Research On Feature Selection Method For Rotor Fault Data Set

With the development of modern industry and technology, in the petrochemical, electric power, metallurgy and other industries dominated the rotating machinery is moving towards large-scale, integrated and highly automated direction. The wide application of computer technology and all kinds of intelligent instruments in the monitoring of mechanical equipment, make the data in the running process of the mechanical equipment is collected and stored. But these data often have the characteristics of "mass" and "high dimension". How to use these data effectively, excavate rules in data, has important significance for rotating machinery fault diagnosis.Feature selection algorithm is a kind of algorithm which can select sensitive features in a large number of features, it can removing the irrelevant and redundant features of the rotor fault data set by constructing appropriate feature selection model. It is important to improve the efficiency and accuracy of fault diagnosis. Therefor, the research of feature selection method is carried out in this paper, which provides theoretical support for reducing the dimension of rotor fault data set. The main research contents are as follows:(1) Taking the vibration signal of rotor system as the research object, analyse signal and extract time domain, frequency domain and time domain characteristics. Make the fault feature fusion, and structure multi domain rotor fault feature set. Research on dimension reduction of rotor fault data set, focusing on the characteristics of particle swarm optimization algorithm and Laplace score algorithm and its application in feature selection.(2) Aiming at the problem that rotor fault data contains a large number of irrelevant and redundant features, the fault feature selection method based on particle swarm optimization algorithm and Laplace score was proposed. This method first use Laplace score to select features, then search the optimal feature subset on the compact feature subset by chaos particle swarm optimization algorithm, in the search process used the accuracy of support vector machine as fitness function. The obtained feature subset is input to the classifier for fault classification. The results show that this method can filtrate a most discrimination classification feature set and improve the accuracy and efficiency of the classifier.(3) A simple fault diagnosis system is designed. The particle swarm optimization algorithm and Laplace score algorithm are embedded into the system. This system can achieve data processing, feature extraction, fault diagnosis and other functions.Researches showed that feature data set contains a large number of information which can reflect running state of the rotor, so how to make better use of these data, and obtain a feature set which is more favorable for the identification of faults,will be an important direction of fault diagnosis research.