A multiscale local pattern filtering method for noise mode identification of dynamic signals in bridges

Abstract

The problem of effective information extraction of dynamic signals is a key challenge of vibration-based structural health monitoring. Traditional ensemble empirical mode decomposition (EEMD) and complementary EEMD (CEEMD) are effective filtering methods but have the shortcomings of mode mixing, completeness and the endpoint effect in adaptive decomposition stage. To address the above problems, this paper proposes a noise mode identification method based on multiscale local pattern filtering with a machine learning algorithm. In addition, this method improves the filtering performance of mode decomposition and wavelet threshold denoising methods effectively. To validate the performance of this method, simulation experiment and real signal analysis were used for validation: their results showed that the proposed method has better adaptive decomposition performance than EEMD and CEEMD. Compared with the EEMD, CEEMD and wavelet threshold denoising methods, the proposed method can better balance smoothness and the power reserve, has optimized time- and frequency-domain characteristics, filters with a higher signal-to-noise ratio and reliability, and has more optimized filtering performance. The method proposed in this paper can be applied to bridge health monitoring technology, and has important scientific and engineering significance.