Research On Rolling Bearing Fault Diagnosis Based Mathematical Morphology

Rotating machinery is the main equipment used in industry.Its normal operation is directly related to the work of its key component-rolling bearing.Fault diagnosis has become the top priority of enterprise maintenance equipment.After the failure of rolling bearing,its signal characteristics will show non-stationary,non-linear,non Gaussian and other properties.Traditional signal analysis methods usually ignore some signal characteristics for analysis.The more signal details are ignored,the analysis results will have obvious deviation.How to avoid these errors,not too much ignore the fault signal characteristics,but also effectively extract useful information is the focus of this paper.Mathematical morphology is a nonlinear signal analysis method based on set theory and integral geometry.It has been applied and developed in the field of mechanical fault detection and diagnosis because it is based on a complete mathematical theory and conforms to the processing method of one-dimensional signal characteristics.The work to be carried out in this paper is as follows:(1)Compared with a single traditional structural element,the advantages and disadvantages of the proposed new structural element are obvious.The structure of the proposed new structural element is too complex,so that it can only be applied to a few signal morphological processing that conform to the corresponding characteristics.In this thesis,a total of 32 kinds of first-order operators are tested and 9 kinds of high-quality second-order operators are combined.Finally,the synthesis operator is obtained by selecting the best among the best.This method does not improve the complexity of structural elements and has strong adaptive ability,which is suitable for time-varying structural element filtering method.Adopting time-varying structure element and optimizing the filtering process,the noise suppression is more obvious and the filtering is faster and more effective.(2)Using the grayscale hit-miss transform suitable for one-dimensional signal feature extraction,combined with the multi-scale filtering method,the optimal fault identifier is found and the fault point feature matrix is established for the time domain fault feature,which provides excellent feature vector for subsequent classification.(3)The morphological spectrum of the overall scale characteristics and signal failure characteristics of the signal is proposed-the multi-scale hit failure feature vector is proposed and classified in combination with the neural network,which avoids the spatial insensitivity of the morphological spectrum,finds the data characteristics suitable for the rolling bearing fault signal(with the characteristics of the periodic attenuation shock signal),and improves the training speed and classification accuracy of the neural network.Conclusion: This paper focuses on the application of new synthetic morphological operators of time-varying structural elements in mechanical fault signals,and the time-varying structural element filtering method based on double operators.The traditional feature extraction has less research on the time-domain waveform.The multi-scale gray hit miss transform is used to identify the time-domain impact location and extract and convert it into a two-dimensional feature vector;Morphological spectrum and multi-scale gray hit miss feature vectors are input into neural network for fault classification to avoid the problem of similarity of feature data.