Frequency domain decomposition (FDD) has been widely used in operational modal analysis due to the simplicity and intuitiveness in identifying modes at the spectrum peaks and the efficiency in dealing with large number of sensor data. However, it is difficult to separate closely spaced modes and requires manual peak selection to avoid spurious noise-induced peaks, which prevent the FDD to be well applied in real-time modal identification for long-term structural monitoring. To address these difficulties, an improved and automatic FDD is proposed to accurately identify closely spaced modes and enable automated modal identification. The key ingredients of the method are mainly two-fold. On the one hand, the joint approximate diagonalization (JAD) of power spectral density (PSD) matrices in the resonant frequency band is innovatively introduced to improve the identification accuracy of closely spaced mode shapes and then, the natural frequencies and damping ratios are obtained by simple least-squares regression; this also avoids subjective errors in selecting frequency peaks. On the other hand, to render fully automatic identification, the number of closely spaced modes is determined by the singular values of the PSD matrix at a resonant peak and then, the resonant frequency band is automatically selected through a generalized modal assurance criterion. Numerical, laboratory and field tests are conducted to demonstrate the improved accuracy of the JAD-FDD method in separating closely spaced modes as well as the remarkable effectiveness and efficiency in realizing automatic frequency-domain modal identification. • Improved and automatic FDD is proposed for modal identification with closely spaced modes. • Joint approximate diagonalization is introduced to improve the accuracy of FDD in separating closely spaced modes. • Full-automation ability is achieved by automatic resonant frequency band selection criterion. • More accurate identification results of closely spaced modes are achieved over the existed EFDD and FSDD methods. • Laboratory and field tests unveil the substantial superiority in computation efficiency over the well-established automatic SSI.
Xie et al. (Tue,) studied this question.