Research On Data Enhancement And Fault Diagnosis Technology Of Electro-hydrostatic Actuator

Electro-Hydrostatic Actuator(EHA)has the advantages of heavy-load capability,high efficiency and overcomes the easy pollution and stuck failure mode of the traditional aerospace actuator.It is the main research direction of aerospace actuator in the future which has been applied to the spacecraft abroad.Prognostics and Health Management(PHM)technology can realize real-time monitoring,condition based maintenance and health management of complex mechanical equipment,and improve the overall reliability of the system.PHM system has been deployed in the electronic system of F-35 fighter of the US Army,but the research of related fields in our country is still in the initial stage.In view of the lack of fault state data,the difficulty of service state mathematical modeling and the low recognition rate of the empirical statistical model threshold method for mild fault,this paper uses the Generative Adversarial Network(GAN)to enhance the small sample fault status data firstly.As the core content of PHM of Electro-Hydrostatic Actuator,fault diagnosis method of state data driven is adopted,and the WDCGANSDAE comprehensive fault diagnosis model based on deep learning is proposed.The software and hardware platform of PHM of Electro-Hydrostatic Actuator is built,and the fault diagnosis model is embedded to realize the Online fault diagnosis and health management of Electro-Hydrostatic Actuator.The main work is as follows:(1)In this paper,the research background and status of EHA and PHM technology are summarized.In view of the less fault data,difficult mathematical modeling of service state and low recognition rate of mild fault by empirical statistical model threshold method,this paper enhances the small sample fault state data of EHA by using Generative Adversarial Network firstly,and the fault diagnosis technology research is studied based on state data driven method.(2)According to the transfer function and fault experience,the mathematical model of the working response of the electro-static servo mechanism is established.It is analyzed that the main fault modes of the EHA are oil filter blockage and pressurization tank leakage,and the fault response is obvious which is easy to carry out fault implantation experiment.The simulation model is built in MATLAB,and the corresponding parameters are adjusted to simulate the displacement response trend of the two fault states system,and compared with the experimental prototype data from a research institute of Beijing.It is found that the fault response trend is the same,which verifies the correctness of the model;but the displacement signal alone is not enough for multi fault diagnosis,and the simulation data is too ideal,so it is decided to carry out the research on fault diagnosis technology of EHA based on the experimental data of the prototype.(3)By using different GAN models such as Wasserstein GAN,the data enhancement experiments are carried out on the state data of the EHA prototype,and the generated data are compared and analyzed.The gap between the generated data in data quality and data diversity is measured by MMD distance index,it is found that the MMD index of the data generated by DCGAN and WDCGAN models is closed to the real data distribution and retains a good diversity of samples,so the subsequent fault diagnosis research is carried out based on the data generated by this two models.(4)The EHA state data are encoded by One-Hot.Then the multiple fault diagnosis model training and testing experiments are carried out based on the enhanced data set.The test data and real data are respectively substituted into the model for classification.The classification results are evaluated based on the confusion matrix and three model evaluation indexes,i.e.,accuracy,F1 measure and kappa coefficient.Through comprehensive analysis,the WDCGAN-SDAE combination model has the best effect on the fault diagnosis of the EHA is obtained.(5)The PHM system of EHA based on web is built,including intelligent monitoring hardware and software platform.The hardware based on zynq-7000 platform and dual core cortex A9 processor is responsible for signal acquisition,data conversion,data transmission and other functions,and software platform based on Web front-end and back-end separation architecture is responsible for online data analysis,fault diagnosis,life prediction,system management and other functions.The last,but not the least,the WDCGAN-SDAE model is embedded in the fault diagnosis model library to realize the fault diagnosis function of the EHA.