You are on page 1of 15

Ambient Vibration Testing and Model Updating of a 44-Storey Building in Vancouver, Canada

Martin Turek1, Carlos E. Ventura2 and Sebastin Guerrero3 (1) Graduate Student (2) Professor Room 2010, Department of Civil Engineering, University of British Columbia 6250 Applied Science Lane, Vancouver, BC, Canada V6T 2E7 meturek@civil.ubc.ca; ventura@civil.ubc.ca (3) Engineer Glotman Simpson Vancouver, BC, Canada
The Melville is a 44-Storey reinforced concrete building located in downtown Vancouver, Canada. An ambient vibration test was performed on the building in its final stages of construction. In a typical setup, three measurements were taken on every second floor. The analysis using the frequency-domain decomposition and stochastic subspace identification techniques obtained the first 10 modes of vibration and damping estimates. These results were then used to update a FE model of the structure, to be used for seismic analysis studies. The results discussed in this paper are also used as a case study for the development of a design methodology for SHM systems. INTRODUCTION This paper presents the results of an ambient vibration test performed on a 44-storey concrete building. The Melville is located in downtown Vancouver, Canada, which is a region of high seismic risk. At the time of testing the building was under construction, with all of the concrete in th place up to the 44 floor and cladding up to the 35th floor. The tests provided information on the dynamic characteristics of the building (natural frequencies, mode shapes and damping) and information about the actual vibrations (amplitudes and noise levels, etc.). The test was performed for two reasons: to update an ETABS model used for seismic studies, and also as a case study for the development of a design methodology for ambient vibration based structural health monitoring systems. A description of the design methodology is presented in the next section. This paper will present a description of the building, the ambient vibration test, the results, the initial development of the full FEM and the updating of an equivalent FEM. DEVELOPMENT OF A DESIGN METHODOLOGY FOR SHM SYSTEMS Currently there exist many concepts for various types of SHM systems for civil engineering structures [1,2,3], but few of them are actually in operation. Part of the reason for this is a lack of knowledge of the performance of the damage detection techniques in a real scenario. This leads to a desire to develop a design methodology for SHM systems that will provide the level of confidence needed. The proposed design methodology described here attempts to evaluate these techniques under almost real conditions for a given structure. There are two ways to evaluate a damage detection technique: by a physical test of by a simulation. Physical tests [4,5] tend to be expensive, but can provide valuable insight into the real performance of the techniques. Simulations are easier to implement and are more flexible but tend to be less realistic. The development of any damage detection technique found in the literature is based on some sort of simulation. An example is from a benchmark simulation on the IASC/ASCE steel frame [6]. This design methodology looks at creating a calibrated vibration simulation using the results of an ambient vibration test.

The methodology can be split into two main components: the design of simulations to be used for the evaluation of the system, and the design of the damage detection methodology to be implemented. The process can be described as in Figure 1.

AV Test FEM-up Choose Damage Cases Simulation Add Noise Design of DD Methodology

Choose Sensor Locations Design Iterations

Figure 1: Proposed Design Methodology for SHM Systems

The six main steps of the methodology are as follows: i) ii) iii) iv) v) vi) Perform an ambient vibration test on the structure of interest Design and update a FEM of the structure Simulate the ambient level vibrations using the FEM Add noise to the simulated signals Select the damage detection method(s) appropriate to the application Choose sensor layout

The results of the ambient vibration test are of critical importance to the entire process. The dynamic characteristics are used in the updating of the model (Step ii). The ground level vibrations measured can be used as input to the FEM for the simulation, and the damping estimates obtained are also used in the simulation (Step ii). The noise levels are calibrated from the recorded signals (Step iv). Once the damage detection methodology has been determined, and the simulations created, the design process for the system can begin. The main aspect of the design is the layout of the sensors. The location and number must be chosen. This can be done in an iterative process that includes modifications to the damage detection methodology, and to the simulation, particularly the damage cases used. These iterations are necessary if certain damage cases cannot be identified. The final result of the process is to find the optimal sensor set with the minimum number. This design methodology does not focus on details such as instrumentation and electronics. Items such as choice of instrumentation, connections and computing systems are considered to be outside of this methodology. It is assumed that the technology available is now, or will be, sufficient to successfully implement the designed system. This paper focuses on the first two steps of the design methodology.

MELVILLE BUILDING The Melville is a mixed development in downtown Vancouver, composed of a high rise 46-storey residential tower at the west end of the development and a 10-storey hotel at the east end. Both share a common podium structure which includes three levels above ground and 5 underground parking levels. The total height of the residential tower is 146 m from the slab on grade at parking level 5 to the roof of the tower. The total height underground is 15 m. The podium structure height is 21m. A photo of the tower at the time of testing is shown in Figure 2. The typical tower floor plan and sensor layout is shown in Figure 3. The gravity system only carries vertical loads due to selfweight and occupancy, consisting of flat concrete slabs at the tower levels. The ground floor is a transfer slab, and in this case a slab/slab-band system was used. All the slabs are simply supported by columns distributed along the surface of the building. The foundations for the columns are simple spread footings. Along the perimeter of the building at the underground parking levels, continuous basement walls have the double function of carrying vertical loads as well as working as retaining walls for the soil surrounding the construction. A central core placed approximately at the centre of the tower defines the lateral system. The core foundation is a massive raft footing bearing directly on the natural soil. In addition a slab on grade was poured at the lower parking level directly bearing on the existing soil. AMBIENT VIBRATION TEST The AVT was performed in June of 2006, by a team of 7 people. For the test, the reference sensors were placed th with the data acquisition system on the 39 floor. Three teams of 2 were deployed to move the roving sensors from the top of the building down, and from the bottom of Figure 2: Photo of the Melville Building the parking garage up. Typically every second or third floor was measured. For each floor, a sensor layout as shown in Figure 3 was applied. This is a typical layout for a building that allows the translation in both directions and the torsional movements to be measured. The only measurement that varied from Figure 3 was for level P5, where only a single point was measured in both horizontal directions. For each setup 30 minutes of data was recorded at 500 samples per second. The long length of the tests was required due to the low frequencies of the building. A high sampling rate was chosen for two reasons. First, it has been found that in building tests previously [7] that more samples can provide a better definition of the mode shapes. Also, that for a difficult building to test such as this one, it is better to acquire more data than possibly necessary and decimate at a later time.

AMBIENT VIBRATION TEST RESULTS The data analysis was performed using the ARTeMIS Extractor software [8] using the five methods available. Those methods are the Frequency Domain Decomposition (FDD) Method, Enhanced FDD (EFDD), and three variations of the Stochastic Subspace Identification (SSI) method. The FDD methods are frequency domain methods which are based on taking a singular value decomposition of the cross-spectral density matrices of the signal set. The mode shapes are the singular vectors associated with the singular values, which essentially provide the amplitudes at a given frequency. The singular values at all frequencies are plotted, and these are analogous to the spectral density plots. The advantage of the SVD plots are that the effect of the noise in the signal is reduced, and closely spaced modes are easily identified. Figure 4 shows the plot of the singular value lines from the FDD analysis on the Melville data. The modes are numbered on the plot.

Figure 3: Typical Sensor Layout

The SSI method is a time-domain method which essentially computes the cross-correlations of the signal set, to obtain the matrices of a state-space formulation of the system. Those matrices are then decomposed into the modes and frequencies. Descriptions of the FDD and SSI can be found in [9] and [10] respectively. Damping is estimated from the EFDD and SSI methods in two different ways. In the EFDD, the spectral bell for a given mode is isolated and transformed to the frequency domain to identify its single-degree-of-freedom correlation function. The decay of this function defines the damping estimate. In the SSI technique, damping is estimated from the complex frequency.

Figure 4: SVD Plot of Complete Building Test (With Mode Numbers)

From the results of the analysis, 10 modes were identified under 5 Hz. Before the analysis, the data was decimated in MATLAB to 100 sps for better management of the data. Then during the actual analysis, it was decimated down to 10 sps, which allows for better results using the SSI technique. A total of 512 frequency lines were used. Tables 1 - 4 present the natural frequency, standard deviation of frequency, damping estimate and standard deviation of damping. For the FDD technique there is no information on the last three quantities so they are not presented. A complete set of modes shapes is presented in the Appendix, Figures A1-A10.

Table 1: Melville Building Natural Frequencies Method FDD EFDD SSI-UPC SSI-PC SSI-CVA Mode Descrip. Frequency [Hz] 1 1 E/W 0.313 0.309 0.303 0.303 0.303 2 1 N/S 0.342 0.346 0.330 0.330 0.329 3 1 Tor 0.645 0.648 0.644 0.648 0.649 4 2 E/W 1.230 1.228 1.228 1.227 1.226 5 2 N/S 1.406 1.413 1.416 1.416 1.417 6 2 Tor 2.021 2.022 2.024 2.024 2.023 7 2 E/W 2.803 2.801 2.813 2.823 2.810 8 2 N/S 3.311 3.320 3.315 3.317 3.318 9 3 Tor 3.516 3.518 3.509 3.507 3.510 10 4 Tor 4.805 4.773 N/A N/A N/A E/W East/West; N/S North/South; Tor Torsion; N/A Not Applicable

Table 2: Melville Building Standard Deviation of Natural Frequencies Method FDD EFDD SSI-UPC SSI-PC SSI-CVA -2 Mode Std. Dev. of Frequency [10 Hz] 1 N/A 0.546 0.590 0.563 0. 755 2 N/A 0.345 0.595 0.596 0. 545 3 N/A 0.333 0.381 0.885 1.174 4 N/A 0.309 0.387 0.442 0.708 5 N/A 0.462 0.287 0.170 0.351 6 N/A 0.386 0.300 0.258 0.295 7 N/A 0.852 1.148 3.084 0.897 8 N/A 1.750 1.399 1.270 1.357 9 N/A 1.121 0. 432 0. 695 0. 725 10 N/A 3.243 N/A N/A N/A

Table 3: Melville Building Damping Estimates Method FDD EFDD SSI-UPC SSI-PC Mode Damping Ratio [%] 1 N/A 3.861 1.156 1.038 2 N/A 2.361 2.099 2.034 3 N/A 1.665 1.873 1.828 4 N/A 1.003 1.164 1.239 5 N/A 1.038 0.518 0.668 6 N/A 0.950 1.006 1.096 7 N/A 0.515 0.957 1.137 8 N/A 1.167 1.423 1.391 9 N/A 0.815 1.124 1.059 10 N/A 0.340 N/A N/A

SSI-CVA 1.441 2.148 2.076 1.326 0.614 1.094 1.495 1.320 1.127 N/A

Table 4: Melville Building Standard Deviation of Damping Estimates Method FDD EFDD SSI-UPC SSI-PC SSI-CVA Mode Std. Dev. of Damping Ratio [%] 1 N/A 1.948 0.459 0.544 0.909 2 N/A 0.368 0.395 0.441 0.498 3 N/A 0.561 1.360 1.279 1.531 4 N/A 0.157 0.509 0.463 0.658 5 N/A 0.091 0.147 0.344 0.225 6 N/A 0.207 0.139 0.150 0.143 7 N/A 0.313 0.234 0.682 1.413 8 N/A 0.361 0.252 0.332 0.201 9 N/A 0.131 0.205 0.196 0.151 10 N/A 0.388 N/A N/A N/A

FINITE ELEMENT MODEL AND UPDATING One of the objectives of this study was to generate several computer models of the building and perform a sensitivity analysis until the natural frequencies and modes shapes were close to the ones obtained from the ambient vibration test. Four different models were developed using two different commercial software systems widely known in structural engineering, ETABS version 8.0 [11] for 3D analysis and SAP version 9.0 [12] for 2D. The first two were 2D models, and only the core in each direction (E/W and N/S) was modeled for the tower levels, and the core walls plus the basement walls for the basement levels. The first 3D model contained all of the structural elements present in the building including gravity system, columns, walls and the Lateral Resisting System (LRS) elements. A second 3D model was generated (Figure 5), including the exterior cladding and the partition walls, up to level 35 since this was the state of the construction at the time of testing. The results obtained with the 2D models were very similar to the ones obtained with the 3D model containing only the structural elements. This comparison was used as a verification of the 3D model before going to the more complex full 3D model, and also confirming that it is a valid assumption to just model the LRS elements since the gravity elements in this type of building do not have a large influence on the dynamic response of the building. The results obtained with the complete 3D model including the non-structural elements confirmed their influence at least in the initial response of the building; however, when the building is excited by major shaking such as by an earthquake it is expected that the nonstructural components will not have any significant influence in the response of the building. Table 5 presents a comparison between the frequencies of the various models. Table 6 gives the initial design comparison for the full 3D model. Test modes shown are from EFDD. Examples of the modes obtained with the 3D model are shown in the Appendix, Figure A11. Figure 5: ETABS Full 3D Model

Table 5: Comparison of All Models E/W [Hz] ETABS no cladding 0.175 SAP 2D no columns 0.162 Test 0.312 ETABS cladding + 0.270 drywall

N/S [Hz] 0.186 0.170 0.342 0.301

Torsion [Hz] 0.667 N/A 0.659 0.763

Table 6 Initial FEM Design Correlations Mode ETABS Test 1 0.270 0.309 2 0.301 0.346 3 0.763 0.648 4 1.316 1.228 5 1.471 1.413 6 2.381 2.022 7 3.030 2.801 8 3.571 3.320 9 4.000 3.518 10 5.000 4.773 UPDATING OF AN EQUIVALENT MODEL

direction 1 E/W 1 N/S 1 Tor 2 E/W 2 N/S 2 Tor 2 E/W 2 N/S 3 Tor 4 Tor

An equivalent model (Figure 6) to the one in ETABS was created for the purpose of updating the material parameters, while having a direct correlation to the test results. The equivalent model and updating was performed in FEMTools [13]. The equivalent model focused on the tower only, taking care to correctly model the core, while simplifying the floor layout. Each floor was taken to be a square, with total area equal to the floor area of the ETABS model, and with identical thickness. Similarly, the cladding was modeled with the identical thickness and properties, although the total surface area varied, since it was applied to the perimeter of each floor. To model the effect of the tower built on the parking garage, its base boundary conditions were converted to springs and the spring stiffness adjusted to match the translational modes. Only the first three modes were used for the updating. Table 7 shows the correlation between the equivalent model and the test data before updating for five modes. Note: In Figure 6, the equivalent model is shown with the measured test points (red dots). Those points lie outside of the geometry of the equivalent model; this is due to the fact that the real geometry is highly irregular, and the equivalent model, although having equal floor areas, is slightly smaller in the North/South direction. The points seen below the model are those measured in the parking garage. Figure 6: Equivalent Model

Table 7 Equivalent Model Correlations Before Updating Mode FEM [Hz] EMA [Hz] Diff. [%] MAC [%] 1 0.26 0.31 -15.76 97.1 2 0.31 0.34 -9.38 96.1 3 0.74 0.64 15.05 88.6 4 1.16 1.22 -5.20 90.6 5 1.70 1.41 21.04 82.9 The intention of the model was to examine the properties of the cladding and of the core for updating. In particular, it was of interest to find how each component and each property affects the ratio between the translational modes and the torsional mode. It is seen in Table 8 that for both of the translational modes in the model the frequencies are lower than the test; while for the torsional mode the frequency is higher. It was found when initially constructing the equivalent model that the translational modes between it and the ETABS model were matching, but the torsional mode of the equivalent model was too high (1Hz instead of 0.74 Hz). The ratio was corrected by removing some of the cladding, because the cladding had a more significant effect on the torsional modes than on the translational modes. Therefore it is believed that a similar effect is being observed between the frequencies of the ETABS model and those of the test. First a sensitivity analysis was performed examining the thickness and modulus of elasticity for the core and cladding elements. The sensitivity plot is shown in Figure 7. In the figure, Parameters 1 to 4 represent the modulus of the core and cladding elements, and parameters 5 to 8 represent the thickness of the three core elements and the cladding. All of these are global parameters, meaning a change is applied equally to all elements. Responses 1 to 3 are the natural frequencies of the first 3 modes. It was found that the MAC values were not very sensitive to changes in the parameters, so they were excluded from the analysis. Since the ratio between torisional and translational frequencies is not correct, the model must be updated in such a way that these modes will change at different rates. This is difficult, since most parameters will change both of them (ie either increase or decrease) in a similar way. Therefore, it is useful to examine the sensitivity matrix to find any parameters which affect either the translational or torsional modes only. It is seen that parameter 7, which is a core element thickness, is mostly sensitive to the two translational modes (shown in blue). The ideal process would then be to first update the entire model to match the torsional mode using the cladding (see Table 9) and then update the specific core element set that will change only the translational mode. This is deceptive however, since when the Figure 7: Equivalent Model Sensitivity Diagram thickness of a core element is changed both the stiffness and mass will change; meaning that depending on the location of the element the model frequencies may not change when the parameter is changed. Since the core element is an essential part of the structure, that is the case observed here. So instead, the model was updated in two ways to examine the results (both by adjusting the cladding modulus only): first to match the translational modes, then to match the torsional mode. The results are shown in Tables 8 and 9.

Table 8: Updating Results: Matching Translational Modes Mode FEM [Hz] EMA [Hz] Diff. [%] MAC [%] 1 0.30 0.31 -3.60 97.1 2 0.35 0.34 3.78 96.2 3 0.85 0.64 32.38 88.2 4 1.32 1.22 8.10 90.2 5 1.86 1.41 31.96 83.6 Table 9: Updating Results: Matching Torsional Mode Mode FEM [Hz] EMA [Hz] Diff. [%] 1 0.23 0.31 -25.98 2 0.27 0.34 -19.88 3 0.64 0.64 -0.72 4 1.05 1.22 -14.26 5 1.59 1.41 13.33

MAC [%] 97.1 96.0 88.8 90.6 80.6

In both cases, the mode shape correlation (MAC) did not change significantly. These tables show the expected results: when the translational modes are matched, the torsional frequency increases, with an increase in error of 17%. When the torsional mode is matched, the translational frequencies decrease, with an increase in error of 10% in each. In terms of actual property changes, the modulus increased by 100% for the translational matching, and decreased by 50% for the torsional matching. So it is seen that the torsional matching is slightly more efficient, creating less error and requiring less of a parameter change. A third method was also explored, in which all of the cladding elements were treated individually. Therefore the modulus of elasticity was updated locally for each cladding element. The results are shown in Table 10. Table 10: Local Updating Results: Cladding Modulus of Elasticity Mode FEM [Hz] EMA [Hz] Diff. [%] MAC [%] 1 0.29 0.31 -7.65 97.3 2 0.33 0.34 -3.74 96.1 3 0.74 0.64 14.53 92.9 4 1.16 1.22 -5.38 89.4 It is seen from the results in Table 10 that the error is more evenly distributed, while neither translational nor torsional modes are matching exactly. Also it is noted that while the translational modes have been better matched, the torsional mode did not change (see Table 7). This is different than from Table 8, where the translational modes matched while creating an increased torsional mode. The local updating then gives generally better results, however the parameter changes were seen to be concentrated near the base of the model. From the above results, it is likely that the best matching would come from a change in the arrangement of the cladding. Instead of having a global change in cladding properties for the ETABS model, a variation of properties across each floor will be necessary.

SUMMARY This paper presented the results of an ambient vibration test on a 44-story reinforced concrete building in Vancouver, Canada. The tests were successful at identifying 10 modes of the structure below 5 Hz. The results obtained using the five methods available with the ARTeMIS Extractor software were very consistent, identifying fundamental frequencies in the building of 0.312 and 0.342 Hz in the two translational directions. The damping values were typically between 1 and 2% from the SSI method. The FDD estimates were slightly higher, buy the SSI estimates tend to be more accurate.

The results were first used to validate several models used in a seismic response study. The results confirmed the overall behaviour of the structure, and allowed for a modification of the parameters to get a better match of the frequencies. The initial design of the model showed a relatively good match of the modes. The results of the updating study on the equivalent finite element model suggested that the distribution of the cladding will have the most effect on correctly matching the first three modes. The ETABS model described in this paper will be updated in more detail, and then used for simulations in a design of a SHM system for the real structure. ACKNOWLEDGEMENTS The authors wish to thank Juan Carlos Carvajal, Jose Centeno, Jorge Holmann, Kate Thibert and Shabnam Hoesseni, graduate students from the University of British Columbia for their hard work during the tests. The assistance and permission to access to the building of Keith Mackie of Amacon Construction is greatly appreciated. Micheal OKeefe, P.Eng from Glotman Simpson Consulting Engineers, Vancouver, B.C. for his support and provide of structural and architectural drawings. Celso Mendoza, P.Eng from Glotman Simpson Consulting engineers, Vancouver, B.C. for his time and counseling during the ETABS modeling. REFERENCES [1] Naeim, F., Hagie, S., Alimoradi, A., Automated Post-Earthquake Damage Assessment and Safety Evaluation of Instrumented Buildings, SMIP-05 Seminar on Utilization of Strong-Motion Data, Los Angeles, CA, May 2005 [2] Straser, E., Sohn, H., Kiremidjian, A., Law, K.H., A Framework for Health Monitoring of Structures, Proc. of ASCE Structures Congress, 2000 [3] Celebi, M., Sanli, A., Sinclair, M., Gallant, S., Radulescu, D., Real-Time Seismic Monitoring Needs of a Building Owner and the Solution, Earthquake Spectra, Vol. 20, Issue 2, pp333-346, May 2004 [4] Maeck, J., De Roeck, G., (2003) Description of Z24 Benchmark, Mechanical Systems and Signal Processing, Vol. 17, No. 1, pp. 127-131 [5] Farrar, C.R., Dynamic Characterization and Damage Detection in the I-40 Bridge over the Rio Grande, Los Alamos National Laboratory Report, LA-12767-MS, 1994 [6] Johnson, E.A., Lam, H.F., Katafygiotis, L.S., Beck, J.L., (2004) Phase I IASC-ASCE Structural Health Monitoring Benchmark Problem Using Simulated Data, ASCE Journal of Engineering Mechanics, Vol. 130, No. 1, pp. 3-15 [7] Turek, M., Thibert, K., Ventura, C., Kuan, S., Ambient Vibration Testing of Unreinforced Brick Masonry Buildings in Vancouver, Canada, Proceedings of IMAC XXIV, St. Louis, Mo., 2005, paper 77 [8] ARTeMIS Extractor Software Version 3.5, Structural Vibration Solutions, Inc., (Copyright 19992006) Structural Vibration Solutions, Inc.

[9] Brincker, R., Zhang, L., Andersen, P., Modal Identification from Ambient Responses using th Frequency Domain Decomposition, 18 International Modal Analysis Conference, San Antonio, Texas, pp 625-630, 2000 [10] Pridham, B.A., Wilson, J.C., (2004) Identification of Base Excited Structures Using OutputOnly Parameter Estimation, Earthquake Engineering and Structural Dynamics, Vol. 33, No. 1,pp 133-155 [11] ETABS Version 9.09, (Copyright 1984-2006), Computers and Structures Inc. [12] SAP2000 Advanced Version 10.0.9 (Copyright 1976-2006) Computers and Structures Inc. [13] FEMTools Version 3.1.2, (Copyright 1994-2006), Dynamic Design Solutions N.V.

APPENDIX MODE SHAPES

Figure A1: Mode 1 1 E/W (0.312 Hz)

st

Figure A2: Mode 2 1 N/S (0.342 Hz)

st

Figure A3: Mode 3 1 Torsion (0.645 Hz)

st

Figure A4: Mode 4 2

nd

E/W (1.221 Hz)

Figure A5: Mode 5 2 N/S (1.406 Hz)

nd

Figure A6: Mode 6 2 Torsion (2.031 Hz)

nd

Figure A7: Mode 7 3 E/W (2.803 Hz)

rd

Figure A8: Mode 8 3 N/S (3 Torsion) (3.311 Hz)

rd

rd

Figure A9: Mode 9 3 Torsion (3 N/S) (3.525 Hz)

rd

rd

Figure A10: Mode 10 4 Torsion (4.814 Hz)

th

Figure A10: Mode 1 - 1 E/W; Mode 3 1

st

st

Torsion; Mode 5 2

nd

N/S; Mode 7 3rd E/W

You might also like