Modeling And Reasoning Of Conditional State Fuzzy Petri Nets For Fault Diagnosis Of Industrial Processes

Fuzzy Petri nets can be described structurally in graphical form and also support rich mathematical analysis.Fuzzy Petri nets are of strong knowledge representation and logical reasoning,which is widely used in fault diagnosis of industrial processes.However,the conventional fuzzy Petri nets model rarely considers the significant influence of the correlation between variables on the confidence when using traditional production rules for analysis and reasoning.Additionally,the parameters of fuzzy Petri nets model are usually assigned on the empirical experience.In this context,the fuzzy Petri nets usually suffer limitations in the applications of fault diagnosis reasoning.In this thesis,conditional state fuzzy Petri nets along with the reversed nets are proposed and applied to industrial process fault diagnosis and reasoning combined with association rule mining technologies.The main contents of the study are addressed as follows.1.Based on general fuzzy Petri nets,conditional state fuzzy Petri nets are explicitly introduced together with the related definitions,basic properties and reasoning methods based on matrix operations.2.APRIORI algorithms based on association rules of data mining is suggested to extract the confidence parameters of the conditional state fuzzy Petri nets model.On the basis of conditional state fuzzy Petri nets,reversed conditional state fuzzy Petri nets are established for fault diagnosis and reasoning.At the same time,dynamic confidence reasoning metrics are proposed based on the iterative algorithm of maximal algebra for backward reasoning.Consequently,the probability of industrial process failure is obtained,helping expressing diagnosis information dynamically.3.The proposed method is applied and verified by a reactor reaction process.The results of simulation experiments show that the probability of each fault and the dynamic diagnosis information are revealed.