Research On Fault Diagnosis Based On SDG And Its Application

When fault diagnosis is studied in industry, especially in process industry, it is difficult to build precise quantitative models and diagnose faults by variable residual detection. According to function mechanism and unit relationships of the processes, qualitative models are built which reflect deep knowledge of systems, such as signed directed graph (SDG). Combined with other quantitative methods, SDG-based method is a new approach to study quasi-quantitative fault diagnosis. Yet there are many problems not resolved in the theory of SDG and its application in process industry, including modeling method, inference mechanism, fusion with quantitative methods, etc. The focuses of this dissertation are on the methods of SDG modeling in process industry and SDG-based fault diagnosis, and four problems are studied as below.(1) The characters of SDG model are deeply analyzed. It is demonstrated that SDG is a linear qualitative model of initial response by differential algebraic equation. Then the influences of compensatory response and inverse response caused by negative feedback and feed forward in SDG model are analyzed, which destroy the consistent path. In addition, the mathematic description of SDG model by matrices is given. (2) SDG modeling method is studied. The complementary relationship of node information in model-driven SDG and trend information in data-driven trend analysis is illuminated. Then SDG model is built by trend analysis to associate node trends and meanwhile the qualitative fault diagnosis method based on trend analysis and SDG is proposed. (3) The method of quasi-quantitative fault diagnosis based on SDG is studied. Quantitative information not in SDG model and its sampling is processed by data fusion. One is to calculate fault propagation probabilities according to node quantitative information by fuzzy fusion in inverse inference. Another is to fuse dynamic information of initial response and state...