductility. In stiff and brittle structures, there may be little damage up to the point of effective yield. However, even small demands beyond this point may lead to a rapid increase in deflections and damage, due to loss of stiffness and strength. Such structures might therefore be expected to show either low damage or very high damage, with little in-between (e.g. Crowley et al. 2004). The non-ductile concrete frames with rigid, brittle and weak masonry infill found in the Kocaeli region might have had these characteristics, particularly since in many cases open ground floors were present. A model that gave a damage distribution dependent on ductility as well as yield strength and stiffness might therefore be of value. Unfortunately, the HAZUS methodology is of no direct help here. Damage distributions for a given seismic input are based on the expected deflection (the crossing point of demand and capacity spectra), and no account is taken of how far or near that point is from brittle collapse. The distribution of deflections around that expected point is assumed to be log-normally distributed, based on the central limit theorem and the assumption that both demand (ground motion intensity) and capacity (structural characteristics) are also normally or log-normally distributed. This latter assumption may be reasonable, but in a highly non-linear brittle system, with a cliffedge in response around fracture, the assumption that response is also lognormally distributed is unlikely to apply. To investigate this further, this study developed the model described in the next section. 4 MONTE-CARLO BASED SIMULATIONS FOR THIS STUDY

In order to investigate further the possibility that the Type II damage distributions discussed above were due to a brittle structural response, the HAZUS approach was modified as follows. The capacity spectrum method was still used for determining expected deflection, but the damage distribution was generated from a series of Monte Carlo type simulations, assuming that both the demand (ground) and capacity spectra were lognormally distributed. Figure 6 shows the idealized capacity spectrum, and Figure 7 shows the ground spectrum, which was a smoothed and simplified version of the SKR stiff soil recording from Kocaeli; Figure 7 also shows spectra recorded at the recording sites shown in Figure 3. 7

Figure 6: Idealized capacity spectrum assumed for this study


16 Spectral acceleration (m/s 2) 14 12 10 8 6 4 2 0 0.0 0.5 1.0 Period (secs) 1.5 2.0
Idealization used SKR, stiff, 3km IZT(X), rock, 8km IZT(Y), rock, 8km YPT(X), soft, 5km YPT(Y), soft, 5km

Figure 7: Idealized demand spectrum used for study, and Kocaeli spectra In Figure 6, au and dy are the spectral acceleration and displacement at yield, and 1 and 2 are the displacement ductility ratios to first loss of capacity and to fracture; they are assumed to be log-normally distributed random variables with a coefficient of variation c. c is the ratio of lower to upper yield plateaux strengths, taken as fixed for a given simulation. The mean 5% damped demand spectrum was taken as the idealized spectrum of figure 7, but this was also treated as log-normally distributed. Table 2 shows the various cases considered, and Table 3 the adopted parameter values. For the two base cases, called brittle and ductile, au and dy were given similar mean values to those used for 4-7 storey RC frames in the TCIP study referred to previously, but with the yield strength au increased by about 50% to allow for the effect of masonry infill. The mean values of 1, 2 and c and also coefficient of variation () values were chosen more arbitrarily, but look reasonable, for example in relation to standard values proposed in FEMA 356 or HAZUS. The relationships between peak deflection and damage state shown in Table 4 were also based on HAZUS; note that these values were taken as constant (ie not lognormally varying).

Table 2: Cases considered for this study CASE Description 1 Base case: brittle 2 Base case: ductile 3 Base case: brittle + low variability 4 Base case: ductile + high variability 5 CAR Ltd data for Izmit 6 RMS Inc data for Izmit South East Table 3: Values of parameters for Monte Carlo simulation CASE (see Table 2) 1 2 3 4 5 0.5 0.5 0.25 0.9 0.34 d for demand spectrum 0.5 0.5 0.25 0.9 0.41 c for capacity (au, dy , 1, 2) Viscous damping ratio 7% 0.4 , reduction factor on hysteretic energy au (m/s2) 3 3 3 3 2.5 yield spectral acceln dy (m) 0.11 0.11 0.11 0.11 0.11 yield spectral disp. 1.5 2.5 1.5 2.5 2.6 1 (ductility to loss of strength) 3 4 3 4 5.2 2 (ductility to fracture) c (ratio of yield to 0.2 0.4 0.2 0.4 0.55 residual strength) T (secs) - struct. period 1.2 1.2 1.2 1.2 1.32 6 0.27 0.30

2.4 0.1 1.42 3.5 0.35 1.28

Table 4: Assumed relationship between deflection and damage state Damage state Spectral deflection Approx equiv drift Assumed damage (mm) ratio None 0 - 38 0 0.5% 0% Low 38 68 0.5% - 0.9% 2% Moderate 68 126 0.9% - 1.7% 10% Extensive 126 330 1.7% - 4.5% 50% Complete >330 >4.5% 100% Figure 8a shows the predicted distributions for the brittle and ductile base cases; the Type II distribution shape due to the brittle cliff-edge changes to Type

I for the ductile case, as expected. Figure 8b show that decreasing the variability of the demand and capacity variables changes the brittle structures distribution to a (more-or-less) Type I one, and figure 8c shows that an increase in variability has the opposite effect for the ductile structure.
35% 30% Percentage damaged 25% 20% 15% 10% 5% 0% None Slight Moderate Extensive Complete

Base case: brittle Base case: ductile

a) Brittle & ductile base cases: cases 1 & 2


60% 50% Percentage damaged

Base case: brittle Base case: brittle + low variability


Percentage damaged

40% 35% 30% 25% 20% 15% 10% 5% 0%

Base case: ductile Base case: ductile + high variability

40% 30% 20% 10% 0% None Slight Moderate Extensive Complete

None

Slight

Moderate

Extensive

Complete

b) Brittle + low variability: case 3

c) Ductile plus high variability: case 4

Figure 8: Predicted distributions for brittle and ductile base cases


35% 30% Percentage damaged 25% 20% 15% 10% 5% 0% None Slight Moderate Extensive Complete MDR Base case: brittle Base case: ductile HAZUS: beta =0.7 HAZUS: beta =1.0 HAZUS: beta =1.3

Figure 9: Comparison of predictions with HAZUS methodology for base cases Figure 9 compares the damage predicted by HAZUS for the brittle and ductile base cases with those from the present study. The brittle and ductile structure have the same yield strength and deflection, and the expected deflection is below that 10

corresponding to strength reduction in the brittle structure for the adopted input spectrum. Hence the expected deflection is the same for both brittle and ductile structures, and the only parameter that HAZUS has available to affect the distribution is the value, given fixed values of the threshold deflections (Table 4). As the value changes from 0.7 to 1.3, it can be seen from Fig 4 that the HAZUS distribution shape changes from fully Type I to something approaching Type II, although the shape is not as markedly bi-modal as for the brittle base case from this study. The beta value for the HAZUS calculation represents a combination of the effect of ground motion (demand) and structural variabilities ( d and c in Table 3); the combined effect of d =c =0.5 (the base case considered here) is broadly equivalent to a combined of about 1 to 1.3. It may be noted that there appears to be no obvious reason why brittleness and a high value of structural variability should be linked, and the Monte Carlo model used here shows a marked advantage in this respect. Finally, two actual damage distributions recorded after Kocaeli were chosen (cases 5 & 6 of Tables 2 & 3, but numbered 4 & 7 in Table 1). The parameters available in the model were varied within reasonable bounds to see how closely the model distribution could match the recorded ones. The two right hand columns for cases 5 & 6 in Table 3 show the parameters chosen; the deflection values of Table 4 were assumed to remain applicable. The resulting distributions are shown in Figure 10. They are not of course in any way predictions, but the exercise suggests that reasonable results may be obtained from the model. Whether the building population recorded by CAR Ltd really was so much more ductile than that of the population recorded by RMS Inc. is of course another question.
50%

Simulation Recorded

60% 50% Percentage damaged 40% 30% 20% 10% 0%

Simulation Recorded

40% Percentage damaged

30%

20%

10%

0% None Slight Moderate Extensive Complete

None

Slight

Moderate Extensive Complete

a) CAR Ltd data for Izmit b) RMS Inc data for Izmit South East Figure 10 Comparison of simulations from this study with Izmit data 5. CONCLUSIONS

A weakness in the current HAZUS methodology for predicting earthquake losses is that it cannot directly account for the effect of structural brittleness or ductility on

11

damage distributions, and is no better in this respect than intensity based methods. This study has shown that a modification to HAZUS, using Monte Carlo simulations, redresses this deficiency in HAZUS in a way which (initially at least) appears plausible by reference to observed data from Kocaeli. Further work is needed to develop a viable and reliable tool for predicting losses for the insurance industry, but the approach described here appears a promising one to pursue. 6. REFERENCES

ATC (1985). ATC 13 Earthquake Damage Evaluation Data for California, Advanced Technology Council. AIJ, (2001). Report on Damage Investigation of the 1999 Kocaeli Earthquake in Turkey. Architectural Institute of Japan. Bommer, J.et al (2002), Development of a Turkish Loss Model for Turkish Catastrophe Insurance, Journal of Seismology, vol. 6, no. 2, pp. 431-446. Bird, J.F., Bommer J.J., Bray J.D., Sancio R.B., Spence R.J.S. Comparing Loss Estimation with Observed Damage in a Zone of Ground Failure: A study of the 1999 Kocaeli Earthquake in Turkey. submitted to Bull. Earthquake Eng. March 2004 Bray, J. D. and Stewart, J. P.(2000), Damage patterns and foundation performance in Adapazari, in Kocaeli, Turkey Earthquake of August 17, 1999 Reconnaissance Report, Earthquake Spectra, 16 (Suppl. A), 163-189. Crowley H, Pinho R and Bommer J (2004) A Probabilistic Displacement-Based Vulnerability Assessment Procedure for Earthquake Loss Estimation. Bulletin of Earthquake Engineering 2(2): in press EEFIT, (2003). The Kocaeli, Turkey earthquake of 17 August 1999. A field report by EEFIT. DAyala, D. and Free, M. (eds), Earthquake Engineering Field Investigation Team, Institution of Structural Engineers, London. FEMA, (1999). HAZUS 99 technical and user manuals. Federal Emergency Management Agency, Washington, D.C. FEMA (2000) FEMA 356: Prestandard and commentary for the seismic rehabilitation of buildings. Federal Emergency Management Agency, Washington. Johnson L A, Coburn A and Mohsen R. (2000) Damage survey approach to estimating insurance losses. Turkey Earthquake of August 17, 1999 Reconnaissance Report, Earthquake Spectra, 16 (Suppl. A), 163-189. Kircher C, Nasser A, Kustu O & Holmes W. (1997). Development of building damage functions for loss estimation. Earthquake Spectra 13(4) 663-682. Spence, R.J et al ., (2003). Comparing Loss Estimation with Observed damage: A study of the 1999 Kocaeli Earthquake in Turkey. Bulletin of Earthquake Engineering, 1 (1) 83-113

12

You might also like