Research On Dimension Reduction And Classification Of Rotor Fault Data Set

To make a comprehensive fault diagnosis for large-scale and complicated mechanical equipment,it is necessary to collect a large amount of vibration signal information of mechanical equipment,which inevita bly leads to the mechanical big data problem.However,the development of information science and technology urgently needs to reveal the essential information and laws contained in massive mechanical data.Therefore,how to get the real valuable fault inf ormation from the large amount of information generated by rotating machinery is a challenging task in current fault diagnosis.For a long time,linear model has been the mainstream of machine learning.Many linear feature extraction methods have achieved good results in many fields and have been proved to be effective.However,in engineering practice,vibration signals of mechanical equipment are often non-stationary,resulting in the nonlinear distribution of extracted features,so that different feature s affect and restrict each other.At this point,the traditional linear dimensionality reduction method cannot effectively mine the inherent information in the nonlinear data.The manifold learning method can automatically explore the intrinsic dimension o f low-dimensional manifold,indicating that the dimension reduction method based on manifold learning is reasonable and feasible.In this paper,based on manifold learning theory,the feature set of rotor faults is dimensionally-reduced,and the main research work is as follows:(1)Aiming at the problem that linear discriminant analysis(LDA)is difficult to mine the inner manifold structure of non-stationary vibration signal,a locality margin discriminant projection(LMDP)dimensionality reduction algori thm is proposed.The algorithm defines the local intra-class similarity and the local inter-class similarity,so that neighboring different classes are farther apart in the low-dimensional space,and the same-class neighboring samples are closer in the low-dimensional space.The effectiveness of the proposed method is verified by the vibration signal set of two different double-span rotor systems.(2)In order to improve the recognition accuracy of rotor fault feature set,a fault diagnosis method for fault data set is proposed,which combines local centroid mean minimum-distance discriminant projection(LCMMDP)and K-nearest centroid neighbor classification based on local mean and class mean(KNCNCM).This method extracts the low-dimensional sensitive feature subset by LCMMDP and uses KNCNCM for fault pattern recognition.The proposed method integrates the advantages of LCMMDP in dimension reduction and KNCNCM in pattern recognition and obtained higher fault identification accuracy.The validity of the propo sed method is verified by the instance of the fault diagnosis of a double span rotor system data-set and simulation data-set.(3)In order to solve the problem of unbalanced neighbor relation of sa mple points in rotor fault data,a fault data set reduction method based on local and global Balanced Orthogonal Discriminant Projection(LG BODP)is proposed.First of all,the mixed feature of the rotor vibration signal is extracted from multiple angles in time domain,frequency domain and time-frequency domain,and the high-dimensional feature set is constructed.The low-dimensional fault sensitive feature subsets were extracted by the proposed LGBODP algorithm.Then,the K-nearest neighbor(KNN)method is used as a fault feature classifier to recognize different fault types of rotors.The effectiveness of the proposed algorithm is verified by the vi bration signal sets of two different types of double-span rotor systems.