A noise reduction method for rolling bearing based on improved Wiener filtering

To accurately identify compound faults of bearings, a new noise reduction method is presented. With the new method, input signals and the order of Wiener filtering are adaptively determined according to feature mode decomposition (FMD), signal evaluation index, and Euclidean distance. First, to effectively separate frequency components from vibration signals, vibration signals are decomposed into modal components based on the FMD algorithm; second, kurtosis, root mean square, and variance, which are sensitive to fault information, are selected to build evaluation vectors. Third, the Euclidean distance between the evaluation vectors of the component signal and the original signal are calculated to represent the correlation among signals. By acquiring the two component signals that have the greatest and least correlation to original signals, an actual signal and a mixed signal required by Wiener filtering can be adaptively determined. Furthermore, the order of Wiener filtering is adaptively determined with maximum kurtosis as the criterion. Finally, fault features are extracted through the spectral analysis of signals after Wiener filtering and the type of compound faults is judged based on that. To demonstrate the accuracy and effectiveness of the proposed method, the proposed method is compared with the classical method. The result of comparison shows that the presented method can restrict the noise more effectively and determine the type of complex faults of bearings more accurately.