The Research Of Anomaly Detection And Fault Diagnosis Based On Artificial Immune System

There are some problems in state monitoring and fault diagnosis of complexequipment, such as lack of fault prior knowledge, state monitoring depend onthreshold of parameters and monitoring and diagnosis are separate from each other.Artificial Immune System is inspired by intelligent mechanisms of the BiologicalImmune System which gives us some new ideas. This dissertation focuses on thecore problem of Negative Selection Algorithm: holes or redundancy in the process ofdetector generation. To solve the core problem, the concept of hyper-ring detectorand its generalized form are proposed. Anomaly detection methods which based onhyper-ring detector and generalized hyper-ring detector are discussed. A patternrecognition method based on antigen-antibody recognition mechanism is proposed,and then quick response mechnism between state monitoring and fault diagnosis areset up.To solve the main problem in Negative Selection Algorithm, a new concept ofhyper-ring detector is proposed. The description problem of anomaly degree andlevel of anomaly in anomaly state monitoring is solved, which provide a novelmethod in anomaly degree monitoring of equipment.Generation strategies of hyper-ring detector are analysed, the constructionmethod of reduction memory center hyper-ring detector is proposed based on aiNetimmune network which solved conflict between holes and redundancy, the efficiencyof anomaly detection are greatly increased and the level of anomaly is achieved atthe same time. The efficiency and feasibility of proposed reduction memory centerhyper-ring detector is validated by standard data and bearing fault data.Based on the concept of hyper-ring detector and support vector data descriptionmethod, a more efficient generalized hyper-ring detector is proposed and also itsconstruction method and anomaly degree detection method. The research results areverified by standard data and bearing fault data. To solve monitoring and diagnosis separated problem, the non-self space isdivided into several anomaly subspaces based on generalized hyper-ring detector. Torealize integration between monitoring and diagnosis, a novel pattern recognitionmethod based on antigen-antibody recognition mechanism is proposed, and itsadvantage are validated by Iris and Wine data from UCI database. The proposedpattern recognition method provide guarantee in quick response between statemonitoring and fault diagnosis.