WO-AMNAD: A Novel Angular Deconvolution Approach for Bearing Fault Identification Under Time-Varying Speed Conditions
Abstract
As a widely used tool in vibration analysis, blind deconvolution (BD) has been extensively studied to recover the periodic fault impulses from noise-corrupted signals for diagnosing rotating machinery faults. However, traditional BD algorithms typically assume signals exhibit periodic stationarity. When equipment operates under variable-speed conditions, fluctuations in rotational speed will cause fault intervals to become nonconstant, which will render the method ineffective. Aiming to address the aforementioned issue, this article proposes a novel angular minimum noise amplitude deconvolution (AMNAD). First, the periodic noise amplitude ratio index is redefined in the angular domain, while leveraging a resampling method to reconstruct the original nonperiodic signal. This solution effectively overcomes the limitation of fault intervals changing in the time domain. Subsequently, the walrus optimizer (WO) is used to dynamically determine the optimal key parameter sets, improving the feature extraction performance of the BD algorithm. Finally, envelope order spectrum analyses are performed on the filtered signal to extract fault features. The simulated and experimental signal analyses are employed to confirm the effectiveness and reliability of the proposed WO-AMNAD and compare with other existing methods. Overall, the comparison results show that the proposed method has a noticeable advantage over comparative algorithms.