Research On Rolling Bearing Fault Analysis And Prediction Method Based On HI Curve And GRU

In today’s increasingly complex mechanical structures,it is important to investigate the fault prediction of rolling bearings,which are important components in many rotating equipment.However,traditional prediction methods based on physical models are unable to achieve the results required by industry.In order to effectively predict the faults in rolling bearings and reduce the losses caused by the occurrence of faults.In this paper,based on the degradation principle of rolling bearings,mainly using neural network model,a prediction study including the degradation trend of rolling bearings,the fault start time point and the remaining useful life(RUL)is carried out.The main research contents are as follows:First of all,based on the full life cycle signals of rolling bearings,the fault pattern and the corresponding time and frequency domain indicators are analysed to summarise the degradation trend of rolling bearings from smooth operation to fault occurrence.The health indicator(HI)is constructed by using the feature evaluation method to filter out the root mean square,peak value,root mean square frequency and centre of gravity of the frequency.A dimensionless HI curve is created by proposing a feature fusion method based on spectral clustering,and after calculating the feature evaluation indicators,it is verified that the proposed HI curve reflects the degradation trend with more accurate fitting results.Based on the HI curve created,a Gated Recurrent Unit(GRU)based degradation trend prediction model is proposed.In order to enhance the sensitivity of the network to changes in trend over a longer period of time while mining the hidden information before and after the time series,an attention mechanism is added,improved model resulting in significant improvement in prediction accuracy and a more accurate prediction of the degradation trend of rolling bearings can be predicted.Secondly,to research the RUL of rolling bearings,a method based on HI curve for rapid determination of fault start time point is used;an improved self-attentive convolutional neural network module is proposed to extract the multidimensional features of rolling bearings,Where the temporal relationships between features and their interrelationships can be analyzed more precisely.And then the RUL is first predicted by the GRU network and estimated by the prediction results,which is verified using the XJTU-SY dataset,and the accuracy of the prediction is a big improvement compared to the common model,the proposed method effectively improves the accuracy of RUL prediction.