Research On Fault Diagnosis Of Rotating Machinery Based On Laplacian Eigenmaps Algorithm

Rotor, as the core component of rotating machinery, plays a vital role in the operation of the entire machinery. The operation condition of the rotor system directly relates to the performance of the whole machine. It is extremely vulnerable to all kinds of fault for a rotor system to work with high speed and overload for a long time under bad environment. It needs to make fault diagnosis and machine maintenance in time; otherwise it will cause the failure of the whole machine, and cause huge economic losses for the industrial production process of the construction. There is also a potential safety hazard. Aimed at some shortages of traditional time-frequency analysis in the field of fault diagnosis, this study took the rotor system as the research object, and designed the fault simulation scheme and further studied the Laplacian Eigenmaps feature mapping algorithm, using manifold learning method to extract the effective features of rotor fault system. The main research work is as follows:1) In view of redundancy and incomplete feature extraction in signal of high frequency zone with the traditional time-frequency analysis method, this paper proposed a fault diagnosis method based on multi-scale Laplacian Eigenmaps feature mapping algorithm(MS-LE).The method can make fault signal multi-scale decomposition, improved the resolution to obtain information, and extracted all scales of wavelet entropy. The wavelet entropies of all status of the same fault formed a high dimensional feature sets and made the intrinsic dimension estimation. Used LE method based on intrinsic dimension estimation to reduce dimension, obtained low dimensional feature vector to realize the rotor fault diagnosis, and compared the traditional method PCA, method LLE and LE method. The results proved the effectiveness of the multi-scale feature mapping method for fault recognition.2) In the early compound fault diagnosis of mechanical equipment, there are problems that the interfering noise is strong and the mining capacity of traditional linear method on fault information potential characteristics is not strong. To solve these problems, a diagnosis method based on manifold sub-band feature mapping algorithm was proposed. The method used wavelet packet analysis method on the strong inhibition and multi-resolution signal decomposition of noise willfulness. In order to obtain a complete fault signal characteristic, the original signal phase space should be reconstructed at first, then the signal was decomposed into several manifold sub-bands, and multiple state of the same manifold with the same faults sub-band combined into intrinsic characteristics of high dimensional data space and estimated the intrinsic dimension. Using LE method to map the essential characteristics of high dimensional data space to low dimension feature vector, and finally, to extract the information entropy value, through the experimental analysis compared to the classic LLE,LE and MS-LE method, proved that the manifold sub-band feature mapping algorithm of single fault and compound fault recognition is practical.3) In this paper, it independently developed and designed a set of rotor fault diagnosis system based on Laplacian Eigenmaps, including parameter set, the system’s overall architecture, structure module design, and manifold learning method in the diagnosis, etc It focused on introducing Laplacian Eigenmaps method into the system design, choosing proper methods according to different rotor fault, and comparing various methods of analysis at the same time. It improved the availability and accuracy of fault diagnosis.