Research Of Fault Diagnosis Strategy Based On Testability D Matrix

The structure and technology of sophisticated equipments,such as satellite and military system,are becoming more and more complex,and their functions and performances are improved day after day.The traditional detection methods are unable to perform complex testing tasks,let alone timely represent the operation state of equipments.Therefore,it is necessary to develop the design and research for testability of equipments.Testability is a kind of design characteristic,which can not only reflect the health status and operation state of equipments in time,but also realize the detection and isolation of faults,and has been widely used in aerospace,weapons and other fields.Diagnostic strategy is the focus of design for testability(DFT),which is the key to improve the fault diagnosis ability,enhance the diagnostic efficiency and reduce the full life cycle cost.Dependence matrix(D matrix)is the basis of realizing diagnosis strategy and testability analysis,and it is also the "data bridge" for transforming testability model into diagnosis strategy.Aiming at the key problems and difficulties of fault diagnostic strategy based on testability D matrix,thorough and detailed research has been conducted on the diagnostic strategy of single fault,multiple fault,and multi-valued attribute system(MVAS),and the main contents and achievements are as follows.(1)The weight index for fault detection and the information entropy are mixed together to improve the diagnosis results of one-step look-ahead and multi-step look-ahead optimization algorithms,and the two new greedy algorithms,Mixl algorithm and Mix2 algorithm,are designed to avoid missing important information,which are easy to program.The D matrices of the systems such as the classical example,damper electric mechanism control system,aircraft flap control system,laser strapdown inertial navigation system,are used to verify the correctness of the new algorithms.It is proved by stochastic simulation experiments that compared with traditional greedy algorithms,the Mixl algorithm and Mix2 algorithm have a larger application scope,and in their application scope,the two novel algorithms can obtain a good result in a reasonable time.Rollout algorithm is a traditional multi-step look-ahead optimization algorithm,which only considers information entropy,and thus Mix2 algorithm is used to improve Rollout algorithm.The improved Rollout algorithm can get better diagnosis results and wider application scope based on stochastic simulation experiments.(2)Growing algorithm,a multi-step look-ahead optimization algorithm,is introduced to search for better results of fault diagnosis.This new algorithm simplifies the test sequencing problem as a combinatorial problem comprising a basic test set with unnecessary tests,and then chooses the failure states and finds a suitable test set for the selected failure states,which can avoid the backtracking approach of the traditional algorithms.The validity and stability of growing algorithm are verified by classical example,five groups of actual system and stochastic simulation experiments,and it is suitable for D matrix with density less than 30%.In addition,growing algorithm can be applied to MV AS by extending the heuristic formula of information entropy reasonably,and compared with the traditional algorithms,the extended growing algorithm can obtain better diagnosis results in a shorter time,when the dimension of multi-valued D matrix is greater than 60×60.(3)Multiple fault diagnosis and maintenance strategy by translating into single fault(MFDMSTS)is presented to reduce the expected test cost of multiple fault diagnosis(MFD).The new strategy is divided into two stage:in the stage of translating multiple faults into single faults single fault algorithms is used to deal with the MFD,and in the stage of maintenance for multiple faults a maintenance stategy that in the form of complete binary tree is designed to complete the diagnosis and isolation for multiple faults.The simplified form of MFDMSTS is developed to reduce the running time and improve the efficiency of MFD,and its application scope is determined,One-step look-ahead optimization algorithm and multi-step look-ahead optimization algorithm for MFD are introduced based on information entropy,rollout strategy and the theory of MFDMSTS,and their application scope is discussed.(4)The swarm intelligence algorithms select one test each time for all current ambiguity subsets when they are used to search the optimal diagnosis strategy,but the selected test is not optimal for some subusets.In order to solve this problem,test sequencing problem of MVAS is fused with ant colony optimization(ACO)algorithm and discrete particle swarm optimization(DPSO)algorithm respectively,and then ANT-TS algorithm and PSO-TS algorithm are proposed.This two algorithms have both randomness and directivity,which not only improve the searching ability of the algorithm but also reduce the running time and increase the efficiency.The correctness and stability of ANT-TS algorithm and PSO-TS algorithm are verified by stochastic simulation experiments.The experimental results show that the parameters,expected test cost and cycles of ANT-TS algorithm and PSO-TS algorithm are fewer,and the algorithms run faster than the traditional ACO algorithm.The ANT-TS algorithm can obtain the test sequences with less expected test cost and wide application scope when compared with the multi-valued Rollout algorithm.In conclusion,the weight index for fault detection and the information entropy are mixed together by Mixl algorithm and Mix2 algorithm,which can improve the diagnosis results of one-setp look-ahead optimization algorithms.The improved Rollout algorithm based on Mix2 algorithm can obtain better diagnosis results and wider application scope.Growing algorithm is suitable for D matrix with density less than 30%,and the extended Growing algorithm is applicable to multi-valued D matrix with dimension greater than 60 × 60.MFDMSTS and its simplified form can reduce the expected test cost and improve the fault isolation rate for MFD.The diagnostic strategy of MVAS is realized by ANT-TS algorithm and PSO-TS algorithm with lower expected test cost,and there are few parameters and cycles in the algorithms.The theories and algorithms proposed in this paper are verified in the testsing for complex systems and stochastic simulation experiments,which can provide reliable diagnosis strategy for systems.They also offer certain basic theory and technical support for testability and comprehensive fault diagnosis,and expand the application field of testability theory.