Research On Vibration Fault Diagnosis And State Trend Prediction Method Of Gearbox Under Data Drive

The gearbox is the leading power transmission equipment of rotating machinery,the worse the working environment,the greater the probability of failure of the gearbox,and then affects the reliability and safety of the whole mechanical system.Therefore,it is imperative to study the gearbox’s fault diagnosis and condition monitoring to ensure the automatic system’s stable operation.Since vibration signal can reflect its running state most directly and effectively,this paper explores the de-noising,feature extraction,fault pattern recognition,and state trend prediction of gearbox fault vibration signal based on vibration signal processing.The main contributions and achievements of this thesis are summarized as follows:(1)To address the instability and nonlinearity of the gearbox vibration signal in actual working conditions,and the presence of high-frequency noise distribution,a gearbox vibration signal noise reduction model based on the EEMD and SVD is proposed in this study.Firstly,the signal preprocessing process of EEMD superimposed with white noise is utilized to suppress the impact of impulse noise in the vibration signal of the gearbox and to overcome the mode aliasing problem of EMD.Then,SVD is employed to further denoise the residual white noise in the component,which can effectively eliminate high-frequency noise and the legacy of white noise.This model effectively integrates the complementary advantages of EEMD and SVD.By using the vibration signal noise reduction experiment in the actual working environment,it is proved that the scheme established in this paper can effectively reduce the noise of the original vibration signal of the gearbox.(2)A gearbox fault diagnosis method based on wavelet packet transform and improved LSSVM is proposed to solve the problem that the traditional single LSSVM model is susceptible to the influence of penalty factor and kernel function in the classification process.The fault vibration signals were extracted by wavelet packet transform and its energy distribution,and then the Morlet wavelet and PSO algorithm were used to optimize the penalty factor and Gaussian kernel factor of LSSVM.The PSO algorithm overcomes the problem of easily falling into local extreme value by making full use of the Morlet wavelet to transform the local solution with higher frequency in the solution space and uses the disturbed PSO algorithm to optimize the parameters of LSSVM.A large number of experiments show that the improved LSSVM classification model has higher gearbox fault classification accuracy.The accuracy is 9.21% higher than ELM and 7.14% higher than BP.The comparison experiment shows that the pattern recognition model established in this paper is superior to the existing methods.(3)To enhance the prediction efficiency and minimize error in predicting gearbox vibration signal state trend,the prediction model of LSTM is optimized using the genetic algorithm.The number of neurons,training times,and learning rate of the LSTM are optimized using the global optimization ability of the genetic algorithm.The relationship between the predicted and actual values of the gearbox vibration signal is constructed as a fitness function.The gearbox vibration signal,which has been noise-reduced,is used as input to predict the trend of the subsequent vibration state.Compared with SVR,ELM,RNN prediction models,and existing methods,experiments show that the RMSE,MAE,MAPE,and MSE of the gearbox vibration state trend prediction model established in this paper are all the smallest,indicating that the model has higher prediction accuracy and better performance.