Engineering Geology 79 (2005) 43 59 www.elsevier.

com/locate/enggeo

Evaluation of dynamic response and local soil effects of the Evripos cable-stayed bridge using multi-sensor monitoring systems
Vassilis Lekidisb, Maria Tsakiria,T, Konstantia Makrab, Christos Karakostasb, Nikos Klimisb, Issam Sousb
a b

School of Rural and Surveying Engineering, National Technical University of Athens, Greece Institute of Engineering Seismology and Earthquake Engineering ITSAK, Thessaloniki, Greece

Received 22 December 2003; received in revised form 13 September 2004; accepted 1 October 2004

Abstract The acquisition of response data from large structures and surrounding sites during earthquakes is essential to determine the effect of local soil conditions on seismic ground motion and consequently on the structures response. This paper studies the local soil effects on the ground motion based on recordings from a permanent accelerometer array installed on the cable-stayed bridge of the Evripos channel in Greece. Furthermore, the dynamic characteristics of the bridge using analytical and experimental methods, which include geodetic monitoring techniques, are investigated in order to provide a more realistic modelling of the bridge. This can result in reliable predictions of the bridge response in future earthquake excitations. D 2005 Elsevier B.V. All rights reserved.
Keywords: Bridge monitoring; Seismic response; Local soil effects; GPS; Standard spectral ratio; Horizontal to vertical ratio

1. Introduction The majority of cable-stayed bridges are designed with a relatively flexible structural system, which has a large number of long period eigenmodes. However, a differentiation of ground motion is present in such types of elongated structures. This kind of differentiation of ground motion is very often related with

T Corresponding author. Tel.: +30 210 7722728; fax: +30 210 7722728. E-mail address: mtsakiri@central.ntua.gr (M. Tsakiri). 0013-7952/$ - see front matter D 2005 Elsevier B.V. All rights reserved. doi:10.1016/j.enggeo.2004.10.015

the effect that local geology (dynamic properties of sediments and underlying rock formations as well as the geometry of sedimentsbedrock interface) has on the characteristics (amplitude, frequency content and duration) of recorded ground motion and consequently on the bridge response caused by seismic activity. Investigations to examine the effect of local soil conditions on earthquake ground motion (site effects), quantify the amplification and frequency content of ground motion on soil formations along major bridges in earthquake prone areas and assess the dynamic characteristics of these structures is critical for their

44

V. Lekidis et al. / Engineering Geology 79 (2005) 4359

performance. The use of analytical and experimental methods contributes towards a better understanding of the structures seismic performance by providing a more realistic modelling of the bridge. When a structure is excited dynamically, the response of the structure can be recorded and the parameters describing the dynamic properties of the structure, the modal parameters can be identified. This process, called experimental modal analysis, may be carried out in different ways depending on the size of the structure, the type of loading and the configuration of the instrumentation system. For this reason, special arrays such as permanent accelerometers are deployed covering not only various geological conditions close to active faults but the structure itself. Further to traditional sensors like accelerometers, advances in the use of modern surveying techniques, such as the Global Positioning System (GPS), offer an alternative method to augment existing monitoring regimes for relatively straightforward recovery of absolute displacements in a pre-oriented reference frame (e.g. Wieser and Brunner, 2002; Roberts et al., 2001; Wong et al., 2001; Watson and Coleman, 1998; Santerre and Lamoureux, 1997). Geodetic monitoring can provide measurements of relative displacements in near real-time and help identify the dynamic characteristics of vibrating systems. The information collected on the response of the structure during and after seismic events can be used by building operators and structural engineers to detect changes in the structures dynamic characteristics (such as eigenfrequencies). This paper presents the evaluation of the local soil effects on the ground motion and on the structures response based on recordings from a multi-sensor system, including a permanent accelerometer array and geodetic systems, installed on the cable-stayed bridge of the Evripos channel, in Greece. Due to its location, the Evripos cable-stayed bridge can be affected by far distance earthquakes of large magnitude which are known to contain energy in long periods (Papazachos and Papazachou, 1989; Newmark and Hall, 1987). Following a description of the geological and geotechnical characteristics of the bridge site, the effect of local soil conditions on earthquake ground motion and on the bridges response are examined using essentially empirical approaches. Two typical empirical methods, the

Standard Spectral Ratio (SSR) and the Horizontal to Vertical Ratio (HVSR), have been used on earthquake response data from several recent seismic events that have occurred in the broader area and have triggered the accelerometer network of the bridge. Based on the same seismic data, the analytically and experimentally derived modes of the bridge are compared. An investigation into the effectiveness of GPS measurements used as experimental data to provide the structural parameters from dynamic measurements is also performed. Finally, the paper concludes with a discussion on the agreement between the analytical and experimental approaches.

2. The Evripos cable-stayed bridge 2.1. Bridge geometry and permanent accelerometer structural array The Evripos bridge in Greece connects the island of Evia to the mainland. The cable-stayed section of the bridge measures 395 m in length, with a central span of 215 m and side-spans of 90 m each. The deck, 13.5 m in width, is at 40 m above sea-level, suspended by cables from two 90-m-high pylons. Four additional spans on each side, made of prestressed R/C beams on elastomeric bearings, complete the whole 694.5 m of the bridge length (Fig. 1). The present study concentrates on the dynamic behaviour of the central, cablestayed section of the bridge. A permanent accelerometer network of 43 sensors was installed on the Evripos bridge in 1994 by Institute of Engineering Seismology and Earthquake Engineering (ITSAK). Since then the bridges behaviour has been continuously monitored. The positions of the sensors were selected so that a complete description of the dynamic behaviour of the bridge is possible. Six vertical and two transverse sensors record the response at the middle span of the bridge, whilst six sensors are installed on each of the two piers. There are also four triaxial sensors, two located at the base of each pier and two free-field. The solidstate accelerometers are interconnected to provide common triggering, common sampling and common timing. The sampling rate of the accelerometers is 200 signals per second (sps). The instrumentation layout of the bridge is presented in Fig. 2.

V. Lekidis et al. / Engineering Geology 79 (2005) 4359

45

Fig. 1. The Evripos cable-stayed bridge (photo taken from North, with Evia island and Euboean coast on the left and mainland Boeotean coast on the right).

2.2. Geological and geotechnical characteristics at bridge site The geology of the area along with the geotechnical characteristics of the foundation subsurface, consisting of rockmass and soil layers, is briefly described in this section. The geological structure of the broader area as well as the main tectonic faults,

essentially controlling the broader area of the Evripos bridge, is presented in Fig. 3. The Boeotean coast (cf. Fig. 1) is formatted by thick bedded TriassicJurassic limestone and dolomites, with a mean inclination of 208 to 308 towards the NW. The Euboean coast consists of more recent Cretaceous limestone, essentially medium thick bedded (thickness ranging from 0.3 to 1.5 m), exhibiting an

Fig. 2. The permanent accelerometer structural array on Evripos cable-stayed bridge.

46

V. Lekidis et al. / Engineering Geology 79 (2005) 4359

Fig. 3. Geological models of the Evripos bridge area. (a) Model No. 1 (after Geomechaniki Ltd, 1985). (b) Alternative model No. 2 (after Marinos et al., 1994).

anticlinal structure and a mild inclination of 108 to 158 towards the SE, which coincides with the channels direction. Cretaceous limestone is ruled by three main systems of discontinuities: one of them is bedding and the remaining two main systems are joints (J1 and J2) perpendicular to the first system of bedding, thus delimiting orthogonal blocks of dimensions 1.5 to 5 m. Development of faults along the Evripos channel originates from confrontation of those two limestones.

This family of faults developed along NWSE or WNWESE direction, belongs to the same family that have conditioned the shape of the Euboean gulf. Some of them exhibited partial reactivation during past seismic events (faults of Atalanti). Therefore, development of geological formations at the examined bridge area is highly controlled by these faults. An approach of the geological model where the Evripos bridge is founded (called model No. 1 in Fig. 3a) is

V. Lekidis et al. / Engineering Geology 79 (2005) 4359

47

given in Geomichaniki Ltd. (1985). An alternative geological model (No. 2) is also provided (Marinos et al., 1994) in Fig. 3b, where Cretaceous limestone consists an overthrust nappe over the underlying serpentine complex, and this geological structure has been afterwards fragmented by gravitational faults. This geological model is essentially based on the assumption of a milder activity of faults of Euboean coast in the recent past. An evaluation based on combined results coming from site investigation and laboratory tests on rock cores (Geomechaniki Ltd, 1985; Marinos et al, 1994), concludes in a reliable description of the subsurface foundation of the bridges four piers (M4, M5: piers on Boeotean coast and M6, M7: piers on Euboean coast, cf. Fig. 2), as follows: M4 is located on Triassic limestone and ! Pier dolomites of Boeotean coast, moderately to slightly fragmented exhibiting relatively good mechanical behaviour. There is only a slight possibility of existence of karstic voids. The uniaxial compression strength (UXL) ranges mainly from 25 to 50 MPa, whilst the modulus of elasticity ranges from 1 to 2.2 GPa for rock cores. Pier M5, located on the Boeotean coast, is founded on piles ending in a serpentine complex developed in depths between 17 m and 49 m. The aforementioned complex is characterised by an inhomogeneous structure of poor to very poor mechanical behaviour, underlying a layer of clayey silts with limestone fragments. The bsoundQ bedrock consists of Triassic limestone and dolomites of identical mechanical behaviour as the one of M4 pier. The plan view of the pile cap of bridge pier M5 is 21.5 13.5 m and diameter of foundation piles is 120 cm while their length is about 27 m (Stathopoulos, 1994). Pier M6 belongs to the Euboean coast and is founded, via bored piles of 15 m length, on Cretaceous limestone, highly fragmented and occasionally presenting mechanical behaviour of clay gravel mixture. The plan view of the pile cap and the diameter of the piles of pier M6 is the same as for pier M5. Finally, pier M7 is also located on Cretaceous limestone of the Euboean coast. Those limestones are moderately to relatively high fragmented and

exhibit a fair to rather good mechanical behaviour. The possibility to meet karstic voids filled with clayey material is relatively higher compared to that of pier M4.

3. Effect of local soil conditions on earthquake ground motion and bridge response It is widely recognised that local geological conditions may have a large impact on ground motion at a given site. The geometry of the subsoil structure, the variation of soil types and its properties with depth, the lateral discontinuities, and the surface topography are usually at the origin of large amplification of ground motion and have been repeatedly correlated with damage distribution during destructive earthquakes (Aki, 1993; Bard, 1994; Faccioli, 1991; Cha vez-Garc a et al., 1996). Among the goals of engineering seismology and soil dynamics is to quantify the amplification and frequency content of ground motion on soil formations throughout metropolitan regions or along major infrastructures in earthquake prone areas. For this reason, special arrays such as that of the Evripos bridge are deployed covering various geological conditions close to active faults and record valuable data. In this section, results on the effect of local soil conditions on earthquake ground motion (site effects) are presented along with the possible effect that these may have on the bridge response using data from the four triaxial sensors located either at the base (piers M5 and M6) of each pier or at free-field (Boeotean coast pier M4 and Euboean coast pier M7). 3.1. Empirical approach In the last decade, the developments concerning empirical methods for estimating ground shaking at a particular site have been extensively reported (e.g. Finn, 1991; Aki and Irikura, 1991; Aki, 1993; Bard, 1994). Two of the most common methods implemented in this study are the Standard Spectral Ratio (SSR) and the Horizontal to Vertical Ratio (HVSR), whose applicability, advantages and drawbacks have been discussed in detail in the above references. The Standard Spectral Ratio, SSR (Borcherdt, 1970), is the ratio of Fourier amplitude spectra of a

! !

48

V. Lekidis et al. / Engineering Geology 79 (2005) 4359 Table 2 Earthquake recordings that were used in this study No. 1 M4 M5 M6 M7 No. 2 No. 3 No. 4 No. 5 1998 05 22 1999 09 07 2001 07 26 2003 06 18 2003 06 26

soil-site record to that of a nearby rock-site record from the same earthquake and component of motion. Source information is the same for this pair of records and when the two sites are closely located, the path effect is considered the same between the two. Then, the ratio of the Fourier amplitude spectra expresses the effect of the local soil conditions at the soil site in frequency domain (amplification and resonant frequencies). This technique is applicable only to data from dense, local arrays. A usual option for the selection of the reference station is a site of outcropping rock. The basic prerequisites for the application of this technique in the case of a reference station at the surface are: (a) the existence of simultaneous recordings at a soil site and at a reference site, (b) the reference site to be free of any kind of site effects (sediments and topography) and (c) the distance between the soil site and the reference site to be shorter than the epicentral distance, in order to consider that the effect of the propagating path of the seismic energy is the same for the two sites. In the case of the Evripos bridge array, both free field stations (M4 and M7) are good candidates to be used as reference stations based on the geological and geotechnical description of the sites. However, in order to determine the most suitable, the HVSR technique is applied. The basis of this technique is the spectral ratio of the horizontal to the vertical component of ground motion. The vertical component of the ground motion, in cases where the soil stratigraphy is flat and horizontal, is considered free of any kind of influences related to the soil conditions at the recording site. Theoretically speaking, a site free of any kind of site effects should have a horizontal to vertical spectral ratio equal to 1 over the whole frequency range. However, in practice this is not valid. According to Stiedl et al. (1996), when the horizontal to vertical (H/V) spectral ratio is around 1 for frequencies lower than 5 Hz, then this

site could be safely used as a reference one for the SSR technique. In order to compute the above spectral ratios for the bridge piers, real data were used that were recorded by the permanent accelerometer array of the Evripos bridge during earthquakes (Table 1). There was an effort to select data that were recorded from earthquakes with M wN 4.0 and had simultaneous recordings at all sites (piers M4, M5, M6 and M7). Table 1 shows the earthquakes with M wN 4.0 that have triggered the Evripos bridge. However, earthquake No. 3 was not recorded by the most distant instrument (M4), while the records for earthquake Nos. 4 and 5 at M6 were bnoisyQ and practically of no use. Table 2 summarises the earthquake recordings that were finally used in this study. Peak ground acceleration distribution along the longitudinal axis of the bridge is given in Fig. 4 for the three components of ground motion. Component L is the one parallel to the longitudinal axis of the bridge; T is the one perpendicular to the bridge axis, while component V is the vertical one. For all three components, the lowest peak acceleration values are observed at site M7 at the Euboean coast, independently of magnitude, azimuth and epicentral distance of the earthquake. Peak accelerations at sites M4 and M6 are in general of the same order, while the largest peak accelerations are observed at site M5. Those observations are consistent with local soil conditions at the sites under consideration. Increase

Table 1 Seismological information of the earthquakes that triggered the Evripos Bridge accelerometer array No. 1 2 3 4 5 Date 1998 1999 2001 2003 2003 05 09 07 06 06 22 07 26 18 26 Time GMT 16:22 11:57 00:22 05:25 13:45 uO N 38.495 38.059 39.039 38.62 38.61
O kE

Depth (km) 19 14 7 21 19

Magnitude (M w) 4.3 5.9 6.4 4.9 4.6

Distance (Km) 15 43 93 19 17

23.428 23.571 24.339 23.68 23.65

V. Lekidis et al. / Engineering Geology 79 (2005) 4359

49

Fig. 4. Distribution of peak ground acceleration along the main axis of the bridge (pier M4 to pier M7) for (a) L-component, (b) T-component and (c) V-component.

of peak acceleration values at site M5 seems to be attributed to the existence of a 17-m-thick soil layer (marine deposits) underlined by a 32-m-thick brockQ layer of very poor mechanical behaviour (serpentine complex) as opposed to the other sites that are located on limestone (weathered or highly fragmented). In a following step, computation of horizontal to vertical (H/V) spectral ratios for the free field sites M4 and M7 has been undertaken in order to conclude on the adequacy of one over the other to be used as reference station. For earthquake events in Table 1, the average spectral ratio has been computed for the two horizontal components (Fig. 5). It can be seen that for site M7, the amplification level for both horizontal components is essentially less than 2 for a wide frequency band up to 10 Hz, while respective amplification for site M4 is larger than 2, and in some cases greater than 3, for frequencies exceeding 2 Hz. Thus, the most adequate site to be used as reference is M7. This choice is in good agreement

with geological and geotechnical characteristics of Cretaceous limestone, presenting improved mechanical characteristics compared to Triassic limestone and dolomites. Once a reference site is established, the standard spectral ratio (SSR) technique can be implemented on earthquake recordings of Table 2 for piers M4, M5 and M6. Fig. 6 shows the average transfer functions of the two horizontal components of ground motion for the aforementioned sites, using the data collected during the earthquakes of Tables 1 and 2. These transfer functions represent modification of ground motion with respect to frequency of the recorded motion at site M7. In particular, the following observations can be summarised: 1) At site M4, the amplification of seismic motion is less than 4 for frequencies between 5 and 10 Hz, while per average in this frequency band, amplification is around 3.

50
10

V. Lekidis et al. / Engineering Geology 79 (2005) 4359

10

M4
Comp L Comp T

M7
Comp L Comp T

Amplification factor

Amplification factor
1 1 10

10

Frequency (Hz)

Frequency (Hz)

Fig. 5. Average horizontal to vertical (H/V) spectral ratios for the horizontal components of sites M4 and M7.

M4
10
Comp L Comp T

M5
10
Comp L Comp T

Amplification factor

Amplification factor
1 1 1 10

10

Frequency (Hz)

Frequency (Hz)

M6
10

Amplification factor

Comp L Comp T

10

Frequency (Hz)
Fig. 6. Average standard spectral ratios (SSR) for the two horizontal components of sites M4, M5 and M6.

V. Lekidis et al. / Engineering Geology 79 (2005) 4359

51

2) At site M5, amplification of ground motion is observed for a wider frequency band (N 3 Hz) with amplitudes greater than 5 (per average), while for the longitudinal component L and for a specific frequency (around 4.5 Hz), the observed amplification reaches values exceeding 10. 3) At site M6, amplification is observed for a very narrow frequency band between 6 and 9 Hz with values less than 5. 4) In any case, taking into account the range of the first 23 eigenfrequencies of vibration of the bridge (0.362 1.632 Hz, see Section 4), no amplification of ground motion is observed (transfer functions at this frequency range have almost unit amplitude) especially at sites M5 and M6 in which the instruments are placed at the pile cap (base) of each pier. All the above observations lead to the conclusion that (a) the effect of local geology in seismic motion might be of concern only for the case of site M5, where amplification is larger and for a wider frequency range, a fact that is consistent with the geological and geotechnical information for the area and (b) the effect of local geology on seismic motion does not affect the response of the bridge and vice versa, because in all cases amplification is observed at frequencies greater than those of the first 23 eigenfrequencies of the bridge, as derived from the analytical investigation of the bridge response (see Section 4). 3.2. Theoretical approach Concerning theoretical analyses, it is of common practice in geotechnical earthquake engineering to use the well-known 1-D equivalent linear response analysis in order to estimate the ground response at sites where soil or rock layers can be modelled horizontally and are subjected to transient, vertically

propagating shear waves. Although 2-D models can also be used to depict the combined effects of soil topography and soil properties on evaluating the expected ground motion, their accuracy and reliability are strongly related to the adequate knowledge of geometrical features and soil properties attributed to the model. According to Rassem et al. (1996, 1997), the 1-D model can only be used with good accuracy in shallow and wide valleys when the geometric dimensions seen in Fig. 7 follow the relationships B/D z 10 and B/L z 2. In the case of the Evripos channel, a crude geometrical approximation considers as rock boundaries the Boeotean (Cretaceous) limestone and dolomite, the Euboean (TriassicJurassic) limestone and the serpentine complex, despite their difference in mechanical behaviour. Recent marine deposits (sandy silts) and clayey silts with limestone fragments are also considered as soil deposits of the submarine valley, independently of their differences in mechanical characteristics. In particular, dimensions (Fig. 7) are as follows: 240 V 2B V 300 m, 15 VD V 30 m and 30 VL V 60 m. In this case: 8 V (B/D) V 10 and 2 V (B/ L) V 4, resulting that the use of 1-D model for the Evripos bridge is rather justified for site response analysis. However, the reliability of theoretical results imposes a relatively accurate knowledge of soil dynamic properties, as well as the dynamic properties and the depth of the bseismic bedrockQ. Lack of the relevant properties would imply a need of numerous parametric site-specific response analyses, the validation of which might be ambiguous and contradictory. It is considered, at the present level of knowledge, that the use of site dependent amplification factors according to site classification (Dobry et al., 1997; Klimis et al., 1999; Rodriguez-Marek et al., 2001; Anastasiadis and Klimis, 2002) would be

Fig. 7. The Evripos channel geometry crudely approximated (after Rassem et al., 1997).

52

V. Lekidis et al. / Engineering Geology 79 (2005) 4359

an appropriate index that corroborates the empirical approach results for piers M5 and M6. According to Unified Building Code of 1997 UBC/97, the site at pier M5 is classified in class C whilst the amplification factor ( F V) for periods between 0.4 and 2.0 s (0.5 to 2.5 Hz) is about 1.7 for the same site, thus coinciding with the amplification factors calculated in Klimis et al. (1999) and Anastasiadis and Klimis (2002). In cases where the shaking intensity of the input motion increases, the above amplification factor is expected to decrease. A direct comparison between the empirical approach and a site-dependent approach is of limited interest. This is because the results of the former approach are provided by a site-specific evaluation based on recordings of very low shaking intensity whereas the latter approach is based on a generic site categorization resulting from a large number of 1-D equivalent linear analyses (for the same soil category, but not specific to the examined site). However, the fact that amplification factors concerning site M5 are larger to the respective factors of sites M6 and M4 is an observation conforming to the conclusions of the empirical approach, in that local site effects might only be of concern for pier M5.

bridge, dynamic analyses of finite element models of the bridge have been performed, using both response spectrum and time-history methods (Lekidis et al., 1999). For the above analysis, the SAP2000 structural analysis program was employed. Specifically, 194 beam elements were used to model the piers and cables, and 554 shell elements were used to model the deck, resulting in a 4346 d.o.f. F.E. model with 749 joints. In the model, the piers were assumed clamped at their base, and rollers were assumed at the two deck ends that rest on the abutments. The eigenvalue analysis was run on an Intel Pentium 4, 1.7 GHz PC, requiring 19 s of CPU time. The results showed that for the first 23 eigenmodes (ranging from 2.764 s/ 0.362 Hz to 0.612 s/1.632 Hz), the corresponding cumulative modal participating mass ratios are 90.6% for the longitudinal direction and 80.1% for the transverse direction. For the mode shapes shown in Fig. 8, the period and frequency are respectively 2.764 s (0.362 Hz) for mode 1, 2.601 s (0.385 Hz) for mode 2, and 1.065 s (0.939 Hz) for mode 10. 4.1. Accelerometer data In Figs. 9 and 10, Fourier transforms derived from accelerometer data registering for earthquakes of September 1999 (No. 2) and July 2001 (No. 3) are given. The data illustrated refer to two sensor recordings: the first sensor (14-M5TNT, cf. Fig. 2) records the longitudinal response of the top at the northern end of the main pylon on the Boeotean coast. The second sensor (9-S2RNV, cf. Fig. 2) records the vertical response at the northern pavement at the middle of the central span of the bridge. It can be seen that both positions present a basic frequency at around 0.4 Hz (2.5 s). The same frequency is also observed on accelerometers located on the top of the main pylon at the Euboean coast (sensor 15-M6TNL cf. Fig. 2), indicating a strong interaction between these parts of the structure through the cable connection (Lekidis et al., 1999). From the above spectral analysis of the bridge using experimental data, it is possible to estimate the mode shapes of the bridge and the corresponding modal frequencies. Using phase information from the FFTs, it is found that the experimentally derived eigenvalue at 0.4 Hz corresponds to the first analytical

4. Comparison of analytical methods with experimental data from multi-sensor systems The relatively high seismic activity in the area of the Evripos bridge due to a number of faults as discussed in Section 2.2, and the effect of local soil conditions on the bridge response, make essential the monitoring of this heavily used transportation structure. In this section, results on the dynamic response of the bridge from earthquakes No. 2 and No. 3 of Table 1 are presented using data from the permanent accelerometer network of the bridge. The specific earthquakes were chosen because of their magnitude. In particular, earthquake No. 2 was the most recent catastrophic seismic event in Athens, resulting in the collapse of almost 50 buildings and the death of approximately 150 people. Also, results from GPS data are compared with the analytical response models of the bridge. In relation to the effects of local geology in seismic motions presented in Sections 2 and 3 for the Evripos

V. Lekidis et al. / Engineering Geology 79 (2005) 4359

53

Fig. 8. Three-dimensional structural model and mode shapes of the bridge.

mode of 0.362 Hz. In the same manner, the excited modal frequency at 0.9 Hz is found to correspond to the analytical 10th mode of 0.939 Hz. The results are shown in Fig. 11, which illustrates the analytical modes developed from finite element computations and the equivalent modes derived from the 26/7/2001

earthquake data. Clearly, the graphs show good correlation of the two methods. The derivation of the modal characteristics of long-span bridges using real measurements develops a better understanding of the vibration characteristics and the underlying excitation mechanisms. Therefore, the development

(a)

(b)
0.08

0.06
Event 7/9/1999 M5NL

0.06

Amplitude (g / Hz)

0.04

Amplitude (g / Hz)

Event 26/7/2001 M5NL

0.04

0.02

0.02

0 0 1 2 3 4 5 6 7 8 9 10

0 0 1 2 3 4 5 6 7 8 9 10

Frequency (Hz)

Frequency (Hz)

Fig. 9. FFTs of longitudinal recordings at top of pylon at Boeotean coast (sensor 14-M5TNL): (a) earthquake of September 1999, (b) earthquake of July 2001.

54

V. Lekidis et al. / Engineering Geology 79 (2005) 4359

(a) earthquake of September 1999

0.16 0.25

(b) earthquake of July 2001

0.2 0.12

Amplitude (g / Hz)

Amplitude (g / Hz)

Event 7/9/1999 9-S2RNV

Event 26/7/2001 9-S2RNV

0.15

0.08

0.1

0.04 0.05

0 0 1 2 3 4 5 6 7 8 9 10

0 0 1 2 3 4 5 6 7 8 9 10

Frequency (Hz)

Frequency (Hz)

Fig. 10. FFTs of vertical recordings at middle of the central span (sensor 9-S2RNV).

of a continuous on-line monitoring system collecting measurements under varying operating environments is very useful because it provides important information about the principal mechanisms of the bridge response. It also contributes towards the improvement of the analytical procedures that are used for the prediction of the seismic response, as well as for model updating, damage assessment, and evaluation of the original design assumptions.

4.2. GPS data Advances in the use of modern surveying techniques, such as the Global Positioning System (GPS), offer an alternative method to augment existing monitoring regimes of structures for relatively straightforward recovery of absolute displacements in a pre-oriented reference frame. Experimental geodetic data collected from the deck of the Evripos

Fig. 11. Comparison of analytically and experimentally derived modes of the bridge.

V. Lekidis et al. / Engineering Geology 79 (2005) 4359

55

bridge are used to derive modal frequencies comparable to the analytical modes. Geodetic Javad receivers (GPS/GLONASS Legacy) were used, two of which were employed as reference receivers located on bedrock, on pillars about 300 m away from the bridge, and two were positioned on the middle of two spans of the bridge where large displacements can be experienced. The use of two reference stations improves considerably the geometry and the reliability of the final solutions (Stewart and Tsakiri, 2001). Observations at 10 HZ sampling rate on both L1 and L2 frequencies were recorded in sessions of 4 h per monitoring point for a number of days. Additionally, a robotic total station (Leica TCA 1800) was employed to provide independent observations of the motion of the points. The robotic total station (RTS) was collecting data at a rate of about 8 Hz. Similarly, the instrument was located on a stable pillar outside the bridges area. The two monitoring points S1 and S2 were established on the concrete surface of the bridges deck, each one next to

an accelerometer (Fig. 12). A special prism and antenna mount was used so that simultaneous monitoring using GPS and total station measurements could be obtained, retaining mount verticality without using a tripod. The carrier phase data from both L1 and L2 frequencies were processed using in-house kinematic post-processing GPS software (Forward et al., 2001). The mean formal accuracies of the estimated displacements in the three components of north, east and height were 6 mm, 4 mm and 11 mm, respectively, for GPS. The robotic total station solutions yielded a precision of 56 mm for north, east displacements and 23 mm for absolute height displacements. The response of point S1 is shown in Fig. 13 in the form of spectra for the height component derived from total station and GPS data. The dominant loading effect was due to traffic on the bridge as there was no seismic activity during the experiment. Within the duration of the experiment, other effects such as wind or thermal effects were insignificant.

Fig. 12. Test monitoring points using geodetic techniques.

56

V. Lekidis et al. / Engineering Geology 79 (2005) 4359

Fig. 13. Spectral analysis of height displacement for point S1. (a) Total station power spectra. (b) GPS power spectra.

A dominant peak is seen at about 0.5 Hz in both graphs, which is also evident at the graphs of Fig. 10. This indicates that these geodetic survey systems are capable of identifying structural motions on the bridge of frequencies up to at least 0.5 Hz. It is interesting to notice the power at longer periods present in both graphs. Periods of up to 10 s have been detected with the total station. The GPS spectrum indicates more power at low frequencies caused by the long wavelength biases of the solutions, such as multipath. Furthermore the peaks in the GPS spectrum are more dspreadT than for the total station, as a direct result of the GPS solution being an order of magnitude noisier than the total station solutions. Similar spectral analyses for a suspension bridge have been reported by Xu et al. (2002). Clearly, there are more unmodelled residual errors in the GPS series with the most dominant being the multipath caused by the steel cables of the bridge and problems caused by weak satellite geometry. Poor signal quality is expected in this type of environment, but the results published by other authors for similar situations (e.g. Wieser and Brunner, 2002; Roberts et al., 2001) suggest that GPS may still be useful. Similar results were observed for monitoring point S2.

The components of the mode shapes can be determined from the values of the so-called btransfer functionsQ at the natural frequencies (Xu et al., 2002). Four modes were extracted by the FFT technique from the GPS solutions of point S1 and are given in Table 3. It can be seen that these correspond reasonably well to the first values of the theoretical modal frequencies of the bridge derived from finite element analysis (FEM). However, because the theoretical modal frequencies have very close values to each other, it is difficult to reliably resolve them using only the FFT technique. This is because its frequency resolution is rather low and damping estimates derived for this technique are not always consistent. Nevertheless, the dominant features are close to 0.5 Hz, which corroborate the results of Fig. 13.
Table 3 First four modal frequencies of bridge derived from geodetic data Mode Analytical values (FEM) Experimental values (FFT) Frequency (Hz) 0.369 0.389 0.449 0.543

Period (s) Frequency (Hz) Period (s) 1 2 3 4 2.764 2.601 2.283 1.812 0.362 0.385 0.438 0.552 2.708 2.566 2.227 1.843

V. Lekidis et al. / Engineering Geology 79 (2005) 4359

57

Although there are other attractive techniques available for modal identification and high frequency resolution, such as the auto-regressive moving average vector ARMAV which analyses the time series obtained from the measurements (Piombo et al., 1992) and most advanced such as the stochastic subspace identification (SSI) technique (Peeters and De Roeck, 1999), they are more applicable to the stationary hypothesis. Timefrequency approaches seem to be more efficient for structural identification and quantitative study of fast-changing transient signals. Future work will concentrate on the combination of techniques to identify reliably the vibration modes from free vibration response. This simple test nevertheless highlights that the use of geodetic measurements at sampling rates of 10 Hz can provide a clear measurement of displacement responses at lower frequencies and allows the extraction of the mode frequencies of the structure. Using the FFT technique for the spectral analysis of the measurements is a fast approach to structural identification in the frequency domain and can be easily implemented for on-line health monitoring to verify the quality of the measurements. As the sensors measure different quantities (displacement and acceleration) at different frequencies (8 Hz, 10 Hz for the geodetic sensors and 50 Hz for the accelerometers), it is impossible for the geodetic sensors to resolve the full range displacements. However, with 100 Hz GPS receivers already entered the market, there is an obvious potential for GPS to provide high frequency information along with the capability of measuring low frequencies making it a suitable technology for structural monitoring applications.

5. Concluding remarks The effect of local soil conditions on earthquake ground motion and consequently on the Evripos bridge response has been investigated with a set of earthquake recordings at the four ground stations of the permanent accelerometer array of the bridge. It is shown that the effect of local geology in seismic motion might be of concern only for the case of pier M5, where amplification is larger for a wider frequency range, a fact that is consistent with the geological and geotechnical information for the area.

It is also concluded that the effect of local geology on seismic motion does not affect the response of the bridge and vice versa, because amplification is observed at frequencies significantly higher than those of the 10 first eigenfrequencies of the bridge. For the case of the Evripos channel, a simplified geometrical approximation showed that the use of 1-D model is rather justified for site response analysis. However, due to lack of data on dynamic properties of soil and rock formations at the site, theoretical investigations would require numerous parametric site-specific response analyses, the validation of which might be ambiguous and contradictory. The dynamic characteristics of the bridge, derived from analytical investigations, are found to agree well with experimental results obtained from earthquake recordings of the permanent accelerometer array installed on the bridge. The experimental data are useful for a better understanding of the actual behaviour of the structure, and help in the development of more realistic analytical models. The latter can then be used for more reliable predictions of the bridge response in future earthquake excitations. Experimental data collected using geodetic techniques were also implemented to show that can identify the response signature of some natural or ambient excitations, such as traffic, wind, earthquake and their combination. In these experiments only response data of ambient vibrations were measurable but the actual loading conditions were unknown. While the data have succeeded in identifying key response modes only from traffic loading, these can be applied also to the modal identification of natural-excited responses, such as from earthquakes. The geodetic measurements can offer additional information to accelerometer data for the structural analysis and obtain more precise characteristics of the signal vibration. In this particular case, geodetic measurements can provide direct displacement measurements in the 0.12 Hz frequency range, complementing the accelerometer data which best operates in the high 120 Hz frequency range. Whilst the robotic total station solutions yielded a precision of 23 mm for absolute height displacements, the GPS results indicate that the bridge deck is not well suited for high precision monitoring due to shading and diffraction effects caused by the steel cables. GPS solutions revealed long period biases,

58

V. Lekidis et al. / Engineering Geology 79 (2005) 4359 Cha vez-Garc a, F.J., Sanchez, L.R., Hatzfeld, D., 1996. Topographic site effects and HVSR. A comparison between observations and theory. Bulletin of the Seismological Society of America 86, 1559 1575. Dobry, R., Ramos, R., Power, M.S., 1997. Site factors and site categories in seismic codes: a perspective. Proceedings of the FHWA/NCEER Workshop on the National Representation on Seismic Ground Motion for New and Existing Highway Facilities, Burligame, California May 2930, 1997, Technical Report NCEER-97-0010. Faccioli, E., 1991. Seismic amplification in the presence of geological and topographic irregularities. In: Prakash, S. (Ed.), Proc. 2nd Intern. Conf. On Recent Advances in Geotechnical Earthq. Engrg. and Soil Dynamics, March 1115, St. Louis, Missouri, vol. 2. Rolle, Univ. of Missouri, pp. 1779 1797. Finn, W.D.L., 1991. Geotechnical engineering aspects of seismic microzonation. Proc. 4th Intern. Conf. Seismic Zonation, August 2529, Stanford, California, vol. I. Earthquake Engineering Research Institute (E.E.R.I.), Oakland, CA, pp. 199 250. Forward, T., Stewart, M.P., Penna, N., Tsakiri, M., 2001. Steep slope stability monitoring using switched antenna arrays and permanent GPS networks. Proc. 10th FIG Int. Symposium on Deformation Measurements, 1922 March, Orange, California, USA. Geomechaniki Ltd., 1985. Geological and Geotechnical Research Study, vol. III. Ministry of Environment and Public Works, Greece. (in Greek). Klimis, N.S., Margaris, B.N., Koliopoulos, P.K., 1999. Sitedependent amplification functions and response spectra in Greece. Journal of Earthquake Engineering 3 (2), 237 270. Lekidis, V.A., Karakostas, C.Z., Talaslidis, D.G., 1999. Instrumentation, measurements and numerical analysis of bridges: an example of the cable-stayed bridge on Evripos Channel, Greece. In: Erdik, M., et al. (Eds.), Proc. Advanced NATO Workshop on Strong Motion Instrumentation for Civil Engineering Structures. Kluwer Academic Publishers, pp. 481 493. Marinos, P., Fytrolakis, N., Vainalis, D., 1994. The contribution of engineering geology in the construction of the Halkis high cable-stayed bridge. Proceedings of the 7th Congress, Thessaloniki, Greece, Bulletin of the Geological Society of Greece, vol. XXX/4, pp. 351 360. Newmark, N., Hall, W., 1987. Earthquake Spectra and Design. Earthquake Engineering Research Institute (E.E.R.I.). Papazachos, C.B., Papazachou, K., 1989. bEarthquakes in GreeceQ Book, pp. 8485, p. 356. Peeters, B., De Roeck, G., 1999. Based stochastic subspace identification in civil engineering. Proc. 2nd Int. Conference on Identification in Engineering Systems, Swansea, UK, pp. 639 648. Piombo, B., Giorcelli, E., Garibaldi, L., Fasana, A., 1992. ARMAV approach for large structure identification. Proc. 17th Int. Seminar on Modal Analysis, Leuven, vol. 2, pp. 721 732. Rassem, M., Ghobarah, A., Heiderbrecht, A.C., 1996. Site effects on the seismic response of a suspension bridge. Engineering Structures 18 (5), 363 370.

probably due to multipath and weak satellite geometry. However, GPS is capable of providing higher frequency information than the total station. A possible solution would be the direct integration of the GPS and total station raw data to provide an doptimisedT geodetic displacement solution combining the high frequency GPS data with the lower frequency, but more precise total station data. Integrated geodetic systems could be easily set up for continuous bridge monitoring, providing near real-time displacement solutions. Such systems would have advantages over accelerometer data, which require suitable processing involving double integration and requisite steps of filter choice for signal processing in order to compute displacements. Although it is not desirable to completely replace an accelerometer network with geodetic sensors, judicial placement of geodetic sensors could supply important additional information to allow fast determination of structural modal parameters for online health monitoring. The emergence of receivers with high logging rates (e.g. 100 Hz) is likely to make GPS a technique that can provide data directly comparable with accelerometer data in the high frequency range. The cost of a permanent monitoring network could be significantly reduced by the use of high precision GPS OEM boards as part of an augmented monitoring system, replacing accelerometers located on the top of pylons and in places of the structure with unobstructed satellite visibility. References
Aki, K., 1993. Local site effects on weak and strong ground motion. Tectonophysics 218, 93 111. Aki, K., Irikura, K., 1991. Characterization and mapping of earthquake shaking for seismic zonation. Proc. 4th Intern. Conf. on Seismic Zonation, August 2529, Stanford, California, vol. 1. Earthquake Engineering Research Institute (E.E.R.I.), Oakland, CA, pp. 61 110. Anastasiadis, A.J., Klimis, N.S., 2002. Effect of soil non-linearities and site characteristics on evaluation of site coefficients. Proc. 12th European Conference on Earthquake Engineering, 913 September, p. 498 (on CD-ROM). Bard, P.Y., 1994. Effects of surface geology on ground motion: recent results and remaining issues. Proceedings of the 10th European Conference on Earthquake Engineering, Vienna, Austria 1, 305 323. Borcherdt, R.D., 1970. Effects of local geology on ground motion near San Francisco Bay. Bulletin of the Seismological Society of America 60, 29 61.

V. Lekidis et al. / Engineering Geology 79 (2005) 4359 Rassem, M., Ghobarah, A., Heiderbrecht, A.C., 1997. Engineering perspective for the seismic site response of alluvial valleys. Earthquake Engineering and Structural Dynamics 26, 447 493. Roberts, G.W., Meng, X., Dodson, A.H., 2001. The use of kinematic GPS and triaxial accelerometers to monitor the deflections of large bridges. Proc. 10th FIG Int. Symp on Def. Measurements, 1922 March, California, pp. 268 275. Rodriguez-Marek, A., Bray, J.D., Abrahamson, N.A., 2001. An empirical geotechnical seismic site response procedure. Earthquake Spectra 17 (1), 65 87. Santerre, R., Lamoureux, L., 1997. Modified GPS-OTF algorithms for bridge monitoring: application to the Pierre-Laporte suspension bridge in Quebec City. The IAG Scientific Assembly, 39 September, Brazil. Stathopoulos, S.N., 1994. The construction of the High Cablestayed bridge on Evripos Channel. Proc. of the 11th Greek Reinforced Concrete Conference, May, Corfu, Greece, pp. 326 337 (in Greek).

59

Stewart, M.P., Tsakiri, M., 2001. The application of GPS to dam surface monitoring. Journal of Geospatial Engineering 3 (1), 45 57. Stiedl, J.H., Tumarkin, A.G., Archuleta, R.J., 1996. What is a reference site? Bulletin of the Seismological Society of America 86, 1733 1748. Watson, C., Coleman, R., 1998. The Batman Bridge: structural monitoring using GPS. Proc. of the International Workshop Advances in GPS Deformation Monitoring, 2425 September, Perth, Australia. Wieser, A., Brunner, F.K., 2002. Analysis of bridge deformations using continuous GPS measurements. Proc. of the 2nd Conference of Engineering Surveying INGEO2002, November, Bratislava, pp. 45 52. Wong, K.-Y., Man, K.-L., Chan, W.-Y., 2001. Moving Hong Kongs bridges, real-time kinematic spans the gap. GPSWorld. (July). Xu, L., Guo, J.J., Jiang, J.J., 2002. Timefrequency analysis of a suspension bridge based on GPS. Journal of Sound and Vibration 254 (1), 105 116.