In mechanical equipment,the operating condition of rolling bearings directly affects the safety and reliability of the equipment.Therefore,the fault diagnosis and remaining life prediction of rolling bearings,as an important means of intelligent health management,is of great practical significance to prevent accidents and ensure the safe and stable operation of equipment.This paper uses deep learning methods to carry out research on rolling bearing fault diagnosis and remaining life prediction,and the main research work is as follows:(1)In view of the current big data era,the traditional rolling bearing fault diagnosis relies too much on manual feature extraction,cannot meet the automation and intelligence needs,and cannot efficiently process and analyze a large amount of data,and proposes a combination of Convolutional Neural Networks(CNN)and Cat Boost algorithm based The method is based on a combination of Convolutional Neural Networks(CNN)and Cat Boost algorithm.Firstly,the pre-processed data is extracted from the CNN and the extracted features are fed into the model as input.The effectiveness and accuracy of the method was verified with the published CWRU bearing fault dataset,and the results showed that bearing faults of different damage levels could be accurately identified.(2)To address the problems of current rolling bearing residual prediction methods,i.e.,the prediction accuracy decreases when the network model structure is too complex and is accompanied by gradient disappearance,a bearing residual life prediction method with optimised depth residual network is proposed.First,the collected original fault signal is pre-processed to convert the original time-domain vibration signal to the frequency domain,and then it is normalised to a linear function;the original signal is reconstructed by selecting a VAE encoder,extracting its middle part of data to do the work of dimensionality reduction,replacing the features of the original signal with the reduced features,and inputting them into the convolutional layer of the constructed Res Net network for deep The weighted features were then fed into the convolutional layer of the constructed Res Net network to assign weights to the degenerate features,followed by mapping the weighted features into the fully connected network to obtain the RUL predictions.An optimisation algorithm was chosen to train the final model parameters to prevent overfitting of the model.Finally,the validity and accuracy of the proposed prediction model was verified using the PHM2012 bearing acceleration degradation dataset. |