Research On Fault-diagnosis Knowledge Acquisition Methods For Health Management Of Key Subsystems Of Flight Vehicle

Key subsystems of flight vehicle, such as ngine, power filling system, power transimission system, etc, have crucial effect for filght vehicle’s function, safety and security. Health management(HM) technologies for filght vehicle are of significance for these subsystems to enable operating condition monitoring, degradation trend assessment, remaining useful life prediction and safe operation assurance. Thus HM is considered the premise and foundation that supposes condtion based maintenance(CBM), even autonomic logistics.As one of the key taches in HM system, knowledge acquisition, which provides knowledge used in the whole process of filght vehicle health management, is considered the foundation of intelligent monitoring, diagnosis, prognosis and maintenance. However, the main “bottle-neck” in the research of filght vehicle health management is still shortage of knowledge, especially the issue that fault-diagnosis knowledge is incomplete, tardily updating, lagged accumulation and confined domain. Accordingly, this dissertation researches on fault diagnosis knowledge acquisition methods in HM for key subsystems of flight vehicle. The detailed contents and innovative works can be summarized as follows:1. Basic concepts and meanings of fault-diagnosis knowledge are deeply analyzed. Basic theories of knowledge acquisition are systematically studied. The management manner of fault information with its data structure in filght vechile’s whole useful life period are indagated and analyzed, based on which a knowledge acquisition frame for filght vechile fault diagnosis oriented to multi-information sources mining is constructed; the strategy of knowledge acquisition aimed to three types of fault information sources are proposed; the main challenges in knowledge acquisition are analyzed; and the method of fault diagnosis knowledge encapsulation based on ontology is studied.The research shows that knowledge acquisition model and data mining algorithm are the engines of knowledge acquisition, which have strong influence on precision and efficiency of knowledge acquisition. Some issues, such as requirements of knowledge acquisition, precision of data mining, characters of data set, algorithms’ computational complexity, models’ function, should be considered when choice models and algorithms.2. Knowledge acquisition methods based on testing-information source are deeply studied.(1) Aiming at the non-integrity of condition test data, the method of complete analysis of data samples is studied and a primary component analysis(PCA) based vacant data evaluation algorithm is presented.(2) The method of abnormity detection based on clustering model is studied. A maximum scatter difference based fuzzy clustring algorithm(MSD-FCA) is presented to mining fault modes which hide in data set and have little differenre with normal data.(3) The method of extracting fault features based on association model is studied. A pattern matrix based association rule mining algorithm(PM-ARMA) is presented to analyzing interration of parameters and mining fault information from samples.The validation results with ground test data of liquid rocket engine shows that PCA based vacant data evaluation algorithm can efficiently enhance accuracy of complete analysis of data sample; MSD-FCA is suited for extract fault modes hided in data set in which differenres of data modes is little; PM-ARMA can decrease calculation amount of data mining evidently and ensure efficiency of extracting fault information.3. Knowledge acquisition methods based on statistic-inforamtion source are deeply studied.(1) Several kinds of rough set models are analyzed and compared in detail. Based on semantic analysis of unknown attribute values, the method of processing incomplete information based on characteristic relation is studied.(2) Based on the definition of characteristic realtion based attribute-value set, maximal characteristic similar set(MCSS) is defined, and a method of extracting optimal generalized decision rules for fault diagnosis from incomplete information table based on MCSS is presented.(3) Basic therories of rough set based granular computing(Gr C) is introduced. A method of decision rules for fault diagnosis with inforamtion complete based on Gr C is presented. With the boundary analysis for universe of Gr C, the method can fully mining the classify inforamtion hided in unknown attribute values.The validation results with fault records of helicopter transmission system shows that the rough set extended model based on characteristic relation processes unknown attribute values according to their semantic, which is accordng with factual condition in engineering. The two methods, based on MCSS and Gr C respectively, can efficiently mining fault information from incomplete diagnosis information table to extract concision and refined decision rules for fault diagnosis.4. Knowledge acquisition methods based on simulation-inforamtion source are deeply studied.(1) A model of flight vechile fault diagnosis knowledge acquisition based on simulation is presented for acquisition of knowledge for fault diagnosis using computer simulation technology when fault information is shortage.(2) The methods of fault data acquisition based on simulation and knowledge translation of simulation result, with their procedures are studied in detail.(3) The above model and methods are applied with a power filling system of LRE ground test-bed as object.The research shows that fault diagnosis knowledge can be extracted correctly by simulating of function and dynamic behaviors of system in fault condition, and analyzing simulation results. It provides a valid way for fault knowledge updating and accumulation of complex and high-priced flight vehicle.5. The basic function and structure of fault diagnosis knowledge-base of key subsystems of flight vehicle are analyzed. Based on the ontology model for fault diagnosis knowledge representation defined in this paper, the storage interface and quering interface of knowledge-base are designed. The fault diagnosis knowledge-base of helicopter powertrain is developed.The research shows that the storing construction of knowledge-base is stabilization for the requirement of storing and manageing mass fault diagnosis knowledge. The quering interface can optimize the capability of quering interface to ensure the efficiency and accuracy of knowledge reasoning.