Abstract: Decomposition methods are the most effective means to handle the multi-component separation of mechanical signals. However, the existing decomposition methods do not take typical mechanical fault features as the decomposition target, and the extraction of decomposition modes is difficult to achieve adaptive filtering. Thus, the poor component separation and feature extraction effect of complex signals makes it insufficient to meet the diagnostic needs. In view of this, the harmonic feature mode decomposition (HFMD) is proposed. The signal periodic intensity evaluation index, harmonics-to-noise ratio (HNR) is selected as the decomposition target. The finite impulse response (FIR) filter coefficient updating mechanism is used to achieve adaptive filtering in the extraction of decomposition modes. Firstly, the filter bank is initialized with a tree-based band division method. On this basis, the optimal filter coefficients are solved with HNR as the decomposition target. Furthermore, the correlation coefficient is used to evaluate, compare and reduce the redundant modes. Finally, the extraction of periodic features in complex signals and the separation of harmonic components are realized by setting the number of modes as the convergence criterion. Simulation and experimental cases verify that the proposed HFMD can extract the bearing fault information more accurately and effectively than the traditional decomposition methods.

Key words: decomposition, harmonics-to-noise ratio, fault diagnosis, rolling bearing, adaptive filtering