Elsevier

Journal of Sound and Vibration

Volume 366, 31 March 2016, Pages 117-132
Journal of Sound and Vibration

Underdetermined blind modal identification of structures by earthquake and ambient vibration measurements via sparse component analysis

https://doi.org/10.1016/j.jsv.2015.10.028Get rights and content

Abstract

Sparse component analysis (SCA) approach was adopted to handle underdetermined blind modal identification of structures, where the number of sensors is less than the number of active modes. To exploit the sparsity of structural responses in time–frequency domain, Short Time Fourier Transform (STFT) was used in this study. The proposed SCA-based approach has two main stages: modal matrix estimation and modal displacement estimation. In the first stage, hierarchical clustering algorithm was used to estimate the modal matrix. The clustering algorithm was preceded by a preprocessing step to select the points in time–frequency domain that only one mode makes contribution in the responses. These points were fed to the clustering algorithm as an input. Performing this analysis enhanced the modal matrix estimation accuracy and reduced the computational cost while conducting clustering analysis. Having estimated mixing matrix, the complex-valued modal responses in the transformed domain were recovered via Smoothed zero-norm (SL-0) algorithm. In a broad sense, using the SL-0 algorithm permits researchers to use any kind of transform in seeking sparsity, regardless of obtaining real-valued or complex-valued signals in transformed domain. Natural frequencies and damping ratios were extracted from the recovered modal responses. Performance of the proposed method was investigated using a synthetic example and a benchmark structure with earthquake and ambient excitation, respectively.

Introduction

Modal identification of structural systems when only system responses are available is a challenging issue [1], [2]. Blind source separation (BSS) technique [3] is a statistical signal processing tool that has been widely used to deal with this problem [4], [5], [6], [7]. Introducing modal displacements as virtual sources paved the way for application of BSS in modal identification problems [8], [9]. Independent component analysis (ICA) [10] and second order blind identification (SOBI) [11] are the mostly used techniques in BSS, which are based on fourth order and second order statistics, respectively. Independence of the sources is the basic assumption of the ICA-based methods; however SOBI-based ones only require that the source signals should be uncorrelated.

Sparse component analysis (SCA) is another approach which is well suited to underdetermined problems. This approach comprises two major stages: mixing (modal) matrix estimation and source (modal displacements) recovery. SCA-based methods require no condition on independence or uncorrelatedness of the sources and sparsity is the only requirement that can be satisfied in transformed domain [12]. The choice of underlying transform plays a pivotal role in determining the type of excitation that any identification algorithm can undertake. Trying to exploit sparsity of structural responses in transformed domain, different transforms have been used in the literature. To cite a few, Discrete Cosine Transform (DCT) [13], linear time–frequency analysis [14], [15] and quadratic time–frequency analysis [16], [17] were used successfully in blind modal identification field.

In first stage, clustering algorithms [18] can be used to estimate modal matrix of structural systems. However, considering the fact that in practical situations the sources may have overlap in transformed domain, a criterion should be set to discard these points. Different researchers have investigated this issue exclusively and have reported a number of criteria [19], [20], [21], [22]. Of those, comparing the direction between real and imaginary part of the mixture signals was used in this study to distinguish Single Source Points (SSPs) from Multiple Source Points (MSPs), where only one and multiple sources make contribution in the responses, respectively [21]. Although this criterion was originally developed for speech signals, it showed promising results for vibration problems as well. Using this preprocessing step not only enhanced the estimation accuracy significantly, but it also reduced the computational cost while conducting clustering analysis.

Once the incomplete mixing matrix has been obtained, an underdetermined system of linear equations is formed. Solving this equation, especially when the source signals are complex-valued, is a challenging task [15]. In attempt to avoid complex elements, the authors in Ref. [13] chose DCT as underlying transform and used l1-magic algorithm [23] in recovery stage. But regarding the fact that DCT transform is in frequency domain, their proposed method was incapable of handling non-stationary excitations, like earthquake. Using Smoothed zero norm (SL-0) algorithm [24], [25], [26] in this paper enabled us to use any kind of transform, regardless of obtaining real-valued or complex-valued signals in transformed domain.

In this study, Short Time Fourier Transform (STFT) was chosen to fulfill sparsity requirement and to provide a general framework for tackling wide variety of excitations like earthquake and ambient vibration. In addition, constituent elements of the first and second stages were designed in such a way to render the proposed algorithm robust against noise, computationally efficient and less dependent on type of sparsifying transform that might yield real or complex coefficients. Whereas, constant resolution of the STFT transform, which requires a predefined window length, and restriction of the proposed technique to proportionally damped structures remained as the major limitations of the proposed technique.

The rest of this paper is organized as follows. The modal identification problem and it’s affinity to BSS problem are explored in Section 2. Then, two main stages of the SCA-based approach are presented in 3 First stage: modal matrix estimation, 4 Second stage: modal displacements estimation, independently. Numerical simulations using a synthetic example and a benchmark structure are carried out to investigate the performance of the proposed method in Section 5. Finally, a brief conclusion is given in Section 6 as a closing section.

Section snippets

Modal identification and BSS

A Linear Time Invariant (LTI) N-DOF structural system subjected to an excitation force is governed by the following differential equation [27]:Mx¨(t)+Cẋ(t)+Kx(t)=f(t)where M, C and K are the N×N mass, proportional damping (diagonalizable) and stiffness matrices, respectively, and f(t) is a column-vector that contains excitation force.

x(t) is the absolute nodal displacement vector, which is transformed into modal coordinate as follows:x(t)=Φq(t)where ΦRN×N is the square real-valued modal

First stage: modal matrix estimation

It is aimed to estimate modal matrix of an LTI structural system which is subjected to earthquake and ambient excitation. The key idea is that by scattering all of the responses with respect to each other in sparse domain, the main components (contributing sources) align in certain directions representing the eigenvectors of the system. Then, they are grouped by means of a clustering algorithm to yield columns of the mixing matrix. However, in real life examples source signals may not be fully

Second stage: modal displacements estimation

Modal displacements are recovered with the knowledge of mixing matrix and responses. For determined case, modal displacements could be recovered by using a simple inversion technique (Eq. (14)).q(t)=Φ1x(t)

However, in underdetermined problems complete modal matrix is no longer available. In this case, system responses and its partial modal matrix form an underdetermined system of linear equations.X(t)=AS(t)where Rm×n,m<n , and every m×m sub-matrix of A is of full rank [12].

In general, the Eq.

Performance evaluation and verification

The proposed algorithm was applied to a number of structural models to render the verification process. The capability of the proposed method in handling earthquake and ambient vibrations were all examined through the following simulations.

Modal assurance criterion was used to quantify the accuracy of the obtained mode shapes, where φ~i is the estimated mode shape and φi is the theoretical one.MAC(φ~i,φi)=(φ~iT.φi)2(φ~iT.φ~i)(φiT.φi)

Conclusions

A novel blind modal identification technique—which requires a limited number of sensors—was proposed in this paper. It is capable of identifying modal parameters of structures subjected to earthquake and ambient excitations. Time–frequency nature of the proposed method laid the foundation of such capability in dealing with different sorts of vibration. Time–frequency representation of each response was refined through the SSPs selection process. This preprocessing step paved the way for

Acknowledgment

The authors would like to thank the editor and reviewers for their invaluable comments that made significant improvement to our paper.

References (45)

  • V.G.Ã. Reju et al.

    An algorithm for mixing matrix estimation in instantaneous blind source separation

    Signal Processing

    (2009)
  • C.S. Lin et al.

    Modal identification from nonstationary ambient response data using extended random decrement algorithm

    Computers and Structures

    (2013)
  • G. a Papagiannopoulos et al.

    On the use of the half-power bandwidth method to estimate damping in building structures

    Soil Dynamics and Earthquake Engineering

    (2011)
  • B. Hwa et al.

    A new method to extract modal parameters using output-only responses

    Journal of Sound and Vibration

    (2005)
  • J. Herault, C. Jutten, Space or time adaptive signal processing by neural network models, AIP, Snowbird, UT, USA,...
  • B. Hazra et al.

    Modified cross-correlation method for the blind identification of structures

    Journal of Engineering Mechanics

    (2010)
  • A. Belouchrani et al.

    A blind source separation technique using second-order statistics

    IEEE Transactions on Signal Processing

    (1997)
  • P. Georgiev et al.

    Sparse Component Analysis and Blind Source Separation of Underdetermined Mixtures

    IEEE Transactions on Neural Networks

    (2005)
  • Y. Yang et al.

    Time–frequency blind source separation using independent component analysis for output-only modal identification of highly damped structures

    Journal of Structural Engineering

    (2013)
  • S.F. Ghahari et al.

    Response-only modal identification of structures using strong motion data

    Earthquake Engineering Structural Dynamics

    (2013)
  • S.F. Ghahari et al.

    Blind modal identification of structures from spatially sparse seismic response signals

    Structural Control and Health Monitoring

    (2013)
  • R. Xu et al.

    Survey of clustering algorithms

    IEEE Transactions on Neural Networks/a Publication of the IEEE Neural Networks Council

    (2005)
  • Cited by (56)

    • A review on the application of blind source separation in vibration analysis of mechanical systems

      2024, Measurement: Journal of the International Measurement Confederation
    • A multi-task learning-based automatic blind identification procedure for operational modal analysis

      2023, Mechanical Systems and Signal Processing
      Citation Excerpt :

      Therefore, it would be meaningful to develop an approach that can automatically realize the ability of SCA decoupling. Artificial intelligence (AI) technique has been increasingly applied for solving some typical problems in practical engineering such as crack detection [29–32], data anomaly detection [33,34], and noise reduction [35]. In particular, deep learning technique can automatically learn features from mass data without artificially extracting features or setting feature threshold.

    View all citing articles on Scopus
    View full text