Research On Problems Of Fault Diagnosis Based On Intelligent Algorithm

Pump-jack is a mechanical device applied to exploit the oil and pumps oil through sucker rod. Now the highlighted subject for research of pump-jack fault diagnosis is the method based on intelligent algorithm. Two fault diagnosis methods based on intelligent algorithm are researched in this paper, and use them to diagnosis the fault of the pump-jack. The detail work is as follows:The extension neural network is used for fault diagnosis of the pump-jack. The method fully uses the advantages of qualitative description and quantitative description of extenics, and considers the characteristics of parallel structure of the neural network. The traditional genetic algorithm is improved to present an adaptive genetic algorithm. Adaptive algorithm of crossover and mutation probability is presented. Namely, the probability adjusts according to the actual situation of the population momentarily. The method could overcome the shortcoming of falling into local extreme value of BP algorithm and the convergence speed improved highly. The training successful ratio was high. The weight value of the extension neural network(ENN) is optimized by the genetic algorithm. The ratio of the right output frequency to the total number of input samples of ENN is used as fitness function in the process of optimization. Chromosome is real-coded according to extensional matter element.The method was successful in fault diagnosis of pump-jack and the application effect is ideal. The artificial immune algorithm is combined with the genetic algorithm to present a immune genetic optimizing algorithm. It keeps the searching speed,capability of global and local searching of genetic algorithm, overcomes the flaws of uncontrollable converging direction and low local searching efficiency because of crossover searching, and avoids premature convergence to a large extent. A computation method for affinity degree based on vector distance between antibodies is presented according to the characteristic that the distance is short and the affinity is large when antibodies are similar. The method is relatively simple and there is not parameter needs to be adjusted according to experience. A adjusting factor based on density is increased in the process of promoting and suppressing of antibody. Thus, the best individual can be preserved, the diversity can be ensured, and the phenomenon of premature convergence can be avoided. The characteristic of low learning efficiency of RBF neural network can be overcome when the algorithm is used to optimize the hide centers of RBF network. The approximation accuracy can be also improved and the number of constructing the center of the hide layer of network is dispensable. The network was successful in fault diagnosis of pump-jack.