Performance Degradation Assessment And Remaining Useful Life Prediction Of Rolling Bearings

As an important component of rotating machinery,the health of rolling bearings is related to the normal operation of the whole machinery.Rolling bearings have different bearing life due to the different environments and working positions.There are many reasons for bearing failure,but mainly bearing fracture and damage caused by fatigue damage,improper installation and external pollution.If the whole process monitoring of the bearing from normal state to the final complete failure can be realized and repaired in time to.realize the active maintenance of the bearing,the economic property loss caused by the bearing and the casualties will be greatly reduced.In this paper,the vibration signal of the rolling bearing is obtained by the acceleration sensor,and the original information is processed by wavelet noise reduction.Then the characteristic information in the time domain,frequency domain and time-frequency domain of the bearing is extracted and analyzed,and the health status of the bearing is judged initially.In the time domain,we focus on the characteristics such as root mean square value and kurtosis value,extracting the moment of vibration shock and observing the numerical trend.In the frequency domain,the characteristics of amplitude spectrum,power spectrum and frequency mean are detailed.Explore and mine the frequency domain features in the vibration signal.Regarding the time-frequency domain,the short-time Fourier transform and the Wigner-Viller distribution are mainly used for the impact moment obtained by the above analysis,from which the existence of the impact energy can be clearly seen.The extracted wavelet energy entropy is also an important feature in the time-frequency domain.From the obtained entropy map,both the frequency characteristics and the vibration shock information can be extracted,indicating that the above methods have the practicability of extracting the characteristics of the vibration signal and have the ability to excavate the anomalous signals in the vibration signal.Then,the nonlinear algorithm ISOMAP which has better feature of dimensionality reduction is used to simplify the data of the above features,which not only eliminates unnecessary features,but also prevents the model from being too large due to excessive data redundancy.The Gaussian mixture model is fused which is to establish the degradation index based on log likelihood probability.The local weighted regression is used to smooth the processing to improve the accuracy of the index.According to the rules in probability statistics,the threshold of the recession index is built to complete the degradation assessment of bearing life.Finally,the features are input into the correlation vector machine model for training and prediction,and the remaining life prediction of the bearing is completed.The PCA and T-SNE dimensionality reduction methods are used to compare the experimental results,and the advantages and disadvantages of feature extraction are verified to achieve the optimal results of bearing residual life prediction.