Research And Implementation Of Equipment Health Management System For Aerospace

The rapid development of aerospace has put forward higher requirements for Space TT & C technology.In recent years,space TT & C tasks are becoming more and more complex.The real-time and stability of equipment is an important guarantee for the successful completion of the space mission.During the operation of TT&C system,equipment aging,component damage,parameter drift and other fault phenomena will occur.If we can not find the fault status of the equipment in time and deal with it,it will have a great impact on the TT & C task,and will cause serious losses to the space mission.Therefore,it is imperative to carry out health management for the space TT&C equipment.This paper mainly studies the related methods of equipment condition monitoring and fault type diagnosis,and establishes a health management system for equipment,so as to detect and identify faults in time.Because of the variety of the equipment and the complexity of the system,the monitoring data of process state has the characteristics of nonlinearity and high dimensionality.At the same time,the different frequency of fault occurrence causes the imbalance of historical data.Therefore,this paper chooses Data-Driven methods for condition monitoring and fault diagnosis.They are KPCA and GBDT.In this paper,based on previous studies and practical applications,we have done the following work:(1)Research on equipment condition monitoring methods.Aiming at the time variation of the equipment running,a sliding window based KPCA algorithm is introduced to establish a time-varying dynamic monitoring model.In view of the difficulty of determining the size of the sliding window in the traditional algorithm,a dynamic update window size method is proposed.The traditional KPCA algorithm ignores the different contribution of different samples and variables to the model training.This paper proposes a Two-dimensional Weighted KPCA algorithm from the sample dimension and variable dimension to enhance the influence of normal samples and important variables on the model establishment.Then we applies the method to the modeling process of Variable Moving window KPCA,obtaining a Two-dimensional Weighted KPCA based on Variable Moving window.Simulation results show that the improved algorithm has better fault recognition ability than traditional KPCA,and has higher recognition rate and lower false negative rate for fault variables in the dynamic process of equipment operation.(2)In the research of fault type diagnosis,in order to optimize the classification ability of GBDT on the unbalanced fault sample,the improvement is proposed from the sample level and the model level.Firstly,SMOTE oversampling technology is introduced into GBDT algorithm,and different training samples are generated for each class through SMOTE oversampling,which is applied to each round of weak classifier training.Secondly,the loss function is improved in the stochastic gradient lifting decision tree with random subsampling.The loss balance is achieved by weighting the class probability with the number of samples in each class in the loss function.Through a number of comparative experiments,it is shown that the two improved GBDT algorithms have better classification performance than the traditional GBDT on the unbalanced sample,and have a certain promotion in the accuracy and AUC value.(3)In the system design and implementation,the TW-VMKPCA algorithm and BL-SGBT algorithm proposed in this paper are respectively applied to the detailed implementation of equipment condition monitoring module and fault type diagnosis module.At present,the system has shown good performance in trial operation,and all functions have met the needs of user.