| With the improvement of the technical level,the internal structure of the rotating machinery system becomes more and more complex.In addition,the poor working environment makes it very easy to cause faults.Failure to find faults in time often leads to very serious consequences.Rolling bearing is one of the core components of the rotating machinery system,and its running state directly affects the performance and safety of the entire rotating machinery system.By analyzing the statistical characteristics,frequency,power and energy aggregation of rolling bearing vibration signals,it is possible to judge the working state of rotating machinery equipment at any moment,so as to monitor and adjust the faults that have occurred or will occur,which is of great significance to ensure the continuous,stable and safe operation of machinery equipment.In the process of data acquisition for mechanical vibration signal analysis and research,due to the complex and changeable working environment of the mechanical system,it is easy to generate interference noise,which directly affects the accuracy of the signal analysis results,and the feature extraction of the mechanical vibration signal is not thorough or obvious,resulting in the decline of the accuracy of the diagnosis results.The main research contents of this paper are as follows:(1)A nonlinear filtering(NLF)-logarithmic wavelet threshold method for impulse noise reduction is proposed.In most cases,researchers will assume that the noise is Gaussian noise,which is relatively simple and easy to filter.Unlike Gaussian noise,Alpha stable distribution impulse noise is a non-stationary signal.The performance of traditional signal denoising methods is degraded,and strong impulse noise cannot be removed,and the computational complexity will increase.To solve this problem,this paper proposes a new logarithmic threshold function based on the traditional hard and soft threshold functions,which can overcome the inherent defects of the traditional hard and soft threshold functions,and combines this wavelet logarithmic threshold function de-noising method with the median value flat filtering method.Simulation experiments show that this method can greatly suppress strong impulse noise,further improve the smoothness of the de-noising signal,and enhance the de-noising effect.(2)A bearing fault diagnosis method based on signal inherent mode depth modeling analysis is proposed.The traditional mechanical bearing fault diagnosis model is easily disturbed by system noise,and the efficiency of feature recognition is low.Firstly,the collected bearing vibration signal is decomposed into noise adaptive full empirical mode decomposition(CEEMDAN),and the correlation coefficient is used to determine whether the modal component(IMF)contains noise,and the variable step minimum mean square algorithm(VSSLMS)is used to reduce the noise of the noisy IMF component and reconstruct it;Then,the time-frequency spectrum of the de-noised vibration signal is obtained by discrete wavelet transform(DWT),and the feature is enhanced by using morphological open operation;Finally,the improved GoogLeNet network model is used to train the feature map,and the Softmax classifier is used to complete the feature classification,thus realizing the bearing fault diagnosis.Simulation results show the effectiveness of the algorithm.(3)Using Simulink platform of MATLAB software and App Designer development platform,bearing vibration signal analysis and fault alarm software are designed.The software mainly includes modules such as system login interface,data input,signal preprocessing and analysis,and fault diagnosis based on deep learning.The user logs in through the system,imports the collected rolling bearing fault data,and then performs filtering,time domain analysis,frequency domain analysis,etc.through the signal preprocessing module.The discrete wavelet transform is used to decompose and reconstruct the bearing vibration signal,and the two-dimensional spectrum is obtained as the feature map.Finally,the fault is identified through the fault diagnosis module. |
