This article presents a comparative study of the modal parameter identification of structures based on the continuous wavelet transform (WT) using the modified complex Morlet wavelet function and the improved Hilbert-Huang transform (HHT). Special attention is given to some implementation issues, such as the modal separation and end effect in the WT, the optimal parameter selection of the wavelet function, the new stopping criterion for the empirical mode decomposition (EMD) and the end effect in the HHT. The capabilities of these two techniques are compared and assessed by using three examples, namely a numerical simulation for a damped system with two very close modes, an impact test on an experimental model with three well-separated modes, and an ambient vibration test on the Z24-bridge benchmark problem. The results demonstrate that for the system with well-separated modes both methods are applicable when the time-frequency resolutions are sufficiently taken into account, whereas for the system with very close modes, the WT method seems to be more theoretical and effective than HHT from the viewpoint of parameter design.

No abstract

No abstract

This paper presents the use of time-frequency decomposition based on two novel methods termed wavelet transform (WT) and Hilbert-Huang transform (HHT) for modal parameter extraction of structure. The impacts of modal separation, end effects, and noise levels on the time-frequency resolutions of these two approaches are investigated and compared by using numerical simulations. To verify their effectiveness on the real structure, an experimental study is implemented on a bridge model subjected to impact load and ambient vehicle loadings. The parameter identification results demonstrate that both the WT and HHT are effective as the time-frequency resolutions are given sufficient concerns, whereas from the viewpoints of parameter selection and quantitative analysis, in some sense, the WT method seems to be more theoretical and efficient than the HHT.

This paper is primarily concerned with the application of static test data in conjunction with the probabilistic neural network (PNN) for the classification of damage patterns of a cable-stayed bridge. A total of I I damage patterns are considered by combination of 5 typical damage regions. Both training and testing data, derived from static analysis via finite element method (FEM), are contaminated with different noise level to simulate the FE model and measurement errors. The study ofdamage pattern identification is conducted by taking into account the change ratios ofthe deflection of the main beam and the tower under loading as input neurons of the PNN. The effects of noise levels, the types of damage patterns, and the number of input neurons on the identification accuracy are investigated. Based on the classification results some valuable conclusions were obtained.