The Artificial Immune Theory And Application In The Fault Diagnosis Of Mechanical Equipment

The artificial immune network and algorithm have its unique excellencies in the self and non-self reorganization,data pattern learning,memory and classify, etc. Applies the immune algorithm in the anomaly detection and fault diagnosis of mechanical equipment is an unique topic, which have a great deal of research value and the practical application significance.In order to detect the abnormal state of the equipment,which is lack of fault samples. The abnormal state detection problem to equipment is described and some new terms (for example, state space, self space, non-self space, and so no) are introduced on the basis of the self and non-self concepts of immune system. In order to detect the abnormal state of equipment more efficiently, this paper proposes a modified negative selection algorithm based on analyzing the negative selection mechanism of immune system and the existing negative selection algorithm. A formula has given to evaluate the quantity of detectors, simultaneously; an indirect appraisement has given for the distributed degree of detector in the non-self space. The experimental results of the anomaly detection of C618 lathes gear box indicates this modified negative selection algorithm can generate "more quality" detectors and have good ability to cover the non-self space.In order to combine anomaly detection with fault patter diagnosis, utilize the clonal selection mechanism of immune system and research achievements of artificial immune system, a new fault diagnosis approach with continuously learning and data samples identifiable abilities is investigated. Aim at solve the problem that the existing fault diagnosis approaches lack continuously,self-adaptive learning capabilities, a model of fault diagnosis based on clonal selection mechanism. In order to solve the problem of equipment state data pattern recognition and classified, an information parameter of identifier for sorts is joined in the definition of antigen and antibody. A clonal selection evolutional learning algorithm is used in the training phase of fault diagnosis. Aim at the problem that the clonal operation of antibody is insufficient; this paper proposes a clonal operation operator in the two-dimensional space based on affinity. Thought the experiment towards fault diagnosis of C618 lathe gear box demonstrated the validity of this method and algorithm.In order to solve those problems of nonlinear coupling and multi-faults in fault diagnosis. Combined the immune evolutional learning algorithm and Radial Basis Function Neural Network (RBFN), an Immune Neural Network model is designed. Use the immune evolutional learning algorithm to ascertain the crytic-layers structure and crytic-layers central parameters of RBFN. Utilize the training samples to train the Immune Neural Network. Finally, apply the trained network to fault diagnosis. Thought the experiments of 7216 tapered bearings and Iris datum demonstrated the pattern recognition capability of this Immune Neural Network.