Research On Method Of Remaining Useful Life Prediction Of Bearings

Bearings are common rotating parts in instrument equipment.As a key component in large-scale precision instruments and equipment,their operating status directly affects the safety and stability of the system.Based on this,it is very necessary to accurately predict the remaining service life of the bearing.As the amount of data increases and the requirements for forecast accuracy increase,traditional forecasting methods cannot meet industrial needs,and the application of life prediction methods based on deep learning algorithms is born.This paper first studies the vibration mechanism of the bearing,clarifies the relationship between the fault information and the vibration signal;and calculates the variable flexibility and variable stiffness vibration frequency of the bearing and the general solution of the vibration equation,and determines the sampling frequency required for subsequent experiments based on the natural frequency;Finally,the traditional life prediction model is used to obtain the calculated life of the bearing,which is compared with the real life.The comparison result reflects that the life prediction method based on the traditional model cannot meet the requirements of industrial applications.Aiming at the shortcomings of traditional methods such as poor prediction accuracy and inability to meet real-time online monitoring in practical applications,a prediction method based on vibration data is adopted.First,the preprocessing of the vibration signal collected by the experiment is divided into the degradation stages.This paper proposes a new health index based on time-frequency analysis.This index is a time-frequency domain composite index that can well reflect the degradation process of bearing performance.The accuracy of dividing the degradation stage according to it is higher than the two common health indicators of root mean square and kurtosis.Considering the time correlation between bearing vibration signal sequences,this paper proposes a new remaining service life prediction algorithm,which combines long and short-term memory networks on the basis of traditional convolutional neural networks,and realizes the comparison of the previous one through the gated structure.Selective retention of moment characteristics.And added a convolutional attention mechanism to visualize the convolutional attention mask,understand the input features on which the network makes decisions from the interpretability level of deep learning,and observe the network’s attention to the time-frequency graph.The proposed network is used to train and learn the standard bearing data set.The final prediction results are compared with the life prediction methods proposed in some public documents in recent years.The results show that the accuracy of the life prediction obtained by this method is higher than that of other methods.At the same time,in order to ensure the experimental results Rigorous,cross-validation experiments are performed on all bearings in the data set,and the verification results further show that the prediction accuracy of this method is higher than other methods.