A maximum correntropy criterion based recursive method for output-only modal identification of time-varying structures under non-Gaussian impulsive noise

This work considers the problem of output-only identification for time-varying structures in a recursive manner under non-Gaussian impulsive noise. Presently, the most existing identification methods are based on classical least-squares (LS) criterion due to its mathematical tractability, computational simplicity and optimality under Gaussian assumption. However, the performance of the LS based methods deteriorates seriously in non-Gaussian situations, especially when responses are corrupted by impulsive noise (outliers) which are frequently encountered in real test cases. To deal with this problem, a maximum correntropy criterion based recursive modal identification method is proposed in this paper, and a class of heavy-tailed alpha-stable distribution is adopted to model non-Gaussian impulsive noise. The effect of the kernel bandwidth parameter in proposed method is discussed and a rational value is given for the model structure selection. Finally, the proposed method is comparatively assessed against its LS counterpart via a numerical example and a laboratory time-varying structure experiment. The comparisons have illustrated the advantages of the proposed method on the modal frequency estimates in terms of estimation accuracy and robustness under the non-Gaussian impulsive noise.