Summary The accuracy of fault diagnosis and condition monitoring of mechanical systems depends on the feature extraction of a non‐stationary vibration signals acquired from multiple accelerometer sensors. Extracting fault features of such complex vibration signals is a challengeable task due to the signals masked by an intensive noise. Recently, the multivariate empirical mode decomposition (MEMD) algorithm has been proposed in order to extend empirical mode decomposition (EMD) for the multi‐channel signal and make it suitable for processing multivariate signals. It is found that, likewise, EMD, MEMD is also essentially acting as a dyadic filter bank for the multivariate input signal on each channel. However, different from EMD, MEMD better aligns the same intrinsic mode functions (IMFs) across the same frequency range from different channels, which plays an important role in real‐world applications. However, MEMD still exhibits the degree of mode mixing problem, which affects the accuracy of extracting fault features. In this article, an improved MEMD, namely NAMEMD, is proposed to extract the most meaningful multivariate IMFs by adding uncorrelated white Gaussian noise in separate channels, under certain conditions, to enhance the decomposed multivariate IMFs by minimizing mode mixing problem. After that, a new method is proposed to select the most effective multivariate IMFs related to faults. To optimize the performance of extracting vibration fault features, a proposed noise‐assisted MEMD algorithm is then combined with a competent non‐linear Teager‐Kaiser energy operator, thereby guarantees a superior fault diagnosis performance. To verify the effectiveness of the proposed method, both a synthetic analytic signal and experimental wind turbine benchmark vibration datasets are utilized and tested. The results demonstrate that the proposed approach is suited for capturing a significant fault features in wind turbine multi‐stage gearboxes, thus providing a viable multivariate signal processing tool for wind turbine gearbox condition monitoring.

No abstract

No abstract