Research On The Methods Of Health State Estimation And Prediction For Integrated Modular Avionics

With the development of integrated avionics,the architecture of integrated modular Avionics(IMA)has become the main trend of modern aviation.As an open system architecture on which various avionics functions are hosted,IMA is one of the most important systems to ensure flight safety.Therefore,the performance of IMA directly affects the operation of avionics functions.Once IMA fails,it will not only affect the operation of avionics functions,but also threaten flight safety,and even lead to catastrophic accidents.Prognostics and Health Management(PHM)is a research hotspot,which can effectively prevent aircraft from sudden failure.In order to ensure that flight missions are completed safely,it is imperative to carry out research on PHM of IMA.IMA is a very large scale integration electronic system with many modules and complex structures,whose fault propagation and impacts are extremely intricate.The hierarchical architecture of IMA separates the hardware resources from the avionics functions,which makes it more difficult to acquire the performance state characteristics and capture the performance degradation trend.Therefore,the application of PHM technique to IMA faces many challenges.In view of this,starting with the selection of performance degradation characteristic parameters,this dissertation researches on IMA degradation modeling,health state estimation and remaining useful life prediction.The main work and innovations of this dissertation are as follows:1.Aiming at the issues that IMA performance state characteristics are difficult to obtain and degradation trend is hard to capture,the intermittent failure frequency and function completion time are adopted as IMA health characteristic parameters to monitor health state and research performance change of IMA,under the requirement of hosted function and system maintenance.This dissertation analyzes the IMA operation management mechanism and explores the relationship between IMA failure factors and performance degradation.Based on the aircraft maintenance manual and the fault isolation manual of Boeing 787 airplane,combining error handling method of health monitor function,the health characteristic parameters of IMA are selected.2.Aiming at the difficulties of IMA health state estimation and remaining useful life prediction,an IMA performance degradation Lévy model is proposed,which realizes the modeling of performance degradation for complex electronic systems affected by multiple failure factors.Intermittent faults manifest during the IMA performance degradation process,resulting from which the error handling process will impact on the operation of avionics functions.In view of this,this dissertation proves that the completion time of avionics functions conforms to Lévy process.The IMA performance degradation Lévy model under the influence of single factor is established,based on which,the IMA performance degradation Lévy model influenced by multiple factors is constructed.A hardware-in-the-loop simulation platform is built using Freescale P2020 Reference Design Board to figure out the parameters of IMA performance degradation Lévy model.Then,the performance degradation process of IMA under different flight intensities is simulated.3.Aiming at the problem that the current health state of the system is difficult to accurately evaluate,which is caused by the overlapping boundary that the overlapping data of two adjacent health degrees resulting from health state division,a deep quantum inspired neural network is proposed to identify the current health state of the system and complete the health state estimation of IMA.A novel architecture of deep learning network which fully inherits the advantages of deep belief network and multi-level activation function quantum neural network is designed to recognize overlapping classes of data.Time domain features of function completion time and intermittent fault data are fed into deep quantum inspired neural network for evaluating the health states of IMA.4.Aiming at the issues of feature absence,feature redundancy and unstable feature extraction in artificial feature extraction during life prediction,an algorithm based on deep feature learning with quantum neural network is proposed to accurately predict IMA remaining useful life.The deep extreme learning auto-coding feature extractor is designed to directly retain the complete information from the original data and extract features from discontinuous samples.An enhanced quantum neural network model based on quantum gates is presented to study the inner structure and predict the tendency of data.The effectiveness of the proposed prediction algorithm is verified by the standard datasets.The algorithm is applied to IMA remaining useful life prediction.5.To ensure the prediction accuracy and satisfy the real-time requirement of IMA,a prediction algorithm called ensemble online sequential parallel extreme learning machine is proposed to realize online prediction of IMA remaining useful life.The parallel extreme learning machine with different activation functions is designed to improve the capability for extracting heterogeneous features.The ensemble online sequential parallel extreme learning machine is constructed by integrating multiple enhanced parallel online sequential extreme learning machines by virtue of ensemble learning strategy.The calculation speed and prediction accuracy of the proposed algorithm are verified by the standard datasets.The algorithm is applied to prediction of IMA remaining useful life.