rocking behavior of an instrumented unique building on the

Rocking Behavior of an InstrumentedUnique Building on the MIT CampusIdentified from Ambient Shaking Data

Mehmet Çelebi,a) M.EERI, Nafi Toksöz,b) and Oral Büyüköztürkb)

A state-of-the-art seismic monitoring system comprising 36 accelerometersand a data-logger with real-time capability was recently installed at Building54 on the Massachusetts Institute of Technology’s (MIT) Cambridge, MA,campus. The system is designed to record translational, torsional, and rockingmotions, and to facilitate the computation of drift between select pairs of floors.The cast-in-place, reinforced concrete building is rectangular in plan but has ver-tical irregularities. Heavy equipment is installed asymmetrically on the roof.Spectral analyses and system identification performed on five sets of low-amplitude ambient data reveal distinct and repeatable fundamental translationalfrequencies in the structural NS and EW directions (0.75 Hz and 0.68 Hz, respec-tively), a torsional frequency of 1.49 Hz, a rocking frequency of 0.75 Hz, andvery low damping. Such results from low-amplitude data serve as a baselineagainst which to compare the behavior and performance of the building duringstronger shaking caused by future earthquakes in the region. [DOI: 10.1193/032112EQS102M]

INTRODUCTION

Seismic monitoring of structures is an important part of the National EarthquakeHazards Reduction Program (NEHRP) in the United States. Under the Advanced NationalSeismic System (ANSS; http://earthquake.usgs.gov/monitoring/anss/, last accessed 27January 2011) authorized by U.S. Congress in 1999 (USGS 1998) and administered byUSGS, several structures have been instrumented throughout seismic areas of the UnitedStates. One of the criteria used by ANSS in establishing plans for the geographical dis-tribution of instruments on structures is population exposure to earthquake hazards(e.g., COSMOS 2001). The city of Boston and the vicinity is one of the populatedareas where exposure to seismic hazards is considerable. Ebel and Hart (2001) have com-piled evidence of damage and felt reports from events that occurred at distances of up to800 km away from Boston. New recorded evidence on ground and structures during the 11March 2011 Great East Japan earthquake shows that long-distance effects of earthquakescan be significant (Çelebi et al. 2014). The potential risk to built environments in parts ofBoston and the adjacent city of Cambridge is increased due to very weak geotechnicallayering between the surface and bedrock. As Britton (2003) has shown, amplificationof motions by a factor of 2 to 3 in the weak soil areas of Boston and Cambridge is possible.

Earthquake Spectra, Volume 30, No. 2, pages 705–720, May 2014; © 2014, Earthquake Engineering Research Institute

a) Earthquake Science Center, USGS, Menlo Park, CA. 94025b) Professor, Massachusetts Institute of Technology, Cambridge, MA

705

In the absence of response records from strong ground motions, measurements of structuralresponses to low-amplitude ambient motions can be useful to assess possible behaviors andperformances that might be expected for larger motions.

This paper augments previous results presented by Çelebi et al. (2011) for a state-of-theart seismic monitoring system recently installed with support from ANSS at Building 54 (alsoknown as the Green Building but hereafter referred to as “the building”) on the campus of theMassachusetts Institute of Technology (MIT) in Cambridge (MA). This installation is similarto others funded by ANSS (e.g., the Atwood Building in Anchorage, AK; Çelebi 2006). Ourstudy focuses on identifying dynamic characteristics and behavior of the building revealed byrandom recordings of ambient data from the monitoring system. Throughout the paper, well-established analysis techniques based on Fourier amplitude spectra, cross-spectra, coherencyfunctions, and phase angles (Bendat and Piersol 1980) are used. In addition, system identi-fication techniques (Ljung 1987) are used to validate the spectral analyses results and toextract critical damping ratios. Computation of drift ratios was not performed becausethe amplitudes of the data are too low to ensure reliable results; such comparisons mustwait responses recorded from larger input motions.

In this paper, we present analysis results for five data sets to show the repeatability ofresults derived using only one set of data.

THE BUILDING AND THE SITE

The building is cast-in-place, reinforced concrete with a mat foundation. It was con-structed in the early 1960s and opened to service in 1964. The building has 20 stories(87.3 m, or 286.5 ft., overall height) plus a one-story basement below ground level. Theshape is rectangular in plan—14.6 m ð48 ft:Þ � 34 m ð116.5 ft:Þ—with solid shear wallsextending from the basement-foundation level to the roof level at the two narrow ends.A picture of the building and its typical in-plan view are presented in Figure 1. Significant

(a) (b)

Figure 1. (a) Image of Building 54 at MIT Campus (captured using Google Earth©). (b) Typicalplan view of the building depicting the distribution of shear walls at the western and eastern endsof the building.

706 ÇELEBI ET AL.

features expected to influence the shaking response of the building include the monolithicshear walls at its two narrow ends, large openings in its narrow direction at ground floorlevel, and large and heavy equipment anchored asymmetrically at its roof. Furthermore,there are vertical discontinuities immediately above the ground level and at the levelabove it. The top two floors (19th and 20th levels) are mechanical floors. The narrowends of the building are aligned at 20° ccw from true north, but in this paper, the directionoff the narrow and wider edges of the building will be referred to as the structural NS and EWdirections, respectively.

Geotechnical aspects of the building site and its vicinity are well characterized. Theground surface is approximately ∼6.1 m (20 ft) above sea level, and depth to bedrock isapproximately 30.55–34.0 m (100–110 ft) below sea level. Figure 2a shows a typicaldepth versus shear wave velocity log from a site approximately 50 m (150 ft) from the build-ing (Jen 2011, Haley & Aldrich 1983). From this log, the site transfer function presented inFigure 2b is computed using software developed by Mueller (1997) based on Haskell’s shearwave propagation method (Haskell 1953, 1960). For these calculations, damping (ξ) is intro-duced via the quality factor Q, a term used by geophysicists and is equivalent to damping bythe relationship: ξ ¼ 1∕ð2QÞ; the Q values used range between 25 and 660 for shear wavevelocities between 100 m∕s and 600 m∕s, having been interpolated to vary linearly withinthese bounds. Thus, in the vicinity of the building the fundamental site frequency is estimatedto be ∼1.5 Hz. This value will be useful in assessing any site-related resonance of the build-ing that might occur during the shaking from an earthquake or other significant excitation.More detailed information on site conditions can be found in Britton (2003) and Haley &Aldrich (1983). It is important to add that using microtremor measurements and other meth-ods, Hayles et al. (2001) determined ∼1.4�1.5 Hz resonant frequency for a site cited as MITUniversity Park, which is part of the same campus as the building.

0 100 200 300 400 500 600−35

−30

−25

−20

−15

−10

−5

0

5

Vs (M/S)

(a) (b)

DE

PT

H (

M)

DEPTH−Vs PROFILE: GREEN BUILDING VICINITY

SURFACE AT 20 FT ~ 6.1 M

BEDROCK AT ~ −100 to −110 FT ~ −30.5 to −34M

0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 50

1

2

3

4

5

6

7

8

9

10

SITE TRANSFER FUNCTION: GREEN BUILDING VICINITY

SIT

E T

RA

NS

FE

R F

UN

CT

ION

FREQ (HZ)

Figure 2. (a) Borehole siteVS versus depth profile of site within ∼50 m (∼150 ft) of the buildingis assumed to be representative of the vicinity. (b) Site transfer function computed for the avail-able borehole log.

ROCKING BEHAVIOROF AN INSTRUMENTED UNIQUE BUILDINGON THE MIT CAMPUS 707

BUIILDING INSTRUMENT ARRAY

Figure 3 shows a schematic of the building depicting important dimensions as well aslocations and orientations of the 36 accelerometers deployed throughout the building. Eachaccelerometer channel is connected via cable to a 36-chhannel central recording system.

The accelerometers used in the building are Kinemetrics Episensors1 with a �4 g

full-scale recording capability and the recorder system is Kinemetrics Granite1. The acceler-ometers are powered via cables from the recorder with a �12 volt DDC, have a power con-sumption of 1.6 amperes and �2.55-volt output capability. As in most systems, the analogsignals are digitized at a very high sampling rate within the recorder, and then digital datais multiplied by a calibration constant based on the voltage output and decimated to the desiredsampling rate. The USGS obtains and serves the digitized data at 200 samples per second.

General information on structural monitoring procedures and suggestions for deploy-ments of accelerometers can be found in COSSMOS (2001) and Çelebi (2004). The choiceof locations for sensors are not based on mathematical formulas or computations.

Based on the budget and desired data for prescribed utilization, the locations are selectedbased on known structural behavioral patterns and optimizing the number of channels envi-saged. Therefore, it is no surprise that accelerometers must be located at the basement or thelowest level of a building and at the roof level. At the basement level, generally, at a mini-mum, a triaxial accelerometer is deployed to capture data to be used as input motions in abuilding. However, in this particular building, because of anticipated rocking behavior, wedeployed additional three vertical accelerometers, one at each of the corners in different loca-tions from where the triaxial accelerometer is located. The rest of the accelerometers aredistributed as three per level, at several levels (including the roof). Care is given to deployaccelerometers where there are vertical or horizontal irregularities in order to detect the effectof such irregularities on the response of the building; in this building, the two floors above theground level are such floors. The remaining accelerometers are distributed selectively at sev-eral pairs of neighboring floors to be able to compute the exact drift ratios between adjacentfloors, as well as average drift ratios between any two instrumented floors. In general, minordeployment adjustments are made when there are physical obstacles or objections by occu-pants at a particular floor. Such distribution along the height of the building helps to betteridentify and confirm higher modes and their shapes.

Accurate (<1μs) timing for data sampled at 200 Hz is synchronized to UTC by connectingto a GPS antenna deployed at the roof. The array is designed for recording the (a) transla-tional, (b) torsional, and (c) rocking motions of the building, as well as for computing (d) driftratios between adjacent floors or average drift ratios between non-adjacent floors. The cap-abilities of the state-of-the-art recording system include: (1) local and remote access to real-time data streaming, (2) selectable duration of pre-event signal, (3) local and remote access toany data in the buffer (4) easy transmission or retrieval of event data, (5) system health mon-itoring, and (6) local or remote on-demand ambient data recording.

1 Mentioning commercial names in the manuscript is for information only and does not indicate endorsement of themanufacturer or the products.

708 ÇELEBI ET AL.

SUMMARY OF AMBIENT DATA

To date, no earthquake response data has been recorded by the structural array in thebuilding. In the absence of earthquake responses data, and since completion of the monitor-ing system deployment in October 2010, we have made use of the “on-demand” and “remoterecording” capability of the system and obtained 90 sets of ambient response data with dura-tions of about 120–451 seconds as summarized in Table 1.

Figure 3. Schematic showing accelerometers deployed throughout the various floor levels ofthe building. The arrows indicate the orientation of positive acceleration for each sensor. Accel-erometers are located as close as possible to the east and west shear walls. Vertical accelerometersare deployed at the four corners of the basement. Level 20 (shown in dashed lines) is notinstrumented.

Table 1. Ambient data sets recorded (e.g., 20110201105927 refers the starting time ofrecorded data: Year: 2011, Month: 02, Day: 01, Hour: 10, Minute: 59, Seconds: 27)

Data description No. of sets Length of data

Individual sets

20101019161909 1 120 seconds20101019175021 1 160 seconds20101028210523 1 250 seconds

Once-a-day data20110201105927 8 sets 452 seconds each

to20110208105927

Once-every-hour data20110404113028 79 sets 331 seconds each

to20110407173028


ANALYSIS OF SELECTED AMBIENT DATA

A cursory analysis of several sets of ambient data (Table 1) indicated that the dynamiccharacteristics of the building generally appear to be repeatable for the low-amplitudemotions recorded. Therefore, in this paper, we present detailed data analyses for onlyone set that was recorded on 28 October 2010. Selected results from some other datasets are also presented to show the repeatability of the identified dynamic characteristics.

DATA SET OF 201010028-210533

Figure 4 shows a sample time-history plot of ambient acceleration data from the roof. Aswill be shown, torsion is significant in this building because (1) structurally it is torsionallyunbalanced, most likely due to the asymmetric shear walls around the elevator shafts on theeast end of the building (Figure 1), and (2) heavy equipment is asymmetrically installed at theroof and elsewhere in the building.

Figure 5a shows ambient vertical acceleration data of 250-second duration remotelyrecorded on demand from the four corners of the basement of thee building. Figure 5bshows 20-second window of the same data. These four vertical motions are used to inferrocking behavior of the building.

Spectral Analysis

Figure 6 shows amplitude spectra computed for the ambient acceleration data presentedin Figures 4 and 5. The signal-to-noise ratio of the ambient data is sufficiently high to be ableto clearly distinguish between the lowest (fundamental) frequencies (periods) [NS translation0.75 Hz (1.33 s), EW translation 0.68 Hz (1.47 s), and torsion 1.42 Hz (0.70 s)]. The top threespectra in the Figure 6a clearly depict these identified frequencies. It is noted that torsion

0 50 100 150 200 250

0

5

10

15

20

25

TIME(S)

AC

CE

LER

AT

ION

(CM

/S/S

)

GREEN BUILDING: AMBIENT:20101028−210523

CH35−CH36(TORSION)

CH36(NS)

CH35(NS)

CH34(EW)

5 10 15 20 25

0

5

10

15

20

25

TIME(S)

(a) (b)

AC

CE

LER

AT

ION

(CM

/S/S

)


CH34(EW)

CH35(NS)

CH36(NS)

CH35−CH36(TORSION)

Figure 4. (a) Ambient acceleration data of 250-second duration remotely recorded on demandfrom the roof of the building. Relative torsional motions (red curve at bottom of each frame) areindicated by the difference between two parallel channels at the roof. (b) 20-second window of thesame data.

710 ÇELEBI ET AL.

calculated by taking the difference between the two parallel accelerations recorded by chan-nels (denoted CH) 35 and 36 (bottom curve in the Figure 6a) does indicate same peaks (at1.42 Hz) in common with the fundamental translational frequencies but the translational fre-quencies (0.75 Hz [NS] and 0.68 Hz [EW] disappear in the torsional amplitude spectra.Hence, translational and torsional frequencies are not coupled. The second modes in theNS direction at ∼2.6 Hz (0.39 s) and in the EW direction at ∼2.45 Hz (0.41 s) are also visiblein the spectra. In the Figure 6b, spectra for all four vertical components of acceleration in thebasement are very similar and exhibit frequency peaks in common with those of the hori-zontal translational and torsional modes (e.g., note the minor peak at 2.4 Hz, which is most

0 50 100 150 200 250−0.2

0

0.2

0.4

0.6

0.8

1


AC

CE

LER

AT

ION

(CM

/S/S

)

TIME(S)

CH3(UP)

CH4(UP)

CH5(UP)

CH6(UP)

5 10 15 20 25−0.2

0

0.2

0.4

0.6

0.8

1


AC

CE

LER

AT

ION

(CM

/S/S

)

TIME(S)

(a) (b)

CH3(UP)

CH4(UP)

CH5(UP)

CH6(UP)

Figure 5. (a) Ambient vertical acceleration data of 250-second duration remotely recorded ondemand from the basement of the building. (b) 20-second window of the same data.

0 0.5 1 1.5 2 2.5(a)

(b)

3 3.5 4 4.5 50

1000

2000

3000

4000

AM

PLI

TU

DE

(CM

/S)


CH35−CH36

CH36(NS)

CH35(NS)

CH34(EW) [WEST END]

[WEST END]

[EAST END]

0.68 0.75

0.75

1.42

0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 50

50

100

150

FREQ(HZ)

CH3(UP)

CH4(UP)

CH5(UP)

CH6(UP)

0.68

0.75

1.42

Figure 6. Amplitude spectra of (a) horizontal motions and of torsion (curve at bottom) at theroof, and (b) vertical motions in the four corners of the basement.


evident for CH3, that corresponds to the second mode in EW translation). This clearly indi-cates an interaction between the basement vertical motions and the horizontal translationaland torsional modes, most likely by means of rocking, a form of structure-foundation-soil interaction. This observation is significant in that this interaction takesplace even during low-amplitude ambient shaking. (Further evidence of this interaction isshown later in the plots of cross-spectra, coherency and phase angles in Figures 8 and 9.)

The left frame in Figure 7 shows the cross-spectrum, coherency, and phase angles forCH35 at the roof and CH23 at the 12th level and depicts the fundamental and second-modefrequencies in the NS direction: 0.75 Hz and 2.60 Hz with high coherency and 0° and 180°phases respectively. The peak at ∼1.4 Hz belongs to the fundamental torsional motions (with0° phase angle). This is also more clearly depicted in the right frame of Figure 7 that showscross-spectrum, coherency, and phase angles for torsional motions CH35-CH36 at roof andCH23-CH24 at the 12th level. A small peak at ∼4.7 Hz indicates second torsional mode with180° phase angle.

Rocking

That rocking occurs in the building is demonstrated by cross-spectrum, phase angle, andcoherency functions between pairs of vertical motions s in the basement. Figure 8 showsplots for two pairs of vertical motions (CH3 & CCH4 and CH5 & CH6) around the EWaxis (i.e., rocking in the NS direction). Clearly, the peaks at ∼0.775 Hz and ∼22.5 Hz inthe cross-spectra, which have coherency nearly equal to unity and are out of phase by180°, indicate rocking in the NS direction. In contrast, Figure 9 shows two pairs of verticalmotions (CCH3 & CH5 and CH4 & CH6) around the NS axis, where at the same frequencies,

0

5

10

x 106

Sxy ROOF:CH35 [NS] & 12TH LEVEL(CH23) [NS]

−1000

100

PH

AS

E

0

0.5

1

CO

HE

RE

NC

E

FREQ(HZ)

GREEN BLDG: 20101028210523

0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5

0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5

0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5

02468

x 106

TORSION: ROOF(CH35−CH36) & 12TH LEVEL(CH23−CH24)S

xy

−1000

100

PH

AS

E

0

0.5

1

CO

HE

RE

NC

E

FREQ(HZ)

(a) (b)

GREEN BLDG: 20101028210523

0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5

0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5

0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5

Figure 7. Cross-spectra, coherency, and phase angles for selected components of motion. (a) Thefundamental and second-mode frequencies at 0.75 Hz and 2.60 Hz in the NS direction are clearlyillustrated using CH35 at the roof and CH23 at the 12th Level. The two modes are highly coherentand have phase angles of 0° and 180°, respectively. The peak at ∼14 Hz with 0° phase angle andcoherence of 1 belongs to the fundamental torsional mode. (b) Torsional motions at the roof(CH35–CH36) and at 12th level (CH23–CH24). The small peak at ∼4.7 Hz with 180° phaseangle corresponds to the second torsional mode.

712 ÇELEBI ET AL.

the motions have near-unity coherence but are in phase (0°), thus indicating no rockingaround the NS axis (EW direction).

System Identification

System identification analysis was performed using the ambient data to identify and/orvalidate key frequencies and compare them with those determined by spectral analyses.

0

5000

10000

Sxy

GREEN BLDG: 20101028210523

BASEMENT:CH3 [UP] & CH4 [UP]

−1000

100

PH

AS

E

0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 50

0.5

1

CO

HE

RE

NC

E

FREQ(HZ)

0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5

0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 50

5000

10000

Sxy


−1000

100

PH

AS

E

0

0.5

1

CO

HE

RE

NC

E

FREQ(HZ)

(a) (b)

GREEN BLDG: 20101028210523

0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5

0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5

0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5

Figure 8. Cross-spectra (Sxy), phase angle, and coherence for pairs (W end on the left, E end onthe right) of vertical sensors in the basement. Rocking around the EW axis (NS direction) isidentified on the basis of near unity coherency and 180° phase angle at ∼0.75 Hz and∼2.5 Hz in both plots.

0

5000

10000

Sxy


−1000

100

PH

AS

E

0

0.5

1

CO

HE

RE

NC

E

FREQ(HZ)

GREEN BLDG: 20101028210523

0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5

0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5

0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 50

2000

4000

Sxy

GREEN BLDG: 20101028210523

BASEMENT: CH4[UP] & CH6 [UP]

−1000

100

PH

AS

E

0

0.5

1

CO

HE

RE

NC

E

FREQ (HZ)

(a) (b)

0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5

0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5

0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5

Figure 9. Cross-spectra (Sxy), phase angle, and coherence for pairs (S side on the left, N side onthe right) of vertical sensors in the basement. Zero phase angles and near unity coherency at∼0.75 Hz and ∼2.5 Hz (for two pairs of vertical motions around NS axis from the basement)indicate that there is no rocking in the EW direction.


A model is estimated using appropriate pairs of recorded acceleration responses as single-input, single-output. The auto-regressive extra input (ARX) model based on least squaresmethod is used inn this analysis. The reader is referred to Ljung (1987) and the MatlabUser’s Guide (1998 and newer versions) for detailed formulations of the ARX and othersystem identification models. Some of the key frequencies for four of the importantmodes were identified along with the corresponding modal damping ratios. The resultsare summarized in Table 2.

REPEATABILITY (ANALYSES OF FOUR ADDITIONAL DATA SETS)

Four additional sets of data (20110201105927, 200110208105927, 201104405103028,and 201104406223028) are analyzed to assure that results are repeatable. For the sake ofbrevity, only limited results from each one of these data sets are presented.

The data set 20110201105927 (Figure 10) is representative of the amplitude spectra ofhorizontal motions at the roof and vertical motions at the basement. The results are similar to

Table 2. Summary of frequencies ( f ), periods (T), and damping percentages (ξ) determinedby system identification

MODE

TRANSLATIONAL

TORSIONALNS EW

f (Hz) T (s) ξ (%) f (Hz) T (s) ξ (%) f (Hz) T (s) ξ (%)

1 0.75 1.33 0.03 0.68 1.47 0.04 1.49 0.33 0.192 2.63 0.38 0.06 2.49 0.4 0.01 – – –

0

0.2

0.4

0.6

0.8

1 CH34(EW)

CH35(NS)

CH36(NS)

CH35−CH36

[WEST END]

[EAST END]

[WEST END]

.68

.75

1.42

GREEN BLDG:201102011059

AM

PLI

TU

DE

(CM

/S)

0

0.02

0.04

0.06

0.08

FREQ(HZ)

CH3[UP]

CH4[UP]

CH5[UP]

CH6[UP]0.75

0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5

0 0.5 1 1.5 2 2.5(a)

(b)

3 3.5 4 4.5 5

Figure 10. Relative amplitude spectra of (a) horizontal motions and of torsion (curve at bottom)at the roof, and (b) vertical motions in the four corners of the basement.

714 ÇELEBI ET AL.

those inn Figure 6. Again, similar to Figure 7, cross-spectrum, coherency, and phase anglespresented in Figure 11a for CH35 at the roof and CH23 at the 12th level depict with highcoherency the same fundamental 0.75 Hz and second-mode frequency 2.60 Hz in the NSdirection and with 0° and 180° phases respectively. The frequency peak at ∼1.4 Hz with0° phase angle belongs to the fundamental tensional motions, which is again more clearlydepicted in the right frame off Figure 11, which shows the cross-spectrum, coherency, andphase angles for torsional motions CH35–CH36 at the roof and CH23–CH24 at the12th level.

Data set 20110208105927 is used to demonstrate in Figure 12 that, as inn Figures 7and 11, the fundamental NS translational and torsional are similar and coherent as thoseidentified from the previous data sets.

Data set 20110405103028 is used to demonstrate in Figure 13 that, as in Figure 8, twopairs off vertical motions (CH3 and CH4 and CH5 and CH6) around the EW axis (that is,rocking in the NS direction) clearly exhibit the peaks at ∼0.75 Hz and ∼22.5 Hz in the cross-spectra, which have coherency nearly equal to unity and are out of phase by 180°, indicaterocking in the NS direction.

Data set 20110406223028 is used to demonstrate in Figure 14, as in Figure 9, that showstwo pairs of vertical motions (CH3 and CH55 and CH4 and CH6) around NS axis, where atthe same frequencies, the motions have near-unity coherence but are in phase (0°), thus indi-cating no rocking around the NS axis (EW direction).

EXTRACTION OF MODE SHAPES

In addition to system identification techniques previously used in this paper and based onsingle-input single-output (SISO) auto-regressive models based on least square methods,

00.5

11.5

2x 10−7

Sxy

GREEN BLDG:201102011059

ROOF:CH35(NS)&12TH.FL. CH23(NS)

-1000

100

PH

AS

E

0

0.5

1

CO

HE

RE

NC

E

FREQ(HZ)

0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5

0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5

0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 50

0.51

1.52

x 10−7

Sxy

GREEN BLDG:201102011059

TORSION:ROOF(CH35-CH36)& 12TH.LEVEL(CH23-CH24)

0100

PH

AS

E

0

0.5

1

FREQ(HZ)

(a) (b)

-100

0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5

0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5

0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5

Figure 11. Similar to Figure 7, from data set 201102011059, the cross-spectra, coherency, andphase angles for selected components of motion illustrate repeatability of the fundamental andsecond-mode frequencies at 0.75 Hz and 2.60 Hz in the NS direction, as well ∼1.4 Hz for thefundamental torsional mode.


more modern tools include time and frequency domain methods to extract mode shapes andassociated frequencies and damping. These modern methods are well-described in the litera-ture of the last two decades. Developing a long, comprehensive list of references and descrip-tions of methods is outside of the scope of this paper. Some pertinent methods (andreferences) include but are not limited to such methods as subspace identification for linearsystems as described by Van Overschee and De Moor (1996), and out-only frequency

0

2

4

6 x 10−10

Sxy

GREEN BLDG:20110405103028BASEMENT:CH3(UP)&CH4(UP)

-1000

100

PH

AS

E

0

0.5

1

CO

HE

RE

NC

E

FREQ(HZ)

0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5

0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5

0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 50

0.5

1 x 10−9

Sxy

GREEN BLDG:20110405103028

BASEMENT:CH5(UP)&CH6(UP)

-1000

100

PH

AS

E

0

0.5

1

CO

HE

RE

NC

E

FREQ(HZ)

(a) (b)

0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5

0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5

0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5

Figure 13. Similar to Figure 8, from data Set 20110405103028, the cross-spectra, coherency,and phase angles for pairs of motions (CH3 and CH4) and (CH5 and CH6) illustrate rocking withhigh coherency around the EW axis at ∼0.75 Hz and 1,800 out of phase.

0

1

2

x 10−8

Sxy

GREEN BLDG:201102081059

ROOF:CH35(NS)& 12TH.FL.CH23(NS)

-1000

100

PH

AS

E

0

0.5

1

CO

HE

RE

NC

E

FREQ(HZ)0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5

0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5

0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 501234

x 10−8

Sxy

GREEN BLDG:201102081059TORSION:ROOF(CH35-CH36)&12TH.FL.(CH23-CH24)

-1000

100

PH

AS

E

0

0.5

1

CO

HE

RE

NC

EFREQ(HZ)

(a) (b)

0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5

0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5

0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5

Figure 12. Similar to Figures 7 and 11, from data set 20110208105927, the cross-spectra, coher-ency, and phase angles for selected components of motion illustrate repeatability of the funda-mental and second mode frequencies at 0.75 Hz and 2.60 Hz in the NS direction as well ∼1.4 Hzfor the fundamental torsional mode.

716 ÇELEBI ET AL.

domain decomposition methods as described by Basserville et al. (2001) and Brinckner et al.(2001). A review of these methods is in Peeters and DeRocek (2001).

Based on these methods, there are several commercially (e.g., Artemis2 and MACEC2)and non-commercially developed software and methods available to extract modal shapesassociated with modal frequencies. There are functions in Matlab that will do the taskalso (e.g., N4SID). While acknowledging the availability of many choices in softwareand methods, needless to say, a list of references and descriptions is too long to list withinthe scope of this paper. Therefore, due to cost issues, we used a non-commercially developedand Matlab-based code by Johnson (1997). The code uses the eigen realization algorithmwith data correlation (ERA/DDC) method for extracting frequencies and mode shapes asdeveloped by Juang and Pappa (1985) and Juang et al. (1987). The software utilizes singularvalue decomposition (SVD) of the system Hankel matrix (also available in Matlab). Thematrix has columns populated by the correlation functions between all channels and a refer-ence channel. Simply stated, in order to get the best results, an analysis can be repeated fordifferent reference channels and increasing order of the model which is the number ofsingular values which are retained in SVD. In Figure 15, we show results of applyingthe Johnson ERA software for the extracted mode shapes of the building for the data setindicated in the figure. In the analyses, for the reference channel, data from channels inthe lower parts of the building (e.g., lower level in Figure 3) are used. It is noted that a secondtorsional mode could be extracted with the ERA method. For all the other modes, the fre-quencies determined compare well with those identified by spectral techniques as summar-ized in Table 2.

0

0.5

1 x 10−9

Sxy

GREEN BLDG:20110406223028

BASEMENT:CH3(UP)&CH5(UP)

-1000

100

PH

AS

E

0

0.5

1

CO

HE

RE

NC

E

FREQ(HZ)

0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5

0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5

0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5

0

0.5

1x 10−9

Sxy

GREEN BLDG:20110406223028BASEMENT:CH4(UP)&CH6(UP)

-1000

100

PH

AS

E

0

0.5

1

CO

HE

RE

NC

E

FREQ(HZ)(a) (b)

0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5

0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5

0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5

Figure 14. Similar to Figure 9, from data Set 20110405103028, the cross-spectra, coherency,and phase angles for pairs of motions (CH3 and CH5) and (CH4 and CH6) illustrate withhigh coherency and 00 in phase that there is no rocking at ∼0.75 Hz around NS axis in theEW direction.

2 Mentioning commercial names in this manuscript is for information only and does not indicate endorsement of themanufacturer or the products.


DISCUSSION AND CONCLUSIONS

Five sets of low-amplitude ambient vibration data remotely accessed from a state-of-the-art seismic monitoring system installed in a unique building on the MIT campus areused to infer vibration behavior of the building and its dynamic characteristics. One set ofdata is used to display detailed results of analyses, and the other four sets are used to showthat the results are repeatable. The data reveal distinct translational frequencies for the twomajor orthogonal axes (0.75 Hz NS, 0.68 Hz EW), a torsional frequency of ∼11.49 Hz, anda rocking frequency of 0.75 Hz. Very stiff shear walls at the east and west ends of thebuilding and weak geotechnical layers underlying its foundation provide ideal conditionsfor soil-structure interaction in the form of rocking behavior around the EW axis of thebuilding. Rocking around the EW axis and vertical motions around the NS axis is sche-matically depicted in Figure 16. Clear evidence of such rocking behavior is rarely observedin low-amplitude data.

In addition, a site frequency of ∼1.5 Hz is determined from a shear wave velocity versusdepth profile obtained from a borehole in the vicinity of the building. While the translationalfrequencies of the building are not close to the site frequency, the torsional frequency isalmost identical and may have contributed to resonant behavior during which the torsionalfrequency of the building is injected into the horizontal and vertical motions in the basementbecause of the rocking. In addition, the observation that the fundamental structural frequencyin the NS direction (0.75 Hz) also appears in the vertical motions of the basement suggeststhat these spectral peaks reflect rocking motions of the building at this frequency.

Finally, it has been observed that dynamic characteristics of structures duringstrong shaking can vary considerably from those that characterize low-amplitude shakingÇelebi 2007). Likely, this will be true for this building as well. Thus, the study resultsserve primarily as a guide and baseline to possible behaviors that might be expected inresponse to stronger shaking.

−1 0 10

10

20

30

40

50

60

70

80

90

HE

IGH

T(M

)

MODE1

MODE2

f1=0.74HZ

f2=2.56HZ

−1 0 10

10

20

30

40

50

60

70

80

90

NORMALIZED MODAL AMPLITUDE

GREEN BUILDING MODE SHAPES [DATA: 20101028210523]

MODE1

MODE2

f1=0.67HZ

f2=2.46HZ

−1 0 10

10

20

30

40

50

60

70

80

90 MODE1

MODE2

f1=1.46HZf2=2.56HZ

Figure 15. Extracted mode shapes using the eigen realization algorithm coded by Johnson(1997).

718 ÇELEBI ET AL.

ACKNOWLEDGEMENTS

The authors thankMIT for granting permission to USGS to deploy the monitoring systemin the building. Scott Wade, the building manager, provided access to the building, andMaryla E. Walters, the MIT Facilities Archivist, provided facility information. Jim Smithand Jason De Cristofaro of the USGS successfully installed the monitoring system. Theauthors gratefully acknowledge critical reviews by Roger Borcherdt and Chris Stephensof the USGS. Dr. Lucy Jen and Professor John Ebel provided invaluable geotechnical infor-mation. Dr. Rashanat Omrani helped with processing eigen realization algorithm used inextracting mode shapes.

REFERENCES

Basseville, M., Benveniste, A., Goursat, M., Hermans, L., Mevel, L., and Van der Auweraer, H.,2001. Output-only subspace-based structural identification: From theory to industrial\testingpractice, ASME Journal of Dynamic Systems Measurement and Control 123, 668–676.

Bendat, J. S., and Piersol, A. G., 1980. Engineering Applications of Correlation and SpectralAnalysis, John Wiley and Sons, New York, 472 pp.

Brincker, R., Zhang, L., and Andersen, P., 2001. Modal identification of output-only systemsusing frequency domain decomposition, Smart Materials and Structures 10, 441–445.

Britton, J., 2003. Microzonation of the Boston Area, M.S. Thesis, Boston College, MA.Çelebi, M., 2004. Structural Monitoring arrays – Past, present, and future, Proceedings, NATO

SFP Workshop on Future Directions on Strong Motion and Engineering Seismology,Kusadasi, Izmir, Turkey, 17–24 May 2004, in Future Directions in Strong Motion Instrumen-tation (P. Gulkan and J. Anderson, Eds.), NATO Science Series: IV, Earth and EnvironmentalSciences, Kluwer Academic Publishers, v. 58, 157–179.

Çelebi, M., 2006. Recorded earthquake responses from the integrated seismic monitoring net-work of the Atwood Building, Anchorage, (AK), Earthquake Spectra 22, 847–864.

E-W DIRECTIONVERTICAL MOTIONS ONLY NO ROCKING AROUND N-S AXIS

(b)(a)

N-S DIRECTION ROCKING AROUND E-W AXIS

cL

Figure 16. (a) Schematic depiction of rocking around EW axis and (b) no rocking around NSaxis (no scale).


Çelebi, M., 2007. On the variation of fundamental frequency (period) of an undamaged building–a continuing discussion, in Proceedings, Int’l Conf. on Experimental Vibration Analysis forCivil Engineering Structures (EVACES’07), Porto, Portugal, 24–27 October 2007, 317–326.

Çelebi, M., Toksöz, N., and Büyüköztürk, O., 2011. Revelations from ambient shaking data of arecently instrumented unique building at MIT Campus, in Proceedings, IOMAC 2011 (Int’lOperational Modal Analysis Conference, Istanbul, Turkey, 9–11 May 2011.

Çelebi, M., Okawa, I., Kashima, T., Koyama, S., and Iba, M., 2014. Response of a tall buildingfar from the epicenter of the March 11, 2011 M ¼ 9.0 Great East Japan earthquake andits aftershocks, Journal of the Structural Design of Tall Buildings and Special Buildings23, 427–441.

Consortium of Organizations for Strong-Motion Observation Systems (COSMOS), 2001. Pro-ceedings of Invited Workshop on Strong-Motion Instrumentation of Buildings, CP-2001/04.

Ebel, J. E., and Hart, K., 2001. Observational evidence of amplification of earthquake groundmotions in Boston and vicinity, Civil Engineering Practice 16, 5–16.

Haley, and Aldrich, 1983. Isoseismal/Geological Conditions Maps for Eastern Massachusetts,Unpublished report to the Massachusetts Civil Defense Agency (now the MassachusettsEmergency Management Agency), Framingham, MA.

Haskell, N. A., 1953. The dispersion of surface waves on multi-layered media, Bull. Seismolo-gical Soc. Am. 43, 17–34.

Haskell, N. A., 1960. Crustal reflection of plane SH waves, J. Geophysical Res. 65, 4147–4150.Hayles, K., Ebel, J. E., and Urzua, A., 2001. Microtremor measurements to obtain resonant fre-

quencies and dround shaking amplification for soil sites in Boston, Civil Engineering Practice16, 17–36.

Jen, L., 2011. Personal communication.Johnson, E. A., 1997. On-line Monitoring, Control, and Reliability of Structural Dynamical Sys-

tems, Ph.D. Thesis, University of Illinois at Urbana–Champaign.Juang, J. N., and Pappa, R. S., 1985. An eigensystem realization algorithm for modal parameter

identification and model reduction, Journal of Guidance, Control, and Dynamics 8, 620–627.Juang, J. N., Cooper, J. E., and Wright, J. R., 1987. An eigensystem realization algorithm using

data correlations (ERA /DC) for modal parameter identification, in Proceedings of the 5thInternational Modal Analysis Conference, 6–9 April 1987, Imperial College of Science &Technology, London, England.

Ljung, L., 1987. System Identification: Theory and User, Prentice Hall, Englewood Cliffs, NJ.Matlab Users Guide (1988 and newer versions). System Identification Toolbox for use with

Matlab, The Mathworks Inc., South Natick, MA.Mueller, C., 1997. Personal communication.Peeters, B., and De Roeck, G., 2001. Stochastic system identification for operational modal

analysis: A review, ASME Journal of Dynamic Systems, Measurement, and Control 123,659–667.

U.S. Geological Survey (USGS), 1998. An assessment of seismic monitoring in the United States:Requirement for an Advanced National Seismic System, USGS Circular 1188.

Van Overschee, P., and De Moor, B., 1996. Subspace identification for linear systems. KluwerAcademic Publishers, Dordrecht, The Netherlands.

(Received 21 March 2012; accepted 8 October 2012)

720 ÇELEBI ET AL.

rocking behavior of an instrumented unique building on the

Documents