Application Research Of Deep Learning In Health Management Of B777 Flight Control System

The fault characteristics are extracted from the parameters recorded during the operation of the flight control system,and an effective health management system is constructed by means of various diagnostic algorithms and prediction algorithms,which is beneficial to real-time fault diagnosis,state prediction and life assessment of the flight control system.It is of great significance to ensure flight safety.However,due to the complex coupling relationship inside the flight control system,it is difficult to effectively establish the accurate physical model and fault model of the system,which causes the accuracy of the flight control system fault diagnosis is not high,and the prediction ability has greater uncertainty.Deep learning is a research hotspot in the field of machine learning in recent years.The multi-hidden layer structure of the algorithm can extract the abstract features of the data layer by layer and obtain the most complete feature representation of the data,which is an effective means to solve the above problems.Therefore,this research focuses on fault diagnosis and state prediction in health management,and studies the health management technology of B777 flight control system based on deep learning.Firstly,aiming at the problem that traditional fault diagnosis algorithm cannot effectively extract fault features,a fault diagnosis algorithm CNN-LSTM is proposed,which combines CNN and RNN.The algorithm has the ability to extract local features and time series features of data space.By analyzing the hydraulic fault of the rudder servo control system and constructing the fault model under AMESim hydraulic simulation software to obtain the fault data for the fault diagnosis experiment,the experiment proves that the proposed algorithm can extract the fault features more completely and thus has higher Fault classification ability.Secondly,aiming at the problem that the traditional multivariate state prediction algorithm has low prediction accuracy,poor generalization performance for non-stationary sequences and cannot be iteratively predicted,a multi-state prediction algorithm MTL-LSTM combining multi-task learning and LSTM is proposed.Two dimensions model the forecasting task and predict multiple parameters simultaneously.Multivariate state prediction experiments in the pitch loop of flight control system show that MTL-LSTM can greatly improve the prediction accuracy.The research work of this subject provides a new research idea for the health management of large complex systems and the processing of industrial big data.