Quantitative Evaluation And Fault Diagnosis Of Controller System Fault Tolerance For Soft Error

Soft error refers to a temporary,random state change or transient caused by the interaction between high-energy particles and silicon elements in a semiconductor device.In the field of spaceflight,the occurrence of soft errors in satellite-borne systems become a high-frequency event due to the changeable and high radiation environment in space,Soft errors must be fully considered in the design of reliable satellite-borne systems.Based on the theory of internal reliability threat chain of complex systems,this thesis proposes a soft error quantification evaluation method for the controller system,analyzes the key nodes sensitive to soft error in the controller system and their correlation with the system failure state,and quantitatively evaluates the propagation mechanism of soft error in the controller system.The core techniques of this thesis mainly include data set acquisition,abnormal state determination,multi-node analysis and method validity test.Firstly,aiming at the data samples required by the analysis of the threat chain of the system,this thesis establishes a virtual system model for the controller system,simulates the occurrence of soft errors by using fault injection,collects the timing values of system components during the operation of the controller system,forms a data set.The controller system is modeled using the full system simulator Simics,which includes the microprocessor,operating system,and application.Taking PID as the application program,the automatic injection and timing value collection platform is built based on the target controller system model.During a fault injection,the timing values of system components are collected,and the state of the system is monitored to obtain the system failure type.A data set covering the temporal and spatial information is formed.Secondly,based on the problem that it is difficult to directly determine the state of a single component in the system with the original value,this thesis proposes an abnormal state determination method based on generation antagonism,converts the sequence data into a cyclic recursive graph,and completes the calibration of the state of the sequence graph.The classical statistical method is also used to calibrate the component state to verify the effectiveness of our proposed method.Finally,in view of the internal node number and the correlation between the control system of the characteristics of the complex,puts forward the modeling method based on bayesian network node correlation,using data driven bayesian structure learning and parameter learning independently determine the soft error propagation path construction and quantitative characterization,then on this basis,completes the fault source location and system failure type inference.Based on the classical statistical methods and generate against the determination methods of bayesian network modeling,respectively,draw the bayesian classifier of the ROC curve,verified based on the generated against abnormal judgement method has a better effect.