| Prognostics and Health Management(PHM)can effectively improve equipment service reliability and further implement condition based maintenance.Electro-hydrostatic actuators(EHA)have the benefits of heavy-load capacity,high reliability and easy maintenance.Both technologies have been applied in foreign aircrafts,and a preliminary PHM has been used in the EHAs of the American F-35 fighters.Nonetheless,both are in the R&D stage in our country,and there is little research for the PHM in EHAs,which is the topic in this program,including the system architecture and their component technological approaches,with the application of Principal Component Analysis(PCA)method to extract the health features as the focus,the basic mathematical tool in the machine learning.The status quo PHM methods that can be applied in EHAs were investigated,and main mathematical tools studied and PCA selected as a starter tool to extract the features of EHA health and faults,a data-driven approach rather than the conventional threshold models based on one or a few directly measured data.The system architecture of EHA-PHM was analyzed to obtain its main components and their approaches,including fault mode database,sensor layout,feature extraction,diagnosis,fault prediction and maintenance decision.The principle fault modes were analyzed to have the two relatively emulated faults selected as study samples,namely,the gas leakage of the bootstrapped reservoir and the filter blockage.The feature extraction by PCA and its application in fault diagnosis were analyzed.Usually,PCA is used to data dimension reductions by extracting principle components.However,the residue components,as a side product of principle components,are rather used to obtain the Squared Prediction Error(SPE)statistics to extract the health or fault features,to discriminate the subtle changes induced by faults.At the same time,the Hotelling T~2(T~2)statistics from the principle components was used to further tell the pattern of a certain fault mode.The effect of the low pressure at the EHA pump inlet was analyzed by Computational Fluid Dynamics(CFD),indicating that the low pressure induced by the gas leakage of the bootstrapped reservoir can cause a lower pump volume efficiency and therefore worse EHA performances.An AMEsim model was built to implement performance simulation for the filter blockage fault,showing that,due to the throttling effect,the performances also deteriorates.The actuator displacement,velocity,and pressure,as well as the electric motor current,voltage,and the velocity were selected as the multiple variables to implement PCA for the emulated faults in step responses of the EHA,showing that,the SPE and T~2 statistics can represent the health or fault features,and can diagnose the faults earlier than with direct measurements.An EHA testing platform was built.It is validated by experiments that,the SPE and T~2statistics can represent the health or fault features,and shows more prominent with the deteriorating faults.During testing the gas leakage fault of the bootstrapped reservoir,while the fault cannot be detected by directly measuring displacement or velocity,the SPE and T~2statistics can tell the differences when pressure reduces by 88%.During testing the filter blockage fault,while the fault can only be indicated as worse as under an equivalent 91%filter blockage,the SPE statistics can discriminate a much less serious case as a 33%fault.In addition,under the two faults,the discrepancy patterns of SPE and T~2 are different,showing that PCA has the potential of identify a specific EHA fault mode.The approach to the fault prediction or the left life prediction based on a degeneration curve was discussed,to clarify the future research focuses like a bettering fault mode database,a deeper probe to feature extraction methods,extensive experiments for the degeneration models of key fault modes,and the prediction models.In summary,the PCA method is found to be able to be used to extract health features and further to diagnose faults for EHAs,whose effectiveness and accuracy are greatly upgraded,demonstrating that the machine learning is an important technological approach to the PHM for EHAs. |
