Research And Application Of Part Health Condition Assessment Method Based On Logistic Regression

In recent years,with the continuous improvement of industrial level,more and more functions of machinery and equipment have brought convenience to daily life and production,but also brought some problems.The complexity of equipment,for instance,becomes higher than before,thus making much more maintenance cost.Consequently,how to improve the reliability of mechanical equipment have become a hot topic these days.And in the context of relevant situations,the health status of rolling bearings,one of the most important part of mechanical equipment,with no doubt impacts the reliability and stability of the facility.There are three states in the process of a working bearing:normal,degradation and failure.Therefore,if the health status of rolling bearings can be monitored in real time,one can make a reasonable maintenance schemes according to the information in question,for the prevention of bearings’ failure and the improvement of economic efficiency.In view of the above considerations,we propose a health assessment method for the rolling bearing based on logistic regression model in this paper.The main contents are as follows:(1)On the basis of the characteristics of vibration signal,a method of combined noise reduction by EMD and wavelet threshold,which can accurately estimate the variance of the noise and improve the effect of noise reduction,was proposed to deal with the collected vibration signal of rolling bearing.Then,the signal after noise reduction was extracted from the time and frequency domain characteristic indexes,and the obtained index set was used as the sample data of the establishment of the Logistic regression model.(2)Logistic regression model theory was studied,and an initial value selection method based on adaptive control was proposed.This method could avoid the problem of singularity of jacobi matrix,improving the computational stability when Newton iteration method was applied to estimate the model parameters.In addition,considering that the coefficient of the independent variables have practical significance,arctan(·)function was used to constrain the coefficient,thus improving the explanatory ability of the model.(3)Combining the multicollinearity diagnostic index and elastic penalty function to select variables,and using strong rules to optimize its solution path.Leverage,residuals,data deletion models and modified statistics were used to screen the sample points.A complete model evaluation system was established to evaluate the Logistic regression model.(4)Based on Newton and coordinate descent method,the Logistic regression model of health assessment was established,and the parameters of elastic network were compared and selected.Through simulation experiments,numerical results were compared with SPSS to verify the effectiveness,robustness and prediction ability of the proposed algorithm and the fitting model,which provides a very effective method for real-time monitoring of bearing state.