Synthetically Intelligent Diagnosis Approach Of High-speed Trains Based On The Non-canonical Knowledge Processing

Maintenance and diagnostic of the high speed train is the important segments of the high-speed railway system integrated logistic supporting work. It is also important for keeping and improving the transportation capacity. It has accumulated vast diagnosing domain knowledge in the maintaining process of high-speed trains. But the non-canonical knowledge could not be fully utilized because its uncertainty, incompletion and inconsistency. On the other hand, there would be some conflicting decision making among different expert groups because of the complex relationship among the structures and the dynamic behavior of the system failures. A synthetically intelligent diagnosis approach of high-speed trains based on the non-canonical knowledge processing was proposed in this thesis. The content of the research includes three aspects such as complex system and failure modes modeling of high-speed trains, fuzzy fault diagnosis under non-canonical knowledge and comprehensive diagnosing decision making of different allopatric expert groups.The first research aspect includes the maintenance mode analysis and failure mode and effect analysis of the high-speed train. Based on the analysis of the failure characteristics and the current maintenance and overhaul mode of high-speed trains, the advantages and disadvantages of the current application mode and maintenance strategy were summarized. Then failure mode and effect analysis of the typical critical systems were carried out with the analytic hierarchy process of structures and functions. Potential key failure modes and their corresponding fault phenomenon and reason accidents were combed. The mapping relationships among failure modes, fault phenomenon and reason accidents in different key systems were defined explicitly.The next research aspect is the research of the modelling approach and reasoning approach of the complex fault relationships in high-speed trains. Based on the quantitative analysis and qualitative analysis of fault tree analysis method in electromechanical system, fault tree models of key systems were built with the result of the failure mode and effect analysis. Focused on the deficiencies of the traditional fault tree analysis (FTA) of complex system, Petri net modeling was combined with FTA to solve the problem of data explosion and to realize dynamic diagnosis. Types of modeling elements were reduced by the transformation from fault tree modelling to Petri-net modelling. Relational matrix analysis was used to solve the minimal cut set equation of fault tree. Based on the established state equation of Petri net, initial Token and enable-transfer algorithm were used to express the mathematical transfer process of faults. Thus the influence of the dynamic failure to the overall reliability of the entire system could be reasoned and diagnosed mathematically.The third research aspect is fuzzy intelligent diagnosis approach under multi-source heterogeneous knowledge environment. Focused on the multi-source heterogeneous, incompleteness and semantic ambiguity of the fault domain knowledge in the maintenance of the high-speed trains, a fuzzy intelligent diagnosis algorithm based on multi-source heterogeneous knowledge fusion was proposed. Aiming at the deficiencies of multi-source heterogeneous, Resource Description Framework (Schema) of the local database was transferred into local ontology model and finally be transferred into global ontology. Then horn logic rules were adopted to describe the relations such as equivalence, inclusion and collision. Thus the RDF(S) ontologies of the fault domain knowledge were merged into a unified whole. Based on the global ontology of the knowledge and the failure mode and effect analysis of the key systems in high-speed trains, fault tree modellings were established together with T-S fuzzy theory. The sort of the importance degree of the basic events was calculated by importing the fuzzy multiplication and max/min operator. Thus the influence of the non-canonical knowledge to the diagnosis process could be reduced in some degree.The last research aspect is research of comprehensive diagnosing decision making method of different allopatric expert groups. Focused on the fuzzy and uncertain knowledge and decision conflict among allopatric expert groups in the maintenance of the high-speed train, a fuzzy synthesis diagnosis algorithm based on FTA method, fuzzy influence diagrams and D-S evidential theory was proposed. Aiming at the dynamic transferring process and the fuzzy behavior of the decision making process, fuzzy influence diagrams were introduced to combined with fault tress analysis. Fault tree modelling was transformed into fuzzy influence diagrams to express transfer process of faults. Decision nodes and decision-making matrixes of the fault events were built. Then D-S evidential theory was combined with fuzzy influence diagrams to deal with the problem of decision conflict among allopatric expert groups. DSmT algorithm was used to overcome the shortcoming such as the independence and cross-referencing of the evidence in D-S evidential theory. Random sets explanation of the combination rules was given to describe the combining process of decision form different expert groups. The comprehensive diagnosis result from multi-group experts was more positive than from single group experts.To testify the effevtiveness of the aforementioned theories and technologies, the example verifications based on real diagnostic record of the high-voltage traction system, the air supplying and braking system and the bogie system are conducted. Mapping relationship among 27 types of failure modes and 86 types of failure causes were established in those three critical systems. A large amount of the historical statistical data and the established mapping relationships were used for vertifying the effectiveness of all the algorithms proposed in every chapter. Finally, conclusions of all research efforts in the thesis are draw and further research directions are prospected.