Jtam.dvi JOURNAL OF THEORETICAL AND APPLIED MECHANICS 49, 3, pp. 825-839, Warsaw 2011 PERFORMANCE ASSESSMENT OF SELECTED OMA TECHNIQUES FOR DYNAMIC IDENTIFICATION OF GEOTECHNICAL SYSTEMS AND CLOSELY SPACED STRUCTURAL MODES Carlo Rainieri Giovanni Fabbrocino University of Molise, Structural and Geotechnical Dynamics Laboratory StreGa, Campobasso, Italy e-mail: carlo.rainieri@unimol.it; giovanni.fabbrocino@unimol.it Output-only modal analysis techniques and dynamic monitoring for vibration-based structural health assessment are primary tools for the investigation of the dynamic behavior even of very complex systems. An increasing attention towards these techniques and the opportunities they offer is rising also in the civil engineering field. In the present paper, the attention is focused on the possibility to enhance the knowledge about specific civil engineering systems using traditional and innovative opera- tional modal analysis (OMA)methods. Two case studies are presented: one is focused on the assessment of the performance of the Second Or- der Blind Identification for output-only modal analysis in the presence of closely spaced modes. In the other, OMA is applied to the simulated dynamic response of an embedded retaining wall. Encouraging results are obtained, pointing out that OMA can be confidently applied also to the analysis of wall vibrations induced by propagating waves. Key words:OperationalModal Analysis, SecondOrder Blind Identifica- tion, close modes, retaining wall, geotechnical structures 1. Introduction The continuous development of high performance materials for civil engine- ering and the enhancement of numerical methods for static and dynamic ana- lysis of structures have led to an increasing complexity of buildings and infra- structures. Inaddition, higher stress levels for optimal exploitation ofmaterials and structural solutions can be achieved, on the analogy with the approaches 826 C. Rainieri, G. Fabbrocino to structural design optimization of mechanical, aerospace and automotive systems. In a similar context, the technology transfer from mechanical and aero- space towards civil engineering applications has determined also an increasing interest in, and the enhancement of, advanced experimental and analytical techniques, such as those able to analyze the dynamic response of structures and provide the modal parameters to support calibration and validation of numerical models. Theirprimaryrole ispointedoutbychallengingproblemswhichoftenarise, for instance, from unexpected response levels due to dynamic user/structure interaction (Strogatz et al., 2005) and require experimental and theoretical insight in the dynamic response of structures. On the other hand, the evolution of dynamic properties of structures over their service life represents another key aspect in view of ageing and structural deterioration prevention and, above all, maintenance of critical components and systems. Regular identification of modal parameters, in fact, can play a relevant role in the development of effective structural health monitoring systems (Doebling et al., 1996). These circumstances led civil engineers to start exploiting a number of techniques, developed in the system identification and experimental modal analysis field, allowing the experimental identification of dynamic properties. Due to the dimensions of civil structures and the difficulty of exciting them properly, an increasing interest towards output-onlymodal identification (also calledOperationalModalAnalysis,OMA) techniques has risen over the years, with significant developments mainly in the last decade. OMA is based on measurements of the structural response to ambient excitation in order to extract themodal characteristics. This is the reasonwhy it is called also ambient or natural-excitation or output-only modal analysis. It is very attractive due to a number of advantages with respect to traditional input-outputmodal analysis (Mohanty, 2005; Cunha and Caetano, 2005): • it is faster and cheaper than input-output techniques; • no excitation equipment is needed, neither boundary condition simula- tion; • it does not interfere with the normal use of the structure; • it allows the identification ofmodal parameters which are representative of the whole system under real service conditions; • it is more suitable for automation and, therefore, for vibration-based structural health monitoring and damage detection applications (Rainieri and Fabbrocino, 2010). Performance assessment of selected OMA techniques... 827 Even ifmostOMAtechniques are derived fromtraditional input-outputmodal analysis procedures, the main difference is related to the basic assumptions about the input. OMAmethods are based on random responses rather than a deterministic input; thus they rely on a stochastic approach and can be seen as the stochastic counterpart of the deterministicmethodsused in the classical experimental modal analysis. OMA is based on the assumption that the input is aGaussianwhite noise, characterized by a flat spectrum in the frequency range of interest. As a conse- quence,modes are uniformly excited and extracted byappropriate procedures. Other assumptions are: • linearity: the response of the system to a certain combination of inputs is equal to the same combination of the corresponding outputs; • stationarity: the dynamic characteristics of the structure do not chan- ge over time, so that the coefficients of the differential equations are constant with respect to time; • observability: test setup must be defined in a way able to measure the dynamic characteristics of interest (for instance, nodal points must be avoided in order to detect a certain mode). However, practical applications of themethods show a certain tolerance about the compliance of the structural responsewith theabovementioned theoretical requirements. In other words, a sufficiently smooth input spectrum and weak non-stationarities andnon-linearities do not leadOMAmethods to fail and/or produce significant losses in accuracy. Extensive reviews of OMA methods can be found in the literature (Peeters, 2000; Rainieri, 2008; Zhang et al., 2005). They have been successful- ly applied to a large variety of mechanical systems and structures including cars and airplanes (Peeters et al., 2005), towers and bridges (Gentile, 2005). Innovative applications concern the identification, through field tests based on microtremor measurements, of the site period of a soil deposit (Ventura and Thibert, 2007) and the shearwave velocity and shear stiffnessmodulusprofiles (Carvajal and Ventura, 2009) for seismic design purposes. OMA is an active research area also from a theoretical point of view, with thedevelopment of newprocedures or adjustmentof algorithmsborrowed from different contexts and research fields. Among them, an increasing interest is raising towards the application of Blind Source Separation (BSS) techniques (Ans et al., 1985) to output-onlymodal analysis. A very promising procedure in this context is represented by the SecondOrderBlind Identification (SOBI) algorithm (Poncelet et al., 2007). However, limitations in the possibility to 828 C. Rainieri, G. Fabbrocino identify closely spaced modes are remarked in the literature (McNeill and Zimmerman, 2008). In the present paper, the application of traditional and innovative OMA techniques to the identification of challenging structural systems is investi- gated. In particular, the attention is first focused on flexible retaining walls, used to adapt soil profiles and ensure stability of excavated areas. The overall dynamic response of the system is the result of relevant soil-structure interac- tions and wave propagation phenomena. Thus, the dynamic identification of such a system represents a challenging task. Simulated data provided byFEM analysis are processed to assess the ability of OMA techniques to shed light on the dynamic response of such complex geotechnical systems. As a consequ- ence, the obtained results represent a preliminary validation step for the use of OMA techniques to the vibration-based Structural HealthMonitoring even of embedded retaining walls (Fabbrocino et al., 2009). In the second part of the paper, instead, the opportunities and limitations in the identification of closely spaced modes by SOBI are quantitatively inve- stigated through its application to simulated data obtained from a simple FE model. The results are also further validated through its application to a real case study. 2. The modal parameter identification techniques: basics Different methods for the output-onlymodal parameter estimation have been adopted in the present research: the (Covariance Driven) Stochastic Subspace Identification (Cov-SSI) (Peeters, 2000), the Frequency Domain Decomposi- tion (FDD) (Brincker et al., 2000) and the Second Order Blind Identification (SOBI) (Poncelet et al., 2007). The present section briefly reports the basis of the above mentioned reference methods. TheFrequencyDomainDecomposition is a frequencydomain,nonparame- tric output-only modal identification technique based on the Singular Value Decomposition (SVD) of the output Power Spectral Density (PSD) matrix. Structural resonances are identified from the singular value plots through a peak picking process; the corresponding singular vector is a good estimate of the mode shape. In the enhanced version of FDD, damping is estimated by Inverse Fast Fourier Transform of the Auto Power Spectral Density function of the Single Degree Of Freedom (SDOF) system corresponding to the mode, which is identified around the peak of the singular value plot by comparing themode shape estimate at the resonancewith the singular vectors associated Performance assessment of selected OMA techniques... 829 to the frequency lines around the peak and retaining all lines above a preset threshold of vector correlation. The Stochastic Subspace Identification, conversely, is a time domain, pa- rametric modal identification procedure based on a state-space description of the dynamic problem. The modal parameters are extracted from realiza- tions of the state matrix and output matrix of the system obtained from the measurements of its response to ambient vibrations through projections and other algebraic operators. InCov-SSI, the extraction of themodal properties is based on the preliminary computation of the covariance matrix of the respon- ses and on the construction of a Toeplitz matrix of the covariances (Peeters, 2000). When applied to OMA, SOBI can be referred to as a non-parametric time domain method, since no a priori model is fitted to the data but it simply takes advantage of their temporal structure to exploit themodal information. It is aBSS technique that, given a series of observed signals, aims at recovering the underlying sources. It is shown that, under given assumptions, the modal coordinates act as virtual sources (Poncelet et al., 2007), thus allowing applica- tion of such amethodology for modal parameter identification in output-only conditions. The definition of virtual sources (Kerschen et al., 2007) establishes a one-to-one relationship between themixingmatrix and themode shapes on onehand, and the sources and themodal coordinates on the other hand.Thus, the mode shapes are obtained from the columns of the mixing matrix, while natural frequencies and damping ratios are obtained through curve fitting of the sources. Themethod is based on second order statistics; in particular, the JointApproximateDiagonalization (JAD) of a number of correlationmatrices is achieved throughanumerical algorithm (Belouchrani et al., 1997). Themain limitation of themethod lies in the requirement of distinctmodal coordinates. Thus, a quantitative evaluation of the effect of the spectral difference on the accuracy of modal identification results is certainly of interest for practical applications. 3. Case study #1: The retaining wall In the present section, the identification, throughOMAtechniques, of the fun- damental dynamic properties of embedded retaining walls from the vibrations inducedby travellingwaves is analysed.The study is a part of awider research activity focused on static and dynamic monitoring of a full scale embedded retaining wall (Fabbrocino et al., 2009) and it represents a validation of the 830 C. Rainieri, G. Fabbrocino possibility to apply OMA to the analysis of complex geotechnical systems, heavily interacting with soil such as the embedded retaining walls. A sample FE model of a wall-soil, system has been set; soil properties adopted in the dynamic analyses are briefly reported in Table 1. Table 1. Soil properties adopted in the FEmodel Soil Material E0 G0 ν γ Vs HL HR layer type [105kPa] [105kPa] [–] [kN/m3] [m/s] [m] [m] A LE 3.188 1.113 0.432 18.00 246.2 8 3 B LE 4.086 1.433 0.426 19.03 271.6 3 3 C LE 14.51 5.045 0.438 19.47 503.9 5 5 D LE 26.49 9.243 0.433 19.98 673.3 20 4 HL [m] – Left of wall; HR [m] – Right of wall Due to a very low amplitude of ambient vibrations, linear elastic behaviour of the system in operational conditions can be assumed. As a consequence, a linear elastic model has been set, characterized by a medium mesh refined twice nearby the wall, absorbent boundaries, Rayleigh damping (1% for the fundamentalmodes) and a B/H ratio equal to 20, B and H being thewidth and the height of the model, respectively. A dynamic time-history analysis of the system has been carried out by applying a prescribed displacement at the base of themodel (representing the bedrock)as the input.Theapplied input is a zeromean,unitvarianceGaussian white noise, sampled at 100Hz and 3600s long. The dynamic response of the system to the applied input has been collected at ten uniformly distributed locations (2m far off each other) along thewall (modelled as a plate element). Then, the FDD and Cov-SSI methods have been applied to the simulated measurements of the wall response to carry out an output-only identification of the fundamental dynamic properties of the system. When FDD has been applied to the simulated data, a 66% overlap and a Hanning window have been used in spectrum computation. The obtained values of the fundamental frequencies and damping ratios of the system are reported in Table 2, while the mode shapes of the wall at the fundamental frequencies are shown in Fig.1. There are no inversions in the sign of the displacements, but in the second mode shape it is possible to observe a change in the curvature. The MAC (Allemang and Brown, 1982) computed between the shapes provided by two OMA methods (Table 2) and comparisons in terms of estimated fundamental frequencies point out a fairly good agreement. The lower value of the MAC for the second mode can be Performance assessment of selected OMA techniques... 831 addressed to numerical problems: in fact, themode shape vectors provided by Cov-SSI showsome slight imaginary components at thepositions characterized by the lower displacements (bottom part of the wall) and they affect the correlation with the (real) vectors provided by the FDD. Fig. 1. Mode shapes of the wall at its fundamental frequencies Table 2.The embedded retaining wall: summary of the results # fFDD fSSI ∆f ξSSI MAC(ψFDD,ψSSI)[Hz] [Hz] [%] [%] I 3.65 3.65 – 1.03 ≈ 1 II 7.45 7.45 – 0.97 0.98 The consistency of the cross checks reveals that OMA techniques can be successfully applied to the output-only modal identification of embedded re- taining walls. 4. Case study #2: SOBI and closely spaced modes In this section, the attention is focused on the performance of SOBI for the output-onlymodal identification of structures characterized by closely spaced modes. Since the separation of the sources from their mixture by SOBI is based on their temporal structure (the source separation is possible if the sources are stationary and have different autocorrelation functions, apart from their statistical distribution), modes are required to be characterized by a certain spectral difference to be identified and it is not possible to deal with repeated frequencies. The problem of identification of closely spaced or even coincident 832 C. Rainieri, G. Fabbrocino modes is already reported in the literature (see, for instance: Zhou and Che- lidze, 2007; McNeill and Zimmerman, 2008). In particular, it is stated that distinct modal coordinates automatically satisfy the requirement about the spectral difference (Zhou andChelidze, 2007);moreover, it is possible to achie- ve separation even in the case of small spectral differences if themode shapes partitioned to the sensor locations are linearly independent (McNeill andZim- merman, 2008) and this can always be accomplished by a judicious choice of sensor locations. Apart from this qualitative information, a quantitative as- sessment of the performance of SOBI for the output-onlymodal identification of structures characterized by closely spaced modes is still missing. In order to assess the accuracy and reliability of results and the quality of separation in the case of close modes, a simple FE model has been set, and its stiffness changed in order to control the spacing of the first twomodes quantifiedbythemodaloverlap factor (SrikanthaPhaniandWoodhouse, 2007) µn = fnζn fn−fn−1 (4.1) where fn and ζn represent natural the frequency and damping ratio of the n-th mode, respectively. TheFEmodel has been implemented through the SAP2000 v.12 computer code and it is a 1 story, 1 bay 3D reinforced concrete framewith the following characteristics (units: kg andm): • dimensions of the beam section: 0.30×0.30; • dimensions of the column section: minor dimension equal to 0.30; ma- jor dimension ranging from 0.32 to 0.35; the larger inertia is in the y direction; columns are fully restrained; • beams and columns provide only stiffness; the mass is applied to the master joint (center of the mass) at the level of the roof; the nodes at the floor level are constrained by a rigid diaphragm; • the values of masses in the master joint along x, y are mx = 40000, my =44000, respectively, while the rotational mass is mRz =40000; • constant damping ratio equal to 0.01; • Gaussian white noise (3600s long, sampled at 100Hz, zero mean, unit variance) is applied as the input ground motion in both the x and y directions and a linear modal time history analysis is carried out; • responses in the x and y direction are collected at the four corners of the floor; nonoise is added to the simulatedmeasurements; each data set is decimated before processing, obtaining the final sampling frequency equal to 4Hz. Performance assessment of selected OMA techniques... 833 A picture of the FE model is shown in Fig.2. The modal overlap factor between the first two modes, which are close, pure translational modes, ran- ges from 17% to 65% (very close modes); also the case of nearly repeated frequencies is investigated. Fig. 2. FE model for the performance assessment of SOBI in the case of closely spacedmodes: mass assignment (a), sections (b) The output-only modal identification by SOBI has been carried out by adopting p =500 and t =1 ·10−8 as the values of the number of correlation matrices to be jointly diagonalised and the threshold to stop JAD, respectively (Cardoso and Souloumiac, 1996). The assessment of SOBIperformance for the output-onlymodal identifica- tion of closely spacedmodes has been carried out by comparing the its results with the natural frequencies and mode shapes obtained from the numerical model. In particular, the scatter between the natural frequencies of the FE model and the estimated values by SOBI is computed; the correlation betwe- en the mode shapes of the numerical model and the estimates provided by SOBI is evaluated through theMAC index (Allemang and Brown, 1982) as MAC(ψnSOBI ,ψ n FEM)= |{ψnSOBI} ⊤ψ n FEM | 2 ({ψnSOBI} ⊤ψ n SOBI)({ψ n FEM} ⊤ψ n FEM) (4.2) where ψnFEM and ψ n SOBI are the mode shapes of the n-th mode provided by the FE model and SOBI, respectively. A high correlation is pointed out by values close to 1; conversely, the closer the MAC value to 0, the poorer the correlation. The results are summarized in Table 3. They point out that fairly good modal identification results can be obtained even in the case of closely spaced modes. The sources are properly separated (Fig.3) and the natural frequencies are accurately identified. Also the mode shape estimates are generally fairly accurate, if a minimum spectral difference is ensured. In 834 C. Rainieri, G. Fabbrocino Table 3.Closely spaced modes: SOBI performance assessment for OMA Larger Modal M1,1 M2,2 M1,2 section overlap f1,F f2,F f1,S f2,S ξ1,S ξ2,S dimen. factor [Hz] [Hz] [Hz] [Hz] [%] [%] [m] [%] 0.350 17 1.32 1.40 1.32 1.40 0.98 0.99 0.05 1.16 1.07 0.340 27 1.31 1.36 1.30 1.35 0.99 ≈ 1.0 0.01 0.96 0.97 0.330 44 1.29 1.32 1.30 1.32 0.99 0.95 0.09 0.92 1.16 0.325 65 1.28 1.30 1.29 1.30 ≈ 1.0 ≈ 1.0 ≈ 0 1.04 0.86 0.320 NRF 1.27 1.28 1.27 1.28 0.58 0.55 0.98 0.71 0.82 fi,F = fi,FEM , fi,S = fi,SOBI , ξi,S = ξi,SOBI , i =1,2 M1,1 =MAC(ψ1,FEM ,ψ1,SOBI), M2,2 =MAC(ψ2,FEM ,ψ2,SOBI), M1,2 =MAC(ψ1,SOBI ,ψ2,SOBI NRF – nearly repeated frequencies Fig. 3. 17%modal overlap: cross-spectrum between two orthogonal measurement directions (a) and extracted sources corresponding to the first (b) and second (c) mode the case of nearly repeated frequencies, instead, even if twodistinct sources are apparently extracted (Fig.4), the mode shapes are not accurately estimated. The comparison of the mode shape estimates associated to these two sources points out that the obtained vectors are very similar MAC(ψ1SOBI ,ψ 2 SOBI)= = 0.98. This proves, as expected, that SOBI is unable to carry out a proper separation of the modes in the case of repeated frequencies, since the mode Performance assessment of selected OMA techniques... 835 shapevectors, as providedby theFEmodel, should theoretically beorthogonal (MAC≈ 0). This limitation is reflected also by the poor correlation between themode shapes provided by SOBI and the FE model (MAC≈ 0.6). Fig. 4. Nearly repeated frequencies: spectra of the sources at 1.27Hz (a) and 1.28Hz (b) SOBI performance in the case of closely spacedmodes has been tested also against a real data set: the RC0 record of the School of Engineering Main Building in Naples (Rainieri et al., 2010). In this case, the first twomodes are characterized by amoderate overlap factor (17%). The related Singular Value plots provided byFDD is shown inFig.5. The obtained results are reported in Table 4 in comparisonwith those provided byFDDandCov-SSI, pointing out that fairly good results can be obtained also in the case of real measurements which take into account the effect of measurement noise. However, a detailed investigation about the effect of noise on the performance of SOBI for the output-onlymodal identification of structures characterized by close modes is out of the scope of the present paper. Table 4. The School of Engineering Main Building (Naples): RC0 record – comparison of output-only modal identification results Mode SOBI EFDD Cov-SSI Natural Damping Natural Damping Natural Damping frequency ratio frequency ratio frequency ratio [Hz] [%] [Hz] [%] [Hz] [%] I 0.92 0.85 0.92 1.03 0.92 1.07 II 0.98 0.92 0.98 1.25 0.98 1.08 III 1.30 0.76 1.30 1.03 1.30 0.82 836 C. Rainieri, G. Fabbrocino Fig. 5. The School of EngineeringMain Building (Naples): RC0 record – Singular Value plot 5. Conclusions Opportunities offered by traditional and innovative OMA procedures for the analysis of complex structural and geotechnical systems have been discussed. In particular, the applicability of OMA to the identification of the funda- mental dynamic properties of embedded retainingwalls has been investigated. Moreover, in the case of systems characterized by closely spaced modes, the performance of an emerging technique forOMA, the SecondOrderBlind Iden- tification, has been quantitatively assessed through its application to simula- ted and real measurements. The obtained results point out that OMA can be successfully applied also to the characterization of the dynamic behavior of complex geotechnical systems, such as embedded retaining walls in opera- tional conditions. On the other hand, SOBI can be reliably applied even in the case of structures characterized by very closely spacedmodes (65%modal overlap), provided that there are no repeated frequencies. Acknowledgements The present work is carried out in the framework of AT2-LR3.1-Task3.1.3 of the ReLuis-DPCExecutive Project 2010-2013 “RELUIS II”, rep.823. Support of ReLuis Consortium is therefore gratefully acknowledged. Authors express their gratitude to Dr. ArindamDey for his contribution in the numerical analysis of the retaining wall. Performance assessment of selected OMA techniques... 837 References 1. AllemangR.J.,BrownD.L., 1982,Acorrelationcoefficient formodalvector analysis,Proc.The 1st SEMInternationalModal AnalysisConference,Orlando, FL, USA 2. AnsB., Hérault J., JuttenC., 1985,Adaptive neural architectures: detec- tion of primitives,COGNITIVA’85, 593-597 3. Belouchrani A., Abed-Meraim K., Cardoso J.F., Moulines E., 1997, A blind source separation technique using second-order statistics, IEEETrans- actions on Signal Processing, 45, 434-444 4. Brincker R., Zhang L., Andersen P., 2000, Modal identification from ambient responses using frequencydomain decomposition,Proc. The 18th SEM International Modal Analysis Conference, San Antonio, TX, USA 5. Cardoso J.F., Souloumiac A., 1996, Jacobi angles for simultaneous diago- nalization, SIAM Journal of Matrix Analysis and Applications, 17, 161-164 6. Carvajal J.C., VenturaC.E., 2009, Identification of operationalmode sha- pes of a soil deposit using a grid of microtremormeasurements,Proc. The 3rd International Operational Modal Analysis Conference, Portonovo, Italy 7. CunhaA.,CaetanoE., 2005,From input-output to output-onlymodal iden- tificationof civil engineering structures,Proc.The 1st InternationalOperational Modal Analysis Conference, Copenhagen, Denmark 8. Doebling S.W., Farrar C.R., Prime M.B., Shevitz D.W., 1996,Dama- ge Identification and Health Monitoring of Structural and Mechanical Systems from Changes in their Vibration Characteristics: A Literature Review, Tech- nical Report LA-13070-MS, UC-900, Los Alamos National Laboratory, New Mexico 87545, USA 9. Fabbrocino G., LaorenzaC., Rainieri C., Santucci De Magistris F., 2009, Seismic monitoring of structural and geotechnical integrated systems, Materials Forum, 33, 404-419 10. Gentile C., 2005, Operational modal analysis and assessment of historical structures,Proc. The 1st InternationalOperationalModal Analysis Conference, Copenhagen, Denmark 11. Kerschen G., Poncelet F., Golinval J.C., 2007, Physical interpretation of independent component analysis in structural dynamics,Mechanical Systems and Signal Processing, 21, 1561-1575 12. McNeill S.I., ZimmermanD.C., 2008,A framework for blindmodal identifi- cation using joint approximate diagonalization,Mechanical Systems and Signal Processing, 22, 1526-1548 838 C. Rainieri, G. Fabbrocino 13. Mohanty P., 2005, Operational Modal Analysis in the presence of harmonic excitations, Ph.D. Thesis, Technische Universiteit Delft, The Netherlands 14. Peeters B., 2000, System identification and damage detection in civil engine- ering, Ph.D. Thesis, Katholieke Universiteit Leuven, Leuven, Belgium 15. Peeters B., Van der Auweraer H., Pauwels S., Debille J., 2005, An OverviewofOperationalModalAnalysis:MajorDevelopment and Issues,Proc. The 1st International Operational Modal Analysis Conference, Copenhagen, Denmark 16. Poncelet F., KerschenG., Golinval J.C., VerhelstD., 2007,Output- only modal analysis using blind source separation techniques,Mechanical Sys- tems and Signal Processing, 21, 2335-2358 17. Rainieri C., 2008,OperationalModalAnalysis for seismic protection of struc- tures, Ph.D. Thesis, University of Naples “Federico II”, Naples, Italy 18. Rainieri C., Fabbrocino G., 2010, Automated output-only dynamic identi- fication of civil engineering structures,Mechanical Systems and Signal Proces- sing, 24, 678-695 19. Rainieri C., Fabbrocino G., Cosenza E., 2010, Integrated seismic early warning and structural health monitoring of critical civil infrastructures in se- ismicallyproneareas,StructuralHealthMonitoring –An International Journal, doi:10.1177/1475921710373296 20. Srikantha Phani A., Woodhouse J., 2007, Viscous damping identification in linear vibration, Journal of Sound and Vibration, 303, 475-500 21. StrogatzS.H.,AbramsD.M.,McRobieA.,EckhardtB.,OttE., 2005, Theoreticalmechanics: crowdsynchronyontheMillenniumBridge,Nature,438 (7064) 22. Ventura C.E., Thibert K., 2007, Analysis of microtremor measurements by the frequency domain decomposition method, Proc. The 2nd International Operational Modal Analysis Conference, Copenhagen, Denmark 23. Zhang L., Brincker R., Andersen P., 2005, An Overview of Operational Modal Analysis: Major Development and Issues, Proc. The 1st International Operational Modal Analysis Conference, Copenhagen, Denmark 24. Zhou W., Chelidze D., 2007, Blind source separation based vibrationmode identification,Mechanical Systems and Signal Processing, 21, 3072-3087 Ocena sprawności wybranych typów operacyjnej analizy modalnej do identyfikacji bliskich postaci drgań w konstrukcjach geotechnicznych Streszczenie Analiza modalna typu „output-only” i dynamiczne monitorowanie stanu kon- strukcji na podstawie obserwacji drgań stanowią podstawowe narzędzia do badania Performance assessment of selected OMA techniques... 839 dynamiki nawetbardzo złożonychukładów.Wzrastające zainteresowanie takimi tech- nikami i możliwościami, jakie oferują, odnotowuje się również w dziedzinie inżynierii lądowej.W prezentowanej pracy uwagę skupiono namożliwości pozyskiwania lepszej informacji o specyficznych układach konstrukcyjnych, poprzez zastosowanie tradycyj- nychoraz innowacyjnychmetod operacyjnej analizymodalnej (OMA).Przedstawiono dwa przypadki.W pierwszym skoncentrowano się na ocenie efektywności ślepej iden- tyfikacji drugiego rzędu dla przypadku analizy modalnej typu „output-only” ukła- du z bliskimi postaciami drgań własnych. W drugim technikę OMA zastosowano do zasymulowanej dopowiedzi dynamicznej ściany oporowej. Otrzymano zachęcające re- zultaty, które wskazały analizę OMA jako godną zaufania przy badaniu drgań ściany wywołanych propagacją fal. Manuscript received March 10, 2011; accepted for print May 23, 2011