Fault Prognostics Of Civil Actuator Based On Relevance Vector Machine

With the rapid development of national aviation technology,the complexity,integration and intellectualization of modern aircraft airborne components are constantly increasing,which result in higher cost of maintenance and guarantee.At the same time,the probability of fault and functional failure is increasing with the increase of the components and influencing factors of complex systems,so higher requirements are put forward for the security and reliability of aircraft operation.The actuator is the core component that drives the main and auxiliary rudder surfaces of aircraft.Its reliability directly affects the flight control system and even the safety of the whole aircraft.Therefore,the stability,reliability and accuracy of the actuator are very important to improve the overall performance of the aircraft.Actuator fault prognostics is a key technology to reduce the risk of serious accidents.It is of great practical significance and theoretical value to carry out research on the fault prognostics of aircraft actuator.In this paper,an improved Relevance Vector Machine(RVM)dynamic prediction algorithm is proposed for the first time.The specific work is as follows:Firstly,the core idea of RVM,namely,bayesian theory,is introduced and two problems that need to be solved urgently are pointed out,that is,low precision of long-time prediction and dynamic updating the prognostics model,and the derivation steps of RVM model are given.Secondly,an actuator fault prediction method(RVM-ARIMA)based on the fusion of RVM and ARIMA is proposed.This method uses ARIMA technology to correct the errors of the RVM predicted results,which effectively improves the accuracy of long-term prediction of RVM.In addition,the sliding window method is used to update the training sample data in time,so that the RVM-ARIMA fusion model always maintains good prediction performance and improves the applicability of long-term prediction of RVM algorithm.Thirdly,a prediction algorithm based on improved incremental learning Relevance Vector Machine is proposed.In this method,sample entropy is introduced to quantify calculate the fluctuation degree and regularity of the time series,which greatly reduces the size of training samples.In addition,the RVM prediction model is updated in real time by incremental learning technology to realize online dynamic prediction.Finally,two algorithms RVM-ARMA and Optimized-IRVM proposed in this paper are validated on a real electro-hydraulic actuator platform.By comparing and analyzing the prognostics results of different algorithms under the same experimental conditions,the prediction results of RVM-ARMA and Optimized-IRVM are better than the comparison algorithm,which further verifies the effectiveness of the proposed improved algorithm.