Attribute Reduction Of The Fault Data Set Of The Rotor Based On NRST

How to dig out the useful information,acquire the rule of the machinery running state from vast amounts of operation data and establish quantitative character pattern of the fault diagnosis and condition monitoring,solve the problem of modeling complex diagnosis,realize the intelligent mode recognition of fault is becoming the urgent problem to solve.Nevertheless,the acquired industrial data which reflect the running state of the complex mechanical system is often mixed with a lot of noise,strong nonlinear and coupling,seriously influencing the efficiency of the information acquisition,and single fault diagnosis model can not precisely fit to complicated mechanical system nowadays.To solve above problems,a fault diagnosis model of rotor which combines the multiple data-driven methods is proposed,focusing on attribute reduction of neighborhood rough set and researching on the neighborhood rough set respectively combined with statistical analysis and machine learning methods.The main research results are as follows:1)In order to avoid the loss of important attributes causing by dispersing the original data set directly,a new method is proposed which using frequency spectrum as condition attributes and fault categories as decision attributes to setup neighborhood decision table,making full use of advantages of dealing with continuous numeric attributes directly of the neighborhood rough set,also describing the prior to the greedy algorithm and analyzing the feasibility.Experimental results prove that the proposed method can effectively obtain key attributes of different typical faults which having real physical meanings and avoid loss of important attributes.2)Aimed at researching the influence of quadratic dimension,a method combining the neighborhood rough and FDA is proposed on the basis of the attribute reduction of the neighborhood rough set to class the fault,figuring out the discriminant function and cumulative discriminant ability,discussing the influence of quadratic dimension of fault mode recognition,mapping the high dimensional data to low dimension,and implementing the fault classification.The experimental results prove that the method can achieve the same recognition accuracy in the case of less feature attributes,thus can save storage space to improve computation efficiency.3)Aimed at seeking effective and accurate fault diagnosis methods and at the same time in order to research the influence of attribute reduction of the neighborhood rough set on machine learning,a method which combining neighborhood rough set with radial basis neural network model(RBFNetwork)is proposed to identify the fault,choosing the standard Gaussian function is as the radial basis function,using self-organizing selection method to determine the basis function center,width,and the connection weights.Experimental results prove that the method can obviously reduce the computing time and improve the recognition accuracy,worth promoting.4)In order to solve the problem of knowledge storage and recovery and promote the development of intelligence diagnostic technology,a diagnosis system which connecting the WEKA data mining platform with the My SQL database to identify the fault is designed,achieving the storage of the fault knowledge,showing the knowledge flow and successfully accessing to the database using My SQL statements in WEKA.