Research On The Theory And Method Of Testability Design With Imperfect Tests

Testability design has an important significance in reducing life cycle cost of theequipment, improving the performance of the diagnostic systems and the guarantee ofequipment safe running. However, some problems are met when the testability theorybased on perfect test assumption (0-1dependency relationship) is applied to thecomplex equipment such as the space station. For example, the diagnostic timeliness isnot taken into account when designing the diagnostic systems, which may leads to theexistence of critical design defects in the equipment; false alarm rate and missingdetection rate are commonly high because the unreliability and latency of the tests is notfully considered; the optimal generation of the sequential fault diagnostic strategy isonly based on the execution cost of the tests, which may result in a high life cycle cost.The problems discussed above are basically caused by the unreasonable perfect testassumption and the testability modeling, analysis and inference method based on it. Inorder to resolve these problems, the testability design theory and method with imperfecttests is studied in this thesis, where imperfect tests denote those tests that are unreliableand have test latency. The research on testability modeling, test selection, dynamic faultdiagnosis and sequential fault diagnosis based on imperfect tests is carried out accordingto the design process of testability, which forms a whole testability design theoreticalsystem. The details of the content are as follows.(1) Testability modeling and information expression method based on imperfecttest assumption is proposed, where test delay, unreliable test and fault propagation canbe effectively modeled, which lay a good foundation for the consequential testabilitydesign process.(2) Testability evaluation method with imperfect tests is proposed. The imperfecttest selection problem is formulated. A genetic algorithm (GA) and a LagrangianRelaxation Algorithm (LRA) are proposed to solve the problem. A heuristic method togenerate the feasible solution in GA is formulated. GA is a general approach for solvingthe problem with imperfect tests including the scenarios with delayed and multiple testoutcomes, non-independent cost, et al. The LRA overcomes the computationalcomplexity of the primal optimization problem by decomposing it into three tractablesub-problems. Theoretical analysis and simulation experiments show that the twomethods are effective to solve the test selection problem, with different characteristics.(3) A delay dynamic coupled fault diagnosis (DDCFD) model to deal with theproblem of coupled fault diagnosis with fault propagation/transmission delays andobservation delays with imperfect test outcomes is proposed, which has the potential toreduce the false alarm rate and missing detection rate of the diagnostic systems. Sincethe faults are coupled, the problem does not have a decomposable structure. Consequently, we propose a Partial-sampling method and a method based on blockcoordinate ascent and the Viterbi algorithm (BCV) to deal with the coupled-stateproblem. By reducing the number of samples and avoiding redundant computations, thecomputation time of Partial-sampling method is substantially smaller than the regularAnnealed MAP method with no noticeable impact on diagnostic accuracy. This BCValgorithm reduces complexity by assuming all other faults to be constant from theprevious iteration when computing the state sequence of a component. Furthermore, thelocal maximum is escaped by using soft state relaxation. For the state estimationproblem of single Markov chain, a high order Viterbi algorithm is formulated. Finally,we test the inference algorithms proposed in this paper by applying them to systems ofdifferent complexity. Some conclusions are obtained, which indicate that the twomethods have different performances in different applying scenarios. They are suitablefor the systems with long delays and short delays, respectively.(4) The test sequencing problem considering life cycle cost is firstly addressed,which is formulated as a complex optimization problem. For the cases with imperfecttests and its special scenario that the tests are perfect, two algorithms and theirvariations reducing the computational time are proposed based on AND/OR graphsearch. Comparisons based on the simulation and the application results show that thediagnostic strategy generated by the proposed algorithms is better than the one obtainedby the traditional method, which has a lower life cycle cost.(5) The theory and method proposed in this thesis are applied on the thermalcontrol system of a spacecraft. The technical process consisting of testability modeling,test selection, dynamic fault diagnosis and sequential fault diagnosis based on imperfecttest assumption is realized. Meanwhile, the similar process is implemented using theexisting theory based on perfect test assumption. Comparisons are made, which indicatethat the proposed theoretical system has a better performance (e.g., more accuratereal-time diagnostic result, lower life cycle cost of the sequential fault diagnosticstrategy, et al.) in realizing testability design with imperfect tests. It has the greatpotential to be applied to other systems and equipment.