| The safety of underwater unmanned vehicles is an important guarantee for their normal running in the complicated and changeable Marine environment.As the power source of UAVs,the reliability and safety of thrusters play a vital role in the normal running of UAVs.In this paper,the classification,identification and diagnosis of eight fault states of propeller,including weak winding,strong winding,single blade damage,double blade damage,single blade deformation,double blade deformation,blade loss and completely blocked rotation,and nine fault states of normal running were studied.Research was carried out on the preprocessing method of operational data of underwater UAVs,the fault diagnosis method based on sliding window and convolutional neural network,and the fault diagnosis method based on attention mechanism improvement.A fault diagnosis platform of underwater UAVs was built based on the fault diagnosis method,and the operating state test was verified on the platform.The main research contents of this paper are as follows:(1)Aiming at the problem of missing and abnormal operation data of underwater unmanned vehicle,the data preprocessing method is studied.For abnormal data,Gram Angle field of time series data was constructed according to the correlation between time series data,and the mean value and amplitude of correlation were calculated based on Gram Angle field.Then,threshold values were set according to 3δ principles to eliminate abnormal values.For the missing data,the improved Gram Angle field was used to determine the correlation sequence of the time series data.Then the KNN algorithm was used to solve the two sequences with the strongest correlation.Finally,the mean difference was used to repair the missing value.(2)Aiming at the problem that the single scale variation is difficult to reflect the fault state characteristics in the data-driven fault diagnosis method,a fault diagnosis method based on sliding window and convolutional neural network was developed.In this paper,the comprehensive features of time series dimension and sensor dimension are extracted for fault diagnosis.Firstly,the multi-dimensional time series data collected by the sensor is transformed into a two-dimensional tensor through sliding window processing,and then the data noise reduction and data feature extraction are carried out through the convolutional denoising autoencoder.Then,the associated features of the time series and sensor dimension are further extracted through the convolutional neural network.Finally,the classification and diagnosis of data are carried out through the full connection layer and softmax layer.(3)Aiming at the problem of uneven feature distribution in fault diagnosis model,the method of strengthening feature distribution is studied.By introducing a time-domain adaptive module and adopting a two-branch processing method.the feature weighting process is carried out for local features and global features.In the convolutional feature module,the data in the direction of time series is compressed to extract the attention weight,so as to strengthen the attention to the time series changing data and enhance the recognition efficiency of the module.(4)Aiming at the real-time fault diagnosis problem of underwater unmanned vehicles,a fault diagnosis platform of underwater unmanned vehicles is built.The GUI interface is built through PYQT,and various algorithms including data acquisition,data storage,data processing,processing diagnosis and result visualization are realized in the process of fault diagnosis at the back-end.Finally,online diagnosis and offline diagnosis of underwater unmanned vehicles are realized.(5)According to the validity of the fault diagnosis algorithm model and diagnosis platform of underwater unmanned vehicle,the test test of underwater unmanned vehicle fault diagnosis is carried out.Based on the known fault state and the unknown fault state,the fault diagnosis algorithm and platform of the underwater unmanned vehicle are verified. |
PDF Full Text Request