Gearbox Fault Diagnosis And Remaining Life Prediction Based On Bayesian Algorithm

In the era of Internet+,with the development of China Made 2025,industrial production has moved toward networking and intelligence.As an important transmission element,gears play an important role in industrial production.In recent years in order to reduce the costs and errors brought about by manpower maintenance and management of gear equipment,gear fault diagnosis and remaining life methods have become a research hotspot.Most researches are based on analytical models,mainly through the construction of an observer to estimate the system output,and then By comparing it with the output measured value to obtain information,this model often requires the establishment of an accurate system model,but with the complicated equipment such as gears,this model cannot be established,and the correct fault diagnosis and prediction of remaining life cannot be performed.Forecast research.In response to this problem,this paper proposes a data-driven method,and using normal and simulated vibration signals which obtained from gears,gear bearings,Fault diagnosis and prediction of bearing residual life by Bayesian algorithm.The breakdown is as follows:(1)Feature extraction of gear vibration signals through discrete Fourier transform(FFT),statistics of each eigenvalue,using dimensionality reduction and decision tree J48 algorithms to find the optimal number of features,and then using naive Bayesian and Bayesian network classifiers,the classification accuracy of the two is compared,and finally a superior feature classifier is found.The Naive Bayes method obtained the classification accuracy of 83.74%.And utilizes a mixture matrix to perform classifier verification.(2)Using the obtained vibration signal of the gear bearing and performing residual life prediction(RUL)based on the improved Bayesian inference algorithm.In the research of RUL,the accelerometer is mainly used to pick up the bearing vibration signal in the gear box,extract the health indicators from the wavelet threshold denoising,and then adopt the linear dimension reduction method and feature accumulation for the indicators.Finally,using the improved Bayesian inference algorithm to predict bearing RUL,and apply the algorithm to QPZZ-II platform for experiment.Obtained an average accuracy of 93.49%.