The Study On Artificial Immune Method Of Equipment Abnormal Degree Detection And Fault Identification Method

The immune system has aroused intense research interest for its unique features.In recent years, its characteristics and mechanisms have been in-depth researched bypeople and applied to a variety of engineering practice to solve the problems oftraditional clustering or classification algorithms that hardly solve. Positive andnegative selection algorithms have become a research frontier area in one-classclassification. They have got a lot of applications, especially in areas of computeranomaly detection and complex equipment’s state detection without fault samples.Based on re-examination of the positive and negative selection mechanism, thispaper puts forward a positive selection algorithm based on the variable thresholdinformation detector (VTI-detector) and a negative selection algorithm based onbased on the negative potential field group detectors (NPFG-detectors), which aresuccessfully applied to the state detection of bearing failure. These researches haveimportant implications for the development of positive and negative selectionalgorithms and they also have broad application prospects.The positive and negative selection algorithms widely used in anomalydetection have played an important role in describing the computer data and devicestate. However, the current negative and positive selection algorithms still have theirlimitations in describing the normal and abnormal state of equipment. So, this paperputs forward the self-nonself space division method based on VTI-detector and theaffinity expression way based on the decentralized incremental theory, where thedetector’s radius can express the samples’ density distribution information. Thispaper also presents a negative selection algorithm based on negative potential fieldby the aid of data potential field distribution, which get rid of the limitations ofexisting single-sample threshold division method. Its validity also has been verifiedby simulation and examples.The appearance of artificial immune recognition system (AIRS) provides animportant basis for classification based on artificial immune system. However, the memory cells of AIRS are under the premise of fixed weight and have the defect ofbias effect when analysis the classification problem. So a variable weight artificialimmune recognition system (V-AIRS) has been proposed in this paper. The V-AIRSuses quadratic programming to optimize the weights of each memory cell and thecells with minor weights will be removed. After this, the number of memory cells isfurther reduced through combining mutation function. Taking the four commonlyused UCI datasets as example, the number of memory cells and classificationaccuracy of V-AIRS algorithm changing according to the system parameter’sevolution were revealed. In addition, the classification results of bearing failuredatasets also show that the proposed method can accurately classify target datasets.Although the negative and positive selection algorithms can accurately describethe information of abnormalities and abnormal degree but they can’t describe thefault message of device; on the contrary, the V-AIRS can accurately classifyequipment failure but it can’t describe the abnormalities and abnormal degree ofequipment. To compensate their shortcomings, this paper combines the recognitionprocesses of human immune system and proposes a fusion approach, which mergesthe anomaly detection and the fault diagnosis. The equipment failure rapiddiagnostic scheme is also proposed in revealing the fault membership degree ofequipment. Through the analysis of rolling bearing vibration signal, the effectivenessof the proposed fusion method is further verified.