Research On Fault Diagnosis For Switch Machine Based On BN

With the development of China railway transportation to high-speed, high-density running direction, higher requirements was put forward to railway signal equipment, especially to switch machine, as the lines of action.At present, the switch machine repair is still using the traditional model combining fault maintenance repair, maintenance personnel repair equipment only relying on the experience, not only leads to low accuracy of fault diagnosis, and longer maintenance time, has been unable to meet the rapid development of railway transportation, seriously restricted the improvement of efficiency, which also is an important factor affecting traffic safety. In order to improve the accuracy of fault diagnosis of switch machine, reduce the time delay of fault, it is necessary to adopt artificial intelligent methods to assist repair personnel to judge the fault and eliminate it in time.As a way to deal with the uncertain problem, bayesian network(BN) has been used in the field of aviation and military and medical with the strong uncertainty reasoning and learning ability, the fault diagnosis model based on bayesian network can make clear and accurate judgments to the qualitative and quantitative relationship between fault phenomenon and reason. So, the methods is used for the fault diagnosis of switch machine.The thesis selects S700 K electric switch machine as the research object, the working principle of control circuit and the switch is analised carefully. Combine expert knowledge and experience in the field of research of signal, sum up the main types of failure of control circuit and mechanical fault switch machine and the causes.List the fault causation questionnaire according to the prior probability of the structure of bayesian network, then construct the bayesian network model for fault diagnosis of switch machine. And then calculates the conditional probability table for all nodes. In input fault conditions, use clique tree propagation algorithm to infer the posterior probability of the reasons for the failure. Because the results that established according to different expert experience of different fault diagnosis models is also different, so put forward a kind of model using artificial fish swarm algorithm to optimize the structure of bayesian network, which integrate different expert experience knowledge to construct bayesian network structure.The artificial fish swarm algorithm is applied to the bayesian network structure learning, the use of global search ability of fish swarm algorithm get more simple, accurate network structure. Because the artificial fish swarm algorithm will appear optimize the poor precision, slow convergence speed and other defects in the later stage, and is improved by introducing the aspiration criterion to avoid local optimal solution search, and optimizing the random behavior of the artificial fish individuals. Using classic function to prove the feasibility of the optimization algorithm, finally obtains bias optimal network structure by improved artificial fish swarm algorithm.