Rolling Bearing Health Situation Analysis Based On Machine Learning

As the most basic rotating body bearing mechanism in mechanical equipment components,rolling bearings have a direct impact on the performance indicators of related equipment in terms of material properties,structural strength and service life.Therefore,in the process of use,accurate and effective health assessment and reliability maintenance of rolling bearings are an important guarantee for the safe and stable operation of machinery and equipment.In order to make the above assessment and maintenance links more intelligent,the subject of this study is rolling bearings,and studies around the problems of vibration signal denoising,health pattern recognition,performance degradation trend tracking,and residual service life prediction,It is expected to improve the accuracy of bearing index analysis and fault classification algorithms,and the following are the key research findings.(1)Addressing the issue of bearing vibration data including a lot of noise,because it’s likely to lose effective bearing information during denoising when using a single full ensemble empirical mode decomposition with adaptive noise.A denoising approach based on wavelet packet semi-soft threshold and enhanced full ensemble empirical mode decomposition with adaptive noise is suggested.The difficulty of parameter selection of wavelet decomposition is overcome by the adaptive property of the algorithm.It solves the problems such as large reconstruction error and easy to appear aliasing mode in the existing empirical mode decomposition methods.The suggested technique provides apparent advantages in noise reduction,according to simulation findings.(2)Bearing fault features are difficult to be extracted,and a single feature is not sufficient to describe bearing faults,and the identification effect is not good.The sparrow search algorithm has high global search capability,which can adapt to the overlapping and similar characteristics of bearing fault diagnosis,avoid wrong fault judgment caused by falling into local optimization in the algorithm iteration.A fault diagnosis method for least squares support vector machines optimized by improved sparrow search algorithm is proposed.The least square support vector machine is susceptible to kernel function and penalty function,adaptive chaos search and Levy flight strategy are used to improve sparrow search algorithm,construct a new identification model.Simulation results show that the proposed method can effectively identify faults with higher accuracy.(3)In view of the traditional single index analysis method,it is difficult to fully reflect the effective information of the bearing in the analysis of bearing performance degradation and life prediction,a bearing health situation prediction method based on fuzzy information granulation and improved relevance vector machine was proposed.Principal components are used to analyze the time domain,frequency domain and frequency domain indexes of weighted fusion bearings,the fuzzy information graining process is also carried out.The kernel width of the correlation vector machine was optimized using the modified sparrow search method,and a full prediction model was built.This method can effectively predict the health situation of bearings.