Research On Marine Bearing Fault Diagnosis Based On Deep Learning

With the increasing transformation of modern technology,our country is also gradually deepening its exploration of the ocean.As the most important maritime transport,ships play an important role in both military and civilian use.In the era driven by big data,on the one hand,ships are moving in the direction of precision and automation.In order to ensure the safe operation of ships,the health status of main components needs to be monitored.On the other hand,how to effectively use big data has become the focus of research.Bearings are key components in most mechanical equipment.In this paper,CNN in deep learning models is used as the bearing fault diagnosis method,which reduces the difficulty of fault diagnosis,improves the accuracy of fault diagnosis,and achieves end-to-end detection purpose.The main tasks are as follows:First,use RS algorithm to design the traditional CNN diagnostic network.Aiming at the problem of numerous hyperparameters in the CNN diagnosis network,firstly,the types and ranges of hyperparameters to be found are determined based on artificial experience,and then the RS algorithm is used to realize automatic hyperparameter optimization on the computer,and the optimal result is used as the final diagnosis model.Then the diagnosis result of the constructed CNN diagnosis network is analyzed,which shows the feasibility of using CNN as the bearing fault diagnosis modelThen,use LSTM and Ada BN algorithms to improve the traditional CNN diagnostic network.The output of the first half of the traditional CNN diagnostic network is a one-dimensional feature map sequence,and the LSTM network itself has good feature extraction capabilities for sequence data.Therefore,the advantages of two special networks can be fully combined to further improve the feature mining ability of the CNN diagnostic network and improve the final recognition accuracy by using the LSTM layer instead of the fully connected layer.By adding BN layers after each special network layer,on the one hand,the output of each layer is normalized,the coupling degree of each layer in the network is reduced,the overall training speed of the network is accelerated,and the real-time diagnosis of bearing faults is ensured.On the other hand,On the other hand,it is used for domain adaptation,which improves the classification performance of the network on the test set.Finally,by explaining the process of improving the CNN diagnosis network,it shows that the network diagnosis process is different from the traditional CNN diagnosis network after the network structure is changed.Finally,the improving CNN diagnostic network is experimentally verified.Firstly,the diagnosis result analysis,neuron visual analysis and classification process visual analysis are carried out to improve the CNN diagnostic network,and the correctness of the three aspects of network structure,training speed and generalization ability is verified.Then,the simulation data set was used to make a comparative analysis between the improved CNN diagnostic network and the traditional CNN diagnostic network,which shows the improvement of the diagnosis network performance after the improvement from the diagnosis results,anti-noise ability and variable working condition adaptive ability.Finally,the public data set is used to conduct a comparative analysis of the improved CNN diagnostic network and other mainstream diagnostic networks,which further verifies the superior diagnostic performance of the improved CNN diagnostic network and is suitable for fault diagnosis of marine bearings.