Research On Non-Stationary Signal Feature Extraction Technology Based On Improved Morphological Filtering

Nowadays,the most effective and reliable method for mechanical equipment working in engineering practice to understand its operating status and failure is to monitor its vibration signal.However,as far as the signal collected in the real environment is concerned,the mechanical signal is non-stationary.Sex is absolute and widespread.Therefore,to monitor and diagnose the mechanical state,it is necessary to study the non-stationarity of the signal.This paper takes the fault vibration signal of rolling bearing as the research object,and based on the nonlinear analysis method-morphological filtering method,a series of researches have been done to extract the fault characteristic frequency of its non-stationary signal.First,it introduces the background and significance of this topic,and summarizes the research status of non-stationary signal feature extraction methods.On this basis,the research ideas and content of this article are established.The concept and characteristics of non-stationary signals are introduced in detail,and the vibration mechanism of rolling bearings and the non-stationarity of their vibration signals are explained.This explains the limitations of traditional time-frequency analysis for non-stationary signals,and has done a solid foundation for the study of fault diagnosis models for non-stationary vibration signals of rolling bearings.Based on the above research,it was decided to use the Empirical Mode Decomposition(EMD)method as the signal preprocessing method,and in order to improve its end effect: "Based on the extreme time scale BP neural network endpoint extension" method,this method is improving The end effect of EMD makes EMD more suitable for processing non-stationary signals.Secondly,after the signal is preprocessed,it is necessary to extract impact features and reduce noise.In order to achieve these two functions at the same time,a morphological filtering method is proposed as a method to further extract fault features and suppress noise.After in-depth research on the morphological filter,an improved morphological filtering method based on the kurtosis criterion is proposed.Taking the bearing inner ring fault as an example,the effect of the proposed method of noise reduction filtering in extracting impact features is verified.Again,based on the above research,a complete fault diagnosis model for rolling bearings is proposed.The diagnosis idea of this model is to first decompose the signal to extract the useful signal components,and then use the extracted signal as the input of the subsequent analysis to further extract the fault.For the signal decomposition,the improved EMD proposed in this paper "based on the extreme value time scale BP neural network endpoint extension" is used,and then the eigenmodes containing fault information obtained after the improved EMD decomposition are based on the kurtosis criterion.The component(IMF)is extracted to complete the signal preprocessing;then the extracted IMF is reconstructed into a new signal,and this signal is used as the input signal to further extract the fault information,and the improved morphological filtering method proposed in this paper is applied to In this part,the noise is filtered under the premise of retaining the fault characteristics,and finally the filtered signal is demodulated by the Teager energy operator demodulation method to complete the fault type diagnosis.Finally,simulation and experimental verification are carried out.The simulation signals of the rolling bearing fault signals and the data from the Electrical Laboratory of Case Western Reserve University in the United States are used to extract different types of bearing faults to verify the feasibility of the proposed model and the effect of fault extraction.The results show that the method can accurately extract the fault features and has a good noise suppression effect.